SNeurodCNN: Structure-focused Neurodegeneration Convolutional Neural Network for Modeling and Classification of Alzheimer’s Disease

SNeurodCNN: Structure-focused Neurodegeneration Convolutional Neural Network for Modeling and Classification of Alzheimer’s Disease | 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 Article SNeurodCNN: Structure-focused Neurodegeneration Convolutional Neural Network for Modeling and Classification of Alzheimer’s Disease Simisola Odimayo, Chollette C. Olisah, Khadija Mohammed This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-3951099/v1 This work is licensed under a CC BY 4.0 License Status: Published Journal Publication published 02 Jul, 2024 Read the published version in Scientific Reports → Version 1 posted 8 You are reading this latest preprint version Abstract Alzheimer’s disease (AD), the predominant form of dementia, is a growing global challenge, emphasizing the urgent need for accurate and early diagnosis. Current clinical diagnoses rely on radiologist expert interpretation, which is prone to human error. Machine learning has thus far shown promise for early AD diagnosis. However, existing methods often overlook focal structural features critical for understanding cerebral cortex neurodegeneration. This paper proposes a machine learning framework that includes a novel structure-focused neurodegeneration CNN architecture named SNeurodCNN and an image brightness enhancement preprocessor using gamma correction. Leveraging mid-sagittal and para-sagittal brain image viewpoints from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) dataset, our framework demonstrated exceptional performance. The para-sagittal viewpoint achieved 97.8% accuracy, 97.0% specificity, and 98.5% sensitivity, while the mid-sagittal viewpoint offered deeper insights with 98.1% accuracy, 97.2% specificity, and 99.0% sensitivity. Model analysis revealed the ability of SNeurodCNN to capture the structural dynamics of mild cognitive impairment (MCI) and AD in the frontal lobe, occipital lobe, cerebellum, temporal, and parietal lobe, suggesting its potential as a brain structural change digi-biomarker for early AD diagnosis. Health sciences/Diseases Health sciences/Diseases/Neurological disorders Health sciences/Diseases/Neurological disorders/Neurodegeneration Biological sciences/Computational biology and bioinformatics Biological sciences/Computational biology and bioinformatics/Computational models Alzheimer’s disease mild cognitive impairment classification deep learning convolutional neural network Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Introduction Alzheimer’s disease (AD) is a specific type of dementia associated with severe neurological deficits that affect cognitive, visual, sensory, and motor functions in people living with the disease 1 . With AD, neurodegeneration, a progressive loss of structure or function of neurons, is inevitable, and there is currently no cure for reversing this process. However, clinical studies have shown that neurodegeneration progresses. With early diagnosis, treatment, and therapeutic interventions, the process can be slowed. At present, a definitive diagnosis of AD remains a complex task because tests for the presence of amyloid plaques and phosphorylated tau are the true determinants of AD and can mainly be performed posthumously 2 . Other clinical practices depend on a multitude of evaluations, including clinical assessments, medical history reviews, cognitive assessments, and neuroimaging, and many years of study are needed to reach a diagnostic decision 3 . Neuroimaging methods, such as positron emission tomography (PET) and MRI, provide information on the extent of structural changes in the brain relevant for pathological alterations characteristic of the brain during degeneration 4 , particularly MRI. MRIs reveal several broad viewpoints, such as axial, coronal, and sagittal, with different levels of information for analysing brain neurodegeneration. Notably, the axial view revealed substantial atrophy of the cerebral cortex, leading to shrinkage of the outer layer of the cerebrum. This atrophy is accompanied by ventricle enlargement, reduced brain volume, and diminished gray matter 5 . In contrast, the coronal view highlights ventricle enlargement and emphasizes significant temporal lobe and cortical atrophy. This is a window into the widespread loss of neurons throughout the brain, accompanied by sulcus widening and gyrus thinning 5 . The sagittal plane provides the most visible information for AD diagnosis 7 . Brain neurodegeneration is evident in the sagittal plane, particularly in the frontal lobe, cerebellum, occipital lobe, thalamus, and corpus callosum, where learning and memory, mental function, motor function, and sensory function 7 can be significantly impacted. Despite the diagnostic potential of MRI, its sole reliance on early AD diagnosis faces numerous limitations 9 . For example, AD may elude visual detection, especially when numerous samples of MCI and AD patients are analysed, thus necessitating the need for a comprehensive clinical evaluation methodology that is reliable. Additionally, the interpretability of MRI scans varies among radiologists and clinicians, which can introduce inconsistency in diagnosis. However, the emerging techniques in machine learning offer promise in diagnosing AD, particularly in timeliness, thereby paving the way for more effective AD management and intervention. Over the past decade, deep learning algorithms, including both pretrained networks and tailored architectures, have been successfully adopted for AD modelling. Pretrained networks have long-standing relevance in AD diagnostic research. Bae tailored a residual network-50 (ResNet50) 10 for discriminating between MCI and AD patients and achieved an accuracy of 82.4%. The GoogLeNet, AlexNet, and ResNet-18 pretrained networks were exploited for classifying patients into cognitively normal, early mild cognitive impairment, mild cognitive impairment, and late mild cognitive impairment categories 11 . With accuracies of 96.39%, 94.08%, and 97.51% for GoogLeNet, AlexNet, and ResNet-18, respectively, ResNet-18 outperforms the other models in terms of performance. By integrating a 3D mobile inverted bottleneck convolution (MBConv) block in a 3D EfficientNet architecture 12 , accuracy, sensitivity, specificity, and AUC values of 86.67%, 75.00%, 90.91%, 97.16%, and 83.33%, respectively, were achieved for the sMCI and pMCI sets. In another work, the DenseNet-169 and ResNet-50 CNN architectures were exploited for early AD diagnosis 13 . DenseNet-169 exhibited superior accuracy, surpassing ResNet-50, with scores ranging between 97.7% and 88.7%. The ResNet-18 pretrained network was useful for AD classification 14 . With the use of the Mish activation function (MAF) for enhancing the model's learning adaptability and a weighted cross-entropy loss function to ensure equitable consideration of the AD, MCI, and CN classes, the network achieved 88.3% accuracy on the preprocessed ADNI dataset. Tailored deep learning algorithms are now paving the way for AD diagnosis. Basaiaa et al . 15 proposed a 3D CNN consisting of 2 convolutional blocks of 5 × 5 × 5 filter sizes and 10 blocks of 3 × 3 × 3 filter sizes. They utilized strides in place of max-pooling for downsampling. Their work achieved 74.8%, 75.1%, and 75.3% accuracy, sensitivity, and specificity, respectively, on the ADNI stable (s-MCI) and MCI conversion (c-MCI) sets. The model achieved 85.9%, 83.6%, and 88.3% accuracy, sensitivity, and specificity, respectively, on the AD and s-MCI sets. Another study proposed a CNN for AD diagnosis and stratification 16 . This research not only facilitated fast and accurate AD diagnosis but also offered classification for normal, MCI, and AD patients. Additionally, we addressed the challenging task of stratifying MCI into very mild dementia (VMD), mild dementia (MD), and moderate dementia (MoD) stages, akin to prodromal AD. The shallow network 16 achieved an overall testing accuracy of 99.68%, which was greater than that of pretrained networks such as DenseNet121, ResNet50, VGG 16, EfficientNetB7, and InceptionV3. Although the dataset used was the Open Access Series of Imaging Studies (OASIS) dataset, it is important to note that their work shows the importance of custom-trained networks in AD diagnosis. A fine-tuned CNN classifier 17 called AlzheimerNet was shown to be capable of classifying Alzheimer's disease into five stages. With data preprocessing and augmentation, their method achieved 98.67% accuracy using the RMSProp optimizer. Considering the different patient groups used for diagnosing AD, 18 independent binary classifications were proposed: healthy control (HC) vs. AD, HC vs. pMCI, HC vs. sMCI, pMCI vs. AD, sMCI vs. AD, and sMCI vs. pMCI according to the deep belief network (DBN). The modifications to the DBN 18 include dropout and zero-masking for overcoming overfitting, a preprocessing algorithm, a principal component analysis for dimensionality reduction, and a multitask feature selection approach. Using the ADNI dataset, accuracies ranging from 87.78–99.62% were observed. Hazarikar et al . 19 replaced the downsampling layer in the traditional LeNet architecture with a fusion of min-pooling and max-pooling layers to retain both minimum-value and maximum-valued signals. Their model achieved an accuracy, precision, recall, and F1-score of 98%, 96%, 97%, and 98%, respectively, on the ADNI dataset. In another work, a VGG-TSwinformer architecture 20 that combines a VGG-16 convolutional neural network and a transformer network was proposed and validated on the ADNI sMCI and pMCI cohorts. The accuracy, sensitivity, specificity, and AUC were 77.2%, 79.97%, 71.59%, and 0.8153, respectively. Similarly, another work 21 also found architecture useful for their methodology. They created a hybrid architecture by combining AlexNet with LeNet and varying the filter sizes from 1 × 1, 3 × 3, and 5 × 5. Scores as high as 96%, 93%, 93%, and 96% for accuracy, precision, recall, and F1 score, respectively, were reported. Another notable architecture is the multiplane convolutional neural network (Mp-CNN) architecture, which simultaneously processes three planes, axial, coronal, and sagittal, of 3D MRI 22 . The architecture of the Mp-CNN comprises 14 layers with rectified linear unit (ReLU) activation and softmax for multiclass classification, and it outperforms traditional 2D CNNs in multiclass classification associated with AD, MCI, and NC. The Swinformer has also been explored 23 as a transformer-based CNN architecture for AD classification. The Swinformer combines a CNN module for planar feature extraction and a transformer encoder module for 3D semantic connections. They argued that Swinformer can capture local features more accurately. The pipeline included data preprocessing and augmentation strategies such as random rotation and mirror reflection and recorded an accuracy of 88.3%. While it is obvious that transfer learning with the use of state-of-the-art pretrained models is a promising technique for diagnosing AD, the tailored deep learning algorithm outperforms traditional methods in terms of performance and shows that it is better at preserving the underlying structure of the data for diagnosing AD. However, of these works, none have captured the structural dynamics of neurodegeneration in the brain in individuals with MCI or AD, which leaves room for additional work to be done. Therefore, this paper seeks to bridge this research gap and makes the following contributions to AD research. We propose a machine learning framework that integrates the novel SNeurodCNN architecture for modelling the structural neurodegeneration of the brain's cerebral cortex and for the task of discriminating between MCI and AD. Investigate whether the varying viewpoints of the two planes of the sagittal axis, midsagittal and parasagittal, provide differing insights into structural neurodegeneration. Investigating SNeurodCNN sensitives patients to brain neurodegeneration is relevant for identifying digital biomarkers (digi-biomarkers) for the regions of the brain where structural neurodegeneration is prevalent. Results This section presents the results and findings of this paper. We begin by outlining the findings of SNeurodCNN and then progress to its analysis using GradCAM to identify features strongly indicative of the model’s sensitivities to brain neurodegeneration. Evaluation of the Structure-focused ADNI Dataset Experiments on the structure-focused ADNI data version of the brain are relevant for understanding the structural changes that are contributors to brain neurodegeneration. The performance of the SNeurodCNN was evaluated on the midsagittal and parasagittal planes of the sagittal axis of the brain. The parasagittal view of the sagittal plane is the version taken from the off-centre plane, while the midsagittal view is the version taken at exactly the centre of the plane. The use of both planes is to investigate whether the model is impacted differently by neurodegeneration, as it pertains to the parasagittal region, which exposes regions of motor control, sensation, and perception, and the midsagittal region, which exposes regions that are responsible for visuospatial integration, memory, and self-awareness. The results of this experiment are presented in Table 1 and Fig. 1. Meanwhile, we demonstrate in Fig. 2 which displays the training and validation loss and accuracy performance of SNeurodCNN with training sets from the midsagittal and parasagittal viewpoints. The loss curves generated in Fig. 3 (a) and (b) demonstrate an exceptional trend in the training progress for both the parasagittal and midsagittal regions of the brain. As the number of epochs increases, there is a smooth and consistent descent in the loss curves, indicating effective model training. The convergence of training and validation losses to stable values, with a plateau at higher loss levels, signifies successful learning without succumbing to overfitting. Additionally, the parallel curves of both validation and training further attests that SNeurodCNN is learning from the training data but also demonstrates a robust ability to generalize to previously unseen data and simultaneously identify intricate pattens in the data while avoiding the pitfalls of overfitting. Furthermore, akin to the loss curves generated by the model, the accuracy curves depicted in Fig. 2 not only demonstrate the model's adeptness