A Machine Learning Approach for Early Parkinson's Disease Diagnosis Based on Brain Texture Features from T1-Weighted Imaging

A Machine Learning Approach for Early Parkinson's Disease Diagnosis Based on Brain Texture Features from T1-Weighted Imaging | 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 A Machine Learning Approach for Early Parkinson's Disease Diagnosis Based on Brain Texture Features from T1-Weighted Imaging Yuhua Xu, Zhou Feifan, Qianqian Gao, Yazhou Ma, Xin Chen, Dong Li, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5859401/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract This study presents an early diagnostic method for Parkinson’s disease using T1-weighted imaging texture features combined with machine learning models. T1-weighted imaging data from the PPMI database were preprocessed to extract texture features from various brain regions,including the thalamus, hippocampus, caudate nucleus, amygdala, globus pallidus and putamen. The Random Forest (RF) model demonstrated excellent performance in distinguishing Parkinson’s patients from healthy controls, achieving an AUC of 0.90, accuracy of 88.9%, precision of 92.3%, sensitivity of 92.3%, specificity of 80.0%, and an F1 score of 92.3%. A simplified RF model also exhibited strong performance with a prediction accuracy of 77.8%. This method effectively leverages brain texture features to assist in the early diagnosis of Parkinson’s disease. Biological sciences/Computational biology and bioinformatics/Computational neuroscience Biological sciences/Computational biology and bioinformatics/Databases Biological sciences/Computational biology and bioinformatics/Predictive medicine Biological sciences/Computational biology and bioinformatics Biological sciences/Neuroscience Health sciences/Anatomy Health sciences/Biomarkers Health sciences/Neurology Health sciences/Pathogenesis Parkinson's disease PPMI database Machine learning MRI T1-weighted imaging Texture analysis Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Parkinson's disease is a neurodegenerative disorder that not only profoundly impacts patients' quality of life but also imposes a substantial burden on China's economy and healthcare system, especially with the onset of an aging population 1 . The disease was firstly reported by Jame Parkinson in 1817, so it was then named after him. 2 . The clinical features of Parkinson's disease include both motor and non-motor symptoms. The primary motor symptoms consist of the following four: resting tremor, rigidity, bradykinesia, and postural instability 3 .Non-motor symptoms include olfactory dysfunction, cognitive impairment, depression, anxiety, autonomic dysfunction, and others 4 . Over time, the symptoms of Parkinson's disease gradually worsen, and patients may also experience on-off phenomena, dyskinesia, and other complications 5 . The exact cause of Parkinson's disease is not yet fully understood. Possible contributing factors include genetics, environmental factors, age, and others. Currently, there is no cure for Parkinson's disease; treatment focuses on controlling symptoms to improve the patient's quality of life 6 . These treatment approaches include pharmacological therapy (such as levodopa), surgical interventions (such as deep brain stimulation), and rehabilitation exercises as adjunctive therapies. 7 , 8 . As mentioned above, Parkinson's disease patients experience numerous and severe symptoms, which significantly impact their quality of life and cause great distress to both the patients and their families 9 – 11 . Therefore, early diagnosis is especially important. It enables doctors to start treatment as soon as possible, slowing the progression of symptoms and providing the greatest benefit to the patients. To diagnose Parkinson's disease, clinicians typically need to gather a detailed medical history of the patient, observe their symptoms, and integrate the findings with additional auxiliary tests for a comprehensive diagnosis 12 . The tools frequently used in diagnosis include clinical rating scales such as the UPDRS, MoCA, MMSE, and PDQ-39, as well as imaging examinations like cranial ultrasound, MRI, and SPECT 13 – 16 . However, these tests are highly dependent on the clinician's experience and are quite subjective. Additionally, the early symptoms of Parkinson's disease are often subtle, which greatly increases the difficulty of early diagnosis. In recent years, there has been an increasing effort to find a more objective and accurate method. As a result, there has been growing exploration of clinical biomarkers for Parkinson's disease 17 . MRI is a commonly used imaging technique in clinical practice 18 . Although Parkinson's disease cannot be directly diagnosed through MRI, studies have shown that analysis of MRI texture features can facilitate the early detection of Parkinson's disease 19 , 20 . Texture features refer to the spatial arrangement of pixels and the distribution pattern of gray levels in an image. By extracting and analyzing these texture features, computers can understand the surface structure or surface properties of objects in the image, such as smoothness, roughness, periodicity, and so on 21 . Machine learning is an important branch of artificial intelligence (AI) that focuses on how computers can automatically learn from experience and make decisions or predictions based on data and algorithms 22 , 23 . In recent years, the application of machine learning in medical diagnostics