Can we improve prediction of Alzheimer's disease and Mild Cognitive Impairment by combining MMSE score and MRI-based imaging data?

Can we improve prediction of Alzheimer's disease and Mild Cognitive Impairment by combining MMSE score and MRI-based imaging data? | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Can we improve prediction of Alzheimer's disease and Mild Cognitive Impairment by combining MMSE score and MRI-based imaging data? Anna Marcisz, Joanna Polanska This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-1747001/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract BACKGROUND This work aimed to find MRI-based markers for Alzheimer's disease (AD) and mild cognitive impairment (MCI) to improve diagnosis. METHODS The multinomial logistic regression was used to predict diagnosis status: AD, MCI, and normal control (NC) combined with the Bayesian information criterion for model selection. Several T1-weighted MRI-based radiomic features were considered as explanatory variables in the prediction model. RESULTS The best radiomic predictor was the relative brain volume defined as the ratio between the volume of the brain without cerebrospinal fluid and the volume of the whole brain multiplied by 100%. The model was trained on the ADNI dataset and tested on the independent EDSD dataset. The proposed method confirmed its quality by achieving a balanced accuracy of 95.18%, AUC of 93.25%, NPV of 97.93%, and PPV of 90.48% for classifying AD vs NC for the EDSD. The comparison of two models: with the MMSE score only as an independent variable, and corrected for the relative brain value and age, shows that the addition of an MRI-based biomarker improves the quality of MCI detection (AUC: 67.04% vs 71.08%) while maintaining quality for AD (AUC: 93.35% vs 93.25%). Additionally, among MCI patients predicted as AD inconsistently with original diagnosis, 56.25% from ADNI and 54.17% from EDSD were re-diagnosed as AD within a 48-month follow-up. It shows that our model can detect AD patients a few years earlier than a standard medical diagnosis. CONCLUSIONS The created method is non-invasive, inexpensive, clinically accessible, and efficiently supports the AD/MCI diagnosis. Alzheimer's disease Mild Cognitive Impairment MRI multinomial logistic regression Figures Figure 1 Figure 2 1. Background Alzheimer's disease (AD) is a progressive, neurodegenerative brain disease that causes memory loss, changes in behaviour, and problems with everyday tasks. Alzheimer's is the most common form of dementia, and it is responsible for 60–80% of dementia cases [ 1 , 2 ]. The intermediate stage from normal cognition to dementia is a mild cognitive impairment (MCI). People suffering from MCI have a high rate of progression to dementia over a relatively short period, but not everyone will develop Alzheimer's disease [ 3 ]. Within a 3-year follow-up period, about 35% of patients with MCI status progress to AD or dementia [ 4 ]. A yearly conversion rate equals 5%-10% [ 4 ]. The main aim of this work is to find easily-accessible biomarkers for Alzheimer's disease (AD) and a mild cognitive impairment (MCI) to improve the diagnosis process. An additional challenge is to predict the diagnosis of Alzheimer's while a patient is still mildly cognitively impaired. Many different methods to predict the diagnosis have been proposed in recent years. These methods are based on machine learning algorithms [ 5 – 13 ], regression models [ 4 , 14 – 18 ], and other methods [ 19 – 24 ]. Many different biomarkers are used to classify AD and MCI. The first group of biomarkers is based on structural brain atrophy obtained from MRI imaging [ 7 – 9 , 13 ]. The second group of biomarkers uses the evaluation of brain metabolic changes, which are measured by fluorodeoxyglucose positron emission tomography (FDG-PET) imaging [ 25 , 26 ]. Fluid biomarkers are the third group, and this is connected with amyloid and tau obtained from cerebrospinal fluid (CSF) [ 6 , 10 , 27 ]. Moreover, diffusion tensor imaging (DTI) and functional MRI (fMRI) are also applied for the detection of AD and MCI [ 5 , 6 , 28 , 29 ]. Most studies use multiple biomarkers in the early diagnosis of AD and MCI and are based on a combination of two or more following biomarkers: MRI-based biomarkers, fluid biomarkers or PET-based markers [ 5 , 30 , 6 , 10 , 24 ]. The availability of all three biomarkers (PET and CSF and MRI or DTI or fMRI) is limited due to the cost, time and invasiveness of the methods (PET and CSF) [ 31 , 24 ]. This article presents a method that improves an MCI and AD's diagnosis process based on easily-accessible clinical biomarkers like age and Mini-Mental State Examination score (MMSE) [ 32 ], available in medical history for almost every patient with suspicion of dementia. We suggest using, in addition to those clinical predictors, the MRI-based disease progression radiological biomarkers to support the diagnosis. In patients suffering from Alzheimer's disease, the brain shrinks, and therefore the space filled with cerebrospinal fluid increases [ 33 , 34 ]. Moreover, this brain shrinkage causes the brain to be more wrinkled. Because of that, we consider the relative brain volume and global measure of brain wrinkling as the imaging biomarkers. 2. Methods 2.1 MATERIALS Data used in the study were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu) and The European DTI Study on Dementia (EDSD). The ADNI was launched in 2003 as a public-private partnership led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD). For up-to-date information, see www.adni-info.org. The European DTI Study on Dementia (EDSD) is a multicenter framework created to study the diagnostic accuracy and inter-site variability of DTI-derived markers in patients with manifest and prodromal Alzheimer's disease [35]. The standard analysis dataset of the ADNI-1 project was used (collection name: ADNI1: Complete 1Yr 1.5T; subjects who have both 6- and 12-month scans available) to build a statistical model for predicting AD or MCI status [36]. This dataset was randomly split into and five subsets to conduct internal testing and 5-Fold Cross-Validation [37]. In the second stage, the final statistical model was built on the whole dataset, and that model was tested on the independent dataset from the EDSD database. The dataset of the ADNI-1 project includes MPRAGE T1- weighted 3D scans (1.5 T) and several clinical and neuropsychological measures acquired from healthy controls (NC), mild cognitive impaired (MCI) subjects and Alzheimer's disease (AD). The second dataset used in the analysis comes from The European DTI Study on Dementia (EDSD) database [35]. The EDSD was started in 2010. The coordinator of this database is the German Center for Neurodegenerative Diseases (DZNE) in Rostock, Germany. Since 2013, the EDSD has also collected the data of subjects with MCI. Dataset used in the preparation of this article includes data of subjects who were marked as "not dropout". Our analysis was based on T1-weighted MRI. ADNI dataset (dataset 1) is a reference dataset, and the EDSD dataset (dataset 2) is an independent validation dataset. Additionally, the EDSD dataset was divided into two subsets related to MRI scanning options: 1.5T and 3T. ADNI provided intensity normalized and gradient un-warped TI image volumes [36]. The EDSD native data were used, and N4 bias field correction in N4ITK framework was applied [38]. For both datasets: ADNI and EDSD, skull stripping was achieved in the SPM 12 software package (https://www.fil.ion.ucl.ac.uk/spm/) [39]. 2.2 STATISTICAL METHODS The clinical characteristics of subjects from the ADNI and EDSD datasets were summarized by the diagnostic group (NC, MCI, AD) and presented in table 1. The following variables were considered at baseline: age, sex, mini-mental state exam (MMSE) and years of education. For quantitative measures, values of mean and SD were calculated, and for categorical variables, the percentage was presented. The comparisons between groups were conducted using the nonparametric Kruskal-Wallis test for quantitative measures (the Conover test was used in the post-hoc analysis), and the chi 2 test to compare proportions and P- value is presented in table 1. Additionally, table 1 contains effect size h 2 (eta-squared) with 95% confidence interval [40, 41]. Table 1. Clinical characteristics of the ANDI and EDSD dataset. Characteristic NC MCI AD Unadjusted P value Effect size h 2 [95%CI] ADNI, n 194 311 133 - - Age, mean (SD) [years] 75.9 (5.08) 74.9 (7.06) 74.7 (7.59) .4643 * 0.0053 [0; 0.0199] Education, mean (SD) [years] 16.0 (2.79) 15.7 (3.00) 14.7 (3.11) .0003 * .0004 † < .0001 ‡ 0.0253 [0.0057; 0.0523] MMSE score, mean (SD) 29.1 (1.03) 27.0 (1.78) 23.5 (1.91) < .0001 * < .0001 † ‡ § 0.6023 [0.5581; 0.6387] Female [%] 47.9 35.4 48.1 .0053 - EDSD, n 194 152 136 - - Age, mean (SD), [years] 68.7 (5.90) 71.2 (6.76) 72.4 (8.28) < .0001 * .0001 ‡ < .0001 § 0.0497 [0.0170; 0.0900] Education, mean (SD), [years] 13.1 (3.67) n=173 12.4 (3.35) n=132 10.3 (3.33) n=134 < .0001 * < .0001 † ‡ 0.1036 [0.0538; 0.1572] MMSE score, mean (SD) 27.4 (6.49) 26.3 (3.14) 20.8 (5.36) < .0001 * < .0001 † ‡ § 0.2198 [0.1569; 0.2793] Female [%] 51.0 43.4 56.6 .07854 - * Kruskal-Wallis rank sum test; † Conover test: AD vs MCI; ‡ Conover test: AD vs NC; § Conover test: MCI vs NC Segmentation of cerebrospinal fluid (CSF) was conducted for each subject separately using the adjusted MiMSeg algorithm [42]. This procedure was based on the Gaussian mixture model and allowed us to separate CSF from the brain by finding the threshold on the greyscale. Two additional descriptors were defined based on MRI scans to numerically represent the changes in the brain structure. The first variable (called 'relative brain volume' (RBV) and shown as a percentage) was defined as the volume of the brain without CSF (V -CSF ) divided by the volume of the whole brain (V) multiplied by 100%: RBV=V -CSF /V·100%, (1) The second variable is a 'shrinkage factor' (SF). The shrinkage factor was defined as the surface of the brain without CSF (S -CSF ) with reference to a volume of the brain without CSF (V -CSF ) and multiplied by 100%: SF=S -CSF /V -CSF ·100%, (2) The additional descriptor is the volume of lateral ventricles. The Automatic Lateral Ventricle delineatioN (ALVIN) algorithm was used to obtain the volume of lateral ventricles. ALVIN is a fully automated algorithm to segment the lateral ventricles from MRI images (ALVIN works within SPM8) [43]. The multinomial logistic regression was used to predict diagnosis status. The following independent variables were considered: age, sex, years of education, MMSE score, relative brain volume, shrinkage factor, and volume of lateral ventricles. The dependent variable was diagnosis status: Alzheimer's disease (AD), mild cognitive impairment (MCI) and normal control (NC) (reference status). Models with two-way interaction terms were also analyzed. A 5-fold cross-validation was executed. The Bayesian information criterion (BIC) was used to select the best model [44]. The comparison between two nested models was conducted using ANOVA, additionally the Bayes factor (exp(DBIC)) was calculated for two compared models. The parameters of a multinomial logistic regression (polytomous) model were estimated by a maximum likelihood estimation procedure. For values of coefficient, the adjusted odds ratio was calculated with its 95% confidence interval according to the method proposed by Woolf [45]. The receiver operating characteristic curve (ROC) together with the area under the curve (AUC) for the classification problem were estimated for both datasets [46]. The scheme of key steps conducted during data analysis is presented in Figure 1 (Figure S1 shows detailed information). 3. Results The tests on ADNI clinical characteristics indicate that the differences between at least two medians are statistically significant for the following variables: years of education and MMSE score. For the independent EDSD dataset, the differences between at least two medians are statistically significant for all variables: age, years of education, and MMSE score. The effect size of age is very small for both datasets, ADNI and EDSD. The effect size of education is small for ADNI and medium for EDSD, and the effect size of the MMSE score is large for ADNI and very large for EDSD. Results of the chi 2 test inform that the null hypothesis "the proportion of the female is the same in NC, MCI and AD" should be rejected for ANDI but not for the independent EDSD dataset. For all cross-validation analyses, the final model has the same structure. Diagnosis status was best predicted by the synergy of relative brain volume, MMSE score and age, where age has a corrective function. A comparison between the model without relative brain volume and age as a predictors (only MMSE was taken into account) and the model with the relative brain volume and age added showed the statistical significance of the differences (p<0.00001; BIC= 785.54 for the model with versus BIC= 809.68 for the model without relative brain volume and age). The value of Bayes factor for compared models is 132646731.7, which indicates very strong evidence for model. No interaction increases the model performance quality. Table S1 presents average values of coefficients (with a 95% confidence interval) obtained in 5-fold cross-validation. Healthy controls (NC) is a reference group. For each predictor, the adjusted odds ratio was calculated (see table S2). For each one percentage point decrease in relative brain volume, the odds of Alzheimer's disease increase by a factor of 1.35 (95% CI [1.27; 1.44]) and the odds of MCI disease increase by a factor of 1.19 (95% CI [1.15; 1.24]) in reference to healthy controls. Among subjects with mild cognitive impairment, for each one percentage point decrease in relative brain volume, the odds of Alzheimer's disease increase by a factor of 1.13 (95% CI [1.10; 1.16]). The decrease of 1 point in MMSE score multiplies the odds of Alzheimer's disease by 8.15 (95% [7.53; 8.81]) in reference to healthy controls. The odds of MCI disease are predicted to grow about 2.65 times larger (95%CI [2.52; 2.79]) for each reduction of point in MMSE score among healthy controls. For each 1 point decrease in MMSE, the odds of Alzheimer's disease increase by 3.07 (95% CI [2.99; 3.16]) for subjects with MCI. Table 2 contains average values of statistics of prediction (with a 95% confidence interval) obtained in a 5-fold cross-validation for ADNI. Table 2. Quality performance indices of prediction system (with 95% confidence interval). Statistics AD vs others NC vs others AD vs NC MCI vs NC AD vs MCI ADNI (Model: Status of disease ~ Relative brain value + MMSE + Age) Sensitivity [%] 63.99 [48.61; 79.37] 70.08 [61.92; 78.24] 100 80.67 [75.63; 85.72] 63.99 [48.61; 79.37] Specificity [%] 94.06 [90.93; 97.19] 87.84 [85.17; 90.51] 100 70.08 [61.92; 78.24] 88.22 [81.53; 94.9] Positive Predictive Value [%] 74.41 [65.56; 83.26] 71.66 [67.12; 76.20] 100 79.84 [76.71; 82.96] 74.41 [65.56; 83.26] Negative Predictive Value [%] 90.91 [87.34; 94.47] 87.13 [84.19; 90.07] 100 71.66 [67.12; 76.20] 82.71 [76.16; 89.26] Prevalence [%] 20.84 [20.42; 21.26] 30.41 [30.05; 30.77] 38.51 [34.3; 42.73] 59.13 [57.50; 60.76] 34.11 [32.94; 35.29] Balanced Accuracy [%] 79.02 [71.65; 86.4] 78.96 [75.23; 82.69] 100 75.38 [72.10; 78.65] 76.1 [68.52; 83.69] AUC [%] 94.18 [92.09; 96.28] 90.01 [87.03; 92.99] 99.65 [99.18;100.00] 79.30 [74.35; 84.24] 90.78 [87.45; 94.11] Cutoff point 0.16 [0.05; 0.26] 0.30 [0.20; 0.40] 8.38 [3.88; 12.87] 47.24 [35.28; 59.21] 30.13 [10.32; 49.95] EDSD (Model: Status of disease ~ Relative brain value + MMSE + Age) Sensitivity [%] 69.85 73.20 96.94 75.78 71.43 Specificity [%] 90.17 88.19 93.42 77.17 80.17 Positive Predictive Value [%] 73.64 80.68 90.48 69.78 79.83 Negative Predictive Value [%] 88.39 83.01 97.93 82.08 71.85 Prevalence [%] 28.22 40.25 39.20 41.03 52.36 Balanced Accuracy [%] 80.01 80.70 95.18 76.48 75.80 AUC [%] 89.95 85.36 93.25 71.08 85.74 EDSD (Model: Status of disease ~ MMSE) Sensitivity [%] 71.32 69.59 97.98 77.78 72.39 Specificity [%] 89.6 89.58 93.10 73.37 79.03 Positive Predictive Value [%] 72.93 81.82 90.65 66.67 78.86 Negative Predictive Value [%] 88.83 81.39 98.54 82.82 72.59 Prevalence [%] 28.22 40.25 40.57 40.65 51.94 Balanced Accuracy [%] 80.46 79.59 95.54 75.57 75.71 AUC [%] 90.19 85.87 93.35 67.04 86.14 Abbreviations: AUC, the area under the ROC curve. Values of areas under the ROC curve (AUC) were very high for classes AD vs others and NC vs others, and 5-fold cross-validation for ADNI resulted in 94.18% and 90.01%, respectively. The value of three classes (AD vs