in making predictions for both datasets but also showcase its adaptability to the training data, indicating a continuous enhancement in predictive accuracy. As the epochs increase, the upward trajectory of the axis curve for both experiments corresponds to the decrease in loss. Although peak accuracy is reached relatively early on for both the midsagittal and parasagittal datasets, this does not compromise the SNeurodCNN performance as the close alignment between training and validation indicates good generalization without overfitting. Table 1: Performance evaluation of SNeurodCNN classification model in the midsagittal and parasagittal planes Metric Midsagittal Plane Parasagittal Plane Accuracy 98.1% 97.8% Precision 97.2% 97.2% Sensitivity 99.0% 98.5% Specificity 97.2% 97.0% F1-score 98.1% 97.8% AUC 98.1% 97.8% As shown in Table 1 and Fig. 1, the accuracy, precision, recall, specificity, F1 score, and AUC of the SNeurodCNN are outstanding. This is an indication that the model is capable of modelling the structural neurogenerative impacts of MCI and AD occurring at both parts of the sagittal plane—the midsagittal and parasagittal planes. While the accuracies recorded are high, which shows the model’s ability to distinguish between instances of AD and instances of MCI, the midsagittal accuracy was increased by 0.3%. Given that the structure-focused ADNI is used, the performance conforms with the medical statement “midsagittal cerebral morphology provides a homologous geometrical reference for brain shape and cortical vs. subcortical spatial relationships” 24 . In medical diagnosis, a high recall is preferred—the higher the value is, the greater the confidence we have in the model’s ability to minimize false positives, that is, diagnosing AD when in fact it is MCI and vice versa. Furthermore, the high F1-scores for both models reflect well-balanced performance in terms of precision and recall, ensuring a lower rate of both false-negative and false-positive predictions. As an additional measure of our model’s performance, we utilized the AUC metric, which is a metric that provides an unbiased view of a model's performance amidst class imbalances aside from the F1 score since the data presented are 180:105 in proportion for AD patients and MCI patients. Scores of 98.1% and 97.8% were observed for the midsagittal and parasagittal sides, respectively, which underscores the models' discriminatory capability in diagnosing AD. SNeurodCNN Sensitivity to Brain Neurodegeneration To understand the sensitivity of SNeurodCNN to brain neurodegeneration, the same slices from the MCI and AD categories for the midsagittal and parasagittal planes were sampled. Then, with Grad-CAM, the activation maps of the last convolutional layer of SNeurodCNN are visualized to better understand the network’s sensitivity to brain degeneration. The Grad-CAM output was analysed using a heatmap; the brighter the yellow, the more significant the region is, and the more prominent the region is where brain neurodegeneration is. The closer to purple it is, the less significant the difference is. These regions can be visualised in Fig. 4, with more examples for the midsagittal in Fig. 5 and parasagittal in Fig. 6. As can be observed from the MCI (sMCI and pMCI) and AD slices in Fig. 5, the frontal lobe, occipital lobe, and cerebellum are the regions highlighted as being highly significant, with the frontal lobe being the most prominent while the cerebellum the least prominent. Therefore, these regions reveal SNeurodCNN sensitivity to brain neurodegeneration in relation to AD. The findings made with the midsagittal plane are consistent with those made with the parasagittal plane and can be observed in Fig. 6, although the parasagittal plane shows additional prominent regions about the locations of the temporal and parietal lobe. Since we used the structure-focused ADNI, it is only expected that the highlights relate to changes in structure caused by shrinkage of the cerebral cortex, known as cerebral atrophy. An obvious interpretation of these observations is that the SNeurodCNN can sense a significant difference in the structural characteristics of the frontal lobe, occipital lobe, and cerebellum and possibly the parietal regions in individuals with MCI (sMCI and pMCI) and AD. Insights into the functions of the frontal lobe, occipital lobe, cerebellum, temporal, and parietal lobe are important for understanding the roles these regions play in brain neurodegeneration in relation to AD. The frontal lobe is mainly responsible for motor action and the temporal integration of behavior 25 . In a recent clinical study 26 , the frontal lobes of patients with MCI were shown to experience cortical atrophy, but severe cortical volume deficit was evident in patients with AD. The region associated with visual processing for depth perception, colour determination, object, and face recognition and responsible for memory formation is the occipital lobe 27 . According to clinical research, structural changes are evident in the occipital lobe and are an indication of early progressive MCI 28 . A clinical study 29 revealed that there was a degree of functional connectivity disruption in the cerebellum of MCI and AD patients. In this paper, our proposed SNeurodCNN showed that its impact is subtle compared to that of the others highlighted. The cerebellum is responsible for regulating motor movement and controlling balance 30 . The additional highlight from the parasagittal viewpoint, the temporal and parietal lobe, might significantly indicate the impact that neurodegeneration can have on the sensory functions of the brain. The parietal region has not been computationally confirmed in DL-based studies, but a clinical study 31 has shown that the parietal gray matter volumes distinguish between MCI patients and AD patients. Memory loss, to a considerable extent, affects MCI and AD patients, with the former being less pronounced than the latter 32,33 . The temporal lobe that is responsible for memory loss is visible in the parasagittal axis of the brain. As has been presented, there is a clinical corroboration to the brain regions our model showed sensitivity to through the midsagittal and parasagittal axis of the brain. Comparison with the state-of-the-art methods We believe that these studies aimed at discriminating between AD and MCI patients contrast them with the SNeurodCNN model. The comparisons are summarized in Table 2. Basaiaa et al . 15 and the DBN model 18 offer general approaches to diagnosing AD and MCI but with less emphasis on neurodegeneration structures than does SNeurodCNN. DenseNet-169 and ResNet-50 CNN models 13 focus on early AD diagnosis with deep architectures, which are potentially less specialized in structural neurodegeneration. The Mp-CNN model 22 offers a unique approach through advanced multiplanar image processing but differs from the focused technique of SNeurodCNN. Finally, the Hybrid LeNet-AlexNet Model 21 combines various network elements but diverges from SNeurodCNN, which focuses on neurodegeneration. While each model has distinctive strengths for diagnosing AD and MCI, SNeurodCNN stands out for its specific focus on the regions where structural neurodegeneration occurs, making it a significant tool in the field. Table 2: Comparison of the SNeurodCNN model with other state-of-the-art models S/N Model Year Performance Metric Architecture Diagnostic Approach Digi-Biomarker 1 Basaia et al .'s 3D CNN Model 15 2019 74.8% - 88.3% accuracy 2 blocks of 5x5x5 and 10 blocks of 3x3x3 filters, use strides for downsampling AD and MCI diagnosis - 2 Deep Belief Network (DBN) Model 18 2023 87.78% - 99.62% accuracy Dropout and zero-masking, multitask feature selection Classification tasks for AD and MCI - 3 DenseNet-169 and ResNet-50 CNN Models 16 2023 DenseNet-169: 97.7% (training), 83.82% (testing); ResNet-50: Lower performance Deep architectures with advanced feature learning Early AD diagnosis - 4 Multiplane Convolutional Neural Network (Mp-CNN) 22 2022 93% accuracy 14-layer architecture processing axial, coronal, sagittal planes of 3D MRI Advanced multiplane image processing - 5 The proposed 2024 SNeurodCNN 98.1% accuracy 97.8% accuracy Sagittal planes; CNN architecture: 2D Conv, 3x3 filter of varying depth, Maxpool, Dense layers, Dropout Classification tasks for AD and MCI Captures structural brain neurodegeneration about the frontal lobe, occipital lobe, cerebellum, and parietal lobe. Discussion We delve into the discussion to elaborate on the implications of the results of this study. Neurologists traditionally use neuroimaging methods, such as MRI, to assess MCI and AD neurodegeneration, but this paper presents a dimension that automates diagnosis toward classifying the disease stages but can help in identifying the regions susceptible to neurodegeneration. This finding substantiates the relevance of the proposed SNeurodCNN model in the early diagnosis of AD. The evaluation of neurodegeneration using the midsagittal and parasagittal planes of the brain results in remarkable performance. These findings surpass those of previous studies 7,10 . SNeurodCNN has shown promise in capturing neurodegenerative features from MR images. As observed with Grad-CAM, structural changes were prominent in the frontal lobe, occipital lobe, and cerebellum on the midsagittal MR images. Frontal lobe dysfunction has been implicated in various neurodegenerative conditions, including AD and MCI 34 . The parasagittal view showed that an additional structural change was possible in the parietal lobe, which was not obvious in the midsagittal slices. The SNeurodCNN model region is most sensitive to the frontal lobe, occipital lobe, cerebellum, and parietal lobe, which is consistent with the findings of clinical studies 25–31 and, in some cases, consistent with regions identified in one of the DL-based studies 7 ; however, our study further confirms the significance of cerebral atrophy on brain structure. Therefore, these findings show that our proposed SNeurodCNN can significantly contribute to AD diagnosis and prognosis. This necessitates the integration of deep learning into healthcare practices to provide early cues about AD that might be present in MCI patients, minimizing the occurrence of false positives and false negatives that can mitigate misdiagnosis. Furthermore, the structural changes that our proposed SNeurodCNN highlights set a record as a potential brain structural change digi-biomarker for the early diagnosis of AD. The benefits of digi-biomarkers are numerous. 1) These findings can help clinicians develop targeted and personalized treatment plans and interventions through personalized analysis of disease states. These findings could lead to valuable targets for therapeutic interventions and monitoring and the assessment of novel treatments accordingly. Materials and Method This section elucidates the dataset employed in this study and outlines the methodologies encompassing the data preprocessing pipeline and the design of the deep convolutional neural network architecture as illustrated in Fig. 7 . Data The ADNI AD benchmark dataset commonly used for AD analysis is relevant to this study. This dataset was designed from a study in which individuals aged 55 to 90 years were enrolled from 57 sites across the U.S. and Canada after providing informed consent. In all, the ADNI comprises 368 participants, categorized into 180 AD patients (82 females, 98 males; age ± SD = 75.28 ± 7.57 years) and 105 stable sMCI patients (41 females, 64 males; age ± SD = 74.69 ± 7.41 years). The average mini-mental state examination (MMSE) scores were 23 for AD patients, 28 for sMCI patients, and 27 for pMCI patients. All subjects had a T1-weighted baseline from the ADNI1/Go/2 cohort. The sMCI group consisted of individuals who were diagnosed with MCI at baseline and remained so for at least two years. A summary of the demographics of the ADNI participants is provided in Table 2 and the mental examination score of the participants in Fig. 8. We particularly explored skull-free patients version. This version utilized the ADNI dataset preprocessed to correct imaging distortion through gradwarping via gradient inhomogeneity correction , intensity correction, and scaling to address gradient drift. The results are further processed using multiatlas label propagation with expectation maximization (MALPEM) to segment, on local scales, the cross-sectional structural volume changes in the brain, which constitute the structure-focused ADNI dataset. It should be noted that each slice of the structure-focused ADNI dataset is characterized by the region of the gray matter that forms the anatomical structures of the MRI brain images and not the gray matter itself. Table 2 Demographics and clinical characteristics of the study population in the ADNI dataset Group Subject Age Gender M F AD 180 75.28 ± 7.57 98 82 sMCI 105 74.69 ± 7.41 64 41 pMCI 83 73.82 ± 6.65 49 34 Preprocessing Considering that structural changes in the brain and AD pathology are more prominent in the sagittal plane, their efficacy in AD classification has long been established 7,12 . This study explored two sagittal plane views, the midsagittal and parasagittal viewpoints, from the ADNI volumetric data in the sagittal plane. From the 3D structured focused ADNI dataset comprising 368 subjects with 155 slices, only slices showing full brain parts were retained; as such, the number of slices varied from sample to sample. In this study, we considered a different approach, which is an unusual practice in the literature where the disease case of a sample is described by all/fraction of their slices. The slices from the sMCI and pMCI patients were combined and referred to as the MCI set, while the slices from the AD samples were likewise combined into an AD set. We consider AD diagnosis to be a sample-independent task; therefore, it is only logical to teach the deep learning algorithm to capture the factors influencing the two categories, MCI and AD, to better understand the disparity that exists between them. In all, a total of 4228 midsagittal AD and MCI sets were retained, each consisting of 2160 and 2068 participants, respectively. The parasagittal AD and MCI sets included 2700 and 2820 participants, respectively, for a total of 5520 data points. The nonlinearity property of the Gamma function is found to be useful35 for expanding the brightness of the intensity of a poorly illuminated image and contributes greatly to the technique proposed 