has been increasingly widespread 24 , 25 . This study aims to develop a machine learning model that analyzes texture features from MRI T1-weighted sequences for the early diagnosis of Parkinson's disease. This approach offers an objective diagnostic method and holds significant clinical relevance. Materials and Methods Data Acquisition The PPMI (Parkinson's Progression Markers Initiative) data is a global, long-term, prospective clinical research project initiated by the Michael J. Fox Foundation (MJFF) 26 . It aims to provide data for the discovery of early biomarkers and the monitoring of disease progression in Parkinson's disease (PD). The goal of this research initiative is to help scientists better understand the pathophysiology of Parkinson's disease by collecting extensive data, thereby advancing early diagnosis and personalized treatment 27,28 . We applied for and were granted access to the PPMI data(https://www.ppmi-info.org). In the PPMI database, 67 patients of Parkinson's disease (PD) and 22 healthy controls were selected from those enrolled between January 1, 2021, and October 1, 2024. The participants met the following criteria: 1) Underwent MRI T1-weighted imaging; 2) Had relatively complete registration information; 3) Showed no apparent abnormalities on MRI images; 4) Had high-quality MRI images with no artifacts. We downloaded the MRI T1 DICOM files for these participants and converted them to Nii format for subsequent preprocessing. Additionally, we collected basic demographic and clinical information, including sex, age, Moca score, H-Y stage and MDS-UPDRS scores. Image Preprocessing To reduce image quality errors caused by scanner artifacts or other factors, we performed N4 bias field correction on all MRI T1-weighted images(Figure 1). N4 bias field correction is a widely used technique in medical image processing, primarily aimed at correcting the inhomogeneities in MRI images, commonly referred to as bias field effects. These effects are typically caused by non-uniformities in the magnetic field distribution, resulting in uneven intensity distributions in the image, which can subsequently affect the accuracy of image segmentation and subsequent analyses 29–31 . Since each subject's brain MRI is unique, in order to more easily study the relationship between different brain regions and Parkinson's patients, we registered each subject's MRI to the MNI152_1mm template. MNI152 is a standard brain template based on the Montreal Neurological Institute (MNI) space, created by spatially registering, averaging, and normalizing brain imaging data from 152 healthy adults. The MNI152 template can be considered an “average brain”, representing the approximate average morphology of all the samples. This template can be downloaded from the ICBM official website (https:// www.bic.mni.mcgill.ca/ ServicesAtlases/ ICBM152NLin2009) 32,33 , where we selected the 1*1*1 mm resolution template. We registered each subject's MRI to the MNI152_1mm template using the Elastix module in 3D-Slicer version 5.6.2. At the same time, we used the AAL116_1mm template to label the MNI152_1mm template. Extraction of Texture Features Based on prior research, we selected brain regions closely associated with the pathophysiology of Parkinson’s disease, including the thalamus 34–37 , hippocampus 38–40 , caudate nucleus 41 , amygdala 42,43 , Globus pallidus and putamen 44 . For each of these regions, corresponding masks were constructed using the RestPlus data analysis package in Matlab 2020. Specifically, the AAL116 anatomical atlas (provided by the Montreal Neurological Institute, MNI) was used as a basis, and the Image Calculator function in RestPlus was employed to extract the relevant regions, which were then used as masks. Additionally, AAL116 was selected as the functional region label and co-registered with the MNI152 standard brain template. Texture features were subsequently extracted from the pre-registered MRI images. These features included first-order statistical features, second-order texture features, and higher-order texture features derived from wavelet transform 45–47 . The texture feature extraction process was implemented using the Radiomics function in Python, ensuring both efficiency and precision in the data processing. Feature Selection and Machine Learning Classification Before performing machine learning modeling and prediction, we conducted an initial feature selection process on the extracted texture features. First, we applied Student's t-test to assess the significance of each feature. Based on the absolute values of the t-statistics, we selected the top 10% of features. Next, we employed Lasso regression with 5-fold cross-validation to determine the optimal regularization parameter, and then selected the non-zero coefficient features as the most important ones. Finally, the selected features were used for machine learning modeling in Python 3.7 version. The models chosen for this study included K-Nearest Neighbors (KNN) 48 , Support Vector Machine (SVM) 49 , Random Forest (RF) 50 , Logistic Regression (LR) 51 , and Naive Bayes (NB) 52 . To stabilize the results, the SMOTE library in Python was used to balance the data .We compared and analyzed the results of different models, plotted the ROC curves, and visualized the important features selected by Lasso. Additionally, the models were evaluated using performance metrics such as the area under the ROC curve (AUC), accuracy, precision,sensitivity, specificity