others, NC vs others, MCI vs others) of balanced accuracy is 76.10%. Specificity of 94.06% was gained for AD vs others, and it is the highest value; the sensitivity for this class is 63.99%. The value of Negative Predictive Value [%] (NPV) is 90.91% for AD vs others, while the value of Positive Predictive Value [%] (PPV) is 74.41%. The values of specificity, sensitivity, NPV and PPV for NC vs others are 87.84%, 70.08%, 87.13% and 71.66%, respectively. The pairwise analysis gave a very large value of AUC for the classification of AD versus NC (99.65%). The value of specificity, sensitivity, NPV and PPV for AD vs NC is 100%. The multinomial logistic regression model was also trained on the whole ADNI dataset and tested on the independent EDSD dataset. Values of model coefficients are presented in table S3. As before, for each predictor, the adjusted odds ratio was calculated (see table S4). One can notice that for each one percentage point decrease in relative brain volume, the odds of Alzheimer's disease increase by 1.35 (95% CI [1.25; 1.46]) in reference to healthy controls and the odds of MCI disease increase by a factor of 1.19 (95% CI [1.13; 1.26]) which is very similar to the estimates obtained in the first stage. Among subjects with mild cognitive impairment, for each one percentage point decrease in relative brain volume, the odds of Alzheimer's disease increase by a factor of 1.13 (95% CI [1.10; 1.16]). For each reduction of point in MMSE score, the odds are predicted to grow about 8.06 times larger (95%CI [6.48; 10.04]) for Alzheimer's disease and 2.64 (95% [2.27; 3.08]) for mild cognitive impairment in reference to healthy controls. The decrease of 1 point in MMSE score multiplies the odds of Alzheimer's disease by 3.05 (95% [2.85; 2.36]) among subjects with MCI status. The obtained model was tested on the independent validation EDSD dataset, and table 2 presents the results. Additionally, Table S5 contains results for two subsets of EDSD: 1.5T and 3T. The results of validation for the independent dataset (EDSD) have shown that values of areas under the ROC curve (AUC) for classes: AD versus others and NC vs others are 89.95% and 85.36%, respectively. The value of three classes (AD vs others, NC vs others, MCI vs others) of balanced accuracy is 76.83%. Specificity of 90.17%, sensitivity of 69.85%, NPV of 88.39%, and PPV of 73.64% were gained for AD vs others. The values of specificity, sensitivity, NPV and PPV for NC vs others are 88.19%, 73.20%, 83.01%, and 80.68%, respectively. The pairwise analysis confirmed the very large value of AUC for the classification of AD versus NC (93.25%). Specificity of 93.42% was gained for AD vs NC, and it is the highest value, the sensitivity for this class is 96.94%. The value of NPV is 97.93% for AD vs NC, while the value of PPV is 90.48%. Additionally, table 2 contains the results of validation for the independent EDSD dataset for the model built on the whole ADNI dataset with only independent variable MMSE score. Results confirm that adding the relative brain volume and age as a corrective function for natural brain aging improves the model. The value of AUC for MCI vs NC increases from 67.04% (for model with MMSE score only) to 71.08% (for model with the relative brain volume and age added). The ROC curve was used to summarize the prediction of the model for ADNI and ESDS datasets (Figure 2). During the follow-up, some subjects have converted from MCI status to AD. Table 3 contains the number and percentage of subjects with changes in diagnosis during 12 months of the follow-up in association with the prediction (ADNI datasets) and 48 months of follow-up (ADNI and the independent EDSD datasets). The prediction of the multinomial logistic regression model in 5-fold cross-validation of the ADNI dataset indicates that 32 subjects, who have MCI screening diagnosis, are predicted as AD status. Among these subjects, predictions are in line with 12 months of the follow-up diagnosis in 10 subjects (31.25%). A similar calculation was conducted for 48 months of follow-up. Table 3 contains the number and percentage of subjects with changes in diagnosis from MCI to AD during 48 months of the follow-up in association with the prediction (ADNI datasets). Table 3. Compliance of the prediction with the change in diagnosis from MCI to AD. Change from MCI to AD Number of subjects (%) Compliance of AD prediction with the follow-up diagnosis No change in diagnosis from MCI to AD among AD predicted patients * AD prediction among subjects with MCI screen diagnosis ADNI – 12 months 10 (31.25%) 22 (68.75%) 32 (100%) ADNI – 48 months 18 (56.25%) 14 (43.75%) 32 (100%) EDSD – 48 months 13 (54.17%) 11 (45.83%) 24 (100%) * Not for all subjects, follow-up diagnosis after 48 months is available, so “no change” or “missing information about change” occurs. This means that the outcome could improve if information about the change in diagnosis was available. The prediction was consistent within 48 months of the follow-up diagnosis for 18 subjects (56.25%) among 32 subjects with MCI screening diagnosis and model prediction of Alzheimer's disease. One subject changed the diagnosis from AD to MCI during follow-up, and this diagnosis is compliant with the prediction. Three subjects developed MCI among NC subjects; the diagnosis is compliant with the prediction for one subject. Prediction of the multinomial logistic regression model on the independent EDSD dataset shows similar results. Table 3 contains the number and percentage of subjects with changes in diagnosis from MCI to AD during four years of follow-up in association with the prediction (EDSD dataset). The multinomial logistic regression model on the independent EDSD dataset predicts Alzheimer's disease in 24 subjects with MCI screening diagnosis. Among these 24 subjects, 13 (54.17%) patients transited from MCI to AD status, and they confirmed the model prediction. The percentage may be even improved as some patients haven't follow-up on their diagnosis. 4. Discussion And Conclusions Our aim was to improve classical diagnosis process based on MMSE score. We focused on finding the common available biomarker, which improves diagnostics. We obtained that the multinomial logistic regression model was of the same structure for all cross-validation analyses and based on the complete ADNI dataset. Diagnosis status was best predicted by the relative brain volume, MMSE score, and age. The results of comparison models with MMSE score only and the relative brain volume and age added shows that adding the relative brain volume (and age as an adjustive factor for natural brain aging) improves model. The value of Bayes factor indicates strong evidence and we can notice that the quality of MCI detection increases (AUC: 67.04% vs 71.08%) while maintaining the quality for AD (AUC: 93.35% vs 93.25%). The average values of coefficients of the multinomial logistic regression models for 5-fold cross-validation and results for the whole ADNI dataset are very similar, which confirms the homogeneity of the training dataset and consistency of the diagnosis process. Average values of statistics of prediction obtained in 5-fold cross-validation for ADNI show that we have the outstanding results of classification AD vs NC and AD vs others, with AUC equaling 99.65% and 94.18%, respectively. Additionally, the values of AUC for AD vs MCI and for NC vs others are also very high (90.78% and 90.01%, respectively). The moderate value of AUC we have for MCI vs NC (79.30%) is still a very good result if we take into account that the MCI group is heterogeneous and some patients from this group develop AD, and some patients have stable MCI status. The average value of balanced accuracy for three classes (AD vs others, NC vs others, and MCI vs others) is 76.10% for 5-fold cross-validation. As we aim to develop a supporting diagnostic test, detecting patients with the disease is the most important, so Positive Predictive Value (PPV) and Negative Predictive Value (NPV) are the most important. We have compared our classification results with results reported in the literature based on the ADNI dataset (11 studies used the ADNI dataset as a training dataset, one study used the internal locally dataset as a training dataset and ADNI dataset as an independent validation dataset; 6 studies among 12 used additionally independent validation dataset) (table 4). Table 4. Overview of previous studies based on the ADNI dataset. Studies Sample size Method Input Validation Groups Parameters Results Agostinho et al., 2022 [6] The internal locally dataset (n=41): AD (n=20), NC (n=21). SVM MRI, PiB-PET and DTI Internal locally dataset and external dataset (ADNI (n=330): AD (n=166), NC (n = 164)) AD, NC AUC, ACC, SEN, SPEC, BACC Dependent validation: AD vs NC: MRI: AUC=96%, ACC=92.05%, SEN=86.78%, SPEC=86.78%, BACC=92.05%; PiB PET: AUC=93%, ACC=90.53%, SEN=92%, SPEC=82, BACC=79.84%; DTI: AUC=86%, ACC=76.84%, SEN=76.17%, SPEC=82.09%, BACC=79.84%. Independent validation: AD vs NC: MRI: AUC=81%, ACC=78.02%, SEN=74.12%, SPEC=82.29, BACC=78.20%; PiB PET: AUC=81%, ACC=76.87%, SEN=87.9%, SPEC=68.33%, BACC=78.12%; DTI: AUC=69%, ACC=62.79%, SEN=54.31%, SPEC=71.98%, BACC=63.15%. Gao et al., 2022 [7] 1134 subjects: AD (n=454), NC (n=680). 3DMgNet (multigrid and convolutional neural network) MRI 10-fold cross-validation and external in-house dataset (AD (n=75), NC (n=59)) AD, NC AUC, ACC, SEN, SPEC Dependent validation: ACC=92.13%, AUC=94.43%, SEN=88.42%, SPEC=95%. Independent validation: ACC=87.91%, AUC=95.74%, SEN=79.73%, SPEC=98.31%. Goenka et al., 2022 [8] 769 subjects: AD (n=70), MCI (n=224), NC (475) CNN MRI 633 scans from ADNI dataset AD, MCI, NC AUC, ACC Dependent validation: AD vs NC: ACC= 97.83%, AD vs MCI: ACC=98.68%, NC vs MCI: ACC=99.10%, NC vs MCI vs AD: ACC=98.26%. AD vs NC: AUC=94%, AD vs MCI: AUC=97%, NC vs MCI: AUC=99%, NC vs MCI vs AD: AUC=98%. Tang et al., 2021 [9] 560 subjects: AD (n=80), EMCI (n=230), LMCI (n=110), NC (n=140) SVM, RF, DT MRI 10-fold cross-validation AD, EMCI, LMCI, NC AUC, ACC, SEN, SPEC RF: NC vs AD: ACC=96.14%, SEN=88.14, SPE=92.81%, AUC=92%. NC vs EMCI: ACC=77.45%, SEN=79.51%, SPE=33.54%, AUC=59%. NC vs LMCI: ACC=87.56%, SEN=64.71%, SPE=83.94%, AUC=81%. EMCI vs AD: ACC=90.15%, SEN=93.51%, SPE=92.43%, AUC=85%. LMCI va AD: ACC=84.54%, SEN=67.91, SPE=72.46%, AUC=89%. Dyrba et al., 2021 [10] 633 subjects: AD (n=189), MCI (n=220), NC (n=254) CNN MRI and PET 1-fold cross-validation and three independent datasets: ADNI-3 (n=575), AIBL (n=606), DELCODE (n=474). AD, MCI, NC AUC, ACC, SEN, SPEC, BACC, PPV, NPV Dependent validation: AD vs NC: BACC=88.9%, SEN=94.2%, SPE=83.6%, PPV=81.5%, NPV=95.2% AUC=94.9%. MCI vs NC: BACC=74.5%, SEN=65.5%, SPE=83.6%, PPV=78.1%, NPV=74.1%, AUC=78.5%. amyloid-positive AD vs amyloid-negative NC: BACC=94.9%, SEN=95.6%, SPE=94.3%, PPV=92.7%, NPV=96.6%, AUC=98.5%. amyloid-positive MCI vs amyloid-negative NC: BACC=86.7%, SEN=79%, SPE=94.3%, PPV=91.6%, NPV= 96.6%, AUC=92.5%. Independent validation DELCODE: AD vs NC: BACC=85.5%, SEN=94.2%, SPE=76.7%, PPV=66.2%, NPV=96.5% AUC=95.3%. MCI vs NC: BACC=71%, SEN=65.2%, SPE=76.7%, PPV=66.9%, NPV=75.3%, AUC=77.5%. amyloid-positive AD vs amyloid-negative NC: BACC=83.3%, SEN=95.9%, SPE=70.7%, PPV=73.4%, NPV=95.3%, AUC=96.8%. amyloid-positive MCI vs amyloid-negative NC: BACC=72.2%, SEN=73.7%, SPE=70.7%, PPV=71.2%, NPV= 73.2%, AUC=84%. Marzban et al., 2020 [5] 406 subjects: NC (n=185), MCI (n=106), AD (n = 115) CNN MRI and DTI 10-fold cross-validation AD,NC, MCI AUC, ACC, SEN, SPEC AD vs NC: AUC=94%, ACC=93.5%, SEN=92.5%, SPEC=93.9. MCI vs NC: AUC=84%, ACC=79.6%, SEN=62.7%, SPEC=89% Li et al., 2020 [11] 404 subjects: NC (n=268), AD (n=136) SVM MRI 10-fold cross-validation and independent validation dataset (AD (n=41), NC (n=25)) AD, NC ACC, SEN, SPEC Dependent validation dataset: AD vs NC: ACC=93.07%, SEN=94.12%, SPEC=98.51. Independent validation dataset: AD vs NC: ACC=84.85%, SEN=85.36%, SPEC=84% Bae et al., 2020 [12] 390 subjects: AD (n=195), NC (n=195) CNN MRI 5-fold cross-validation and independent validation dataset (AD (n=195), NC (n=195)) AD, NC AUC, ACC, SEN, SPEC Dependent validation dataset: AD vs NC: AUC=94%, ACC= 89%, SEN= 88%, SPEC=91%. Independent validation dataset: AD vs NC: AUC=88%, ACC= 83%, SEN= 76%, SPEC= 89% Liu et al., 2020 [13] 449 subjects: AD (n=97), MCI (n=233), NC (n=119) CNN MRI 5-fold cross-validation and independent dataset (AD (n=45), MCI (n=46), and NC subjects (n=44)). AD, MCI, NC AUC, ACC, SEN, SPEC Dependent validation: AD vs NC: ACC=88.9%, SEN=86.6%, SPE=90.8%, AUC=92.5%. MCI vs NC: ACC=76.2%, SEN=79.5%, SPE=69.8%, AUC=77.5%. Independent validation: AD vs NC: AUC=89.8% MCI vs NC: AUC=72.2% Zhang et al., 2019 [24] 857 subjects: NC (n=322), MCI (n=322), AD (n = 213) Graph Analysis MRI Data are randomly partitioned into 80% and 20% for training and testing. AD, MCI, NC AUC AD vs MCI + NC: AUC=73%, NC vs AD + MCI: AUC=72%, MCI vs AD + NC: AUC= 69%. Westman et al., 2012 [24] 369 subjects: AD (n=96), MCI (n=162) and NC (n=111). Orthogonal Partial Least-Squares (OPLS) MRI, PET, CSF 7-fold cross-validation AD, MCI, NC AUC, ACC, SEN, SPEC, PPV, NPV AD vs NC: MRI with CSF: ACC= 91.8%, SEN=88.5%, SPEC=94.6%, PPV=93.4%, NPV=90.5% and AUC=95.8%. MRI only: ACC=87%, SEN=83.3%, SPEC=90.1%, PPV=87.9%, NPV=86.2% and AUC=93% CSF only: ACC=81.6%, SEN=84.4%, SPEC=79.3%, PPV=77.9%, NPV=85.4% and AUC=86.1%. MCI vs NC: MRI with CSF: ACC=77.6%, SEN=72.8%, SPEC=84.7%, PPV=87.4%, NPV=68.1% and AUC=87.6%. MRI only: ACC=71.8%, SEN=66.7%, SPEC=79.3%, PPV=82.4%, NPV=62.0% and AUC=81.5%. CSF only: ACC=70.3%, SEN=66.7%, SPEC=75.7%, PPV=80.0%, NPV=60.9% and AUC=74.9%. Eskildsen et al., 2012 [47] 808 subjects: AD (n=194), NC (n=226), pMCI (n=161), sMCI (n=227) LDA MRI (cortical thickness and age) leave-one-out (LOO) validation AD, NC, pMCI, sMCI AUC, ACC, SEN, SPEC Independent feature sets: AD vs NC: ACC=85.5%, SEN=80.4%, SPEC=89.8%, AUC=92%. pMCI vs sMCI: ACC=67.8%, SEN=64.6%, SPEC=70%, AUC=68.2%. Dependent feature sets: AD vs NC: ACC=87.4%, SEN=82.5%, SPEC=91.6%, AUC=93.1%. pMCI vs sMCI: ACC=68.3%, SEN=67.7%, SPEC=68.7%, AUC=74.7%. Abbreviations: AIBL, Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing; DELCODE, DZNE multicenter observational study on Longitudinal Cognitive Impairment and Dementia; EMCI, early mild cognitive impairment; LMCI, late mild cognitive impairment; pMCI, progressive MCI; sMCI, stable MCI; CNN, convolutional neural network; LDA, linear discriminant analysis; SVM, support vector machine; RF, random forest; DT, decision tree; AUC, area under receiver operating characteristic curve; ACC, accuracy; SEN, sensitivity; SPEC, specificity; BAAC, balanced accuracy; PPV, positive predictive value; NPV, negative predictive value. Table 4. Our results for the independent validation dataset are not worse and, in many cases, even better than the results from previously published studies. Our model achieved the best balanced accuracy of 95.18% (balanced ACC) for the independent validation dataset when the highest value of balanced accuracy for AD vs NC from reported studies is 85.5% [10]. Although the highest reported value of AUC is 96.8, in this study, the decision is supported by the concentration of amyloid in CSF [10]. The second top reported AUC value is 95.74%, but this study focuses on only two categories: AD and NC, while we consider MCI as a third one [7]. The third value of AUC is 95.3%. This value is slightly bigger than ours, but other performance indicators like balanced accuracy, sensitivity, specificity, PPV, and NPV for AD vs NC are better in our approach [10]. The lowest value of AUC for AD vs NC among publications presented in table 4 is 69% [6]. The highest sensitivity value for AD vs NC is 95.6% for analysis based on the concentration of amyloid in CSF and 94.2% for analysis without amyloid data, while our estimated sensitivity is better and equal to 96.94% [10]. For the prediction specificity, the highest value observed is 98.31%, but this study focuses only on two categories: AD and NC, which means that it is easier to achieve better results than for three categories [7]. The second highest reported value of specificity is 89.8%, which is lower than ours (93.42%) [47]. The lowest value of specificity among publications is 68.33% [6]. Only one study from table 4 contains the results of NPV and PPV for AD vs NC, values of these indicators are 95.3% and 73.4% for analysis based additionally on amyloid data, and 96.5% and 66.2% for analysis without amyloid data, respectively [10]. Our results are again better; the value of NPV is 97.93% for AD vs NC, while PPV is 90.48%. The comparison of results for 5-fold cross-validation shows that our model achieves better results than all reported studies for the classification task of AD versus NC (table 4) [5, 6-13, 23, 24, 47]. The prediction results of AD vs NC from reported studies show that the highest AUC is 98.5% [10], when our result is 99.65%, the highest accuracy for AD vs NC is 97.83% [8], when our result is 100% (we have the balanced ACC). The highest values of sensitivity and specificity are 95.6% [10] and 98.51% [11], respectively, when our model achieved 100% for both parameters. Additionally, among MCI patients predicted as AD inconsistently with original diagnosis, 56.25% from ADNI and 54.17% from EDSD were re-diagnosed as AD within a 48-month follow-up. 5. Conclusions Our work shows that the proposed MRI-based biomarker, combined with MMSE score and adjusted for age, gives excellent AD status predictions. Moreover, our method, as based on MRI, doesn't require invasive and expensive laboratory tests and, as being a classical statistical learning model, doesn't require large calculation power. In this paper, we proved that incorporating the radiological MRI-based biomarker into the standard clinical AD predictors leads to a handy model for daily clinical routine and improves the diagnosis process. Additionally, we demonstrated that our model detects some patients transitioning from MCI to AD as AD patients a few years earlier before regular medical diagnosis. 