35 . Despite the intensity correction on the original ADNI dataset, the structure-focused ADNI dataset shows that images are captured under poor illumination conditions. Therefore, the gamma correction, an inverse of the gamma function, is adopted for expanding the brightness of pixel intensities. AD and MCI samples were subsequently split into training and testing sets at a ratio of 80:20. The testing set is further split into halves to accommodate the set for validation. The SNeurodCNN Architecture As previously stated, each structure-focused ADNI MRI slice is characterized by the formation of gray matter, which forms the anatomical structures of the MR brain images. This structural formation, as opposed to gray matter features, is used to capture focal structural changes associated with neurodegeneration in the cerebral cortex as opposed to focal fine-grain changes within anatomical structures. For this reason, the SNeurodCNN architecture is designed to be a deep convolutional neural network for modelling the structural neurodegeneration of the brain and discriminating between MCI and AD. The SNeurodCNN architecture is presented in Fig. 3 . It comprises two downsampling convolution blocks growing in a number of convolutional layers. The first block consists of a convolutional layer with a 3 \(\times 3\) filter size and 32 filter depths, varying image dimensions, a nonlinear activation function, a rectified linear unit (ReLU), and a 2 \(\times 2\) maxpooling layer. The second convolution block differs from the first by only the two convolutional layers, which are made up of 32 and 64 filter depths. This process feeds into the fully connected layer with 500 hidden neurons and a dropout layer, which is a regularizer that optimizes learning by randomly dropping nodes in the hidden neuron by a designated probability, which in our case is 50%. This is followed by the output layer, which is set to two and represents the two classes that the softmax activation function uses to classify an image as belonging to MCI or AD. Experimental settings The network's hyperparameters are chosen in ways that best optimize the network’s learning capability. For the hyperparameters, the Adam optimizer with a learning rate fixed at 0.0001 was used. The other variables are epochs and batch sizes, which are 100 and 32, respectively. We adopted early stopping, a regularizer that prevents overfitting by immediately halting training when the performance of the network does not improve after several epochs. The parameters of the early stopping agent used were patience set to 5 and restore_best_weights set to the true Boolean value. During training, the model learns through a method called backpropagation 36, which updates the weight of the network toward minimizing the gradient error. Considering the impact of hardware resources on model performance, the hardware configuration of the system used for training the network is reported as follows: A Keras library running on a Python-TensorFlow environment with a Google Collab GPU, Tesla K80 (T4). To evaluate the performance of the SNeurodCNN model, the following metrics were identified and found to be consistent with the metrics used in the literature for diagnosing AD. The variables used were the accuracy, precision, recall, specificity, F1-score, and area under the receiver operating characteristic (ROC) curve (AUC), which provide a comprehensive assessment of the model's performance. Their brief descriptions are as follows. 1. Accuracy: This metric measures the proportion of correct predictions over the total number of predictions, as expressed in ( 1 ). \(Accuracy =\frac{TP+TN}{TP+TN+FP+FN}\) ( 1 ) where TP = true positive, TN = true negative, FP = false positive and FN = false negative. 2. Precision: the proportion of true positives out of all the instances predicted as positive. In this case, precision measures the number of correctly classified AD cases out of all instances classified as ADs and is mathematically expressed as follows \(Precision= \frac{TP}{TP+FP}\) ( 2 ) 3. Recall: also known as sensitivity, this measures the proportion of correctly identified positive instances out of all the actual positive instances. In this case, the recall measures the proportion of correctly classified AD cases out of all actual AD cases. It is given as: \(Recall= \frac{TP}{TP+FN}\) ( 3 ) 4. Specificity: This metric quantifies the model’s ability to make true negative predictions out of all the correctly identified negative instances. With regard to the classification model, specificity will measure the proportion of correctly classified MCI cases out of all actual MCI cases. It is mathematically given as follows: \(Specificity= \frac{TN}{TN+FP}\) ( 4 ) 5. F1-score: The F1-score combines precision and recall into a single metric, providing a balanced measure of the model's performance. It considers both false positives and false negatives, making it valuable for overall performance assessment. \(F1=2 \times \frac{Precision\times Recall}{Precision+Recall}\) ( 5 ) 6. AUROC: The AUROC evaluates the model's performance across different classification thresholds, considering the trade-off between sensitivity and specificity. This approach provides insights into classifier discrimination ability and can be particularly useful for imbalanced datasets. Limitations The growing increase in AD cases underscores the pressing necessity of advancing research in deep learning. Our proposed SNeurodCNN achieves remarkable classification accuracies of 98.1% and 97.8% in both the midline and posterior sagittal planes, respectively. This level of performance surpasses that of many existing studies, highlighting the potential of deep learning, particularly our proposed model, in transforming the landscape of neurodegenerative disease diagnosis. The model’s sensitivities to neurodegeneration in the brain structure not only substantiated our model's efficacy but also deepened our comprehension of the intricate nature of neurodegenerative diseases in relation to MCI and AD, thereby opening doors for a better understanding of the disease. While our proposed SNeurodCNN model significantly contributes to AD diagnostic research, its findings are constrained by several limitations. The trainable parameters are large and computationally expensive, especially with respect to reliability and usefulness in real-time clinical diagnosis. Therefore, efforts to achieve high performance while minimizing the computational cost will be explored. Additionally, this paper is directed toward the structural function of the brain analysis, which led to the use of structure-focused ADNIs. However, it will be interesting to understand how the SNeurodCNN model performs on focal fine-grain features available with skull-free ADNIs. This approach enables us to better analyse which of the features, focal fine-grain or focal structure, are representative of the characteristics of neurodegeneration in the brain in a way that makes early diagnosis of AD feasible and reliable in clinical analysis. We expect that the outcome of the aforementioned study will lead to an understanding of the underlying neurobiological mechanisms involved in MCI and AD, which will help in studying AD progression. Conclusion In this paper, we propose a structure-focused neurodegeneration convolutional neural network (CNN) architecture called the SNeurodCNN, which was integrated into a machine learning framework along with preprocessing techniques for image enhancement and data preparation. The proposed framework leveraged the midsagittal and parasagittal brain image viewpoints of the structure-focused ADNI dataset. Through experiments, the proposed framework achieved 97.8% accuracy, with 97.0% specificity and 98.5% sensitivity on the parasagittal planes. On the midsagittal plane, an accuracy, specificity, and sensitivity of 98.1%, 97.2%, and 99.0%, respectively, were achieved. We further showed that the midsagittal lobe highlights the frontal lobe, occipital lobe, and cerebellum, while the parasagittal lobe extends to the parietal lobe as a region of the brain where structural dynamics are prominent due to MCI and AD. We believe this discovery useful for identifying digi-biomarkers for the early diagnosis of AD. In future work, efforts will be made to minimize the computational cost of the proposed model while achieving the same level of neurodegeneration modelling. Additionally, it will be interesting to apply the proposed model to focal fine-grained feature learning. This approach enables us to better analyse which of the features, focal fine-grain or focal structure, are representative of the characteristics of neurodegeneration in the brain in a way that makes early diagnosis of AD feasible and reliable in clinical analysis. Declarations Conflict of interest The authors declare that there are no conflicts of interest. Author Contribution Data collection, implementation, coding, validation, and testing, S.O.; Conceptualization, methodology, model design, and project supervision, C.C.O.; literature review, project scope, S.O., K.M, C.C.O.; result analysis and interpretation S.O, K.M, C.C.O; manuscript – original draft, C.C.O.,K.M, S.O; All authors reviewed, edited, and approved the final manuscript. Availability of Data The datasets analysed during the current study are available at the following link: https://doi.gin.g-node. org/10.12751/g-node.aa605a/ . References Alzheimer's Association, 2018. 2018 Alzheimer's disease facts and figures. Alzheimer's & Dementia , 14(3), pp.367–429. Gonzalez-Ortiz, F. et al. Plasma phospho-tau in Alzheimer’s disease: towards diagnostic and therapeutic trial applications. Mol Neurodegener 18, 18 (2023). Galvin, J. E. & Sadowsky, C. H. Practical Guidelines for the Recognition and Diagnosis of Dementia. The Journal of the American Board of Family Medicine 25, 367–382 (2012). Zeng, H.-M., Han, H.-B., Zhang, Q.-F. & Bai, H. Application of modern neuroimaging technology in the diagnosis and study of Alzheimer’s disease. Neural Regen Res 16, 73 (2021). Van de Mortel, L. A., Thomas, R. M. & van Wingen, G. A. Grey Matter Loss at Different Stages of Cognitive Decline: A Role for the Thalamus in Developing Alzheimer’s Disease. Journal of Alzheimer’s Disease 83, 705–720 (2021). Lin, H.-Y. et al. Differential Patterns of Gyral and Sulcal Morphological Changes During Normal Aging Process. Front Aging Neurosci 13, (2021). Hoang, G. M., Kim, U.-H. & Kim, J. G. Vision transformers for the prediction of mild cognitive impairment to Alzheimer’s disease progression using mid-sagittal sMRI. Front Aging Neurosci 15, (2023). Apostolova, L. G. et al. Hippocampal Atrophy and Ventricular Enlargement in Normal Aging, Mild Cognitive Impairment (MCI), and Alzheimer Disease. Alzheimer Dis Assoc Disord 26, 17–27 (2012). McEvoy, L. K. & Brewer, J. B. Quantitative structural MRI for early detection of Alzheimer’s disease. Expert Rev Neurother 10, 1675–1688 (2010). Bae, J. et al. Transfer learning for predicting conversion from mild cognitive impairment to dementia of Alzheimer’s type based on a three-dimensional convolutional neural network. Neurobiol Aging 99, 53–64 (2021). Shanmugam, J. V., Duraisamy, B., Simon, B. C. & Bhaskaran, P. Alzheimer’s disease classification using pretrained deep networks. Biomed Signal Process Control 71, 103217 (2022). Zheng, B. et al. A modified 3D EfficientNet for the classification of Alzheimer’s disease using structural magnetic resonance images. IET Image Process 17, 77–87 (2023). Al Shehri, W. Alzheimer’s disease diagnosis and classification using deep learning techniques. PeerJ Comput Sci 8, e1177 (2022). Oktavian, M.W., Yudistira, N. and Ridok, A., 2022. Classification of Alzheimer's Disease Using the Convolutional Neural Network (CNN) with Transfer Learning and Weighted Loss. arXiv preprint arXiv:2207.01584 . Basaia, S., et al. Alzheimer's Disease Neuroimaging Initiative, 2019. Automated classification of Alzheimer's disease and mild cognitive impairment using a single MRI and deep neural networks. NeuroImage: Clinical , 21 , p.101645. Marwa, E.G., Moustafa, H.E.D., Khalifa, F., Khater, H. & AbdElhalim, E., 2023. An MRI-based deep learning approach for accurate detection of Alzheimer’s disease. Alexandria Engineering Journal, 63 , pp.211–221. Shamrat, F. M. J. M. et al. AlzheimerNet: An Effective Deep Learning Based Proposition for Alzheimer’s Disease Stages Classification From Functional Brain Changes in Magnetic Resonance Images. IEEE Access 11, 16376–16395 (2023). Zeng, N., Li, H. & Peng, Y. A new deep belief network-based multitask learning for diagnosis of Alzheimer’s disease. Neural Comput Appl 35, 11599–11610 (2023). Hazarika, R.A., Abraham, A., Kandar, D. & Maji, A.K. An improved LeNet-deep neural network model for Alzheimer’s disease classification using brain magnetic resonance images. IEEE Access, 9 , pp.161194–161207, (2021). Hu, Z., Wang, Z., Jin, Y. and Hou, W., VGG-TSwinformer: Transformer-based deep learning model for early Alzheimer’s disease prediction. Computer Methods and Programs in Biomedicine, 229 , p.107291, (2023). Hazarika, R.A., Maji, A.K., Kandar, D., Jasinska, E., Krejci, P., Leonowicz, Z. and Jasinski, M., 2023. An Approach for Classification of Alzheimer’s Disease Using Deep Neural Network and Brain Magnetic Resonance Imaging (MRI). Electronics , 12 (3), p.676. Angkoso, C. V., Agustin Tjahyaningtijas, H. P., Purnama, I., & Purnomo, M. H. Multiplane Convolutional Neural Network (Mp-CNN) for Alzheimer’s Disease Classification. International Journal of Intelligent Engineering and Systems 15, (2022). Hu, Z., Li, Y., Wang, Z., Zhang, S. & Hou, W. Conv-Swinformer: Integration of CNN and shift window attention for Alzheimer’s disease classification. Comput Biol Med 164, 107304 (2023). Bruner, E., Martin-Loeches, M. and Colom, R. Human midsagittal brain shape variation: patterns, allometry and integration. Journal of Anatomy, 216 (5), pp.589–599, (2010). Hoffmann, M., 2013. The human frontal lobes and frontal network systems: an evolutionary, clinical, and treatment perspective. International Scholarly Research Notices , 2013 . Shi, C., Deng, H., Deng, X., Rao, D. and Yue, W., 2023. The Structural Changes of Frontal Subregions and Their Correlations with Cognitive Impairment in Patients with Alzheimer's Disease. Journal of Integrative Neuroscience , 22 (4), p.99. Rehman A, Al Khalili Y. Neuroanatomy, Occipital Lobe. In: StatPearls. StatPearls Publishing ,PMID: 31335040, (2023). Zhao, L., et al. & Alzheimer Disease Neuroimaging Initiative. Risk estimation before progression to mild cognitive impairment and Alzheimer’s disease: an AD resemblance atrophy index. Aging (Albany NY), 11 (16), p.6217, (2019). Tang, F., Zhu, D., Ma, W., Yao, Q., Li, Q. and Shi, J., 2021. Differences changes in cerebellar functional connectivity between mild cognitive impairment and Alzheimer's disease: a seed-based approach. Frontiers in Neurology , 12 , p.645171. D'Angelo, E., Physiology of the cerebellum. Handbook of clinical neurology, 154 , pp.85–108, (2018). Hänggi, J., Streffer, J., Jäncke, L. and Hock, C. Volumes of lateral temporal and parietal structures distinguish between healthy aging, mild cognitive impairment, and Alzheimer's disease. Journal of Alzheimer's Disease, 26 (4), pp.719–734, (2011). Chetelat, G., Desgranges, B., De La Sayette, V., Viader, F., Eustache, F. and Baron, J.C. Mapping gray matter loss with voxel-based morphometry in mild cognitive impairment. Neuroreport, 13 (15), pp.1939–1943, 2002. Saykin, A.J., Wishart, H.A., Rabin, L.A., Santulli, R.B., Flashman, L.A., West, J.D., McHugh, T.L. and Mamourian, A. Older adults with cognitive complaints show brain atrophy similar to that of amnestic MCI. Neurology, 67 (5), pp.834–842, 2006. Dubois, B., Picard, G. & Sarazin, M. Early detection of Alzheimer’s disease: new diagnostic criteria. Dialogues Clin Neurosci 11, 135–139 (2009). Chude-Olisah, C.C., Sulong, G., Chude-Okonkwo, U.A. and Hashim, S.Z., Illumination normalization for edge-based face recognition using the fusion of RGB normalization and gamma correction. In 2013 IEEE International Conference on Signal and Image Processing Applications , pp. 412–416, (2013). Miller, A.S., Blott, B.H. and Hames, T.K. Review of neural network applications in medical imaging and signal processing. Medical and Biological Engineering and Computing, 30 , pp.449–464, 1992. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Published Journal Publication published 02 Jul, 2024 Read the published version in Scientific Reports → Version 1 posted Editorial decision: Revision requested 24 Mar, 2024 Reviews received at journal 21 Mar, 2024 Reviewers agreed at journal 12 Mar, 2024 Reviewers invited by journal 12 Mar, 2024 Editor assigned by journal 12 Mar, 2024 Editor invited by journal 01 Mar, 2024 Submission checks completed at journal 01 Mar, 2024 First submitted to journal 12 Feb, 2024 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-3951099","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":276369645,"identity":"65277193-e4f1-4615-a341-623ff80ff0fd","order_by":0,"name":"Simisola Odimayo","email":"","orcid":"","institution":"University of the West of England","correspondingAuthor":false,"prefix":"","firstName":"Simisola","middleName":"","lastName":"Odimayo","suffix":""},{"id":276369646,"identity":"ec9ecd32-67cd-4d84-8da8-cf5163f44b29","order_by":1,"name":"Chollette C. Olisah","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABDUlEQVRIie2RMUvDQBTHXwhkCt76svQzPAk4xq9yR6BT4uIiWPSg0G66BvolBCGurxTiEnHN4BA/gQFBOoTipYXSIaG6dbgfHLzj3o/3/hyAxXKKIDhc70qHJYAU+yd/UAHTSdjVWyXQ/1LMkcRHFLGYMsub9k6Ay1y30VVYJec1TCKgknsV/Cgky5Iw0J5kNYuvL6okJChioDfdqxAmZDoJiX1TaFflVZojeAz03r/YTtkQXnaKbO/Vc5a+rGFzTNFmCnSKt1JPmObgzHhwMazGJksRBtlqm+VVZeXXN6qH2A8G4ossXtbNZCTEfLr8XLe36nE+Vk3zE43OStm/2R738CKHP9JisVgsf+AXS3NmxQmWn/QAAAAASUVORK5CYII=","orcid":"","institution":"University of the West of England","correspondingAuthor":true,"prefix":"","firstName":"Chollette","middleName":"C.","lastName":"Olisah","suffix":""},{"id":276369647,"identity":"d7b28e8f-81b2-43ce-a033-b70375f1af04","order_by":2,"name":"Khadija Mohammed","email":"","orcid":"","institution":"Baze University","correspondingAuthor":false,"prefix":"","firstName":"Khadija","middleName":"","lastName":"Mohammed","suffix":""}],"badges":[],"createdAt":"2024-02-12 13:31:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-3951099/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-3951099/v1","draftVersion":[],"editorialEvents":[{"content":"https://doi.org/10.1038/s41598-024-60611-8","type":"published","date":"2024-07-03T00:00:00+00:00"}],"editorialNote":"","failedWorkflow":false,"files":[{"id":52031424,"identity":"fde36f61-94c0-496a-9de0-e488e665c018","added_by":"auto","created_at":"2024-03-05 16:25:14","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":25996,"visible":true,"origin":"","legend":"\u003cp\u003eGraphical illustration of classification performance of SNeurodCNN model in the midsagittal and parasagittal planes\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/2a63f9a0ae902ed741995d1e.png"},{"id":52031798,"identity":"36ffb6fe-c13a-4856-8e07-ab590f8090cc","added_by":"auto","created_at":"2024-03-05 16:33:15","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":103027,"visible":true,"origin":"","legend":"\u003cp\u003ePlot of the training and validation loss and accuracy curves generated during the learning process. Graphs (a) and (c) depict the graphs achieved by using parasagittal region-focused images while (b) and (d) depict the graphs achieved by the midsagittal region.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/c5e777b73382870cc30d01c9.png"},{"id":52031425,"identity":"705c7715-441e-47f1-af8a-6d7bd0740326","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":57778,"visible":true,"origin":"","legend":"\u003cp\u003ePlot of the Receiver Operating Characteristic for Parasagittal dataset (a) and Midsagittal dataset (b). The ROC curves illustrate the discrimination of the classification model on the respective dataset while showcasing the model’s excellent trade-off between sensitivity and specificity with a steep ascent and high AUC scores. These high AUC values of 0.978 (97.8%) and 0.981 (98.1%) further demonstrate the model’s superior predictive accuracy in identifying AD and differentiating between MCI patients.\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/a1c1d6a50579677daaf8b95d.png"},{"id":52031430,"identity":"bf9e2625-23fb-45d8-b3df-e894ae05b6fc","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":640680,"visible":true,"origin":"","legend":"\u003cp\u003eRegions of brain neurodegeneration. (a) Midsagittal brain internal regions \u0026nbsp;\u0026nbsp;(Source: ‘Internal Brain Regions’ by Casey Henley, licensed under CC BY-NC-SA \u0026nbsp;\u0026nbsp;4.0 International License, (b) Highlights of SNeurodCNN neurodegeneration \u0026nbsp;\u0026nbsp;sensitivity in the midsagittal (up) and parasagittal (down).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/568b6d4caaeaf5c82c764209.png"},{"id":52031429,"identity":"65e714a2-0976-46ac-8ad2-2dfa5679057e","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":883313,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of regions of interest on the midsagittal plane identified by the CNN model in AD (a), pMCI (b) and sMCI patients (c) subjects.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/642399e40166c5116e52b4ab.png"},{"id":52031432,"identity":"7a1a1d6e-3753-43d6-b5ba-6043c3fe6348","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":834835,"visible":true,"origin":"","legend":"\u003cp\u003eVisualization of regions of interest on the parasagittal plane identified by the CNN model in AD patients (a), pMCI patients (b), and sMCI patients (c) subjects.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/6dd7573c277c348bab31c28d.png"},{"id":52031428,"identity":"40183bf2-72f0-46ea-846a-02c8bfae2bbb","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":184029,"visible":true,"origin":"","legend":"\u003cp\u003eOverview of the proposed machine learning framework for AD diagnosis.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/8b5c329e3fccff2af95b2a2d.png"},{"id":52031427,"identity":"61a89e00-5ba1-472a-9802-74a30160d22c","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":48922,"visible":true,"origin":"","legend":"\u003cp\u003eA chart of the average mini-mental state examination scores of participants across the disease states. The sMCI and pMCI have similar mini-mental scores, but are significantly different from AD.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/d6ec2235d4a6583a6ee2e8c3.png"},{"id":52031426,"identity":"73b3df07-c16e-4bd8-b5ab-ae1da0c1aa1c","added_by":"auto","created_at":"2024-03-05 16:25:15","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":70616,"visible":true,"origin":"","legend":"\u003cp\u003eFigure 3: Architecture of the SNeurodCNN model.\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/915c975d275edde447841702.png"},{"id":64096597,"identity":"701d8071-925f-44b1-8e50-bbd8fe143e0d","added_by":"auto","created_at":"2024-09-06 17:32:50","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2987833,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-3951099/v1/b03744df-27fe-4873-a0bb-6d3fc5f81ac3.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"SNeurodCNN: Structure-focused Neurodegeneration Convolutional Neural Network for Modeling and Classification of Alzheimer’s Disease","fulltext":[{"header":"Introduction","content":"\u003cp\u003eAlzheimer\u0026rsquo;s disease (AD) is a specific type of dementia associated with severe neurological deficits that affect cognitive, visual, sensory, and motor functions in people living with the disease\u003csup\u003e1\u003c/sup\u003e. With AD, neurodegeneration, a progressive loss of structure or function of neurons, is inevitable, and there is currently no cure for reversing this process. However, clinical studies have shown that neurodegeneration progresses. With early diagnosis, treatment, and therapeutic interventions, the process can be slowed. At present, a definitive diagnosis of AD remains a complex task because tests for the presence of amyloid plaques and phosphorylated tau are the true determinants of AD and can mainly be performed posthumously\u003csup\u003e2\u003c/sup\u003e. Other clinical practices depend on a multitude of evaluations, including clinical assessments, medical history reviews, cognitive assessments, and neuroimaging, and many years of study are needed to reach a diagnostic decision\u003csup\u003e3\u003c/sup\u003e. Neuroimaging methods, such as positron emission tomography (PET) and MRI, provide information on the extent of structural changes in the brain relevant for pathological alterations characteristic of the brain during degeneration\u003csup\u003e4\u003c/sup\u003e, particularly MRI. MRIs reveal several broad viewpoints, such as axial, coronal, and sagittal, with different levels of information for analysing brain neurodegeneration. Notably, the axial view revealed substantial atrophy of the cerebral cortex, leading to shrinkage of the outer layer of the cerebrum. This atrophy is accompanied by ventricle enlargement, reduced brain volume, and diminished gray matter\u003csup\u003e5\u003c/sup\u003e. In contrast, the coronal view highlights ventricle enlargement and emphasizes significant temporal lobe and cortical atrophy. This is a window into the widespread loss of neurons throughout the brain, accompanied by sulcus widening and gyrus thinning\u003csup\u003e5\u003c/sup\u003e. The sagittal plane provides the most visible information for AD diagnosis\u003csup\u003e7\u003c/sup\u003e. Brain neurodegeneration is evident in the sagittal plane, particularly in the frontal lobe, cerebellum, occipital lobe, thalamus, and corpus callosum, where learning and memory, mental function, motor function, and sensory function\u003csup\u003e7 can\u003c/sup\u003e be significantly impacted.\u003c/p\u003e \u003cp\u003eDespite the diagnostic potential of MRI, its sole reliance on early AD diagnosis faces numerous limitations\u003csup\u003e9\u003c/sup\u003e. For example, AD may elude visual detection, especially when numerous samples of MCI and AD patients are analysed, thus necessitating the need for a comprehensive clinical evaluation methodology that is reliable. Additionally, the interpretability of MRI scans varies among radiologists and clinicians, which can introduce inconsistency in diagnosis. However, the emerging techniques in machine learning offer promise in diagnosing AD, particularly in timeliness, thereby paving the way for more effective AD management and intervention.\u003c/p\u003e \u003cp\u003eOver the past decade, deep learning algorithms, including both pretrained networks and tailored architectures, have been successfully adopted for AD modelling. Pretrained networks have long-standing relevance in AD diagnostic research. Bae tailored a residual network-50 (ResNet50)\u003csup\u003e10\u003c/sup\u003e for discriminating between MCI and AD patients and achieved an accuracy of 82.4%. The GoogLeNet, AlexNet, and ResNet-18 pretrained networks were exploited for classifying patients into cognitively normal, early mild cognitive impairment, mild cognitive impairment, and late mild cognitive impairment categories\u003csup\u003e11\u003c/sup\u003e. With accuracies of 96.39%, 94.08%, and 97.51% for GoogLeNet, AlexNet, and ResNet-18, respectively, ResNet-18 outperforms the other models in terms of performance. By integrating a 3D mobile inverted bottleneck convolution (MBConv) block in a 3D EfficientNet architecture\u003csup\u003e12\u003c/sup\u003e, accuracy, sensitivity, specificity, and AUC values of 86.67%, 75.00%, 90.91%, 97.16%, and 83.33%, respectively, were achieved for the sMCI and pMCI sets. In another work, the DenseNet-169 and ResNet-50 CNN architectures were exploited for early AD diagnosis \u003csup\u003e13\u003c/sup\u003e. DenseNet-169 exhibited superior accuracy, surpassing ResNet-50, with scores ranging between 97.7% and 88.7%. The ResNet-18 pretrained network was useful for AD classification\u003csup\u003e14\u003c/sup\u003e. With the use of the Mish activation function (MAF) for enhancing the model's learning adaptability and a weighted cross-entropy loss function to ensure equitable consideration of the AD, MCI, and CN classes, the network achieved 88.3% accuracy on the preprocessed ADNI dataset.