and F1 score. Developing a Simplified Model In the modeling process, the initial features we used included all relevant features of the thalamus, hippocampus, caudate nucleus, amygdala, and putamen. To simplify the diagnostic process, we selected the top ten features with the highest absolute weight coefficients from the LASSO regression analysis to construct “simplified models”. We then compared those performance with that of the full models. Results Baseline characteristics In the image preprocessing stage, low-quality images with artifacts were excluded. A total of 34 patients of Parkinson's disease (PD) and 19 healthy controls (HC) were selected for the study, and their MRI T1-weighted images were analyzed. Additionally, clinical data including sex, age, Moca score, H-Y stage, and MDS-UPDRS scores were collected for all participants. The statistical analysis revealed a significant difference in the UPDRS II scores between PD patients and healthy controls (P < 0.001), with the detailed results presented in Table 1. Tabel 1. Basic information Feature Mean±Std Mean±Std P Sex M:20 F:14 M: 9 F:10 0.606 Age 63.264±10.924 64.947±7.524 0.554 Moca 26.382±2.243 27.158±1.537 0.187 H-Y 1.853±0.558 - - UPDRSⅠ 2.029±2.611 1.000±1.491 0.121 UPDRSⅡ 8.735±5.539 0.632±1.065 ＜0.001 UPDRSⅢ 23.206±10.959 - - UPDRSⅣ 1.118±2.772 - - Model evaluation We compared the performance of five machine learning models: KNN, SVM, RF, LR, and NB, using evaluation metrics including AUC, accuracy, precision, sensitivity, specificity and F1 score. The results showed that, except for the LR model with an AUC of 0.84 and the NB model with an AUC of 0.88, all other models had an AUC greater than 0.90, with the KNN model performing the best, achieving an AUC of 0.95. The RF model excelled across all metrics, with an accuracy of 0.889, precision of 0.923, sensitivity of 0.923, specificity of 0.800, and F1 score of 0.923, demonstrating superior performance (Table 2). Table 2.Model Evaluation Results AUC Accuracy Precision Sensitivity Specificity F1 score KNN 0.95 0.833 0.846 0.917 0.667 0.880 SVM 0.92 0.833 0.833 0.909 0.714 0.870 RF 0.90 0.889 0.923 0.923 0.800 0.923 LR 0.84 0.778 0.769 0.909 0.571 0.833 NB 0.88 0.833 0.923 0.857 0.750 0.889 Results presentation of each model In the KNN algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure4A) was employed to determine the optimal regularization parameter, which was found to be λ = 0.01105. At this optimal λ value, 30 non-zero coefficient features were further selected (Figure4B), which represent the most important features, and their visualization is shown in Figure4C. These important features include: Amygdala-related features: wavelet-LLL_firstorder_10Percentile_AMYG wavelet-LLL_glcm_Correlation_AMYG log-sigma-2-0-mm-3D_firstorder_Kurtosis_AMYG log-sigma-5-0-mm-3D_gldm_DependenceVariance_AMYG log-sigma-2-0-mm-3D_glcm_Idmn_AMYG wavelet-HLH_firstorder_Median_AMYG wavelet-LHL_firstorder_Mean_AMYG wavelet-LLH_firstorder_Skewness_AMYG wavelet-HLH_glszm_SmallAreaEmphasis_AMYG log-sigma-5-0-mm-3D_glszm_GrayLevelNonUniformity_AMYG Thalamus-related features: log-sigma-3-0-mm-3D_glrlm_GrayLevelNonUniformity_THA wavelet-LHL_gldm_SmallDependenceLowGrayLevelEmphasis_THA log-sigma-2-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA log-sigma-2-0-mm-3D_glszm_LowGrayLevelZoneEmphasis_THA Putamen-related features: log-sigma-5-0-mm-3D_gldm_DependenceNonUniformity_PUT wavelet-HHL_glcm_Imc1_PUT log-sigma-2-0-mm-3D_firstorder_Skewness_PUT log-sigma-2-0-mm-3D_glszm_SizeZoneNonUniformityNormalized_PUT wavelet-LLL_glcm_Correlation_PUT wavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis_PUT wavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis_PUT wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis_PUT Hippocampus-related features: log-sigma-5-0-mm-3D_glcm_Imc1_HIP log-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformityNormalized_HIP original_glcm_InverseVariance_HIP log-sigma-5-0-mm-3D_firstorder_10Percentile_HIP Globus pallidus-related features: log-sigma-5-0-mm-3D_glcm_ClusterShade_PAL log-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL wavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis_PAL Caudate nucleus-related features: wavelet-LHH_glcm_Idmn_CAU In the SVM algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure5A) was used to determine the optimal regularization parameter, which was found to be λ = 0.01650. At this optimal λ value, 30 non-zero coefficient features were further selected (Figure5B), which represent the most important features, and their visualization is shown in Figure5C. These important features include: Putamen-related features log-sigma-5-0-mm-3D_gldm_DependenceNonUniformity_PUT log-sigma-5-0-mm-3D_gldm_DependenceNonUniformityNormalized_PUT wavelet-HHL_glcm_Imc1_PUT log-sigma-2-0-mm-3D_firstorder_Skewness_PUT log-sigma-2-0-mm-3D_glszm_SizeZoneNonUniformityNormalized_PUT wavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis_PUT Thalamus-related features log-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA log-sigma-5-0-mm-3D_gldm_DependenceVariance_THA wavelet-LLL_firstorder_Energy_THA wavelet-LLL_glcm_MaximumProbability_THA wavelet-LLL_firstorder_TotalEnergy_THA log-sigma-5-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA log-sigma-5-0-mm-3D_gldm_SmallDependenceEmphasis_THA wavelet-HHL_glcm_Imc1_THA Globus pallidus-related features wavelet-HHL_glcm_Imc1_PAL wavelet-LHL_glszm_SmallAreaEmphasis_PAL