6. List Of Abbreviations ACC - accuracy; AD - Alzheimer's disease; ADNI – Alzheimer's Disease Neuroimaging Initiative; AIBL - Australian Imaging, Biomarker & Lifestyle Flagship Study of Ageing; ALVIN - Automatic Lateral Ventricle delineatioN; AUC – area under the curve; BAAC - balanced accuracy; BIC - Bayesian information criterion; CI – confidence interval; CNN - convolutional neural network; CSF - cerebrospinal fluid; DELCODE - DZNE multicenter observational study on Longitudinal Cognitive Impairment and Dementia; DT - decision tree; DTI - diffusion tensor imaging; EDSD – European DTI Study on Dementia; EMCI - early mild cognitive impairment; FDG-PET - fluorodeoxyglucose positron emission tomography imaging; fMRI - functional MRI; LDA - linear discriminant analysis; LMCI - late mild cognitive impairment; MCI - Mild Cognitive Impairment; MMSE – Mini-Mental State Examination score; MRI - magnetic resonance imaging; NC – Normal Control; NPV – negative predictive value; PET - positron emission tomography; pMCI - progressive MCI; PPV – positive predictive value; RBV - relative brain volume; RF - random forest; ROC - receiver operating characteristic curve; SEN - sensitivity; SF - shrinkage factor; sMCI - stable MCI; SPEC - specificity; SVM - support vector machine; Declarations Ethics approval and consent to participate: Not applicable. Consent for publication: Not applicable. Availability of data and materials: Data used in the analysis were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (https://adni.loni.usc.edu/) and The European DTI Study on Dementia (EDSD) (https://www.neugrid2.eu/ and https://neugrid4you.eu/ ). Competing interests: The authors declare that they have no competing interests. Funding: This research was partially funded by the Silesian University of Technology, grant number 02/070/BK_22/0033. Author’s Contributions: Conceptualization, J. P.; methodology, J. P. and A. M.; software, A. M.; formal analysis, A. M.; writing—original draft preparation, A. M.; writing—review and editing, J. P. and A. M.; visualization, A. M.; supervision, J. P.; funding acquisition, J. P. Acknowledgements: Data collection and sharing for this project was funded by the Alzheimer's Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer's Association; Alzheimer's Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research & Development, LLC.; Johnson & Johnson Pharmaceutical Research & Development LLC.; Lumosity; Lundbeck; Merck & Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer's Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California. References Oboudiyat C, Glazer H, Seifan A, Greer C, Isaacson RS. Alzheimer's disease. Semin Neurol. 2013 Sep; 33(4):313-29. doi: 10.1055/s-0033-1359319. Alz.org. Available online: https://www.alz.org/media/Documents/alzheimers-dementia-about-alzheimers-disease-ts.pdf (accessed on 24 October 2021). Roberts R, Knopman DS. Classification and epidemiology of MCI. Clin Geriatr Med. 2013 Nov; 29(4):753-72. doi: 10.1016/j.cger.2013.07.003. Moody JN, Valerio KE, Hasselbach AN, Prieto S, Logue MW, Hayes SM, et al.; Alzheimer's Disease Neuroimaging Initiative (ADNI). Body Mass Index and Polygenic Risk for Alzheimer's Disease Predict Conversion to Alzheimer's Disease. J Gerontol A Biol Sci Med Sci. 2021 Jul 13; 76(8):1415-1422. doi: 10.1093/gerona/glab117. Marzban EN, Eldeib AM, Yassine IA, Kadah YM. Alzheimer's Disease Neurodegenerative Initiative. Alzheimer's disease diagnosis from diffusion tensor images using convolutional neural networks. PLoS One. 2020 Mar 24; 15(3):e0230409. doi: 10.1371/journal.pone.0230409. Agostinho D, Caramelo F, Moreira AP, Santana I, Abrunhosa A, Castelo-Branco M. Combined Structural MR and Diffusion Tensor Imaging Classify the Presence of Alzheimer's Disease With the Same Performance as MR Combined With Amyloid Positron Emission Tomography: A Data Integration Approach. Front Neurosci. 2022 Jan 5; 15:638175. doi: 10.3389/fnins.2021.638175. Gao Y, Huang H, Zhang L. Predicting Alzheimer's Disease Using 3DMgNet. arXiv 2022; arXiv2201.04370v1 [eess.IV]. doi: 10.48550/arXiv.2201.04370 Goenka N, Tiwari S. AlzVNet: A volumetric convolutional neural network for multiclass classification of Alzheimer's disease through multiple neuroimaging computational approaches. Biomed Signal Process Control. 2022; 74: 103500. doi: 10.1016/j.bspc.2022.103500 Tang X, Liu J. Comparing different algorithms for the course of Alzheimer's disease using machine learning. Ann Palliat Med. 2021 Sep; 10(9):9715-9724. doi: 10.21037/apm-21-2013. Dyrba M, Hanzig M, Altenstein S, Bader S, Ballarini T, Brosseron F, et al.; ADNI, AIBL, DELCODE study groups. Improving 3D convolutional neural network comprehensibility via interactive visualization of relevance maps: evaluation in Alzheimer's disease. Alzheimers Res Ther. 2021 Nov 23; 13(1):191. doi: 10.1186/s13195-021-00924-2. Li B, Zhang M, Riphagen J, Morrison Yochim K, Li B, Liu J, Salat DH; Alzheimer's Disease Neuroimaging Initiative. Prediction of clinical and biomarker conformed Alzheimer's disease and mild cognitive impairment from multi-feature brain structural MRI using age-correction from a large independent lifespan sample. Neuroimage Clin. 2020; 28:102387. doi: 10.1016/j.nicl.2020.102387. Bae JB, Lee S, Jung W, Park S, Kim W, Oh H, et al. Identification of Alzheimer's disease using a convolutional neural network model based on T1-weighted magnetic resonance imaging. Sci Rep. 2020 Dec 17; 10(1):22252. doi: 10.1038/s41598-020-79243-9. Liu M, Li F, Yan H, Wang K, Ma Y; Alzheimer's Disease Neuroimaging Initiative, Shen L, Xu M. A multi-model deep convolutional neural network for automatic hippocampus segmentation and classification in Alzheimer's disease. Neuroimage. 2020 Mar; 208:116459. doi: 10.1016/j.neuroimage.2019.116459. Di Stefano F, Epelbaum S, Coley N, Cantet C, Ousset PJ, Hampel H, et al.; GuidAge study group. Prediction of Alzheimer's Disease Dementia: Data from the GuidAge Prevention Trial. J Alzheimers Dis. 2015;48(3):793-804. doi: 10.3233/JAD-150013. Buckley RF, Maruff P, Ames D, Bourgeat P, Martins RN, Masters CL, Rainey-Smith S, Lautenschlager N, Rowe CC, Savage G, Villemagne VL, Ellis KA; AIBL study. Subjective memory decline predicts greater rates of clinical progression in preclinical Alzheimer's disease. Alzheimers Dement. 2016 Jul; 12(7):796-804. doi: 10.1016/j.jalz.2015.12.013. Ampuero I, Ros R, Royuela A, Abraira V, del Ser T, García-Ribas G, et al. Risk factors for dementia of Alzheimer type and aging-associated cognitive decline in a Spanish population based sample, and in brains with pathology confirmed Alzheimer's disease. J Alzheimers Dis. 2008 Jun; 14(2):179-91. doi: 10.3233/jad-2008-14206. Buratti L, Balestrini S, Altamura C, Viticchi G, Falsetti L, Luzzi S, et al. Markers for the risk of progression from mild cognitive impairment to Alzheimer's disease. J Alzheimers Dis. 2015; 45(3):883-90. doi: 10.3233/JAD-143135. Haris M, Yadav SK, Rizwan A, Singh A, Cai K, Kaura D, et al. T1rho MRI and CSF biomarkers in diagnosis of Alzheimer's disease. Neuroimage Clin. 2015 Feb 26; 7:598-604. doi: 10.1016/j.nicl.2015.02.016. Seixas FL, Zadrozny B, Laks J, Conci A, Muchaluat Saade DC. A Bayesian network decision model for supporting the diagnosis of dementia, Alzheimer׳s disease and mild cognitive impairment. Comput Biol Med. 2014 Aug; 51:140-58. doi: 10.1016/j.compbiomed.2014.04.010. Iddi S, Li D, Aisen PS, Rafii MS, Thompson WK, Donohue MC; Alzheimer's Disease Neuroimaging Initiative. Predicting the course of Alzheimer's progression. Brain Inform. 2019 Jun 28; 6(1):6. doi: 10.1186/s40708-019-0099-0. Beheshti I, Demirel H, Matsuda H; Alzheimer's Disease Neuroimaging Initiative. Classification of Alzheimer's disease and prediction of mild cognitive impairment-to-Alzheimer's conversion from structural magnetic resource imaging using feature ranking and a genetic algorithm. Comput Biol Med. 2017 Apr 1; 83:109-119. doi: 10.1016/j.compbiomed.2017.02.011. Zheng C, Xia Y, Pan Y, Chen J. Automated identification of dementia using medical imaging: a survey from a pattern classification perspective. Brain Inform. 2016 Mar; 3(1):17-27. doi: 10.1007/s40708-015-0027-x. Zhang R, Giancardo L, Pena D, Kim Y, Tong H, Jiang X; Alzheimer's Disease Neuroimaging Initiative. From Brain Imaging to Graph Analysis: a study on ADNI's patient cohort. arXiv 2019; arXiv1905.05861v1. doi: 10.48550/ARXIV.1905.05861 Westman E, Muehlboeck JS, Simmons A. Combining MRI and CSF measures for classification of Alzheimer's disease and prediction of mild cognitive impairment conversion. Neuroimage. 2012 Aug 1; 62(1):229-38. doi: 10.1016/j.neuroimage.2012.04.056. Nozadi SH, Kadoury S, The Alzheimer's Disease Neuroimaging Initiative. Classification of Alzheimer's and MCI Patients from Semantically Parcelled PET Images: A Comparison between AV45 and FDG-PET. Int J Biomed Imaging. 2018 Mar 15; 2018:1247430. doi: 10.1155/2018/1247430. Rice L, Bisdas S. The diagnostic value of FDG and amyloid PET in Alzheimer's disease-A systematic review. Eur J Radiol. 2017 Sep; 94:16-24. doi: 10.1016/j.ejrad.2017.07.014. Lee JC, Kim SJ, Hong S, Kim Y. Diagnosis of Alzheimer's disease utilizing amyloid and tau as fluid biomarkers. Exp Mol Med. 2019 May 9; 51(5):1-10. doi: 10.1038/s12276-019-0250-2. Becerra-Laparra I, Cortez-Conradis D, Garcia-Lazaro HG, Martinez-Lopez M, Roldan-Valadez E. Radial diffusivity is the best global biomarker able to discriminate healthy elders, mild cognitive impairment, and Alzheimer's disease: A diagnostic study of DTI-derived data. Neurol India. 2020 Mar-Apr; 68(2):427-434. doi: 10.4103/0028-3886.284376. Chandra A, Dervenoulas G, Politis M; Alzheimer's Disease Neuroimaging Initiative. Magnetic resonance imaging in Alzheimer's disease and mild cognitive impairment. J Neurol. 2019 Jun; 266(6):1293-1302. doi: 10.1007/s00415-018-9016-3. Hansson O, Seibyl J, Stomrud E, Zetterberg H, Trojanowski JQ, Bittner T, Lifke V, Corradini V, et al.; Swedish BioFINDER study group; Alzheimer's Disease Neuroimaging Initiative. CSF biomarkers of Alzheimer's disease concord with amyloid-β PET and predict clinical progression: A study of fully automated immunoassays in BioFINDER and ADNI cohorts. Alzheimers Dement. 2018 Nov; 14(11):1470-1481. doi: 10.1016/j.jalz.2018.01.010. Wolz R, Julkunen V, Koikkalainen J, Niskanen E, Zhang DP, Rueckert D, Soininen H, Lötjönen J; Alzheimer's Disease Neuroimaging Initiative. Multi-method analysis of MRI images in early diagnostics of Alzheimer's disease. PLoS One. 2011; 6(10):e25446. doi: 10.1371/journal.pone.0025446. Folstein MF, Folstein SE, McHugh PR. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975 Nov; 12(3):189-98. doi: 10.1016/0022-3956(75)90026-6. Frisoni GB, Fox NC, Jack CR Jr, Scheltens P, Thompson PM. The clinical use of structural MRI in Alzheimer disease. Nat Rev Neurol. 2010 Feb; 6(2):67-77. doi: 10.1038/nrneurol.2009.215. Zetterberg H, Bendlin BB. Biomarkers for Alzheimer's disease-preparing for a new era of disease-modifying therapies. Mol Psychiatry. 2021 Jan; 26(1):296-308. doi: 10.1038/s41380-020-0721-9. Brueggen K, Grothe MJ, Dyrba M, Fellgiebel A, Fischer F, Filippi M, et al. The European DTI Study on Dementia - A multicenter DTI and MRI study on Alzheimer's disease and Mild Cognitive Impairment. Neuroimage. 2017 Jan; 144(Pt.B):305-308. doi: 10.1016/j.neuroimage.2016.03.067. Wyman BT, Harvey DJ, Crawford K, Bernstein MA, Carmichael O, Cole PE, et al.; Alzheimer's Disease Neuroimaging Initiative. Standardization of analysis sets for reporting results from ADNI MRI data. Alzheimers Dement. 2013 May; 9(3):332-7. doi: 10.1016/j.jalz.2012.06.004. Stone M. Cross-Validatory Choice and Assessment of Statistical Predictions. J. Royal Statistical Soc. Series B; 1974:36:111-147. Tustison NJ, Avants BB, Cook PA, Zheng Y, Egan A, Yushkevich PA, et al. N4ITK: improved N3 bias correction. IEEE Trans Med Imaging. 2010 Jun; 29(6):1310-20. doi: 10.1109/TMI.2010.2046908. Penny WD, Ashburner J, Kiebel S, Henson R, Glaser DE, Phillips C, et al. Statistical Parametric Mapping: An Annotated Bibliography. Wellcome Department of Cognitive Neurology, University College London; 2001. https://www.fil.ion.ucl.ac.uk/spm/doc/spmbib.pdf Cohen J. Eta-Squared and Partial Eta-Squared in Fixed Factor Anova Designs. Educational and Psychological Measurement. 1973; 33(1):107-112. doi: 10.1177/001316447303300111. Ellis PD. The Essential Guide to Effect Sizes: Statistical Power, Meta-Analysis, and the Interpretation of Research Results. Cambridge University Press; 2010:12. Binczyk F, Stjelties B, Weber C, Goetz M, Maier-Hein K, Meinzer HP, et al. MiMSeg - an algorithm for automated detection of tumor tissue on NMR apparent diffusion coefficient maps. Information Sciences, 2017; 384:235-248. doi:10.1016/j.ins.2016.07.052. Kempton MJ, Underwood TS, Brunton S, Stylios F, Schmechtig A, Ettinger U, et al. A comprehensive testing protocol for MRI neuroanatomical segmentation techniques: Evaluation of a novel lateral ventricle segmentation method. Neuroimage. 2011 Oct 15; 58(4):1051-9. doi: 10.1016/j.neuroimage.2011.06.080. Wit E, Heuvel EVD, Romeyn JW. 'All models are wrong...': an introduction to model uncertainty. Statistica Neerlandica; 2012: 66 (3): 217–236. Agresti A. On logit confidence intervals for the odds ratio with small samples. Biometrics. 1999 Jun; 55(2):597-602. doi: 10.1111/j.0006-341x.1999.00597.x. García Barrado L, Coart E, Burzykowski T; Alzheimer's Disease Neuroimaging Initiative. Development of a diagnostic test based on multiple continuous biomarkers with an imperfect reference test. Stat Med. 2016 Feb 20; 35(4):595-608. doi: 10.1002/sim.6733. Eskildsen SF, Coupé P, García-Lorenzo D, Fonov V, Pruessner JC, Collins DL; Alzheimer's Disease Neuroimaging Initiative. Prediction of Alzheimer's disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. Neuroimage. 2013 Jan 15; 65:511-21. doi: 10.1016/j.neuroimage.2012.09.058. Additional Declarations No competing interests reported. Supplementary Files Supplementarymaterial.docx FigS1.jpg Scheme of the preprocessing and analysis Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-1747001","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":113158440,"identity":"d0e580db-b2f2-487b-abb3-759ab62f647a","order_by":0,"name":"Anna Marcisz","email":"","orcid":"","institution":"Silesian University of Technology","correspondingAuthor":false,"prefix":"","firstName":"Anna","middleName":"","lastName":"Marcisz","suffix":""},{"id":113158441,"identity":"af6b613d-cb2b-4b45-b664-350f31c9c9f0","order_by":1,"name":"Joanna Polanska","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCklEQVRIiWNgGAWjYBACxgYYix1EHGCQY2BgbmB4wMAgQ1gLM0SLMVgsgYGBh7B9UC2JDYS0MLc3P3tcUMOQx8/M/OzhjzN26dvZG5s/JDDcwamFseeYufGMYwzFks1s5sY8N5Jzd/YcbJNIYHiGW8uMBDNpHjaGxA2HGcykGT4w5264kdgGdNhhPFrSv0nz/ANpYf8m+eNDfbrBjUSQw/BpyTGT5m0DaeExk+C5cTgBqKVBAq+WnjNl0rx9EkC/8JRJ85w5bgjxiwFuvxi2t2+T5vlmk8fP3r5N8sexanlz9ubDHz5U3JHDqaUBTAFdAgMGEPIALh0M8lAaXQsDbi2jYBSMglEw4gAAwS9XhkgO7UoAAAAASUVORK5CYII=","orcid":"","institution":"Silesian University of Technology","correspondingAuthor":true,"prefix":"","firstName":"Joanna","middleName":"","lastName":"Polanska","suffix":""}],"badges":[],"createdAt":"2022-06-10 23:29:02","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-1747001/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-1747001/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":22896464,"identity":"eed51973-e069-4e89-bc2b-fbb609699bb3","added_by":"auto","created_at":"2022-06-21 16:07:49","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":51956,"visible":true,"origin":"","legend":"\u003cp\u003eThe scheme of key steps of data preprocessing and data analysis.