\u003c/p\u003e \u003cp\u003eTailored deep learning algorithms are now paving the way for AD diagnosis. Basaiaa \u003cem\u003eet al\u003c/em\u003e. \u003csup\u003e15\u003c/sup\u003e proposed a 3D CNN consisting of 2 convolutional blocks of 5 \u0026times; 5 \u0026times; 5 filter sizes and 10 blocks of 3 \u0026times; 3 \u0026times; 3 filter sizes. They utilized strides in place of max-pooling for downsampling. Their work achieved 74.8%, 75.1%, and 75.3% accuracy, sensitivity, and specificity, respectively, on the ADNI stable (s-MCI) and MCI conversion (c-MCI) sets. The model achieved 85.9%, 83.6%, and 88.3% accuracy, sensitivity, and specificity, respectively, on the AD and s-MCI sets. Another study proposed a CNN for AD diagnosis and stratification\u003csup\u003e16\u003c/sup\u003e. This research not only facilitated fast and accurate AD diagnosis but also offered classification for normal, MCI, and AD patients. Additionally, we addressed the challenging task of stratifying MCI into very mild dementia (VMD), mild dementia (MD), and moderate dementia (MoD) stages, akin to prodromal AD. The shallow network\u003csup\u003e16\u003c/sup\u003e achieved an overall testing accuracy of 99.68%, which was greater than that of pretrained networks such as DenseNet121, ResNet50, VGG 16, EfficientNetB7, and InceptionV3. Although the dataset used was the Open Access Series of Imaging Studies (OASIS) dataset, it is important to note that their work shows the importance of custom-trained networks in AD diagnosis. A fine-tuned CNN classifier \u003csup\u003e17\u003c/sup\u003e called AlzheimerNet was shown to be capable of classifying Alzheimer's disease into five stages. With data preprocessing and augmentation, their method achieved 98.67% accuracy using the RMSProp optimizer. Considering the different patient groups used for diagnosing AD, 18 independent binary classifications were proposed: healthy control (HC) vs. AD, HC vs. pMCI, HC vs. sMCI, pMCI vs. AD, sMCI vs. AD, and sMCI vs. pMCI according to the deep belief network (DBN). The modifications to the DBN\u003csup\u003e18\u003c/sup\u003e include dropout and zero-masking for overcoming overfitting, a preprocessing algorithm, a principal component analysis for dimensionality reduction, and a multitask feature selection approach. Using the ADNI dataset, accuracies ranging from 87.78\u0026ndash;99.62% were observed. Hazarikar \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e19\u003c/sup\u003e replaced the downsampling layer in the traditional LeNet architecture with a fusion of min-pooling and max-pooling layers to retain both minimum-value and maximum-valued signals. Their model achieved an accuracy, precision, recall, and F1-score of 98%, 96%, 97%, and 98%, respectively, on the ADNI dataset. In another work, a VGG-TSwinformer architecture\u003csup\u003e20\u003c/sup\u003e that combines a VGG-16 convolutional neural network and a transformer network was proposed and validated on the ADNI sMCI and pMCI cohorts. The accuracy, sensitivity, specificity, and AUC were 77.2%, 79.97%, 71.59%, and 0.8153, respectively. Similarly, another work\u003csup\u003e21\u003c/sup\u003e also found architecture useful for their methodology. They created a hybrid architecture by combining AlexNet with LeNet and varying the filter sizes from 1 \u0026times; 1, 3 \u0026times; 3, and 5 \u0026times; 5. Scores as high as 96%, 93%, 93%, and 96% for accuracy, precision, recall, and F1 score, respectively, were reported. Another notable architecture is the multiplane convolutional neural network (Mp-CNN) architecture, which simultaneously processes three planes, axial, coronal, and sagittal, of 3D MRI \u003csup\u003e22\u003c/sup\u003e. The architecture of the Mp-CNN comprises 14 layers with rectified linear unit (ReLU) activation and softmax for multiclass classification, and it outperforms traditional 2D CNNs in multiclass classification associated with AD, MCI, and NC. The Swinformer has also been explored\u003csup\u003e23\u003c/sup\u003e as a transformer-based CNN architecture for AD classification. The Swinformer combines a CNN module for planar feature extraction and a transformer encoder module for 3D semantic connections. They argued that Swinformer can capture local features more accurately. The pipeline included data preprocessing and augmentation strategies such as random rotation and mirror reflection and recorded an accuracy of 88.3%.\u003c/p\u003e \u003cp\u003eWhile it is obvious that transfer learning with the use of state-of-the-art pretrained models is a promising technique for diagnosing AD, the tailored deep learning algorithm outperforms traditional methods in terms of performance and shows that it is better at preserving the underlying structure of the data for diagnosing AD. However, of these works, none have captured the structural dynamics of neurodegeneration in the brain in individuals with MCI or AD, which leaves room for additional work to be done. Therefore, this paper seeks to bridge this research gap and makes the following contributions to AD research.\u003c/p\u003e \u003cp\u003e \u003cul\u003e \u003cli\u003e \u003cp\u003eWe propose a machine learning framework that integrates the novel SNeurodCNN architecture for modelling the structural neurodegeneration of the brain's cerebral cortex and for the task of discriminating between MCI and AD.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eInvestigate whether the varying viewpoints of the two planes of the sagittal axis, midsagittal and parasagittal, provide differing insights into structural neurodegeneration.\u003c/p\u003e \u003c/li\u003e \u003cli\u003e \u003cp\u003eInvestigating SNeurodCNN sensitives patients to brain neurodegeneration is relevant for identifying digital biomarkers (digi-biomarkers) for the regions of the brain where structural neurodegeneration is prevalent.\u003c/p\u003e \u003c/li\u003e \u003c/ul\u003e \u003c/p\u003e"},{"header":"Results","content":"\u003cp\u003eThis section presents the results and findings of this paper. We begin by outlining the findings of SNeurodCNN and then progress to its analysis using GradCAM to identify features strongly indicative of the model\u0026rsquo;s sensitivities to brain neurodegeneration.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eEvaluation of the Structure-focused ADNI Dataset\u003c/p\u003e\n\u003cp\u003eExperiments on the structure-focused ADNI data version of the brain are relevant for understanding the structural changes that are contributors to brain neurodegeneration. The performance of the SNeurodCNN was evaluated on the midsagittal and parasagittal planes of the sagittal axis of the brain. The parasagittal view of the sagittal plane is the version taken from the off-centre plane, while the midsagittal view is the version taken at exactly the centre of the plane. The use of both planes is to investigate whether the model is impacted differently by neurodegeneration, as it pertains to the parasagittal region, which exposes regions of motor control, sensation, and perception, and the midsagittal region, which exposes regions that are responsible for visuospatial integration, memory, and self-awareness. The results of this experiment are presented in Table 1 and Fig. 1. Meanwhile, we demonstrate in Fig. 2 which displays the training and validation loss and accuracy performance of SNeurodCNN with training sets from the midsagittal and parasagittal viewpoints. The loss curves generated in Fig. 3 (a) and (b) demonstrate an exceptional trend in the training progress for both the parasagittal and midsagittal regions of the brain. As the number of epochs increases, there is a smooth and consistent descent in the loss curves, indicating effective model training. The convergence of training and validation losses to stable values, with a plateau at higher loss levels, signifies successful learning without succumbing to overfitting. Additionally, the parallel curves of both validation and training further attests that SNeurodCNN is learning from the training data but also demonstrates a robust ability to generalize to previously unseen data and simultaneously identify intricate pattens in the data while avoiding the pitfalls of overfitting.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eFurthermore, akin to the loss curves generated by the model, the accuracy curves depicted in Fig. 2 not only demonstrate the model\u0026apos;s adeptness in making predictions for both datasets but also showcase its adaptability to the training data, indicating a continuous enhancement in predictive accuracy. As the epochs increase, the upward trajectory of the axis curve for both experiments corresponds to the decrease in loss. Although peak accuracy is reached relatively early on for both the midsagittal and parasagittal datasets, this does not compromise the SNeurodCNN performance as the close alignment between training and validation indicates good generalization without overfitting.\u003c/p\u003e\n\u003cp\u003eTable 1: Performance evaluation of SNeurodCNN classification model in the midsagittal and parasagittal planes\u003c/p\u003e\n\u003cdiv align=\"\"\u003e\n \u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"374\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003e\u003cstrong\u003eMetric\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e\u003cstrong\u003eMidsagittal Plane\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e\u003cstrong\u003eParasagittal Plane\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e98.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e97.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e97.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e97.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e99.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e98.5%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e97.2%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e97.0%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003eF1-score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e98.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e97.8%\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"23.262032085561497%\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"37.70053475935829%\"\u003e\n \u003cp\u003e98.1%\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"39.037433155080215%\"\u003e\n \u003cp\u003e97.8%\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\u003eAs shown in Table 1 and Fig. 1, the accuracy, precision, recall, specificity, F1 score, and AUC of the SNeurodCNN are outstanding. This is an indication that the model is capable of modelling the structural neurogenerative impacts of MCI and AD occurring at both parts of the sagittal plane\u0026mdash;the midsagittal and parasagittal planes. While the accuracies recorded are high, which shows the model\u0026rsquo;s ability to distinguish between instances of AD and instances of MCI, the midsagittal accuracy was increased by 0.3%. Given that the structure-focused ADNI is used, the performance conforms with the medical statement \u0026ldquo;midsagittal cerebral morphology provides a homologous geometrical reference for brain shape and cortical vs. subcortical spatial relationships\u0026rdquo; \u003csup\u003e24\u003c/sup\u003e. In medical diagnosis, a high recall is preferred\u0026mdash;the higher the value is, the greater the confidence we have in the model\u0026rsquo;s ability to minimize false positives, that is, diagnosing AD when in fact it is MCI and vice versa. Furthermore, the high F1-scores for both models reflect well-balanced performance in terms of precision and recall, ensuring a lower rate of both false-negative and false-positive predictions.\u003c/p\u003e\n\u003cp\u003eAs an additional measure of our model\u0026rsquo;s performance, we utilized the AUC metric, which is a metric that provides an unbiased view of a model\u0026apos;s performance amidst class imbalances aside from the F1 score since the data presented are 180:105 in proportion for AD patients and MCI patients. Scores of 98.1% and 97.8% were observed for the midsagittal and parasagittal sides, respectively, which underscores the models\u0026apos; discriminatory capability in diagnosing AD.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eSNeurodCNN Sensitivity to Brain Neurodegeneration\u003c/p\u003e\n\u003cp\u003eTo understand the sensitivity of SNeurodCNN to brain neurodegeneration, the same slices from the MCI and AD categories for the midsagittal and parasagittal planes were sampled. Then, with Grad-CAM, the activation maps of the last convolutional layer of SNeurodCNN are visualized to better understand the network\u0026rsquo;s sensitivity to brain degeneration. The Grad-CAM output was analysed using a heatmap; the brighter the yellow, the more significant the region is, and the more prominent the region is where brain neurodegeneration is. The closer to purple it is, the less significant the difference is. These regions can be visualised in Fig. 4, with more examples for the midsagittal in Fig. 5 and parasagittal in Fig. 6.\u003c/p\u003e\n\u003cp\u003e\u0026nbsp;As can be observed from the MCI (sMCI and pMCI) and AD slices in Fig. 5, the frontal lobe, occipital lobe, and cerebellum are the regions highlighted as being highly significant, with the frontal lobe being the most prominent while the cerebellum the least prominent. Therefore, these regions reveal SNeurodCNN sensitivity to brain neurodegeneration in relation to AD. The findings made with the midsagittal plane are consistent with those made with the parasagittal plane and can be observed in Fig. 6, although the parasagittal plane shows additional prominent regions about the locations of the temporal and parietal lobe. Since we used the structure-focused ADNI, it is only expected that the highlights relate to changes in structure caused by shrinkage of the cerebral cortex, known as cerebral atrophy. An obvious interpretation of these observations is that the SNeurodCNN can sense a significant difference in the structural characteristics of the frontal lobe, occipital lobe, and cerebellum and possibly the parietal regions in individuals with MCI (sMCI and pMCI) and AD.