log-sigma-5-0-mm-3D_glcm_ClusterShade_PAL log-sigma-5-0-mm-3D_glszm_ZoneEntropy_PAL original_firstorder_Median_PAL wavelet-HHH_glszm_SizeZoneNonUniformityNormalized_PAL Hippocampus-related features wavelet-HHL_firstorder_Mean_HIP wavelet-LLL_glcm_Imc1_HIP log-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformityNormalized_HIP wavelet-HHH_glcm_ClusterShade_HIP wavelet-LHH_firstorder_Skewness_HIP Amygdala-related features wavelet-HHH_firstorder_Skewness_AMYG wavelet-LHL_firstorder_Kurtosis_AMYG wavelet-HLH_firstorder_Median_AMYG wavelet-LHL_glcm_Idn_AMYG wavelet-LLH_firstorder_Skewness_AMYG In the RF algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure6A) was used to determine the optimal regularization parameter, which was found to be λ = 0.01444. At this optimal λ value, 26 non-zero coefficient features were further selected (Figure6B), representing the most important features, and their visualization is shown in Figure6C. These important features include: Thalamus-related features log-sigma-3-0-mm-3D_glszm_SizeZoneNonUniformity_THA log-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA log-sigma-2-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_THA wavelet-LLH_glcm_InverseVariance_THA wavelet-LLH_glcm_ClusterShade_THA log-sigma-2-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA Globus pallidus-related features log-sigma-5-0-mm-3D_glcm_ClusterShade_PAL log-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL wavelet-HHH_glszm_SizeZoneNonUniformityNormalized_PAL wavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis_PAL Putamen-related features log-sigma-2-0-mm-3D_firstorder_Skewness_PUT log-sigma-5-0-mm-3D_glcm_Idn_PUT wavelet-HLH_firstorder_Kurtosis_PUT wavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis_PUT wavelet-LLL_firstorder_Energy_PUT wavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis_PUT Hippocampus-related features log-sigma-2-0-mm-3D_glcm_Imc1_HIP wavelet-HHH_glcm_ClusterShade_HIP Caudate nucleus-related features wavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis_CAU wavelet-HHH_firstorder_Kurtosis_CAU Amygdala-related features wavelet-HHL_glcm_Imc1_PAL wavelet-LLL_firstorder_Minimum_AMYG log-sigma-2-0-mm-3D_glcm_Idmn_AMYG wavelet-LHL_firstorder_Kurtosis_AMYG wavelet-LLL_firstorder_Maximum_AMYG wavelet-HLH_glszm_SizeZoneNonUniformityNormalized_AMYG In the LR algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting,5-fold cross-validation (Figure7A) was used to determine the optimal regularization parameter, which was found to be λ = 0.01263. At this optimal λ value, 32 non-zero coefficient features were further selected ( Figure7B), representing the most important features, and their visualization is shown in Figure7C. These important features include: Hippocampus-related features ： log-sigma-5-0-mm-3D_firstorder_10Percentile_HIP Putamen-related features ： wavelet-HHL_glcm_Imc1_PUT log-sigma-2-0-mm-3D_firstorder_Skewness_PUT wavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis_PUT log-sigma-2-0-mm-3D_glszm_SizeZoneNonUniformityNormalized_PUT Thalamus-related features ： log-sigma-3-0-mm-3D_glszm_SizeZoneNonUniformity_THA log-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA wavelet-LLL_firstorder_Minimum_THA wavelet-LLL_glcm_MaximumProbability_THA log-sigma-2-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_THA log-sigma-3-0-mm-3D_glszm_SmallAreaHighGrayLevelEmphasis_THA log-sigma-2-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA Globus pallidus-related features ： wavelet-LHL_glszm_SmallAreaEmphasis_PAL log-sigma-5-0-mm-3D_glcm_ClusterShade_PAL log-sigma-3-0-mm-3D_glcm_Correlation_PAL log-sigma-4-0-mm-3D_firstorder_RobustMeanAbsoluteDeviation_PAL original_glrlm_RunEntropy_PAL original_firstorder_Median_PAL wavelet-LLL_glrlm_RunEntropy_PAL log-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL wavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis_PAL log-sigma-5-0-mm-3D_glrlm_LongRunEmphasis_PAL Amygdala-related features ： wavelet-LHL_glcm_Idmn_AMYG log-sigma-5-0-mm-3D_gldm_DependenceNonUniformityNormalized_AMYG wavelet-LHL_firstorder_Kurtosis_AMYG log-sigma-4-0-mm-3D_glszm_ZoneEntropy_AMYG wavelet-HLH_firstorder_Median_AMYG wavelet-LHL_firstorder_Mean_AMYG wavelet-LLL_gldm_DependenceNonUniformityNormalized_AMYG wavelet-LHL_gldm_LowGrayLevelEmphasis_AMYG wavelet-LHL_glrlm_LowGrayLevelRunEmphasis_AMYG Caudate nucleus-related features ： wavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis_CAU In the NB algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure8A) was used to determine the optimal regularization parameter, which was found to be λ = 0.02154. At this optimal λ value, 26 non-zero coefficient features were further selected (Figure8B), representing the most important features, and their visualization is shown in Figure8C. These important features include: Putamen-related features log-sigma-5-0-mm-3D_glcm_Idn_PUT log-sigma-2-0-mm-3D_firstorder_Skewness_PUT wavelet-HLH_firstorder_Kurtosis_PUT wavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis_PUT wavelet-HHL_glcm_Imc1_PUT Thalamus-related features log-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA wavelet-LLL_glcm_MaximumProbability_THA wavelet-LLH_glcm_ClusterShade_THA Amygdala-related features log-sigma-2-0-mm-3D_glcm_Idmn_AMYG wavelet-HHH_firstorder_Skewness_AMYG wavelet-LLL_firstorder_Minimum_AMYG wavelet-LHL_glcm_Idmn_AMYG wavelet-LHL_firstorder_Kurtosis_AMYG wavelet-HLH_firstorder_Median_AMYG wavelet-LHL_gldm_LowGrayLevelEmphasis_AMYG Globus