\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-1747001/v1/789c00118b0d38b56286447b.jpeg"},{"id":22895133,"identity":"2c6a6a0b-ca5b-4218-a2dc-f840491a2793","added_by":"auto","created_at":"2022-06-21 16:02:49","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":248431,"visible":true,"origin":"","legend":"\u003cp\u003eThe ROC curve for classification between AD, MCI and NC: (a) The ROC curve for classification with average values of 5-fold cross-validation (ADNI data): AD vs others, NC vs others; (b) The ROC curve for classification with average values of 5-fold cross-validation (ADNI data): Ad vs NC, MCI vs NC, AD vs MCI; (c) The ROC curve for classification using ADNI data as training data and EDSD data (whole dataset) as test data: AD vs others, NC vs others; (d) The ROC curve for classification using ADNI data as training data and EDSD data (whole dataset) as test data: AD vs NC, MCI vs NC, AD vs MCI.\u003c/p\u003e","description":"","filename":"floatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-1747001/v1/a764f17feaa70ee0c075ac90.png"},{"id":22896477,"identity":"82e8d0ba-34ec-4e89-bfa1-ff9a3827915d","added_by":"auto","created_at":"2022-06-21 16:07:53","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":814007,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-1747001/v1/19331849-deef-4f55-8d52-2c483aa18f19.pdf"},{"id":22895131,"identity":"867c1b8c-ca10-47b8-939d-e14c3df9fa0a","added_by":"auto","created_at":"2022-06-21 16:02:49","extension":"docx","order_by":1,"title":"","display":"","copyAsset":false,"role":"supplement","size":15333,"visible":true,"origin":"","legend":"","description":"","filename":"Supplementarymaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-1747001/v1/21a5748b5415e380879a8e0d.docx"},{"id":22896462,"identity":"4a85ab70-e5d7-4248-bad2-8c00cf29eb1e","added_by":"auto","created_at":"2022-06-21 16:07:49","extension":"jpg","order_by":2,"title":"","display":"","copyAsset":false,"role":"supplement","size":109428,"visible":true,"origin":"","legend":"\u003cp\u003eScheme of the preprocessing and analysis\u003c/p\u003e","description":"","filename":"FigS1.jpg","url":"https://assets-eu.researchsquare.com/files/rs-1747001/v1/670dde62685031805ba495d4.jpg"}],"financialInterests":"No competing interests reported.","formattedTitle":"Can we improve prediction of Alzheimer's disease and Mild Cognitive Impairment by combining MMSE score and MRI-based imaging data?","fulltext":[{"header":"1. Background","content":"\u003cp\u003eAlzheimer's disease (AD) is a progressive, neurodegenerative brain disease that causes memory loss, changes in behaviour, and problems with everyday tasks. Alzheimer's is the most common form of dementia, and it is responsible for 60\u0026ndash;80% of dementia cases [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. The intermediate stage from normal cognition to dementia is a mild cognitive impairment (MCI). People suffering from MCI have a high rate of progression to dementia over a relatively short period, but not everyone will develop Alzheimer's disease [\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Within a 3-year follow-up period, about 35% of patients with MCI status progress to AD or dementia [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. A yearly conversion rate equals 5%-10% [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThe main aim of this work is to find easily-accessible biomarkers for Alzheimer's disease (AD) and a mild cognitive impairment (MCI) to improve the diagnosis process. An additional challenge is to predict the diagnosis of Alzheimer's while a patient is still mildly cognitively impaired.\u003c/p\u003e \u003cp\u003eMany different methods to predict the diagnosis have been proposed in recent years. These methods are based on machine learning algorithms [\u003cspan additionalcitationids=\"CR6 CR7 CR8 CR9 CR10 CR11 CR12\" citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e], regression models [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e, \u003cspan additionalcitationids=\"CR15 CR16 CR17\" citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e], and other methods [\u003cspan additionalcitationids=\"CR20 CR21 CR22 CR23\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. Many different biomarkers are used to classify AD and MCI. The first group of biomarkers is based on structural brain atrophy obtained from MRI imaging [\u003cspan additionalcitationids=\"CR8\" citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The second group of biomarkers uses the evaluation of brain metabolic changes, which are measured by fluorodeoxyglucose positron emission tomography (FDG-PET) imaging [\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e, \u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]. Fluid biomarkers are the third group, and this is connected with amyloid and tau obtained from cerebrospinal fluid (CSF) [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]. Moreover, diffusion tensor imaging (DTI) and functional MRI (fMRI) are also applied for the detection of AD and MCI [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e, \u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. Most studies use multiple biomarkers in the early diagnosis of AD and MCI and are based on a combination of two or more following biomarkers: MRI-based biomarkers, fluid biomarkers or PET-based markers [\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e, \u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e, \u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e, \u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. The availability of all three biomarkers (PET and CSF and MRI or DTI or fMRI) is limited due to the cost, time and invasiveness of the methods (PET and CSF) [\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eThis article presents a method that improves an MCI and AD's diagnosis process based on easily-accessible clinical biomarkers like age and Mini-Mental State Examination score (MMSE) [\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e], available in medical history for almost every patient with suspicion of dementia. We suggest using, in addition to those clinical predictors, the MRI-based disease progression radiological biomarkers to support the diagnosis. In patients suffering from Alzheimer's disease, the brain shrinks, and therefore the space filled with cerebrospinal fluid increases [\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e, \u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e]. Moreover, this brain shrinkage causes the brain to be more wrinkled. Because of that, we consider the relative brain volume and global measure of brain wrinkling as the imaging biomarkers.\u003c/p\u003e"},{"header":"2. Methods","content":"\u003cp\u003e\u003cstrong\u003e2.1 MATERIALS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eData used in the study were obtained from the Alzheimer\u0026apos;s Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu) and The European DTI Study on Dementia (EDSD). The ADNI was launched in 2003 as a public-private partnership led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer\u0026apos;s disease (AD). For up-to-date information, see www.adni-info.org. The European DTI Study on Dementia (EDSD) is a multicenter framework created to study the diagnostic accuracy and inter-site variability of DTI-derived markers in patients with manifest and prodromal Alzheimer\u0026apos;s disease [35].\u003c/p\u003e\n\u003cp\u003eThe standard analysis dataset of the ADNI-1 project was used (collection name: ADNI1: Complete 1Yr 1.5T; subjects who have both 6- and 12-month scans available) to build a statistical model for predicting AD or MCI status [36]. This dataset was randomly split into and five subsets to conduct internal testing and 5-Fold Cross-Validation [37]. In the second stage, the final statistical model was built on the whole dataset, and that model was tested on the independent dataset from the EDSD database. The dataset of the ADNI-1 project includes MPRAGE T1- weighted 3D scans (1.5 T) and several clinical and neuropsychological measures acquired from healthy controls (NC), mild cognitive impaired (MCI) subjects and Alzheimer\u0026apos;s disease (AD).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe second dataset used in the analysis comes from The European DTI Study on Dementia (EDSD) database [35]. The EDSD was started in 2010. The coordinator of this database is the German Center for Neurodegenerative Diseases (DZNE) in Rostock, Germany. Since 2013, the EDSD has also collected the data of subjects with MCI. Dataset used in the preparation of this article includes data of subjects who were marked as \u0026quot;not dropout\u0026quot;. Our analysis was based on T1-weighted MRI.\u003c/p\u003e\n\u003cp\u003eADNI dataset (dataset 1) is a reference dataset, and the EDSD dataset (dataset 2) is an independent validation dataset. Additionally, the EDSD dataset was divided into two subsets related to MRI scanning options: 1.5T and 3T.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eADNI provided intensity normalized and gradient un-warped TI image volumes [36]. The EDSD native data were used, and N4 bias field correction in N4ITK framework was applied [38]. For both datasets: ADNI and EDSD, skull stripping was achieved in the SPM 12 software package (https://www.fil.ion.ucl.ac.uk/spm/) [39].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.2 STATISTICAL METHODS\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe clinical characteristics of subjects from the ADNI and EDSD datasets were summarized by the diagnostic group (NC, MCI, AD) and presented in table 1. The following variables were considered at baseline: age, sex, mini-mental state exam (MMSE) and years of education. For quantitative measures, values of mean and SD were calculated, and for categorical variables, the percentage was presented. The comparisons between groups were conducted using the nonparametric Kruskal-Wallis test for quantitative measures (the Conover test was used in the post-hoc analysis), and the chi\u003csup\u003e2\u003c/sup\u003e test to compare proportions and \u003cem\u003eP-\u003c/em\u003evalue is presented in table 1. Additionally, table 1 contains effect size\u0026nbsp;h\u003csup\u003e2\u0026nbsp;\u003c/sup\u003e(eta-squared)\u003csup\u003e\u0026nbsp;\u003c/sup\u003ewith 95% confidence interval [40, 41].\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1.\u003c/strong\u003e Clinical characteristics of the ANDI and EDSD dataset.\u003c/p\u003e\n\u003ctable border=\"1\" cellpadding=\"0\" cellspacing=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003e\u003cstrong\u003eCharacteristic\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e\u003cstrong\u003eNC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e\u003cstrong\u003eMCI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAD\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e\u003cstrong\u003eUnadjusted\u0026nbsp;\u003cbr\u003e \u003cem\u003eP\u0026nbsp;\u003c/em\u003evalue\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e\u003cstrong\u003eEffect size\u0026nbsp;\u003c/strong\u003e\u003cstrong\u003eh\u003csup\u003e2\u003c/sup\u003e\u0026nbsp; \u0026nbsp; \u0026nbsp;[95%CI]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eADNI, n\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e311\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e133\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eAge, mean (SD) [years]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e75.9 (5.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e74.9 (7.06)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e74.7 (7.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e.4643 \u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e0.0053\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0; 0.0199]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eEducation, mean (SD) [years]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e16.0 (2.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e15.7 (3.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e14.7 (3.11)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e.0003 \u003csup\u003e*\u003c/sup\u003e\u003cbr\u003e.0004 \u003csup\u003e\u0026dagger;\u003c/sup\u003e\u003cbr\u003e\u0026lt; .0001 \u003csup\u003e\u0026Dagger;\u003cbr\u003e\u0026nbsp;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e0.0253\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.0057; 0.0523]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eMMSE score, mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e29.1 (1.03)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e27.0 (1.78)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e23.5 (1.91)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e\u0026lt; .0001 \u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003cp\u003e\u0026lt; .0001 \u003csup\u003e\u0026dagger; \u0026Dagger; \u0026sect;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e0.6023\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.5581; 0.6387]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eFemale [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e47.9\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e35.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e48.1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e.0053\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eEDSD, n\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e194\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e152\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"12.88135593220339%\"\u003e\n \u003cp\u003e136\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"18.305084745762713%\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"21.694915254237287%\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eAge, mean (SD), [years]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e68.7 (5.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e71.2 (6.76)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e72.4 (8.28)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e\u0026lt; .0001 \u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003cp\u003e.0001 \u003csup\u003e\u0026Dagger;\u003cbr\u003e\u0026nbsp;\u003c/sup\u003e\u0026lt; .0001 \u003csup\u003e\u0026sect;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e0.0497\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.0170; 0.0900]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eEducation, mean (SD), [years]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e13.1 (3.67) n=173\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e12.4 (3.35) n=132\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e10.3 (3.33) n=134\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e\u0026lt; .0001 \u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003cp\u003e\u0026lt; .0001 \u003csup\u003e\u0026dagger; \u0026Dagger;\u003c/sup\u003e\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e0.1036\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.0538; 0.1572]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eMMSE score, mean (SD)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e27.4 (6.49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e26.3 (3.14)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e20.8 (5.36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"18.305084745762713%\"\u003e\n \u003cp\u003e\u0026lt; .0001 \u003csup\u003e*\u003cbr\u003e\u0026nbsp;\u003c/sup\u003e\u0026lt; .0001 \u003csup\u003e\u0026dagger; \u0026Dagger; \u0026sect;\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.694915254237287%\"\u003e\n \u003cp\u003e0.2198\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.1569; 0.2793]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"22.372881355932204%\"\u003e\n \u003cp\u003eFemale [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"12.88135593220339%\"\u003e\n \u003cp\u003e51.0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.864406779661017%\"\u003e\n \u003cp\u003e43.4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"12.88135593220339%\"\u003e\n \u003cp\u003e56.