\u003c/p\u003e\n\u003cp\u003eInsights into the functions of the frontal lobe, occipital lobe, cerebellum, temporal, and parietal lobe are important for understanding the roles these regions play in brain neurodegeneration in relation to AD. The frontal lobe is mainly responsible for motor action and the temporal integration of behavior\u003csup\u003e\u0026nbsp;25\u003c/sup\u003e. In a recent clinical study\u003csup\u003e\u0026nbsp;26\u003c/sup\u003e, the frontal lobes of patients with MCI were shown to experience cortical atrophy, but severe cortical volume deficit was evident in patients with AD. The region associated with visual processing for depth perception, colour determination, object, and face recognition and responsible for memory formation is the occipital lobe\u003csup\u003e\u0026nbsp;27\u003c/sup\u003e. According to clinical research, structural changes are evident in the occipital lobe and are an indication of early progressive MCI \u003csup\u003e28\u003c/sup\u003e. A clinical study\u003csup\u003e\u0026nbsp;29\u003c/sup\u003e revealed that there was a degree of functional connectivity disruption in the cerebellum of MCI and AD patients. In this paper, our proposed SNeurodCNN showed that its impact is subtle compared to that of the others highlighted. The cerebellum is responsible for regulating motor movement and controlling balance\u003csup\u003e\u0026nbsp;30\u003c/sup\u003e. The additional highlight from the parasagittal viewpoint, the temporal and parietal lobe, might significantly indicate the impact that neurodegeneration can have on the sensory functions of the brain. The parietal region has not been computationally confirmed in DL-based studies, but a clinical study\u003csup\u003e\u0026nbsp;31\u003c/sup\u003e has shown that the parietal gray matter volumes distinguish between MCI patients and AD patients. Memory loss, to a considerable extent, affects MCI and AD patients, with the former being less pronounced than the latter \u003csup\u003e32,33\u003c/sup\u003e. The temporal lobe that is responsible for memory loss is visible in the parasagittal axis of the brain. As has been presented, there is a clinical corroboration to the brain regions our model showed sensitivity to through the midsagittal and parasagittal axis of the brain. \u0026nbsp;\u003c/p\u003e\n\u003cp\u003eComparison with the state-of-the-art methods\u003c/p\u003e\n\u003cp\u003eWe believe that these studies aimed at discriminating between AD and MCI patients contrast them with the SNeurodCNN model. The comparisons are summarized in Table 2. Basaiaa \u003cem\u003eet al\u003c/em\u003e.\u003csup\u003e15\u003c/sup\u003e and the DBN model\u003csup\u003e18\u003c/sup\u003e offer general approaches to diagnosing AD and MCI but with less emphasis on neurodegeneration structures than does SNeurodCNN. DenseNet-169 and ResNet-50 CNN models\u003csup\u003e13\u003c/sup\u003e focus on early AD diagnosis with deep architectures, which are potentially less specialized in structural neurodegeneration. The Mp-CNN model\u003csup\u003e22\u003c/sup\u003e offers a unique approach through advanced multiplanar image processing but differs from the focused technique of SNeurodCNN. Finally, the Hybrid LeNet-AlexNet Model\u003csup\u003e21\u003c/sup\u003e combines various network elements but diverges from SNeurodCNN, which focuses on neurodegeneration. While each model has distinctive strengths for diagnosing AD and MCI, SNeurodCNN stands out for its specific focus on the regions where structural neurodegeneration occurs, making it a significant tool in the field.\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTable 2: Comparison of the SNeurodCNN model with other state-of-the-art models\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\" width=\"862\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd width=\"4.408352668213457%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eS/N\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.37354988399072%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eModel\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.004640371229698%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eYear\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003ePerformance Metric\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.081206496519721%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eArchitecture\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDiagnostic Approach\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.737819025522043%\" valign=\"top\"\u003e\n \u003cp\u003e\u003cstrong\u003eDigi-Biomarker\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"4.408352668213457%\" valign=\"top\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.37354988399072%\" valign=\"top\"\u003e\n \u003cp\u003eBasaia \u003cem\u003eet al\u003c/em\u003e.\u0026apos;s 3D CNN Model\u003csup\u003e15\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.004640371229698%\" valign=\"top\"\u003e\n \u003cp\u003e2019\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003e74.8% - 88.3% accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.081206496519721%\" valign=\"top\"\u003e\n \u003cp\u003e2 blocks of 5x5x5 and 10 blocks of 3x3x3 filters, use strides for downsampling\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eAD and MCI diagnosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.737819025522043%\" valign=\"top\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"4.408352668213457%\" valign=\"top\"\u003e\n \u003cp\u003e2\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.37354988399072%\" valign=\"top\"\u003e\n \u003cp\u003eDeep Belief Network (DBN) Model\u003csup\u003e18\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.004640371229698%\" valign=\"top\"\u003e\n \u003cp\u003e2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003e87.78% - 99.62% accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.081206496519721%\" valign=\"top\"\u003e\n \u003cp\u003eDropout and zero-masking, multitask feature selection\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eClassification tasks for AD and MCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.737819025522043%\" valign=\"top\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"4.408352668213457%\" valign=\"top\"\u003e\n \u003cp\u003e3\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.37354988399072%\" valign=\"top\"\u003e\n \u003cp\u003eDenseNet-169 and ResNet-50 CNN Models\u003csup\u003e16\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.004640371229698%\" valign=\"top\"\u003e\n \u003cp\u003e2023\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eDenseNet-169: 97.7% (training), 83.82% (testing); ResNet-50: Lower performance\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.081206496519721%\" valign=\"top\"\u003e\n \u003cp\u003eDeep architectures with advanced feature learning\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eEarly AD diagnosis\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.737819025522043%\" valign=\"top\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"4.408352668213457%\" valign=\"top\"\u003e\n \u003cp\u003e4\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.37354988399072%\" valign=\"top\"\u003e\n \u003cp\u003eMultiplane Convolutional Neural Network (Mp-CNN)\u003csup\u003e22\u003c/sup\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.004640371229698%\" valign=\"top\"\u003e\n \u003cp\u003e2022\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003e93% accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.081206496519721%\" valign=\"top\"\u003e\n \u003cp\u003e14-layer architecture processing axial, coronal, sagittal planes of 3D MRI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eAdvanced multiplane image processing\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.737819025522043%\" valign=\"top\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd width=\"4.408352668213457%\" valign=\"top\"\u003e\n \u003cp\u003e5\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"19.37354988399072%\" valign=\"top\"\u003e\n \u003cp\u003eThe proposed\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"8.004640371229698%\" valign=\"top\"\u003e\n \u003cp\u003e2024\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eSNeurodCNN\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003cp\u003e98.1% accuracy\u003c/p\u003e\n \u003cp\u003e97.8% accuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.081206496519721%\" valign=\"top\"\u003e\n \u003cp\u003eSagittal planes; CNN architecture: \u0026nbsp;2D Conv, 3x3 filter of varying depth, Maxpool, Dense layers, Dropout\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"15.197215777262182%\" valign=\"top\"\u003e\n \u003cp\u003eClassification tasks for AD and MCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd width=\"22.737819025522043%\" valign=\"top\"\u003e\n \u003cp\u003eCaptures structural brain neurodegeneration about the frontal lobe, occipital lobe, cerebellum, and parietal lobe.\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Discussion","content":"\u003cp\u003eWe delve into the discussion to elaborate on the implications of the results of this study. Neurologists traditionally use neuroimaging methods, such as MRI, to assess MCI and AD neurodegeneration, but this paper presents a dimension that automates diagnosis toward classifying the disease stages but can help in identifying the regions susceptible to neurodegeneration. This finding substantiates the relevance of the proposed SNeurodCNN model in the early diagnosis of AD. The evaluation of neurodegeneration using the midsagittal and parasagittal planes of the brain results in remarkable performance. These findings surpass those of previous studies\u003csup\u003e7,10\u003c/sup\u003e. SNeurodCNN has shown promise in capturing neurodegenerative features from MR images. As observed with Grad-CAM, structural changes were prominent in the frontal lobe, occipital lobe, and cerebellum on the midsagittal MR images. Frontal lobe dysfunction has been implicated in various neurodegenerative conditions, including AD and MCI\u003csup\u003e34\u003c/sup\u003e. The parasagittal view showed that an additional structural change was possible in the parietal lobe, which was not obvious in the midsagittal slices. The SNeurodCNN model region is most sensitive to the frontal lobe, occipital lobe, cerebellum, and parietal lobe, which is consistent with the findings of clinical studies \u003csup\u003e25\u0026ndash;31\u003c/sup\u003e and, in some cases, consistent with regions identified in one of the DL-based studies\u003csup\u003e7\u003c/sup\u003e; however, our study further confirms the significance of cerebral atrophy on brain structure. Therefore, these findings show that our proposed SNeurodCNN can significantly contribute to AD diagnosis and prognosis. This necessitates the integration of deep learning into healthcare practices to provide early cues about AD that might be present in MCI patients, minimizing the occurrence of false positives and false negatives that can mitigate misdiagnosis.\u003c/p\u003e \u003cp\u003eFurthermore, the structural changes that our proposed SNeurodCNN highlights set a record as a potential brain structural change digi-biomarker for the early diagnosis of AD. The benefits of digi-biomarkers are numerous. 1) These findings can help clinicians develop targeted and personalized treatment plans and interventions through personalized analysis of disease states. These findings could lead to valuable targets for therapeutic interventions and monitoring and the assessment of novel treatments accordingly.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003c/p\u003e"},{"header":"Materials and Method","content":"\u003cp\u003eThis section elucidates the dataset employed in this study and outlines the methodologies encompassing the data preprocessing pipeline and the design of the deep convolutional neural network architecture as illustrated in Fig. \u003cspan class=\"InternalRef\"\u003e7\u003c/span\u003e.\u003c/p\u003e\n\u003cp\u003eData\u003c/p\u003e\n\u003cp\u003eThe ADNI AD benchmark dataset commonly used for AD analysis is relevant to this study. This dataset was designed from a study in which individuals aged 55 to 90 years were enrolled from 57 sites across the U.S. and Canada after providing informed consent. In all, the ADNI comprises 368 participants, categorized into 180 AD patients (82 females, 98 males; age\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u0026thinsp;=\u0026thinsp;75.28\u0026thinsp;\u0026plusmn;\u0026thinsp;7.57 years) and 105 stable sMCI patients (41 females, 64 males; age\u0026thinsp;\u0026plusmn;\u0026thinsp;SD\u0026thinsp;=\u0026thinsp;74.69\u0026thinsp;\u0026plusmn;\u0026thinsp;7.41 years). The average mini-mental state examination (MMSE) scores were 23 for AD patients, 28 for sMCI patients, and 27 for pMCI patients. All subjects had a T1-weighted baseline from the ADNI1/Go/2 cohort. The sMCI group consisted of individuals who were diagnosed with MCI at baseline and remained so for at least two years. A summary of the demographics of the ADNI participants is provided in Table\u0026nbsp;\u003cspan class=\"InternalRef\"\u003e2\u003c/span\u003e and the mental examination score of the participants in Fig.\u0026nbsp;8.\u003c/p\u003e\n\u003cp\u003eWe particularly explored skull-free patients version. This version utilized the ADNI dataset preprocessed to correct imaging distortion through gradwarping via \u003cem\u003egradient inhomogeneity correction\u003c/em\u003e, intensity correction, and scaling to address gradient drift. The results are further processed using multiatlas label propagation with expectation maximization (MALPEM) to segment, on local scales, the cross-sectional structural volume changes in the brain, which constitute the structure-focused ADNI dataset. It should be noted that each slice of the structure-focused ADNI dataset is characterized by the region of the gray matter that forms the anatomical structures of the MRI brain images and not the gray matter itself.