pallidus-related features log-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL log-sigma-5-0-mm-3D_glcm_ClusterShade_PAL wavelet-HHH_glszm_SmallAreaEmphasis_PAL wavelet-LHL_firstorder_Median_PAL wavelet-HHL_glcm_Imc1_PAL Hippocampus-related features log-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformityNormalized_HIP original_glcm_InverseVariance_HIP log-sigma-5-0-mm-3D_firstorder_10Percentile_HIP wavelet-LHH_firstorder_Skewness_HIP Caudate nucleus-related features wavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis_CAU wavelet-LHH_gldm_LargeDependenceHighGrayLevelEmphasis_CAU Simplified Models Evaluation We selected the top ten features with the highest absolute weight coefficients, as determined by t-tests and Lasso regression analysis for each model, to build simplified models.The results are presented in Table 3. The findings indicate that both the RF and SVM models achieved AUC values exceeding 0.8, while the AUC values of the remaining models also reached 0.7. The RF model, which performed well in the full models, continued to perform excellently in the simplified models, achieving a prediction accuracy of 0.778, still making it the best-performing model. Table3.Simple Model Evaluation AUC Accuracy Precision Sensitivity Specificity F1 score KNN 0.73 0.722 0.818 0.750 0.667 0.783 SVM 0.84 0.667 0.692 0.818 0.429 0.750 RF 0.88 0.778 0.769 0.909 0.571 0.833 LR 0.73 0.722 0.714 0.909 0.429 0.800 NB 0.70 0.722 0.846 0.786 0.500 0.815 Discussion The Parkinson's Progression Markers Initiative (PPMI) database is a global, multi-center clinical research project designed to collect and share data and biomarkers related to Parkinson's disease (PD), with the aim of advancing the understanding of the disease, improving early diagnosis, monitoring disease progression, and developing potential therapies 53 . Although numerous researchers have focused on identifying early clinical biomarkers for PD 54 – 56 , this area still faces significant challenges and gaps. In an effort to explore more objective and accessible early diagnostic methods, we utilized only MRI T1-weighted imaging in this study. This approach not only reduces the reliance on clinicians' personal experience but also eliminates the subjective bias introduced by scales (such as UPDRS, MoCA, H-Y, etc.) 57 – 59 . As a more objective clinical biomarker, imaging-based methods provide a powerful tool for early disease differentiation, making it an ideal choice.In this study, we first performed N4 bias field correction on the MRI images to minimize errors introduced by equipment variability. To facilitate feature extraction, we registered all subject MRI scans to a standard brain space (MNI space). Based on the MNI standard brain space, we extracted several key brain regions (such as the thalamus, hippocampus, caudate nucleus, putamen, amygdala, and globus pallidus) as regions of interest (ROIs). These ROIs were then paired with each registered subject's MRI scan, and texture features were extracted using the AAL116 label and analyzed with the radiomics library in Python. To stabilize the results, the SMOTE library in Python was also used to balance the data .This method has not been widely discussed in the existing literature and offers certain innovation and practicality. Through this approach, we not only effectively extract diagnostically valuable features from an imaging perspective but also provide a portable and comprehensive technological pathway for the early diagnosis and monitoring of Parkinson's disease. The application of machine learning in the medical field has been widely validated across various subdomains 60 – 62 , with its efficiency and accuracy being recognized by numerous researchers. This study aims to leverage machine learning models in conjunction with MRI T1-weighted imaging features for the early diagnosis of PD and has achieved promising diagnostic results. Prior to model training, we performed feature selection using T-tests and Lasso regression analysis, identifying key features associated with PD, which were then visualized.Experimental results indicate that, except for the LR model with an AUC of 0.84 and the NB model with an AUC of 0.88, all other models had an AUC greater than 0.90,with the KNN model performing the best, achieving an AUC of 0.95. Furthermore, the RF model outperformed other models across all evaluation metrics, with an accuracy of 0.889, precision of 0.923, sensitivity of 0.923, specificity of 0.800,and F1 score of 0.923, all of which were the highest among the models tested. These results suggest that the RF model exhibits significant advantages and reliability in the early diagnosis of PD.In the RF model, important features selected through Lasso regression primarily focused on the texture characteristics of multiple brain regions, including 6 thalamic features, 4 globus pallidus-related features, 6 putamen-related features, 2 hippocampal features, 2 caudate nucleus features, and 6 amygdala-related features. These features played a crucial role in the diagnostic process and provide valuable insights for future research. Through the extraction and analysis of these features, this study offers new approaches for the early diagnosis of Parkinson's disease and provides essential technical support and theoretical foundations for researchers in the field. Moreover, the simplified models, constructed using the top ten features with the highest absolute weight coefficients from each model, also