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"18.305084745762713%\"\u003e\n \u003cp\u003e.07854\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"21.694915254237287%\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003e\u003csup\u003e*\u003c/sup\u003e Kruskal-Wallis rank sum test; \u003csup\u003e\u0026dagger;\u0026nbsp;\u003c/sup\u003eConover test:\u003csup\u003e\u0026nbsp;\u003c/sup\u003eAD vs MCI; \u003csup\u003e\u0026Dagger;\u003c/sup\u003e Conover test:\u003csup\u003e\u0026nbsp;\u003c/sup\u003eAD vs NC;\u003csup\u003e\u0026nbsp; \u0026nbsp;\u0026sect;\u003c/sup\u003e Conover test:\u003csup\u003e\u0026nbsp;\u003c/sup\u003eMCI vs NC\u003c/p\u003e\n\u003cp\u003eSegmentation of cerebrospinal fluid (CSF) was conducted for each subject separately using the adjusted MiMSeg algorithm [42]. This procedure was based on the Gaussian mixture model and allowed us to separate CSF from the brain by finding the threshold on the greyscale.\u003c/p\u003e\n\u003cp\u003eTwo additional descriptors were defined based on MRI scans to numerically represent the changes in the brain structure. The first variable (called \u0026apos;relative brain volume\u0026apos; (RBV) and shown as a percentage) was defined as the volume of the brain without CSF (V\u003csub\u003e-CSF\u003c/sub\u003e) divided by the volume of the whole brain (V) multiplied by 100%:\u0026nbsp;\u003c/p\u003e\n\u003ctable border=\"0\" cellpadding=\"0\" cellspacing=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"90.64885496183206%\"\u003e\n \u003cp\u003eRBV=V\u003csub\u003e-CSF\u003c/sub\u003e/V\u0026middot;100%,\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.351145038167939%\"\u003e\n \u003cp\u003e(1)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eThe second variable is a \u0026apos;shrinkage factor\u0026apos; (SF). The shrinkage factor was defined as the surface of the brain without CSF (S\u003csub\u003e-CSF\u003c/sub\u003e) with reference to a volume of the brain without CSF (V\u003csub\u003e-CSF\u003c/sub\u003e) and multiplied by 100%:\u003c/p\u003e\n\u003ctable border=\"0\" cellpadding=\"0\" cellspacing=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"90.64885496183206%\"\u003e\n \u003cp\u003eSF=S\u003csub\u003e-CSF\u003c/sub\u003e/V\u003csub\u003e-CSF\u003c/sub\u003e\u0026middot;100%,\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"9.351145038167939%\"\u003e\n \u003cp\u003e(2)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eThe additional descriptor is the volume of lateral ventricles. The Automatic Lateral Ventricle delineatioN (ALVIN) algorithm was used to obtain the volume of lateral ventricles. ALVIN is a fully automated algorithm to segment the lateral ventricles from MRI images (ALVIN works within SPM8) [43].\u003c/p\u003e\n\u003cp\u003eThe multinomial logistic regression was used to predict diagnosis status. The following independent variables were considered: age, sex, years of education, MMSE score, relative brain volume, shrinkage factor, and volume of lateral ventricles. The dependent variable was diagnosis status: Alzheimer\u0026apos;s disease (AD), mild cognitive impairment (MCI) and normal control (NC) (reference status). Models with two-way interaction terms were also analyzed. A 5-fold cross-validation was executed. The Bayesian information criterion (BIC) was used to select the best model [44]. The comparison between two nested models was conducted using ANOVA, additionally the Bayes factor (exp(DBIC)) was calculated for two compared models. The parameters of a multinomial logistic regression (polytomous) model were estimated by a maximum likelihood estimation procedure. For values of coefficient, the adjusted odds ratio was calculated with its 95% confidence interval according to the method proposed by Woolf [45]. The receiver operating characteristic curve (ROC) together with the area under the curve (AUC) for the classification problem were estimated for both datasets [46].\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eThe scheme of key steps conducted during data analysis is presented in Figure 1 (Figure S1 shows detailed information).\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003eThe tests on ADNI clinical characteristics indicate that the differences between at least two medians are statistically significant for the following variables: years of education and MMSE score. For the independent EDSD dataset, the differences between at least two medians are statistically significant for all variables: age, years of education, and MMSE score. The effect size of age is very small for both datasets, ADNI and EDSD. The effect size of education is small for ADNI and medium for EDSD, and the effect size of the MMSE score is large for ADNI and very large for EDSD. Results of the chi\u003csup\u003e2\u003c/sup\u003e test inform that the null hypothesis \u0026quot;the proportion of the female is the same in NC, MCI and AD\u0026quot; should be rejected for ANDI but not for the independent EDSD dataset.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor all cross-validation analyses, the final model has the same structure. Diagnosis status was best predicted by the synergy of relative brain volume, MMSE score and age, where age has a corrective function. A comparison between the model without relative brain volume and age as a predictors (only MMSE was taken into account) and the model with the relative brain volume \u0026nbsp;and age added showed the statistical significance of the differences (p\u0026lt;0.00001; BIC= 785.54 for the model with versus BIC= 809.68 for the model without relative brain volume and age). The value of Bayes factor for compared models is 132646731.7, which indicates very strong evidence for model. \u0026nbsp;No interaction increases the model performance quality.\u003c/p\u003e\n\u003cp\u003eTable S1 presents average values of coefficients (with a 95% confidence interval) obtained in 5-fold cross-validation. Healthy controls (NC) is a reference group.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFor each predictor, the adjusted odds ratio was calculated (see table S2). For each one percentage point decrease in relative brain volume, the odds of Alzheimer\u0026apos;s disease increase by a factor of 1.35 (95% CI [1.27; 1.44]) and the odds of MCI disease increase by a factor of 1.19 (95% CI [1.15; 1.24]) in reference to healthy controls. Among subjects with mild cognitive impairment, for each one percentage point decrease in relative brain volume, the odds of Alzheimer\u0026apos;s disease increase by a factor of 1.13 (95% CI [1.10; 1.16]). The decrease of 1 point in MMSE score multiplies the odds of Alzheimer\u0026apos;s disease by 8.15 (95% [7.53; 8.81]) in reference to healthy controls. The odds of MCI disease are predicted to grow about 2.65 times larger (95%CI [2.52; 2.79]) for each reduction of point in MMSE score among healthy controls. For each 1 point decrease in MMSE, the odds of Alzheimer\u0026apos;s disease increase by 3.07 (95% CI [2.99; 3.16]) for subjects with MCI.\u003c/p\u003e\n\u003cp\u003eTable 2 contains average values of statistics of prediction (with a 95% confidence interval) obtained in a 5-fold cross-validation for ADNI.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.\u0026nbsp;\u003c/strong\u003eQuality performance indices of prediction system (with 95% confidence interval).\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellpadding=\"0\" cellspacing=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" width=\"17.877906976744185%\"\u003e\n \u003cp\u003e\u003cstrong\u003eStatistics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAD vs others\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e\u003cstrong\u003eNC vs others\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAD vs NC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e\u003cstrong\u003eMCI vs NC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAD vs MCI\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" width=\"100%\"\u003e\n \u003cp\u003e\u003cstrong\u003eADNI (Model: Status of disease ~ Relative brain value + MMSE + Age)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eSensitivity [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e63.99\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e[48.61; 79.37]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e70.08\u003c/p\u003e\n \u003cp\u003e[61.92; 78.24]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e80.67\u0026nbsp;\u003cbr\u003e\u0026nbsp;[75.63; 85.72]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e63.99\u0026nbsp;\u003cbr\u003e\u0026nbsp;[48.61; 79.37]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eSpecificity [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e94.06\u0026nbsp;\u003cbr\u003e\u0026nbsp;[90.93; 97.19]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e87.84\u0026nbsp;\u003cbr\u003e\u0026nbsp;[85.17; 90.51]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e70.08\u0026nbsp;\u003cbr\u003e\u0026nbsp;[61.92; 78.24]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e88.22\u0026nbsp;\u003cbr\u003e\u0026nbsp;[81.53; 94.9]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003ePositive Predictive Value [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e74.41\u0026nbsp;\u003cbr\u003e\u0026nbsp;[65.56; 83.26]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e71.66\u0026nbsp;\u003cbr\u003e\u0026nbsp;[67.12; 76.20]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e79.84\u0026nbsp;\u003cbr\u003e\u0026nbsp;[76.71; 82.96]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e74.41\u0026nbsp;\u003cbr\u003e\u0026nbsp;[65.56; 83.26]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eNegative Predictive Value [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.91\u0026nbsp;\u003cbr\u003e\u0026nbsp;[87.34; 94.47]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e87.13\u0026nbsp;\u003cbr\u003e\u0026nbsp;[84.19; 90.07]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e71.66\u0026nbsp;\u003cbr\u003e\u0026nbsp;[67.12; 76.20]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e82.71\u0026nbsp;\u003cbr\u003e\u0026nbsp;[76.16; 89.26]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003ePrevalence [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e20.84\u0026nbsp;\u003cbr\u003e\u0026nbsp;[20.42; 21.26]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e30.41\u0026nbsp;\u003cbr\u003e\u0026nbsp;[30.05; 30.77]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e38.51\u0026nbsp;\u003cbr\u003e\u0026nbsp;[34.3; 42.73]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e59.13\u0026nbsp;\u003cbr\u003e\u0026nbsp;[57.50; 60.76]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e34.11\u0026nbsp;\u003cbr\u003e\u0026nbsp;[32.94; 35.29]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eBalanced\u0026nbsp;\u003cbr\u003e\u0026nbsp;Accuracy [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e79.02\u0026nbsp;\u003cbr\u003e\u0026nbsp;[71.65; 86.4]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"bottom\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e78.96\u0026nbsp;\u003cbr\u003e\u0026nbsp;[75.23; 82.69]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e100\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e75.38\u0026nbsp;\u003cbr\u003e\u0026nbsp;[72.10; 78.65]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e76.1\u0026nbsp;\u003cbr\u003e\u0026nbsp;[68.52; 83.69]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eAUC [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e94.18\u0026nbsp;\u003cbr\u003e\u0026nbsp;[92.09; 96.28]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.01\u0026nbsp;\u003cbr\u003e\u0026nbsp;[87.03; 92.99]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e99.65\u003c/p\u003e\n \u003cp\u003e[99.18;100.00]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e79.30\u0026nbsp;\u003cbr\u003e\u0026nbsp;[74.35; 84.24]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.78\u0026nbsp;\u003cbr\u003e\u0026nbsp;[87.45; 94.11]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eCutoff point\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e0.16\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.05; 0.26]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e0.30\u0026nbsp;\u003cbr\u003e\u0026nbsp;[0.20; 0.40]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e8.38\u0026nbsp;\u003cbr\u003e\u0026nbsp;[3.88; 12.87]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e47.24\u0026nbsp;\u003cbr\u003e\u0026nbsp;[35.28; 59.21]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" width=\"16.424418604651162%\"\u003e\n \u003cp\u003e30.13\u0026nbsp;\u003cbr\u003e\u0026nbsp;[10.32; 49.95]\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" width=\"100%\"\u003e\n \u003cp\u003e\u003cstrong\u003eEDSD (Model: Status of disease ~ Relative brain value + MMSE + Age)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eSensitivity [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e69.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e73.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e96.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e75.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e71.43\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eSpecificity [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e88.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e93.42\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e77.17\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e80.17\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003ePositive Predictive Value [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e73.64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e80.68\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.48\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e69.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e79.83\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eNegative Predictive Value [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e88.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e83.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e97.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e82.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e71.85\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003ePrevalence [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e28.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e40.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e39.20\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e41.03\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e52.36\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eBalanced\u0026nbsp;\u003cbr\u003e\u0026nbsp;Accuracy [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e80.01\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e80.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e95.18 \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e76.48 \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e75.80 \u0026nbsp; \u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eAUC [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e89.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e85.36\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e93.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e71.08\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e85.74\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd colspan=\"6\" width=\"100%\"\u003e\n \u003cp\u003e\u003cstrong\u003eEDSD (Model: Status of disease ~ MMSE)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eSensitivity [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e71.32\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e69.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e97.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e77.78\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e72.39\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eSpecificity [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e89.6\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e89.58\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e93.10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e73.37\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e79.03\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003ePositive Predictive Value [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e72.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e81.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e66.67\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e78.86\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eNegative Predictive Value [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e88.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e81.39\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e98.