\u003c/p\u003e\n\n\u003ctable id=\"Tab2\" border=\"1\"\u003e\n \u003ccaption language=\"En\"\u003e\n \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e\n \u003cdiv class=\"CaptionContent\"\u003e\n \u003cp\u003eDemographics and clinical characteristics of the study population in the ADNI dataset\u003c/p\u003e\n \u003c/div\u003e\n \u003c/caption\u003e\n \u003cthead\u003e\n \u003ctr\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eGroup\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eSubject\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/th\u003e\n \u003cth align=\"left\" colspan=\"2\"\u003e\n \u003cp\u003eGender\u003c/p\u003e\n \u003c/th\u003e\n \u003c/tr\u003e\n \u003c/thead\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\u0026nbsp;\u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eF\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003eAD\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e180\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e75.28\u0026thinsp;\u0026plusmn;\u0026thinsp;7.57\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e82\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003esMCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e105\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e74.69\u0026thinsp;\u0026plusmn;\u0026thinsp;7.41\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e64\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e41\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003epMCI\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e73.82\u0026thinsp;\u0026plusmn;\u0026thinsp;6.65\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e49\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e34\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cbr\u003e\u003c/p\u003e\n\u003cp\u003ePreprocessing\u003c/p\u003e\n\u003cp\u003eConsidering that structural changes in the brain and AD pathology are more prominent in the sagittal plane, their efficacy in AD classification has long been established\u003csup\u003e7,12\u003c/sup\u003e. This study explored two sagittal plane views, the midsagittal and parasagittal viewpoints, from the ADNI volumetric data in the sagittal plane. From the 3D structured focused ADNI dataset comprising 368 subjects with 155 slices, only slices showing full brain parts were retained; as such, the number of slices varied from sample to sample. In this study, we considered a different approach, which is an unusual practice in the literature where the disease case of a sample is described by all/fraction of their slices. The slices from the sMCI and pMCI patients were combined and referred to as the MCI set, while the slices from the AD samples were likewise combined into an AD set. We consider AD diagnosis to be a sample-independent task; therefore, it is only logical to teach the deep learning algorithm to capture the factors influencing the two categories, MCI and AD, to better understand the disparity that exists between them. In all, a total of 4228 midsagittal AD and MCI sets were retained, each consisting of 2160 and 2068 participants, respectively. The parasagittal AD and MCI sets included 2700 and 2820 participants, respectively, for a total of 5520 data points.\u003c/p\u003e\n\u003cp\u003eThe nonlinearity property of the Gamma function is found to be \u003csup\u003euseful35\u003c/sup\u003e for expanding the brightness of the intensity of a poorly illuminated image and contributes greatly to the technique proposed\u003csup\u003e35\u003c/sup\u003e. Despite the intensity correction on the original ADNI dataset, the structure-focused ADNI dataset shows that images are captured under poor illumination conditions. Therefore, the gamma correction, an inverse of the gamma function, is adopted for expanding the brightness of pixel intensities. AD and MCI samples were subsequently split into training and testing sets at a ratio of 80:20. The testing set is further split into halves to accommodate the set for validation.\u003c/p\u003e\n\u003cp\u003eThe SNeurodCNN Architecture\u003c/p\u003e\n\u003cp\u003eAs previously stated, each structure-focused ADNI MRI slice is characterized by the formation of gray matter, which forms the anatomical structures of the MR brain images. This structural formation, as opposed to gray matter features, is used to capture focal structural changes associated with neurodegeneration in the cerebral cortex as opposed to focal fine-grain changes within anatomical structures. For this reason, the SNeurodCNN architecture is designed to be a deep convolutional neural network for modelling the structural neurodegeneration of the brain and discriminating between MCI and AD.\u003c/p\u003e\n\u003cp\u003eThe SNeurodCNN architecture is presented in Fig. \u003cspan class=\"InternalRef\"\u003e3\u003c/span\u003e. It comprises two downsampling convolution blocks growing in a number of convolutional layers. The first block consists of a convolutional layer with a 3\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\times 3\\)\u003c/span\u003e\u003c/span\u003e filter size and 32 filter depths, varying image dimensions, a nonlinear activation function, a rectified linear unit (ReLU), and a 2\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(\\times 2\\)\u003c/span\u003e\u003c/span\u003e maxpooling layer. The second convolution block differs from the first by only the two convolutional layers, which are made up of 32 and 64 filter depths. This process feeds into the fully connected layer with 500 hidden neurons and a dropout layer, which is a regularizer that optimizes learning by randomly dropping nodes in the hidden neuron by a designated probability, which in our case is 50%. This is followed by the output layer, which is set to two and represents the two classes that the softmax activation function uses to classify an image as belonging to MCI or AD.\u003c/p\u003e\n\u003cp\u003eExperimental settings\u003c/p\u003e\n\u003cp\u003eThe network\u0026apos;s hyperparameters are chosen in ways that best optimize the network\u0026rsquo;s learning capability. For the hyperparameters, the Adam optimizer with a learning rate fixed at 0.0001 was used. The other variables are epochs and batch sizes, which are 100 and 32, respectively. We adopted early stopping, a regularizer that prevents overfitting by immediately halting training when the performance of the network does not improve after several epochs. The parameters of the early stopping agent used were patience set to 5 and restore_best_weights set to the true Boolean value. During training, the model learns through a method called backpropagation\u003csup\u003e36,\u003c/sup\u003e which updates the weight of the network toward minimizing the gradient error. Considering the impact of hardware resources on model performance, the hardware configuration of the system used for training the network is reported as follows: A Keras library running on a Python-TensorFlow environment with a Google Collab GPU, Tesla K80 (T4).\u003c/p\u003e\n\u003cp\u003eTo evaluate the performance of the SNeurodCNN model, the following metrics were identified and found to be consistent with the metrics used in the literature for diagnosing AD. The variables used were the accuracy, precision, recall, specificity, F1-score, and area under the receiver operating characteristic (ROC) curve (AUC), which provide a comprehensive assessment of the model\u0026apos;s performance. Their brief descriptions are as follows.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e1. Accuracy: This metric measures the proportion of correct predictions over the total number of predictions, as expressed in (\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tabe\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(Accuracy =\\frac{TP+TN}{TP+TN+FP+FN}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(\u003cspan class=\"CitationRef\"\u003e1\u003c/span\u003e)\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\u003ewhere \u003cem\u003eTP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;true positive, \u003cem\u003eTN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;true negative, \u003cem\u003eFP\u003c/em\u003e\u0026thinsp;=\u0026thinsp;false positive and \u003cem\u003eFN\u003c/em\u003e\u0026thinsp;=\u0026thinsp;false negative.\u003c/p\u003e\n\u003cp\u003e\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e2. Precision: the proportion of true positives out of all the instances predicted as positive. In this case, precision measures the number of correctly classified AD cases out of all instances classified as ADs and is mathematically expressed as follows\u003c/p\u003e\n\u003ctable id=\"Tabf\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(Precision= \\frac{TP}{TP+FP}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(\u003cspan class=\"CitationRef\"\u003e2\u003c/span\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\u003e3. Recall: also known as sensitivity, this measures the proportion of correctly identified positive instances out of all the actual positive instances. In this case, the recall measures the proportion of correctly classified AD cases out of all actual AD cases. It is given as:\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tabg\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(Recall= \\frac{TP}{TP+FN}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(\u003cspan class=\"CitationRef\"\u003e3\u003c/span\u003e)\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\u003e4. Specificity: This metric quantifies the model\u0026rsquo;s ability to make true negative predictions out of all the correctly identified negative instances. With regard to the classification model, specificity will measure the proportion of correctly classified MCI cases out of all actual MCI cases. It is mathematically given as follows:\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\u0026nbsp;\u003ctable id=\"Tabh\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(Specificity= \\frac{TN}{TN+FP}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(\u003cspan class=\"CitationRef\"\u003e4\u003c/span\u003e)\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\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e5. F1-score: The F1-score combines precision and recall into a single metric, providing a balanced measure of the model\u0026apos;s performance. It considers both false positives and false negatives, making it valuable for overall performance assessment.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cdiv class=\"gridtable\"\u003e\n \u003ctable id=\"Tabi\" border=\"1\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e\u003cspan class=\"InlineEquation\"\u003e\u003cspan class=\"mathinline\"\u003e\\(F1=2 \\times \\frac{Precision\\times Recall}{Precision+Recall}\\)\u003c/span\u003e\u003c/span\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd align=\"left\"\u003e\n \u003cp\u003e(\u003cspan class=\"CitationRef\"\u003e5\u003c/span\u003e)\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\u003cspan\u003e\u003c/span\u003e\u003c/p\u003e\n\u003cp\u003e6. AUROC: The AUROC evaluates the model\u0026apos;s performance across different classification thresholds, considering the trade-off between sensitivity and specificity. This approach provides insights into classifier discrimination ability and can be particularly useful for imbalanced datasets.\u003c/p\u003e\n\u003cp\u003e\u003c/p\u003e\n\u003cp\u003eLimitations\u003c/p\u003e\n\u003cp\u003eThe growing increase in AD cases underscores the pressing necessity of advancing research in deep learning. Our proposed SNeurodCNN achieves remarkable classification accuracies of 98.1% and 97.8% in both the midline and posterior sagittal planes, respectively. This level of performance surpasses that of many existing studies, highlighting the potential of deep learning, particularly our proposed model, in transforming the landscape of neurodegenerative disease diagnosis. The model\u0026rsquo;s sensitivities to neurodegeneration in the brain structure not only substantiated our model\u0026apos;s efficacy but also deepened our comprehension of the intricate nature of neurodegenerative diseases in relation to MCI and AD, thereby opening doors for a better understanding of the disease.\u003c/p\u003e\n\u003cp\u003eWhile our proposed SNeurodCNN model significantly contributes to AD diagnostic research, its findings are constrained by several limitations. The trainable parameters are large and computationally expensive, especially with respect to reliability and usefulness in real-time clinical diagnosis. Therefore, efforts to achieve high performance while minimizing the computational cost will be explored. Additionally, this paper is directed toward the structural function of the brain analysis, which led to the use of structure-focused ADNIs. However, it will be interesting to understand how the SNeurodCNN model performs on focal fine-grain features available with skull-free ADNIs. This approach enables us to better analyse which of the features, focal fine-grain or focal structure, are representative of the characteristics of neurodegeneration in the brain in a way that makes early diagnosis of AD feasible and reliable in clinical analysis. We expect that the outcome of the aforementioned study will lead to an understanding of the underlying neurobiological mechanisms involved in MCI and AD, which will help in studying AD progression.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eIn this paper, we propose a structure-focused neurodegeneration convolutional neural network (CNN) architecture called the SNeurodCNN, which was integrated into a machine learning framework along with preprocessing techniques for image enhancement and data preparation. The proposed framework leveraged the midsagittal and parasagittal brain image viewpoints of the structure-focused ADNI dataset. Through experiments, the proposed framework achieved 97.8% accuracy, with 97.0% specificity and 98.5% sensitivity on the parasagittal planes. On the midsagittal plane, an accuracy, specificity, and sensitivity of 98.1%, 97.2%, and 99.0%, respectively, were achieved. We further showed that the midsagittal lobe highlights the frontal lobe, occipital lobe, and cerebellum, while the parasagittal lobe extends to the parietal lobe as a region of the brain where structural dynamics are prominent due to MCI and AD. We believe this discovery useful for identifying digi-biomarkers for the early diagnosis of AD. In future work, efforts will be made to minimize the computational cost of the proposed model while achieving the same level of neurodegeneration modelling. Additionally, it will be interesting to apply the proposed model to focal fine-grained feature learning. This approach enables us to better analyse which of the features, focal fine-grain or focal structure, are representative of the characteristics of neurodegeneration in the brain in a way that makes early diagnosis of AD feasible and reliable in clinical analysis.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e \u003ch2\u003eConflict of interest\u003c/h2\u003e \u003cp\u003eThe authors declare that there are no conflicts of interest.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eData collection, implementation, coding, validation, and testing, S.O.; Conceptualization, methodology, model design, and project supervision, C.C.O.; literature review, project scope, S.O., K.M, C.C.O.; result analysis and interpretation S.O, K.M, C.C.O; manuscript \u0026ndash; original draft, C.C.O.,K.M, S.O; All authors reviewed, edited, and approved the final manuscript.\u003c/p\u003e\u003ch2\u003eAvailability of Data\u003c/h2\u003e \u003cp\u003eThe datasets analysed during the current study are available at the following link: https://doi.gin.g-node.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003eorg/10.12751/g-node.aa605a/\u003c/span\u003e\u003cspan address=\"10.12751/g-node.aa605a/\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eAlzheimer's Association, 2018. 