demonstrated strong predictive capabilities, effectively streamlining our prediction process. This study has some limitations. Firstly, although we employed the MNI (Montreal Neurological Institute) standard brain space registration method, which significantly enhanced the efficiency of the study and improved the accuracy of extracting specific brain regions, this approach inevitably led to the loss of some feature information. For example, the volume of brain regions in MRI scans has been considered an important feature in previous studies; however, this information was not fully preserved during the MNI registration process. Secondly, this study did not consider the extraction of features from the substantia nigra, primarily because this region is difficult to extract directly in standard brain space 63 , 64 . However, the substantia nigra is a critical brain area closely associated with Parkinson’s disease, and omitting this region from feature extraction may impact the diagnostic accuracy of the model. Additionally, recent studies have increasingly focused on the relationship between the cerebellum and Parkinson’s disease 65 , 66 . While we previously analyzed cerebellar texture features, the results were not as expected, and therefore, the cerebellum was not further explored in this study. Finally, due to the relatively small sample size and the fact that data were only obtained from the PPMI database, future studies should aim to increase the sample size and conduct multi-center external validation to enhance the generalizability and reliability of the findings. Conclusion This study presents an early Parkinson's disease diagnostic model based on brain texture features and machine learning methods. By extracting texture features from T1-weighted images and combining them with various machine learning algorithms such as KNN, SVM, RF, LR, and NB, we successfully differentiated Parkinson's disease patients from healthy controls. The experimental results show that the proposed method effectively identifies early signs of Parkinson's disease, providing strong support for early diagnosis. A comparison of several machine learning algorithms (KNN, SVM, RF, LR, and NB) revealed that both the KNN and RF models demonstrated superior performance across various metrics, including AUC, accuracy, precision, sensitivity, and F1 score. These results suggest that the KNN and RF algorithms have a significant advantage in using brain texture features for the early diagnosis of Parkinson's disease, with high clinical applicability. This study shows that brain texture features, as a non-invasive and reliable type of imaging data, have broad potential for early diagnosis of neurodegenerative diseases. Particularly in the early stages of Parkinson's disease, texture features can reveal changes in the brain's microstructure, providing scientific evidence for early intervention and treatment of the disease. Early diagnosis is key to improving the quality of life for Parkinson's disease patients. The method demonstrated in this study, which combines brain texture features from T1-weighted images with machine learning algorithms, effectively distinguishes Parkinson's disease patients from healthy controls, suggesting its broad clinical application potential. Early detection and accurate diagnosis can facilitate early intervention and personalized treatment for Parkinson's disease, ultimately improving patient outcomes. Declarations Acknowledgments： The authors would like to thank the investigators of the original studies for providing the summary statistics of the PPMI database. Declaration of interest statement： The authors report no conflict of interest. Funding ： None. Consent for publication ： No consent for publication is needed. Author Contribution X.Y.H. was primarily responsible for conducting the main experiments and drafting the manuscript.Z.F.F., G.Q.Q., and L.D. assisted in the experimental procedures and contributed to the creation of figures and tables.M.Y.Z., C.X., and L.X.G. provided valuable input on the experimental design and offered constructive feedback on the manuscript revision. References Li, G.; Ma, J.; Cui, S.; He, Y.; Xiao, Q.; Liu, J.; Chen, S. Parkinson’s Disease in China: A Forty-Year Growing Track of Bedside Work. Transl Neurodegener 2019, 8 , 22. https://doi.org/10.1186/ s40035-019-0162-z. Parkinson, J. An Essay on the Shaking Palsy. JNP 2002, 14 (2), 223–236. https:// doi.org/ 10.1176/jnp.14.2.223. Jankovic, J. Parkinson’s Disease: Clinical Features and Diagnosis. J Neurol Neurosurg Psychiatry 2008, 79 (4), 368–376. https: //doi.org /10.1136/jnnp.2007.131045. Sun, J.; Lai, Z.; Ma, J.; Gao, L.; Chen, M.; Chen, J.; Fang, J.; Fan, Y.; Bao, Y.; Zhang, D.; Chan, P.; Yang, Q.; Ye, C.; Wu, T.; Ma, T. Quantitative Evaluation of Iron Content in Idiopathic Rapid Eye Movement Sleep Behavior Disorder. Mov Disord 2020, 35 (3), 478–485. https:// doi.org/10.1002/mds.27929. Santos-García, D.; Fuente-Fernández, R. de la. Impact of Non-Motor Symptoms on Health-Related and Perceived Quality of Life in Parkinson’s Disease. Journal of the Neurological Sciences 2013, 332 (1), 136–140. https://doi.org/10.1016/j.jns.2013.07.005. Armstrong, M. J.; Okun, M. S. Diagnosis and Treatment of Parkinson Disease: A Review. JAMA 2020, 323 (6), 548–560. https://doi.org/10.1001/jama.2019.22360. Whone, A. L.; Watts, R. L.; Stoessl, A. J.; Davis, M.; Reske, S.; Nahmias, C.; Lang, A. E.; Rascol, O.; Ribeiro, M. J.; Remy, P.; Poewe, W. H.; Hauser, R. A.; Brooks, D. J.; REAL-PET Study Group. Slower Progression of Parkinson’s Disease with Ropinirole versus Levodopa: The REAL-PET Study. Ann Neurol 2003, 54 (1), 93–101. https://doi.org/10.1002/ana.10609. Chen, W.; Xiao, Q.; Shao, M.; Feng, T.; Liu, W.