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e82.82\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e72.59\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003ePrevalence [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e28.22\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e40.25\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e40.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e40.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e51.94\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eBalanced\u0026nbsp;\u003cbr\u003e\u0026nbsp;Accuracy [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e80.46\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e79.59\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e95.54\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e75.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e75.71\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"17.877906976744185%\"\u003e\n \u003cp\u003eAUC [%]\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e90.19\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e85.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e93.35\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e67.04\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"16.424418604651162%\"\u003e\n \u003cp\u003e86.14\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eAbbreviations: AUC, the area under the ROC curve.\u003c/p\u003e\n\u003cp\u003eValues of areas under the ROC curve (AUC) were very high for classes AD vs others and NC vs others, and 5-fold cross-validation for ADNI resulted in 94.18% and 90.01%, respectively. The value of three classes (AD vs others, NC vs others, MCI vs others) of balanced accuracy is 76.10%. Specificity of 94.06% was gained for AD vs others, and it is the highest value; the sensitivity for this class is 63.99%. The value of Negative Predictive Value [%] (NPV) is 90.91% for AD vs others, while the value of Positive Predictive Value [%] (PPV) is 74.41%. The values of specificity, sensitivity, NPV and PPV for NC vs others are 87.84%, 70.08%, 87.13% and 71.66%, respectively. The pairwise analysis gave a very large value of AUC for the classification of AD versus NC (99.65%). The value of specificity, sensitivity, NPV and PPV for AD vs NC is 100%.\u003c/p\u003e\n\u003cp\u003eThe multinomial logistic regression model was also trained on the whole ADNI dataset and tested on the independent EDSD dataset. Values of model coefficients are presented in table S3.\u003c/p\u003e\n\u003cp\u003eAs before, for each predictor, the adjusted odds ratio was calculated (see table S4). One can notice that for each one percentage point decrease in relative brain volume, the odds of Alzheimer\u0026apos;s disease increase by 1.35 (95% CI [1.25; 1.46]) in reference to healthy controls and the odds of MCI disease increase by a factor of 1.19 (95% CI [1.13; 1.26]) which is very similar to the estimates obtained in the first stage. Among subjects with mild cognitive impairment, for each one percentage point decrease in relative brain volume, the odds of Alzheimer\u0026apos;s disease increase by a factor of 1.13 (95% CI [1.10; 1.16]). For each reduction of point in MMSE score, the odds are predicted to grow about 8.06 times larger (95%CI [6.48; 10.04]) for Alzheimer\u0026apos;s disease and 2.64 (95% [2.27; 3.08]) for mild cognitive impairment in reference to healthy controls. The decrease of 1 point in MMSE score multiplies the odds of Alzheimer\u0026apos;s disease by 3.05 (95% [2.85; 2.36]) among subjects with MCI status.\u003c/p\u003e\n\u003cp\u003eThe obtained model was tested on the independent validation EDSD dataset, and table 2 presents the results. Additionally, Table S5 contains results for two subsets of EDSD: 1.5T and 3T.\u003c/p\u003e\n\u003cp\u003eThe results of validation for the independent dataset (EDSD) have shown that values of areas under the ROC curve (AUC) for classes: AD versus others and NC vs others are 89.95% and 85.36%, respectively. The value of three classes (AD vs others, NC vs others, MCI vs others) of balanced accuracy is 76.83%. Specificity of 90.17%, sensitivity of 69.85%, NPV of 88.39%, and PPV of 73.64% were gained for AD vs others. The values of specificity, sensitivity, NPV and PPV for NC vs others are 88.19%, 73.20%, 83.01%, and 80.68%, respectively. The pairwise analysis confirmed the very large value of AUC for the classification of AD versus NC (93.25%). Specificity of 93.42% was gained for AD vs NC, and it is the highest value, the sensitivity for this class is 96.94%. The value of NPV is 97.93% for AD vs NC, while the value of PPV is 90.48%. Additionally, table 2 contains the results of validation for the independent EDSD dataset for the model built on the whole ADNI dataset with only independent variable MMSE score. Results confirm that adding the relative brain volume and age as a corrective function for natural brain aging improves the model. The value of AUC for MCI vs NC increases from 67.04% (for model with MMSE score only) to 71.08% (for model with the relative brain volume and age added).\u003c/p\u003e\n\u003cp\u003eThe ROC curve was used to summarize the prediction of the model for ADNI and ESDS datasets (Figure 2).\u003c/p\u003e\n\u003cp\u003eDuring the follow-up, some subjects have converted from MCI status to AD. Table 3 contains the number and percentage of subjects with changes in diagnosis during 12 months of the follow-up in association with the prediction (ADNI datasets) and 48 months of follow-up (ADNI and the independent EDSD datasets). The prediction of the multinomial logistic regression model in 5-fold cross-validation of the ADNI dataset indicates that 32 subjects, who have MCI screening diagnosis, are predicted as AD status. Among these subjects, predictions are in line with 12 months of the follow-up diagnosis in 10 subjects (31.25%). A similar calculation was conducted for 48 months of follow-up. Table 3 contains the number and percentage of subjects with changes in diagnosis from MCI to AD during 48 months of the follow-up in association with the prediction (ADNI datasets).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3.\u0026nbsp;\u003c/strong\u003eCompliance of the prediction with the change in diagnosis from MCI to AD.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellpadding=\"0\" cellspacing=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"30.28846153846154%\"\u003e\n \u003cp\u003e\u003cstrong\u003eChange from MCI to AD\u003cbr\u003e\u0026nbsp;\u003c/strong\u003eNumber of subjects (%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.153846153846153%\"\u003e\n \u003cp\u003e\u003cstrong\u003eCompliance of AD prediction with the follow-up diagnosis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.80128205128205%\"\u003e\n \u003cp\u003e\u003cstrong\u003eNo change in diagnosis from MCI to AD among AD predicted patients\u003c/strong\u003e\u003csup\u003e*\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.756410256410255%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAD prediction among subjects\u0026nbsp;\u003cbr\u003e\u0026nbsp;with MCI screen\u0026nbsp;\u003cbr\u003e\u0026nbsp;diagnosis\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" width=\"30.28846153846154%\"\u003e\n \u003cp\u003eADNI \u0026ndash; 12 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.153846153846153%\"\u003e\n \u003cp\u003e10 (31.25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.80128205128205%\"\u003e\n \u003cp\u003e\u0026nbsp;22 (68.75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.756410256410255%\"\u003e\n \u003cp\u003e32 (100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" width=\"30.28846153846154%\"\u003e\n \u003cp\u003eADNI \u0026ndash; 48 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.153846153846153%\"\u003e\n \u003cp\u003e18 (56.25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.80128205128205%\"\u003e\n \u003cp\u003e\u0026nbsp;14 (43.75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.756410256410255%\"\u003e\n \u003cp\u003e32 (100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"bottom\" width=\"30.28846153846154%\"\u003e\n \u003cp\u003eEDSD \u0026ndash; 48 months\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"21.153846153846153%\"\u003e\n \u003cp\u003e13 (54.17%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"25.80128205128205%\"\u003e\n \u003cp\u003e11 (45.83%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.756410256410255%\"\u003e\n \u003cp\u003e24 (100%)\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003e\u003csup\u003e*\u003c/sup\u003e Not for all subjects, follow-up diagnosis after 48 months is available, so \u0026ldquo;no change\u0026rdquo; or \u0026ldquo;missing information about change\u0026rdquo; occurs. This means that the outcome could improve if information about the change in diagnosis was available.\u003c/p\u003e\n\u003cp\u003eThe prediction was consistent within 48 months of the follow-up diagnosis for 18 subjects (56.25%) among 32 subjects with MCI screening diagnosis and model prediction of Alzheimer\u0026apos;s disease. One subject changed the diagnosis from AD to MCI during follow-up, and this diagnosis is compliant with the prediction. Three subjects developed MCI among NC subjects; the diagnosis is compliant with the prediction for one subject.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePrediction of the multinomial logistic regression model on the independent EDSD dataset shows similar results. Table 3 contains the number and percentage of subjects with changes in diagnosis from MCI to AD during four years of follow-up in association with the prediction (EDSD dataset). The multinomial logistic regression model on the independent EDSD dataset predicts Alzheimer\u0026apos;s disease in 24 subjects with MCI screening diagnosis. Among these 24 subjects, 13 (54.17%) patients transited from MCI to AD status, and they confirmed the model prediction. The percentage may be even improved as some patients haven\u0026apos;t follow-up on their diagnosis.\u003c/p\u003e"},{"header":"4. Discussion And Conclusions","content":"\u003cp\u003eOur aim was to improve classical diagnosis process based on MMSE score. We focused on finding the common available biomarker, which improves diagnostics. We obtained that the multinomial logistic regression model was of the same structure for all cross-validation analyses and based on the complete ADNI dataset. Diagnosis status was best predicted by the relative brain volume, MMSE score, and age. The results of comparison models with MMSE score only and the relative brain volume and age added shows that adding the relative brain volume (and age as an adjustive factor for natural brain aging) improves model. The value of Bayes factor indicates strong evidence and we can notice that the quality of MCI detection increases (AUC: 67.04% vs 71.08%) while maintaining the quality for AD (AUC: 93.35% vs 93.25%). \u0026nbsp; The average values of coefficients of the multinomial logistic regression models for 5-fold cross-validation and results for the whole ADNI dataset are very similar, which confirms the homogeneity of the training dataset and consistency of the diagnosis process. Average values of statistics of prediction obtained in 5-fold cross-validation for ADNI show that we have the outstanding results of classification AD vs NC and AD vs others, with AUC equaling 99.65% and 94.18%, respectively. Additionally, the values of AUC for AD vs MCI and for NC vs others are also very high (90.78% and 90.01%, respectively). The moderate value of AUC we have for MCI vs NC (79.30%) is still a very good result if we take into account that the MCI group is heterogeneous and some patients from this group develop AD, and some patients have stable MCI status. The average value of balanced accuracy for three classes (AD vs others, NC vs others, and MCI vs others) is 76.10% for 5-fold cross-validation. As we aim to develop a supporting diagnostic test, detecting patients with the disease is the most important, so Positive Predictive Value (PPV) and Negative Predictive Value (NPV) are the most important.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe have compared our classification results with results reported in the literature based on the ADNI dataset (11 studies used the ADNI dataset as a training dataset, one study used the internal locally dataset as a training dataset and ADNI dataset as an independent validation dataset; 6 studies among 12 used additionally independent validation dataset) (table 4).\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 4.\u0026nbsp;\u003c/strong\u003eOverview of previous studies based on the ADNI dataset.\u003c/p\u003e\n\u003cdiv\u003e\n \u003ctable border=\"1\" cellpadding=\"0\" cellspacing=\"0\" width=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eStudies\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e\u003cstrong\u003eSample size\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003e\u003cstrong\u003eMethod\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003e\u003cstrong\u003eInput\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e\u003cstrong\u003eValidation\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003e\u003cstrong\u003eGroups\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003e\u003cstrong\u003eParameters\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003e\u003cstrong\u003eResults\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eAgostinho et al., 2022 [6]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003eThe internal locally dataset (n=41): AD (n=20), NC (n=21).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI, PiB-PET and DTI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003eInternal locally dataset and external dataset (ADNI (n=330): AD (n=166), NC (n = 164))\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC, BACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation: AD vs NC: MRI: AUC=96%, ACC=92.05%, SEN=86.78%, SPEC=86.78%, BACC=92.05%; PiB PET: AUC=93%, ACC=90.53%, SEN=92%, SPEC=82, BACC=79.84%; DTI: AUC=86%, ACC=76.84%, SEN=76.17%, SPEC=82.09%, BACC=79.84%.\u0026nbsp;\u003cbr\u003e\u0026nbsp;Independent validation: AD vs NC: MRI: AUC=81%, ACC=78.02%, SEN=74.12%, SPEC=82.29, BACC=78.20%; PiB PET: AUC=81%, ACC=76.87%, SEN=87.9%, SPEC=68.33%, BACC=78.12%; DTI: AUC=69%, ACC=62.79%, SEN=54.31%, SPEC=71.98%, BACC=63.15%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eGao et al., 2022 [7]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e1134 subjects: AD (n=454), NC (n=680).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003e3DMgNet (multigrid and convolutional neural network)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e10-fold cross-validation and external in-house dataset (AD (n=75), NC (n=59))\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation: ACC=92.13%, AUC=94.43%, SEN=88.42%, SPEC=95%.\u0026nbsp;\u003cbr\u003e\u0026nbsp;Independent validation: ACC=87.91%, AUC=95.74%, SEN=79.73%, SPEC=98.31%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eGoenka et al., 2022 [8]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e769 subjects: AD (n=70), MCI (n=224), NC (475)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e633 scans from ADNI dataset\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, MCI, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation: AD vs NC: ACC= 97.83%, AD vs MCI: ACC=98.68%, NC vs MCI: ACC=99.10%, NC vs MCI vs AD: ACC=98.26%.\u003cbr\u003e\u0026nbsp;AD vs NC: AUC=94%, AD vs MCI: AUC=97%, NC vs MCI: AUC=99%, NC vs MCI vs AD: AUC=98%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eTang et al., 2021 [9]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e560 subjects: AD (n=80), EMCI (n=230), LMCI (n=110), NC (n=140)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eSVM, RF, DT\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e10-fold cross-validation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, EMCI, LMCI, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eRF: NC vs AD: ACC=96.14%, SEN=88.14, SPE=92.81%, AUC=92%.\u003cbr\u003e\u0026nbsp;NC vs EMCI: ACC=77.45%, SEN=79.51%, SPE=33.54%, AUC=59%.\u003cbr\u003e\u0026nbsp;NC vs LMCI: ACC=87.56%, SEN=64.71%, SPE=83.94%, AUC=81%.\u003cbr\u003e\u0026nbsp;EMCI vs AD: ACC=90.15%, SEN=93.51%, SPE=92.43%, AUC=85%.\u003cbr\u003e\u0026nbsp;LMCI va AD: ACC=84.54%, SEN=67.91, SPE=72.46%, AUC=89%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eDyrba et al., 2021 [10]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e633 subjects: AD (n=189), MCI (n=220), NC (n=254)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI and PET\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e1-fold cross-validation and three independent datasets: ADNI-3 (n=575), AIBL (n=606), DELCODE (n=474).