2018 Alzheimer's disease facts and figures. \u003cem\u003eAlzheimer's \u0026amp; Dementia\u003c/em\u003e, 14(3), pp.367\u0026ndash;429.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGonzalez-Ortiz, F. \u003cem\u003eet al.\u003c/em\u003e Plasma phospho-tau in Alzheimer\u0026rsquo;s disease: towards diagnostic and therapeutic trial applications. Mol Neurodegener 18, 18 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGalvin, J. E. \u0026amp; Sadowsky, C. H. Practical Guidelines for the Recognition and Diagnosis of Dementia. The Journal of the American Board of Family Medicine 25, 367\u0026ndash;382 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng, H.-M., Han, H.-B., Zhang, Q.-F. \u0026amp; Bai, H. Application of modern neuroimaging technology in the diagnosis and study of Alzheimer\u0026rsquo;s disease. Neural Regen Res 16, 73 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVan de Mortel, L. A., Thomas, R. M. \u0026amp; van Wingen, G. A. Grey Matter Loss at Different Stages of Cognitive Decline: A Role for the Thalamus in Developing Alzheimer\u0026rsquo;s Disease. Journal of Alzheimer\u0026rsquo;s Disease 83, 705\u0026ndash;720 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLin, H.-Y. \u003cem\u003eet al.\u003c/em\u003e Differential Patterns of Gyral and Sulcal Morphological Changes During Normal Aging Process. Front Aging Neurosci 13, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoang, G. M., Kim, U.-H. \u0026amp; Kim, J. G. Vision transformers for the prediction of mild cognitive impairment to Alzheimer\u0026rsquo;s disease progression using mid-sagittal sMRI. Front Aging Neurosci 15, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eApostolova, L. G. \u003cem\u003eet al.\u003c/em\u003e Hippocampal Atrophy and Ventricular Enlargement in Normal Aging, Mild Cognitive Impairment (MCI), and Alzheimer Disease. Alzheimer Dis Assoc Disord 26, 17\u0026ndash;27 (2012).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMcEvoy, L. K. \u0026amp; Brewer, J. B. Quantitative structural MRI for early detection of Alzheimer\u0026rsquo;s disease. Expert Rev Neurother 10, 1675\u0026ndash;1688 (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBae, J. \u003cem\u003eet al.\u003c/em\u003e Transfer learning for predicting conversion from mild cognitive impairment to dementia of Alzheimer\u0026rsquo;s type based on a three-dimensional convolutional neural network. Neurobiol Aging 99, 53\u0026ndash;64 (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShanmugam, J. V., Duraisamy, B., Simon, B. C. \u0026amp; Bhaskaran, P. Alzheimer\u0026rsquo;s disease classification using pretrained deep networks. Biomed Signal Process Control 71, 103217 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng, B. \u003cem\u003eet al.\u003c/em\u003e A modified 3D EfficientNet for the classification of Alzheimer\u0026rsquo;s disease using structural magnetic resonance images. IET Image Process 17, 77\u0026ndash;87 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl Shehri, W. Alzheimer\u0026rsquo;s disease diagnosis and classification using deep learning techniques. PeerJ Comput Sci 8, e1177 (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOktavian, M.W., Yudistira, N. and Ridok, A., 2022. Classification of Alzheimer's Disease Using the Convolutional Neural Network (CNN) with Transfer Learning and Weighted Loss. \u003cem\u003earXiv preprint arXiv:2207.01584\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBasaia, S., \u003cem\u003eet al.\u003c/em\u003e Alzheimer's Disease Neuroimaging Initiative, 2019. Automated classification of Alzheimer's disease and mild cognitive impairment using a single MRI and deep neural networks. \u003cem\u003eNeuroImage: Clinical\u003c/em\u003e, \u003cem\u003e21\u003c/em\u003e, p.101645.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarwa, E.G., Moustafa, H.E.D., Khalifa, F., Khater, H. \u0026amp; AbdElhalim, E., 2023. An MRI-based deep learning approach for accurate detection of Alzheimer\u0026rsquo;s disease. Alexandria Engineering Journal, \u003cem\u003e63\u003c/em\u003e, pp.211\u0026ndash;221.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShamrat, F. M. J. M. \u003cem\u003eet al.\u003c/em\u003e AlzheimerNet: An Effective Deep Learning Based Proposition for Alzheimer\u0026rsquo;s Disease Stages Classification From Functional Brain Changes in Magnetic Resonance Images. IEEE Access 11, 16376\u0026ndash;16395 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZeng, N., Li, H. \u0026amp; Peng, Y. A new deep belief network-based multitask learning for diagnosis of Alzheimer\u0026rsquo;s disease. Neural Comput Appl 35, 11599\u0026ndash;11610 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHazarika, R.A., Abraham, A., Kandar, D. \u0026amp; Maji, A.K. An improved LeNet-deep neural network model for Alzheimer\u0026rsquo;s disease classification using brain magnetic resonance images. IEEE Access, \u003cem\u003e9\u003c/em\u003e, pp.161194\u0026ndash;161207, (2021).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, Z., Wang, Z., Jin, Y. and Hou, W., VGG-TSwinformer: Transformer-based deep learning model for early Alzheimer\u0026rsquo;s disease prediction. Computer Methods and Programs in Biomedicine, \u003cem\u003e229\u003c/em\u003e, p.107291, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHazarika, R.A., Maji, A.K., Kandar, D., Jasinska, E., Krejci, P., Leonowicz, Z. and Jasinski, M., 2023. An Approach for Classification of Alzheimer\u0026rsquo;s Disease Using Deep Neural Network and Brain Magnetic Resonance Imaging (MRI). \u003cem\u003eElectronics\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e(3), p.676.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAngkoso, C. V., Agustin Tjahyaningtijas, H. P., Purnama, I., \u0026amp; Purnomo, M. H. Multiplane Convolutional Neural Network (Mp-CNN) for Alzheimer\u0026rsquo;s Disease Classification. International Journal of Intelligent Engineering and Systems 15, (2022).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu, Z., Li, Y., Wang, Z., Zhang, S. \u0026amp; Hou, W. Conv-Swinformer: Integration of CNN and shift window attention for Alzheimer\u0026rsquo;s disease classification. Comput Biol Med 164, 107304 (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBruner, E., Martin-Loeches, M. and Colom, R. Human midsagittal brain shape variation: patterns, allometry and integration. Journal of Anatomy, \u003cem\u003e216\u003c/em\u003e(5), pp.589\u0026ndash;599, (2010).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoffmann, M., 2013. The human frontal lobes and frontal network systems: an evolutionary, clinical, and treatment perspective. \u003cem\u003eInternational Scholarly Research Notices\u003c/em\u003e, \u003cem\u003e2013\u003c/em\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eShi, C., Deng, H., Deng, X., Rao, D. and Yue, W., 2023. The Structural Changes of Frontal Subregions and Their Correlations with Cognitive Impairment in Patients with Alzheimer's Disease. \u003cem\u003eJournal of Integrative Neuroscience\u003c/em\u003e, \u003cem\u003e22\u003c/em\u003e(4), p.99.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRehman A, Al Khalili Y. Neuroanatomy, Occipital Lobe. \u003cem\u003eIn: StatPearls. StatPearls Publishing\u003c/em\u003e,PMID: 31335040, (2023).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhao, L., \u003cem\u003eet al.\u003c/em\u003e \u0026amp; Alzheimer Disease Neuroimaging Initiative. Risk estimation before progression to mild cognitive impairment and Alzheimer\u0026rsquo;s disease: an AD resemblance atrophy index. Aging (Albany NY), \u003cem\u003e11\u003c/em\u003e(16), p.6217, (2019).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTang, F., Zhu, D., Ma, W., Yao, Q., Li, Q. and Shi, J., 2021. Differences changes in cerebellar functional connectivity between mild cognitive impairment and Alzheimer's disease: a seed-based approach. \u003cem\u003eFrontiers in Neurology\u003c/em\u003e, \u003cem\u003e12\u003c/em\u003e, p.645171.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eD'Angelo, E., Physiology of the cerebellum. Handbook of clinical neurology, \u003cem\u003e154\u003c/em\u003e, pp.85\u0026ndash;108, (2018).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eH\u0026auml;nggi, J., Streffer, J., J\u0026auml;ncke, L. and Hock, C. Volumes of lateral temporal and parietal structures distinguish between healthy aging, mild cognitive impairment, and Alzheimer's disease. Journal of Alzheimer's Disease, \u003cem\u003e26\u003c/em\u003e(4), pp.719\u0026ndash;734, (2011).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChetelat, G., Desgranges, B., De La Sayette, V., Viader, F., Eustache, F. and Baron, J.C. Mapping gray matter loss with voxel-based morphometry in mild cognitive impairment. Neuroreport, \u003cem\u003e13\u003c/em\u003e(15), pp.1939\u0026ndash;1943, 2002.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaykin, A.J., Wishart, H.A., Rabin, L.A., Santulli, R.B., Flashman, L.A., West, J.D., McHugh, T.L. and Mamourian, A. Older adults with cognitive complaints show brain atrophy similar to that of amnestic MCI. Neurology, \u003cem\u003e67\u003c/em\u003e(5), pp.834\u0026ndash;842, 2006.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDubois, B., Picard, G. \u0026amp; Sarazin, M. Early detection of Alzheimer\u0026rsquo;s disease: new diagnostic criteria. Dialogues Clin Neurosci 11, 135\u0026ndash;139 (2009).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChude-Olisah, C.C., Sulong, G., Chude-Okonkwo, U.A. and Hashim, S.Z., Illumination normalization for edge-based face recognition using the fusion of RGB normalization and gamma correction. In \u003cem\u003e2013 IEEE International Conference on Signal and Image Processing Applications\u003c/em\u003e, pp. 412\u0026ndash;416, (2013).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller, A.S., Blott, B.H. and Hames, T.K. Review of neural network applications in medical imaging and signal processing. Medical and Biological Engineering and Computing, \u003cem\u003e30\u003c/em\u003e, pp.449\u0026ndash;464, 1992.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":true,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Alzheimer’s disease, mild cognitive impairment, classification, deep learning, convolutional neural network","lastPublishedDoi":"10.21203/rs.3.rs-3951099/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-3951099/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eAlzheimer\u0026rsquo;s disease (AD), the predominant form of dementia, is a growing global challenge, emphasizing the urgent need for accurate and early diagnosis. Current clinical diagnoses rely on radiologist expert interpretation, which is prone to human error. Machine learning has thus far shown promise for early AD diagnosis. However, existing methods often overlook focal structural features critical for understanding cerebral cortex neurodegeneration. This paper proposes a machine learning framework that includes a novel structure-focused neurodegeneration CNN architecture named SNeurodCNN and an image brightness enhancement preprocessor using gamma correction. Leveraging mid-sagittal and para-sagittal brain image viewpoints from the Alzheimer\u0026rsquo;s Disease Neuroimaging Initiative (ADNI) dataset, our framework demonstrated exceptional performance. The para-sagittal viewpoint achieved 97.8% accuracy, 97.0% specificity, and 98.5% sensitivity, while the mid-sagittal viewpoint offered deeper insights with 98.1% accuracy, 97.2% specificity, and 99.0% sensitivity. Model analysis revealed the ability of SNeurodCNN to capture the structural dynamics of mild cognitive impairment (MCI) and AD in the frontal lobe, occipital lobe, cerebellum, temporal, and parietal lobe, suggesting its potential as a brain structural change digi-biomarker for early AD diagnosis.\u003c/p\u003e","manuscriptTitle":"SNeurodCNN: Structure-focused Neurodegeneration Convolutional Neural Network for Modeling and Classification of Alzheimer’s Disease","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-03-05 16:25:10","doi":"10.21203/rs.3.rs-3951099/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-03-24T15:08:29+00:00","index":"","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2024-03-21T15:54:31+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"85b4c4d9-9059-4be4-b17c-6e8844e8f692","date":"2024-03-13T03:07:32+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2024-03-12T13:21:54+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-03-12T13:19:30+00:00","index":"","fulltext":""},{"type":"editorInvited","content":"","date":"2024-03-01T11:01:57+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-03-01T09:51:09+00:00","index":"","fulltext":""},{"type":"submitted","content":"Scientific Reports","date":"2024-02-12T13:21:00+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"scientific-reports","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"scirep","sideBox":"Learn more about [Scientific Reports](http://www.nature.com/srep/)","snPcode":"","submissionUrl":"","title":"Scientific Reports","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Scientific Reports","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"71f2bcad-5673-4e28-8aa2-160752042e93","owner":[],"postedDate":"March 5th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"published-in-journal","subjectAreas":[{"id":29130100,"name":"Health sciences/Diseases"},{"id":29130101,"name":"Health sciences/Diseases/Neurological disorders"},{"id":29130102,"name":"Health sciences/Diseases/Neurological disorders/Neurodegeneration"},{"id":29130103,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":29130104,"name":"Biological sciences/Computational biology and bioinformatics/Computational models"}],"tags":[],"updatedAt":"2024-09-06T17:32:43+00:00","versionOfRecord":{"articleIdentity":"rs-3951099","link":"https://doi.org/10.1038/s41598-024-60611-8","journal":{"identity":"scientific-reports","isVorOnly":false,"title":"Scientific Reports"},"publishedOn":"2024-07-03 00:00:00","publishedOnDateReadable":"July 3rd, 2024"},"versionCreatedAt":"2024-03-05 16:25:10","video":"","vorDoi":"10.1038/s41598-024-60611-8","vorDoiUrl":"https://doi.org/10.1038/s41598-024-60611-8","workflowStages":[]},"version":"v1","identity":"rs-3951099","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-3951099","identity":"rs-3951099","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}