-G.; Luo, X.-G.; Chen, X.-C.; Xie, A.-M.; Liu, C.-F.; Liu, Z.-G.; Liu, Y.-M.; Wang, J.; Chen, S.-D. Prevalence of Wearing-off and Dyskinesia among the Patients with Parkinson’s Disease on Levodopa Therapy: A Multi-Center Registry Survey in Mainland China. Transl Neurodegener 2014, 3 (1), 26. https://doi.org/10.1186/ 2047-9158-3-26. Woo, J.; Lau, E.; Ziea, E.; Chan, D. K. Y. Prevalence of Parkinson’s Disease in a Chinese Population. Acta Neurol Scand 2004, 109 (3), 228–231. https://doi.org/10.1034/ j.1600-0404.2003.00220.x. Chen, R. C.; Chang, S. F.; Su, C. L.; Chen, T. H.; Yen, M. F.; Wu, H. M.; Chen, Z. Y.; Liou, H. H. Prevalence, Incidence, and Mortality of PD: A Door-to-Door Survey in Ilan County, Taiwan. Neurology 2001, 57 (9), 1679–1686. https://doi.org/10.1212/wnl.57.9.1679. Yang, J.-X.; Chen, L. Economic Burden Analysis of Parkinson’s Disease Patients in China. Parkinsons Dis 2017, 2017 , 8762939. https://doi.org/10.1155/2017/8762939. Gelb, D. J.; Oliver, E.; Gilman, S. Diagnostic Criteria for Parkinson Disease. Archives of Neurology 1999, 56 (1), 33–39. https://doi.org/10.1001/archneur.56.1.33. Makkos, A.; Kovács, M.; Aschermann, Z.; Harmat, M.; Janszky, J.; Karádi, K.; Kovács, N. Are the MDS-UPDRS-Based Composite Scores Clinically Applicable? Mov Disord 2018, 33 (5), 835–839. https://doi.org/10.1002/mds.27303. Gallagher, D. A.; Goetz, C. G.; Stebbins, G.; Lees, A. J.; Schrag, A. Validation of the MDS-UPDRS Part I for Nonmotor Symptoms in Parkinson’s Disease. Mov Disord 2012, 27 (1), 79–83. https://doi.org/10.1002/mds.23939. Dalrymple-Alford, J. C.; MacAskill, M. R.; Nakas, C. T.; Livingston, L.; Graham, C.; Crucian, G. P.; Melzer, T. R.; Kirwan, J.; Keenan, R.; Wells, S.; Porter, R. J.; Watts, R.; Anderson, T. J. The MoCA: Well-Suited Screen for Cognitive Impairment in Parkinson Disease. Neurology 2010, 75 (19), 1717–1725. https://doi.org/10.1212/WNL.0b013e3181fc29c9. Seifert, K. D.; Wiener, J. I. The Impact of DaTscan on the Diagnosis and Management of Movement Disorders: A Retrospective Study. Am J Neurodegener Dis 2013, 2 (1), 29–34. Chen, Y.; Gao, C.; Sun, Q.; Pan, H.; Huang, P.; Ding, J.; Chen, S. MicroRNA-4639 Is a Regulator of DJ-1 Expression and a Potential Early Diagnostic Marker for Parkinson’s Disease. Front Aging Neurosci 2017, 9 , 232. https://doi.org/10.3389/fnagi.2017.00232. Chougar, L.; Faucher, A.; Faouzi, J.; Lejeune, F.-X.; Gama Lobo, G.; Jovanovic, C.; Cormier, F.; Dupont, G.; Vidailhet, M.; Corvol, J.-C.; Colliot, O.; Lehéricy, S.; Grabli, D.; Degos, B. Contribution of MRI for the Early Diagnosis of Parkinsonism in Patients with Diagnostic Uncertainty. Movement Disorders 2024, 39 (5), 825–835. https://doi.org/10.1002/ mds.29760. Chen, Y.; Qi, Y.; Li, T.; Lin, A.; Ni, Y.; Pu, R.; Sun, B. A More Objective PD Diagnostic Model: Integrating Texture Feature Markers of Cerebellar Gray Matter and White Matter through Machine Learning. Front Aging Neurosci 2024, 16 , 1393841. https://doi.org/10.3389/ fnagi.2024.1393841. Ryman, S. G.; Poston, K. L. MRI Biomarkers of Motor and Non-Motor Symptoms in Parkinson’s Disease. Parkinsonism Relat Disord 2020, 73 , 85–93. https://doi.org/10.1016/ j.parkreldis.2019.10.002. Castellano, G.; Bonilha, L.; Li, L. M.; Cendes, F. Texture Analysis of Medical Images. Clin Radiol 2004, 59 (12), 1061–1069. https://doi.org/10.1016/j.crad.2004.07.008. Manickam, P.; Mariappan, S. A.; Murugesan, S. M.; Hansda, S.; Kaushik, A.; Shinde, R.; Thipperudraswamy, S. P. Artificial Intelligence (AI) and Internet of Medical Things (IoMT) Assisted Biomedical Systems for Intelligent Healthcare. Biosensors (Basel) 2022, 12 (8), 562. https://doi.org/10.3390/bios12080562. Greener, J. G.; Kandathil, S. M.; Moffat, L.; Jones, D. T. A Guide to Machine Learning for Biologists. Nat Rev Mol Cell Biol 2022, 23 (1), 40–55. https://doi.org/10.1038/ s41580-021-00407-0. Deo, R. C. Machine Learning in Medicine. Circulation 2015, 132 (20), 1920–1930. https:// doi.org/10.1161/CIRCULATIONAHA.115.001593. Choi, R. Y.; Coyner, A. S.; Kalpathy-Cramer, J.; Chiang, M. F.; Campbell, J. P. Introduction to Machine Learning, Neural Networks, and Deep Learning. Transl Vis Sci Technol 2020, 9 (2), 14. https://doi.org/10.1167/tvst.9.2.14. The Parkinson Progression Marker Initiative (PPMI). Progress in neurobiology 2011, 95 (4). https://doi.org/10.1016/j.pneurobio.2011.09.005. A, S.; L, C.-M.; De, L.; Cm, F.; Y, M.; Pa, U.; H, N.; Rn, A.; Lm, C.; T, F.; D, G.; K, K.; K, M.; B, M.; Kl, P.; J, S.; T, S.; Cm, T.; D, W.; A, V.; Sh, C.; R, K.; C, C.-G.; Cs, C.; M, F.; Lma, O.; Sj, H.; T, S.; K, M.; C, S. Assessment of Heterogeneity among Participants in the Parkinson’s Progression Markers Initiative Cohort Using α-Synuclein Seed Amplification: A Cross-Sectional Study. The Lancet. Neurology 2023, 22 (5). https://doi.org/10.1016/S1474-4422(23)00109-6. K, T.; H, S.-T.; S, K.; L, Z.; M, M.; P, S.