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, MCI, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC, BACC, PPV, NPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation:\u0026nbsp;\u003cbr\u003e\u0026nbsp;AD vs NC: BACC=88.9%, SEN=94.2%, SPE=83.6%, PPV=81.5%, NPV=95.2% AUC=94.9%.\u003cbr\u003e\u0026nbsp;MCI vs NC: BACC=74.5%, SEN=65.5%, SPE=83.6%, PPV=78.1%, NPV=74.1%, AUC=78.5%.\u003cbr\u003e\u0026nbsp;amyloid-positive AD vs amyloid-negative NC: BACC=94.9%, SEN=95.6%, SPE=94.3%, PPV=92.7%, NPV=96.6%, AUC=98.5%.\u003cbr\u003e\u0026nbsp;amyloid-positive MCI vs amyloid-negative NC: BACC=86.7%, SEN=79%, SPE=94.3%, PPV=91.6%, NPV= 96.6%, AUC=92.5%.\u003cbr\u003e\u0026nbsp;Independent validation DELCODE:\u003cbr\u003e\u0026nbsp;AD vs NC: BACC=85.5%, SEN=94.2%, SPE=76.7%, PPV=66.2%, NPV=96.5% AUC=95.3%.\u003cbr\u003e\u0026nbsp;MCI vs NC: BACC=71%, SEN=65.2%, SPE=76.7%, PPV=66.9%, NPV=75.3%, AUC=77.5%.\u003cbr\u003e\u0026nbsp;amyloid-positive AD vs amyloid-negative NC: BACC=83.3%, SEN=95.9%, SPE=70.7%, PPV=73.4%, NPV=95.3%, AUC=96.8%.\u003cbr\u003e\u0026nbsp;amyloid-positive MCI vs amyloid-negative NC: BACC=72.2%, SEN=73.7%, SPE=70.7%, PPV=71.2%, NPV= 73.2%, AUC=84%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eMarzban et al., 2020 [5]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e406 subjects: NC (n=185), MCI (n=106), AD (n = 115)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI and DTI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e10-fold cross-validation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD,NC, MCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eAD vs NC: AUC=94%, ACC=93.5%, SEN=92.5%, SPEC=93.9.\u0026nbsp;\u003cbr\u003e\u0026nbsp;MCI vs NC: AUC=84%, ACC=79.6%, SEN=62.7%, SPEC=89%\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eLi et al., 2020 [11]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e404 subjects: NC (n=268), AD (n=136)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e10-fold cross-validation and independent validation dataset (AD (n=41), NC (n=25))\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation dataset: AD vs NC: ACC=93.07%, SEN=94.12%, SPEC=98.51.\u0026nbsp;\u003cbr\u003e\u0026nbsp;Independent validation dataset: AD vs NC: ACC=84.85%, SEN=85.36%, SPEC=84%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eBae et al., 2020 [12]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e390 subjects: AD (n=195), NC (n=195)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e5-fold cross-validation and independent validation dataset (AD (n=195), NC (n=195))\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation dataset: AD vs NC: AUC=94%, ACC= 89%, SEN= 88%, SPEC=91%.\u0026nbsp;\u003cbr\u003e\u0026nbsp;Independent validation dataset: AD vs NC: AUC=88%, ACC= 83%, SEN= 76%, SPEC= 89%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eLiu et al., 2020 [13]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e449 subjects: AD (n=97), MCI (n=233), NC (n=119)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eCNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e5-fold\u003c/p\u003e\n \u003cp\u003ecross-validation and independent dataset (AD (n=45),\u003c/p\u003e\n \u003cp\u003eMCI (n=46), and NC subjects (n=44)).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, MCI, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eDependent validation:\u003c/p\u003e\n \u003cp\u003eAD vs NC: ACC=88.9%, SEN=86.6%, SPE=90.8%, AUC=92.5%.\u003c/p\u003e\n \u003cp\u003eMCI vs NC: ACC=76.2%, SEN=79.5%, SPE=69.8%, AUC=77.5%.\u003c/p\u003e\n \u003cp\u003eIndependent validation:\u003c/p\u003e\n \u003cp\u003eAD vs NC: AUC=89.8%\u003c/p\u003e\n \u003cp\u003eMCI vs NC: AUC=72.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eZhang et al., 2019 [24]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e857 subjects: NC (n=322), MCI (n=322), AD (n = 213)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eGraph Analysis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003eData are\u003cbr\u003e\u0026nbsp;randomly partitioned into 80% and 20% for training and testing.\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, MCI, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eAD vs MCI + NC: AUC=73%,\u0026nbsp;\u003cbr\u003e\u0026nbsp;NC vs AD + MCI: AUC=72%,\u0026nbsp;\u003cbr\u003e\u0026nbsp;MCI vs AD + NC: AUC= 69%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eWestman et al., 2012 [24]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e369 subjects: AD (n=96), MCI (n=162) and NC (n=111).\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eOrthogonal Partial Least-Squares (OPLS)\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI, PET, CSF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003e7-fold cross-validation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, MCI, NC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC, PPV, NPV\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eAD vs NC: MRI with CSF: ACC= 91.8%, SEN=88.5%, SPEC=94.6%, PPV=93.4%, NPV=90.5% and AUC=95.8%.\u003cbr\u003e\u0026nbsp;MRI only: ACC=87%, SEN=83.3%, SPEC=90.1%, PPV=87.9%, NPV=86.2% and AUC=93%\u003cbr\u003e\u0026nbsp;CSF only: ACC=81.6%, SEN=84.4%, SPEC=79.3%, PPV=77.9%, NPV=85.4% and AUC=86.1%.\u003cbr\u003e\u0026nbsp;MCI vs NC: MRI with CSF: ACC=77.6%, SEN=72.8%, SPEC=84.7%, PPV=87.4%, NPV=68.1% and AUC=87.6%.\u003cbr\u003e\u0026nbsp;MRI only: ACC=71.8%, SEN=66.7%, SPEC=79.3%,\u003cbr\u003e\u0026nbsp;PPV=82.4%, NPV=62.0% and AUC=81.5%.\u003cbr\u003e\u0026nbsp;CSF only: ACC=70.3%, SEN=66.7%, SPEC=75.7%, PPV=80.0%, NPV=60.9% and AUC=74.9%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"6.9109947643979055%\"\u003e\n \u003cp\u003e\u003cstrong\u003eEskildsen et al., 2012 [47]\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"13.193717277486911%\"\u003e\n \u003cp\u003e808 subjects: AD (n=194), NC (n=226), pMCI (n=161), sMCI (n=227)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.586387434554974%\"\u003e\n \u003cp\u003eLDA\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eMRI (cortical thickness and age)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"11.518324607329843%\"\u003e\n \u003cp\u003eleave-one-out (LOO) validation\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAD, NC, pMCI, sMCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"7.958115183246074%\"\u003e\n \u003cp\u003eAUC, ACC, SEN, SPEC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"35.916230366492144%\"\u003e\n \u003cp\u003eIndependent feature sets: AD vs NC: ACC=85.5%, SEN=80.4%, SPEC=89.8%, AUC=92%. pMCI vs sMCI: ACC=67.8%, SEN=64.6%, SPEC=70%, AUC=68.2%.\u003cbr\u003e\u0026nbsp;Dependent feature sets: AD vs NC: ACC=87.4%, SEN=82.5%, SPEC=91.6%, AUC=93.1%. pMCI vs sMCI: ACC=68.3%, SEN=67.7%, SPEC=68.7%, AUC=74.7%.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n \u003c/table\u003e\n\u003c/div\u003e\n\u003cp\u003eAbbreviations: AIBL, Australian Imaging, Biomarker \u0026amp; Lifestyle Flagship Study of Ageing; DELCODE, DZNE multicenter observational study on Longitudinal Cognitive Impairment and Dementia; EMCI, early mild cognitive impairment; LMCI, late mild cognitive impairment; pMCI, progressive MCI; sMCI, stable MCI; CNN, convolutional neural network; LDA, linear discriminant analysis; SVM, support vector machine; RF, random forest; DT, decision tree; AUC, area under receiver operating characteristic curve; ACC, accuracy; SEN, sensitivity; SPEC, specificity; BAAC, balanced accuracy; PPV, positive predictive value; NPV, negative predictive value.\u003c/p\u003e\n\u003cp\u003eTable 4.\u003c/p\u003e\n\u003cp\u003eOur results for the independent validation dataset are not worse and, in many cases, even better than the results from previously published studies. Our model achieved the best balanced accuracy of 95.18% (balanced ACC) for the independent validation dataset when the highest value of balanced accuracy for AD vs NC from reported studies is 85.5% [10]. Although the highest reported value of AUC is 96.8, in this study, the decision is supported by the concentration of amyloid in CSF [10]. The second top reported AUC value is 95.74%, but this study focuses on only two categories: AD and NC, while we consider MCI as a third one [7]. The third value of AUC is 95.3%. This value is slightly bigger than ours, but other performance indicators like balanced accuracy, sensitivity, specificity, PPV, and NPV for AD vs NC are better in our approach [10]. The lowest value of AUC for AD vs NC among publications presented in table 4 is 69% [6]. The highest sensitivity value for AD vs NC is 95.6% for analysis based on the concentration of amyloid in CSF and 94.2% for analysis without amyloid data, while our estimated sensitivity is better and equal to 96.94% [10]. For the prediction specificity, the highest value observed is 98.31%, but this study focuses only on two categories: AD and NC, which means that it is easier to achieve better results than for three categories [7]. The second highest reported value of specificity is 89.8%, which is lower than ours (93.42%) [47]. The lowest value of specificity among publications is 68.33% [6]. Only one study from table 4 contains the results of NPV and PPV for AD vs NC, values of these indicators are 95.3% and 73.4% for analysis based additionally on amyloid data, and 96.5% and 66.2% for analysis without amyloid data, respectively [10]. Our results are again better; the value of NPV is 97.93% for AD vs NC, while PPV is 90.48%. The comparison of results for 5-fold cross-validation shows that our model achieves better results than all reported studies for the classification task of AD versus NC (table 4) [5, 6-13, 23, 24, 47]. The prediction results of AD vs NC from reported studies show that the highest AUC is 98.5% [10], when our result is 99.65%, the highest accuracy for AD vs NC is 97.83% [8], when our result is 100% (we have the balanced ACC). The highest values of sensitivity and specificity are 95.6% [10] and 98.51% [11], respectively, when our model achieved 100% for both parameters.\u003c/p\u003e\n\u003cp\u003eAdditionally, among MCI patients predicted as AD inconsistently with original diagnosis, 56.25% from ADNI and 54.17% from EDSD were re-diagnosed as AD within a 48-month follow-up.\u003c/p\u003e"},{"header":"5. Conclusions","content":"\u003cp\u003eOur work shows that the proposed MRI-based biomarker, combined with MMSE score and adjusted for age, gives excellent AD status predictions. Moreover, our method, as based on MRI, doesn\u0026apos;t require invasive and expensive laboratory tests and, as being a classical statistical learning model, doesn\u0026apos;t require large calculation power.\u003c/p\u003e\n\u003cp\u003eIn this paper, we proved that incorporating the radiological MRI-based biomarker into the standard clinical AD predictors leads to a handy model for daily clinical routine and improves the diagnosis process. Additionally, we demonstrated that our model detects some patients transitioning from MCI to AD as AD patients a few years earlier before regular medical diagnosis.\u0026nbsp;\u003c/p\u003e"},{"header":"6. List Of Abbreviations","content":"\u003cp\u003eACC - accuracy;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eAD - Alzheimer\u0026apos;s disease;\u003c/p\u003e\n\u003cp\u003eADNI \u0026ndash; Alzheimer\u0026apos;s Disease Neuroimaging Initiative;\u003c/p\u003e\n\u003cp\u003eAIBL - Australian Imaging, Biomarker \u0026amp; Lifestyle Flagship Study of Ageing;\u003c/p\u003e\n\u003cp\u003eALVIN - Automatic Lateral Ventricle delineatioN;\u003c/p\u003e\n\u003cp\u003eAUC \u0026ndash; area under the curve;\u003c/p\u003e\n\u003cp\u003eBAAC - balanced accuracy;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eBIC - Bayesian information criterion;\u003c/p\u003e\n\u003cp\u003eCI \u0026ndash; confidence interval;\u003c/p\u003e\n\u003cp\u003eCNN - convolutional neural network;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eCSF - cerebrospinal fluid;\u003c/p\u003e\n\u003cp\u003eDELCODE - DZNE multicenter observational study on Longitudinal Cognitive Impairment and Dementia;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDT - decision tree;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eDTI - diffusion tensor imaging;\u003c/p\u003e\n\u003cp\u003eEDSD \u0026ndash; European DTI Study on Dementia;\u003c/p\u003e\n\u003cp\u003eEMCI - early mild cognitive impairment;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFDG-PET - fluorodeoxyglucose positron emission tomography imaging;\u003c/p\u003e\n\u003cp\u003efMRI - functional MRI;\u003c/p\u003e\n\u003cp\u003eLDA - linear discriminant analysis;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eLMCI - late mild cognitive impairment;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eMCI - Mild Cognitive Impairment;\u003c/p\u003e\n\u003cp\u003eMMSE \u0026ndash; Mini-Mental State Examination score;\u003c/p\u003e\n\u003cp\u003eMRI - magnetic resonance imaging;\u003c/p\u003e\n\u003cp\u003eNC \u0026ndash; Normal Control;\u003c/p\u003e\n\u003cp\u003eNPV \u0026ndash; negative predictive value;\u003c/p\u003e\n\u003cp\u003ePET - positron emission tomography;\u003c/p\u003e\n\u003cp\u003epMCI - progressive MCI;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003ePPV \u0026ndash; positive predictive value;\u003c/p\u003e\n\u003cp\u003eRBV - relative brain volume;\u003c/p\u003e\n\u003cp\u003eRF - random forest;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eROC - receiver operating characteristic curve;\u003c/p\u003e\n\u003cp\u003eSEN - sensitivity;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSF - shrinkage factor;\u003c/p\u003e\n\u003cp\u003esMCI - stable MCI;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSPEC - specificity;\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSVM - support vector machine;\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate:\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication:\u0026nbsp;\u003c/strong\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials:\u0026nbsp;\u003c/strong\u003eData used in the analysis were obtained from the Alzheimer\u0026apos;s Disease Neuroimaging Initiative (ADNI) database (https://adni.loni.usc.edu/) and The European DTI Study on Dementia (EDSD) (https://www.neugrid2.eu/ and https://neugrid4you.eu/\u003cu\u003e).\u003c/u\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests:\u0026nbsp;\u003c/strong\u003eThe authors declare that they have no competing interests.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding:\u003c/strong\u003e This research was partially funded by the Silesian University of Technology, grant number 02/070/BK_22/0033.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor\u0026rsquo;s Contributions:\u003c/strong\u003e Conceptualization, J. P.; methodology, J. P. and A. M.; software, A. M.; formal analysis, A. M.; writing\u0026mdash;original draft preparation, A. M.; writing\u0026mdash;review and editing, J. P. and A. M.; visualization, A. M.; supervision, J. P.; funding acquisition, J. P.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements:\u003c/strong\u003e Data collection and sharing for this project was funded by the Alzheimer\u0026apos;s Disease Neuroimaging Initiative (ADNI) (National Institutes of Health Grant U01 AG024904) and DOD ADNI (Department of Defense award number W81XWH-12-2-0012). ADNI is funded by the National Institute on Aging, the National Institute of Biomedical Imaging and Bioengineering, and through generous contributions from the following: AbbVie, Alzheimer\u0026apos;s Association; Alzheimer\u0026apos;s Drug Discovery Foundation; Araclon Biotech; BioClinica, Inc.; Biogen; Bristol-Myers Squibb Company; CereSpir, Inc.; Cogstate; Eisai Inc.; Elan Pharmaceuticals, Inc.; Eli Lilly and Company; EuroImmun; F. Hoffmann-La Roche Ltd and its affiliated company Genentech, Inc.; Fujirebio; GE Healthcare; IXICO Ltd.; Janssen Alzheimer Immunotherapy Research \u0026amp; Development, LLC.; Johnson \u0026amp; Johnson Pharmaceutical Research \u0026amp; Development LLC.; Lumosity; Lundbeck; Merck \u0026amp; Co., Inc.; Meso Scale Diagnostics, LLC.; NeuroRx Research; Neurotrack Technologies; Novartis Pharmaceuticals Corporation; Pfizer Inc.; Piramal Imaging; Servier; Takeda Pharmaceutical Company; and Transition Therapeutics. The Canadian Institutes of Health Research is providing funds to support ADNI clinical sites in Canada. Private sector contributions are facilitated by the Foundation for the National Institutes of Health (www.fnih.org). The grantee organization is the Northern California Institute for Research and Education, and the study is coordinated by the Alzheimer\u0026apos;s Therapeutic Research Institute at the University of Southern California. ADNI data are disseminated by the Laboratory for Neuro Imaging at the University of Southern California.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eOboudiyat C, Glazer H, Seifan A, Greer C, Isaacson RS. Alzheimer\u0026apos;s disease. Semin Neurol. 2013 Sep; 33(4):313-29. doi: 10.1055/s-0033-1359319.\u003c/li\u003e\n \u003cli\u003eAlz.org. Available online: https://www.alz.org/media/Documents/alzheimers-dementia-about-alzheimers-disease-ts.pdf (accessed on 24 October 2021).\u003c/li\u003e\n \u003cli\u003eRoberts R, Knopman DS. Classification and epidemiology of MCI. Clin Geriatr Med. 2013 Nov; 29(4):753-72. doi: 10.1016/j.cger.2013.07.003.\u003c/li\u003e\n \u003cli\u003eMoody JN, Valerio KE, Hasselbach AN, Prieto S, Logue MW, Hayes SM, et al.; Alzheimer\u0026apos;s Disease Neuroimaging Initiative (ADNI). Body Mass Index and Polygenic Risk for Alzheimer\u0026apos;s Disease Predict Conversion to Alzheimer\u0026apos;s Disease. J Gerontol A Biol Sci Med Sci. 2021 Jul 13; 76(8):1415-1422. doi: 10.1093/gerona/glab117.\u003c/li\u003e\n \u003cli\u003eMarzban EN, Eldeib AM, Yassine IA, Kadah YM. Alzheimer\u0026apos;s Disease Neurodegenerative Initiative. Alzheimer\u0026apos;s disease diagnosis from diffusion tensor images using convolutional neural networks. PLoS One. 2020 Mar 24; 15(3):e0230409. doi: 10.1371/journal.pone.0230409.\u003c/li\u003e\n \u003cli\u003eAgostinho D, Caramelo F, Moreira AP, Santana I, Abrunhosa A, Castelo-Branco M. Combined Structural MR and Diffusion Tensor Imaging Classify the Presence of Alzheimer\u0026apos;s Disease With the Same Performance as MR Combined With Amyloid Positron Emission Tomography: A Data Integration Approach. Front Neurosci. 2022 Jan 5; 15:638175. doi: 10.3389/fnins.2021.638175.\u003c/li\u003e\n \u003cli\u003eGao Y, Huang H, Zhang L. Predicting Alzheimer\u0026apos;s Disease Using 3DMgNet. arXiv 2022; arXiv2201.04370v1 [eess.IV]. doi: 10.48550/arXiv.2201.04370\u003c/li\u003e\n \u003cli\u003eGoenka N, Tiwari S. AlzVNet: A volumetric convolutional neural network for multiclass classification of Alzheimer\u0026apos;s disease through multiple neuroimaging computational approaches. Biomed Signal Process Control. 2022; 74: 103500. doi: 10.1016/j.bspc.2022.103500\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eTang X, Liu J. Comparing different algorithms for the course of Alzheimer\u0026apos;s disease using machine learning. Ann Palliat Med. 2021 Sep; 10(9):9715-9724. doi: 10.21037/apm-21-2013.\u003c/li\u003e\n \u003cli\u003eDyrba M, Hanzig M, Altenstein S, Bader S, Ballarini T, Brosseron F, et al.; ADNI, AIBL, DELCODE study groups. Improving 3D convolutional neural network comprehensibility via interactive visualization of relevance maps: evaluation in Alzheimer\u0026apos;s disease. Alzheimers Res Ther. 2021 Nov 23; 13(1):191. doi: 10.1186/s13195-021-00924-2.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eLi B, Zhang M, Riphagen J, Morrison Yochim K, Li B, Liu J, Salat DH; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Prediction of clinical and biomarker conformed Alzheimer\u0026apos;s disease and mild cognitive impairment from multi-feature brain structural MRI using age-correction from a large independent lifespan sample. Neuroimage Clin. 2020; 28:102387. doi: 10.1016/j.nicl.2020.102387.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eBae JB, Lee S, Jung W, Park S, Kim W, Oh H, et al. Identification of Alzheimer\u0026apos;s disease using a convolutional neural network model based on T1-weighted magnetic resonance imaging. Sci Rep. 2020 Dec 17; 10(1):22252. doi: 10.1038/s41598-020-79243-9.\u003c/li\u003e\n \u003cli\u003eLiu M, Li F, Yan H, Wang K, Ma Y; Alzheimer\u0026apos;s Disease Neuroimaging Initiative, Shen L, Xu M. A multi-model deep convolutional neural network for automatic hippocampus segmentation and classification in Alzheimer\u0026apos;s disease. Neuroimage. 2020 Mar; 208:116459. doi: 10.1016/j.neuroimage.2019.116459.\u003c/li\u003e\n \u003cli\u003eDi Stefano F, Epelbaum S, Coley N, Cantet C, Ousset PJ, Hampel H, et al.; GuidAge study group. Prediction of Alzheimer\u0026apos;s Disease Dementia: Data from the GuidAge Prevention Trial. J Alzheimers Dis. 2015;48(3):793-804. doi: 10.3233/JAD-150013.\u003c/li\u003e\n \u003cli\u003eBuckley RF, Maruff P, Ames D, Bourgeat P, Martins RN, Masters CL, Rainey-Smith S, Lautenschlager N, Rowe CC, Savage G, Villemagne VL, Ellis KA; AIBL study. Subjective memory decline predicts greater rates of clinical progression in preclinical Alzheimer\u0026apos;s disease. Alzheimers Dement. 2016 Jul; 12(7):796-804. doi: 10.1016/j.jalz.2015.12.013.\u003c/li\u003e\n \u003cli\u003eAmpuero I, Ros R, Royuela A, Abraira V, del Ser T, Garc\u0026iacute;a-Ribas G, et al. Risk factors for dementia of Alzheimer type and aging-associated cognitive decline in a Spanish population based sample, and in brains with pathology confirmed Alzheimer\u0026apos;s disease. J Alzheimers Dis. 2008 Jun; 14(2):179-91. doi: 10.3233/jad-2008-14206.\u003c/li\u003e\n \u003cli\u003eBuratti L, Balestrini S, Altamura C, Viticchi G, Falsetti L, Luzzi S, et al. Markers for the risk of progression from mild cognitive impairment to Alzheimer\u0026apos;s disease. J Alzheimers Dis. 2015; 45(3):883-90. doi: 10.3233/JAD-143135.\u003c/li\u003e\n \u003cli\u003eHaris M, Yadav SK, Rizwan A, Singh A, Cai K, Kaura D, et al. T1rho MRI and CSF biomarkers in diagnosis of Alzheimer\u0026apos;s disease. Neuroimage Clin. 2015 Feb 26; 7:598-604. doi: 10.1016/j.nicl.2015.02.016.\u003c/li\u003e\n \u003cli\u003eSeixas FL, Zadrozny B, Laks J, Conci A, Muchaluat Saade DC. A Bayesian network decision model for supporting the diagnosis of dementia, Alzheimer׳s disease and mild cognitive impairment. Comput Biol Med. 2014 Aug; 51:140-58. doi: 10.1016/j.compbiomed.2014.04.010.\u003c/li\u003e\n \u003cli\u003eIddi S, Li D, Aisen PS, Rafii MS, Thompson WK, Donohue MC; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Predicting the course of Alzheimer\u0026apos;s progression. Brain Inform. 2019 Jun 28; 6(1):6. doi: 10.1186/s40708-019-0099-0.\u003c/li\u003e\n \u003cli\u003eBeheshti I, Demirel H, Matsuda H; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Classification of Alzheimer\u0026apos;s disease and prediction of mild cognitive impairment-to-Alzheimer\u0026apos;s conversion from structural magnetic resource imaging using feature ranking and a genetic algorithm. Comput Biol Med. 2017 Apr 1; 83:109-119. doi: 10.1016/j.compbiomed.2017.02.011.\u003c/li\u003e\n \u003cli\u003eZheng C, Xia Y, Pan Y, Chen J. Automated identification of dementia using medical imaging: a survey from a pattern classification perspective. Brain Inform. 2016 Mar; 3(1):17-27. doi: 10.1007/s40708-015-0027-x.\u003c/li\u003e\n \u003cli\u003eZhang R, Giancardo L, Pena D, Kim Y, Tong H, Jiang X; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. From Brain Imaging to Graph Analysis: a study on ADNI\u0026apos;s patient cohort. arXiv 2019; arXiv1905.05861v1. doi: 10.48550/ARXIV.1905.05861\u003c/li\u003e\n \u003cli\u003eWestman E, Muehlboeck JS, Simmons A. Combining MRI and CSF measures for classification of Alzheimer\u0026apos;s disease and prediction of mild cognitive impairment conversion. Neuroimage. 2012 Aug 1; 62(1):229-38. doi: 10.1016/j.neuroimage.2012.04.056.\u003c/li\u003e\n \u003cli\u003eNozadi SH, Kadoury S, The Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Classification of Alzheimer\u0026apos;s and MCI Patients from Semantically Parcelled PET Images: A Comparison between AV45 and FDG-PET. Int J Biomed Imaging. 2018 Mar 15; 2018:1247430. doi: 10.1155/2018/1247430.\u003c/li\u003e\n \u003cli\u003eRice L, Bisdas S. The diagnostic value of FDG and amyloid PET in Alzheimer\u0026apos;s disease-A systematic review. Eur J Radiol. 2017 Sep; 94:16-24. doi: 10.1016/j.ejrad.2017.07.014.\u003c/li\u003e\n \u003cli\u003eLee JC, Kim SJ, Hong S, Kim Y. Diagnosis of Alzheimer\u0026apos;s disease utilizing amyloid and tau as fluid biomarkers. Exp Mol Med. 2019 May 9; 51(5):1-10. doi: 10.1038/s12276-019-0250-2.\u003c/li\u003e\n \u003cli\u003eBecerra-Laparra I, Cortez-Conradis D, Garcia-Lazaro HG, Martinez-Lopez M, Roldan-Valadez E. Radial diffusivity is the best global biomarker able to discriminate healthy elders, mild cognitive impairment, and Alzheimer\u0026apos;s disease: A diagnostic study of DTI-derived data. Neurol India. 2020 Mar-Apr; 68(2):427-434. doi: 10.4103/0028-3886.284376.\u003c/li\u003e\n \u003cli\u003eChandra A, Dervenoulas G, Politis M; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Magnetic resonance imaging in Alzheimer\u0026apos;s disease and mild cognitive impairment. J Neurol. 2019 Jun; 266(6):1293-1302. doi: 10.1007/s00415-018-9016-3.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eHansson O, Seibyl J, Stomrud E, Zetterberg H, Trojanowski JQ, Bittner T, Lifke V, Corradini V, et al.; Swedish BioFINDER study group; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. CSF biomarkers of Alzheimer\u0026apos;s disease concord with amyloid-\u0026beta; PET and predict clinical progression: A study of fully automated immunoassays in BioFINDER and ADNI cohorts. Alzheimers Dement. 2018 Nov; 14(11):1470-1481. doi: 10.1016/j.jalz.2018.01.010.\u003c/li\u003e\n \u003cli\u003eWolz R, Julkunen V, Koikkalainen J, Niskanen E, Zhang DP, Rueckert D, Soininen H, L\u0026ouml;tj\u0026ouml;nen J; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Multi-method analysis of MRI images in early diagnostics of Alzheimer\u0026apos;s disease. PLoS One. 2011; 6(10):e25446. doi: 10.1371/journal.pone.0025446.\u003c/li\u003e\n \u003cli\u003eFolstein MF, Folstein SE, McHugh PR. \u0026quot;Mini-mental state\u0026quot;. A practical method for grading the cognitive state of patients for the clinician. J Psychiatr Res. 1975 Nov; 12(3):189-98. doi: 10.1016/0022-3956(75)90026-6.\u003c/li\u003e\n \u003cli\u003eFrisoni GB, Fox NC, Jack CR Jr, Scheltens P, Thompson PM. The clinical use of structural MRI in Alzheimer disease. Nat Rev Neurol. 2010 Feb; 6(2):67-77. doi: 10.1038/nrneurol.2009.215.\u003c/li\u003e\n \u003cli\u003eZetterberg H, Bendlin BB. Biomarkers for Alzheimer\u0026apos;s disease-preparing for a new era of disease-modifying therapies. Mol Psychiatry. 2021 Jan; 26(1):296-308. doi: 10.1038/s41380-020-0721-9.\u003c/li\u003e\n \u003cli\u003eBrueggen K, Grothe MJ, Dyrba M, Fellgiebel A, Fischer F, Filippi M, et al. The European DTI Study on Dementia - A multicenter DTI and MRI study on Alzheimer\u0026apos;s disease and Mild Cognitive Impairment. Neuroimage. 2017 Jan; 144(Pt.B):305-308. doi: 10.1016/j.neuroimage.2016.03.067.\u003c/li\u003e\n \u003cli\u003eWyman BT, Harvey DJ, Crawford K, Bernstein MA, Carmichael O, Cole PE, et al.; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Standardization of analysis sets for reporting results from ADNI MRI data. Alzheimers Dement. 2013 May; 9(3):332-7. doi: 10.1016/j.jalz.2012.06.004.\u0026nbsp;\u003c/li\u003e\n \u003cli\u003eStone M. Cross-Validatory Choice and Assessment of Statistical Predictions. J. Royal Statistical Soc. Series B; 1974:36:111-147.\u003c/li\u003e\n \u003cli\u003eTustison NJ, Avants BB, Cook PA, Zheng Y, Egan A, Yushkevich PA, et al. N4ITK: improved N3 bias correction. IEEE Trans Med Imaging. 2010 Jun; 29(6):1310-20. doi: 10.1109/TMI.2010.2046908.\u003c/li\u003e\n \u003cli\u003ePenny WD, Ashburner J, Kiebel S, Henson R, Glaser DE, Phillips C, et al. Statistical Parametric Mapping: An Annotated Bibliography. Wellcome Department of Cognitive Neurology, University College London; 2001. https://www.fil.ion.ucl.ac.uk/spm/doc/spmbib.pdf\u003c/li\u003e\n \u003cli\u003eCohen J. Eta-Squared and Partial Eta-Squared in Fixed Factor Anova Designs. Educational and Psychological Measurement. 1973; 33(1):107-112. doi: 10.1177/001316447303300111.\u003c/li\u003e\n \u003cli\u003eEllis PD. The Essential Guide to Effect Sizes: Statistical Power, Meta-Analysis, and the Interpretation of Research Results. Cambridge University Press; 2010:12.\u003c/li\u003e\n \u003cli\u003eBinczyk F, Stjelties B, Weber C, Goetz M, Maier-Hein K, Meinzer HP, et al. MiMSeg - an algorithm for automated detection of tumor tissue on NMR apparent diffusion coefficient maps. Information Sciences, 2017; 384:235-248. doi:10.1016/j.ins.2016.07.052.\u003c/li\u003e\n \u003cli\u003eKempton MJ, Underwood TS, Brunton S, Stylios F, Schmechtig A, Ettinger U, et al. A comprehensive testing protocol for MRI neuroanatomical segmentation techniques: Evaluation of a novel lateral ventricle segmentation method. Neuroimage. 2011 Oct 15; 58(4):1051-9. doi: 10.1016/j.neuroimage.2011.06.080.\u003c/li\u003e\n \u003cli\u003eWit E, Heuvel EVD, Romeyn JW. \u0026apos;All models are wrong...\u0026apos;: an introduction to model uncertainty. Statistica Neerlandica; 2012: 66 (3): 217\u0026ndash;236.\u003c/li\u003e\n \u003cli\u003eAgresti A. On logit confidence intervals for the odds ratio with small samples. Biometrics. 1999 Jun; 55(2):597-602. doi: 10.1111/j.0006-341x.1999.00597.x.\u003c/li\u003e\n \u003cli\u003eGarc\u0026iacute;a Barrado L, Coart E, Burzykowski T; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Development of a diagnostic test based on multiple continuous biomarkers with an imperfect reference test. Stat Med. 2016 Feb 20; 35(4):595-608. doi: 10.1002/sim.6733.\u003c/li\u003e\n \u003cli\u003eEskildsen SF, Coup\u0026eacute; P, Garc\u0026iacute;a-Lorenzo D, Fonov V, Pruessner JC, Collins DL; Alzheimer\u0026apos;s Disease Neuroimaging Initiative. Prediction of Alzheimer\u0026apos;s disease in subjects with mild cognitive impairment from the ADNI cohort using patterns of cortical thinning. Neuroimage. 2013 Jan 15; 65:511-21. doi: 10.1016/j.neuroimage.2012.09.058.\u0026nbsp;\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Alzheimer's disease, Mild Cognitive Impairment, MRI, multinomial logistic regression","lastPublishedDoi":"10.21203/rs.3.rs-1747001/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-1747001/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBACKGROUND\u003c/h2\u003e \u003cp\u003eThis work aimed to find MRI-based markers for Alzheimer's disease (AD) and mild cognitive impairment (MCI) to improve diagnosis.\u003c/p\u003e\u003ch2\u003eMETHODS\u003c/h2\u003e \u003cp\u003eThe multinomial logistic regression was used to predict diagnosis status: AD, MCI, and normal control (NC) combined with the Bayesian information criterion for model selection. Several T1-weighted MRI-based radiomic features were considered as explanatory variables in the prediction model.\u003c/p\u003e\u003ch2\u003eRESULTS\u003c/h2\u003e \u003cp\u003eThe best radiomic predictor was the relative brain volume defined as the ratio between the volume of the brain without cerebrospinal fluid and the volume of the whole brain multiplied by 100%. The model was trained on the ADNI dataset and tested on the independent EDSD dataset. The proposed method confirmed its quality by achieving a balanced accuracy of 95.18%, AUC of 93.25%, NPV of 97.93%, and PPV of 90.48% for classifying AD vs NC for the EDSD. The comparison of two models: with the MMSE score only as an independent variable, and corrected for the relative brain value and age, shows that the addition of an MRI-based biomarker improves the quality of MCI detection (AUC: 67.04% vs 71.08%) while maintaining quality for AD (AUC: 93.35% vs 93.25%). Additionally, among MCI patients predicted as AD inconsistently with original diagnosis, 56.25% from ADNI and 54.17% from EDSD were re-diagnosed as AD within a 48-month follow-up. It shows that our model can detect AD patients a few years earlier than a standard medical diagnosis.\u003c/p\u003e\u003ch2\u003eCONCLUSIONS\u003c/h2\u003e \u003cp\u003eThe created method is non-invasive, inexpensive, clinically accessible, and efficiently supports the AD/MCI diagnosis.\u003c/p\u003e","manuscriptTitle":"Can we improve prediction of Alzheimer's disease and Mild Cognitive Impairment by combining MMSE score and MRI-based imaging data?","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2022-06-21 16:02:47","doi":"10.21203/rs.3.rs-1747001/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"de37940c-e749-45c2-be2b-7b535009369c","owner":[],"postedDate":"June 21st, 2022","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2022-06-21T16:02:48+00:00","versionOfRecord":[],"versionCreatedAt":"2022-06-21 16:02:47","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-1747001","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-1747001","identity":"rs-1747001","version":["v1"]},"buildId":"_2-kVJe1T_tPrBINL-cwx","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}