-F.; R, T. High-Throughput CSF Proteomics and Machine Learning to Identify Proteomic Signatures for Parkinson Disease Development and Progression. Neurology 2023, 101 (14). https://doi.org/10.1212/WNL.0000000000207725. Sled, J. G.; Zijdenbos, A. P.; Evans, A. C. A Nonparametric Method for Automatic Correction of Intensity Nonuniformity in MRI Data. IEEE Trans Med Imaging 1998, 17 (1), 87–97. https://doi.org/10.1109/42.668698. Vovk, U.; Pernus, F.; Likar, B. A Review of Methods for Correction of Intensity Inhomogeneity in MRI. IEEE Trans Med Imaging 2007, 26 (3), 405–421. https://doi.org/10.1109/ TMI.2006.891486. Belaroussi, B.; Milles, J.; Carme, S.; Zhu, Y. M.; Benoit-Cattin, H. Intensity Non-Uniformity Correction in MRI: Existing Methods and Their Validation. Med Image Anal 2006, 10 (2), 234–246. https://doi.org/10.1016/j.media.2005.09.004. VS Fonov, AC Evans, K Botteron, CR Almli, RC McKinstry, DL Collins and BDCG, Unbiased Average Age-Appropriate Atlases for Pediatric Studies, NeuroImage,Volume 54, Issue 1, January 2011, ISSN 1053–8119. VS Fonov, AC Evans, RC McKinstry, CR Almli and DL Collins, Unbiased Nonlinear Average Age-Appropriate Brain Templates from Birth to Adulthood, NeuroImage, Volume 47, Supplement 1, July 2009, Page S102 Organization for Human Brain Mapping 2009 Annual Meeting. Ye, Z.; Zhang, G.; Zhang, Y.; Li, S.; Liu, N.; Zhou, X.; Xiao, W.; Münte, T. F. The Role of the Subthalamic Nucleus in Sequential Working Memory in De Novo Parkinson’s Disease. Mov Disord 2021, 36 (1), 87–95. https://doi.org/10.1002/mds.28344. van Nuland, A. J. M.; den Ouden, H. E. M.; Zach, H.; Dirkx, M. F. M.; van Asten, J. J. A.; Scheenen, T. W. J.; Toni, I.; Cools, R.; Helmich, R. C. GABAergic Changes in the Thalamocortical Circuit in Parkinson’s Disease. Hum Brain Mapp 2020, 41 (4), 1017–1029. https://doi.org/10.1002/hbm.24857. Lauro, P. M.; Lee, S.; Akbar, U.; Asaad, W. F. Subthalamic-Cortical Network Reorganization during Parkinson’s Tremor. J Neurosci 2021, 41 (47), 9844–9858. https://doi.org/10.1523/ JNEUROSCI.0854-21.2021. Dirkx, M. F.; den Ouden, H. E. M.; Aarts, E.; Timmer, M. H. M.; Bloem, B. R.; Toni, I.; Helmich, R. C. Dopamine Controls Parkinson’s Tremor by Inhibiting the Cerebellar Thalamus. Brain 2017, 140 (3), 721–734. https://doi.org/10.1093/brain/aww331. Belloso-Iguerategui, A.; Zamarbide, M.; Merino-Galan, L.; Rodríguez-Chinchilla, T.; Gago, B.; Santamaria, E.; Fernández-Irigoyen, J.; Cotman, C. W.; Prieto, G. A.; Quiroga-Varela, A.; Rodríguez-Oroz, M. C. Hippocampal Synaptic Failure Is an Early Event in Experimental Parkinsonism with Subtle Cognitive Deficit. Brain 2023, 146 (12), 4949–4963. https:// doi.org/10.1093/brain/awad227. Le, L.; M, K.; H, P.; T, I. Hippocampal Functional Connectivity in Parkinson’s Disease. Neuro-degenerative diseases 2022, 22 (1). https://doi.org/10.1159/000526377. Camicioli, R.; Moore, M. M.; Kinney, A.; Corbridge, E.; Glassberg, K.; Kaye, J. A. Parkinson’s Disease Is Associated with Hippocampal Atrophy. Mov Disord 2003, 18 (7), 784–790. https://doi.org/10.1002/mds.10444. Elevated caudate connectivity in cognitively normal Parkinson’s disease patients - PubMed . https://pubmed.ncbi.nlm.nih.gov/33087833/ (accessed 2024-12-04). Zhang, L.; Zhang, P.; Dong, Q.; Zhao, Z.; Zheng, W.; Zhang, J.; Hu, X.; Yao, Z.; Hu, B. Fine-Grained Features Characterize Hippocampal and Amygdaloid Change Pattern in Parkinson’s Disease and Discriminate Cognitive-Deficit Subtype. CNS Neurosci Ther 2024, 30 (1), e14480. https://doi.org/10.1111/cns.14480. Common and distinct roles of amygdala subregional functional connectivity in non-motor symptoms of Parkinson’s disease - PubMed . https://pubmed.ncbi.nlm.nih.gov/36806219/ (accessed 2024-12-04). Sigirli, D.; Ozdemir, S. T.; Erer, S.; Sahin, I.; Ercan, I.; Ozpar, R.; Orun, M. O.; Hakyemez, B. Statistical Shape Analysis of Putamen in Early-Onset Parkinson’s Disease. Clin Neurol Neurosurg 2021, 209 , 106936. https://doi.org/10.1016/j.clineuro.2021.106936. Jiang, Z.; Yin, J.; Han, P.; Chen, N.; Kang, Q.; Qiu, Y.; Li, Y.; Lao, Q.; Sun, M.; Yang, D.; Huang, S.; Qiu, J.; Li, K. Wavelet Transformation Can Enhance Computed Tomography Texture Features: A Multicenter Radiomics Study for Grade Assessment of COVID-19 Pulmonary Lesions. Quant Imaging Med Surg 2022, 12 (10), 4758–4770. https://doi.org/10.21037/qims-22-252. Wavelet transforms in separation science for denoising and peak overlap detection . https://pubmed.ncbi.nlm.nih.gov/32108426/ (accessed 2024-12-04). Chaddad, A.; Daniel, P.; Niazi, T. Radiomics Evaluation of Histological Heterogeneity Using Multiscale Textures Derived From 3D Wavelet Transformation of Multispectral Images. Front Oncol 2018, 8 , 96. https://doi.org/10.3389/fonc.2018.00096. Wang, J.; Geng, X. Large Margin Weighted K-Nearest Neighbors Label Distribution Learning for Classification. IEEE Trans Neural Netw Learn Syst 2024, 35 (11), 16720–16732. https://doi.org/10.1109/TNNLS.2023.3297261. Wang, H.; Shao, Y.; Zhou, S.; Zhang, C.; Xiu, N. Support Vector Machine Classifier via L0/1 Soft-Margin Loss. IEEE Trans Pattern Anal Mach Intell 2022, 44 (10), 7253–7265. https://doi.org/10.1109/TPAMI.2021.3092177. A review on longitudinal data analysis with random forest . https://pubmed.ncbi.nlm.nih.gov/36653905/ (accessed 2024-12-04). Logistic Regression in Medical Research - https://pubmed.ncbi.nlm.nih.gov/33449558/ (accessed 2024-12-04). Sugahara, S.; Ueno, M. Exact Learning Augmented Naive Bayes Classifier. Entropy (Basel) 2021, 23 (12), 1703. https://doi.org/10.3390/e23121703. Parkinson Progression Marker Initiative. The Parkinson Progression Marker Initiative (PPMI). Prog Neurobiol 2011, 95 (4), 629–635. https://doi.org/10.1016/j.pneurobio.2011.09.005. Tsukita, K.; Sakamaki-Tsukita, H.; Kaiser, S.; Zhang, L.; Messa, M.; Serrano-Fernandez, P.; Takahashi, R. High-Throughput CSF Proteomics and Machine Learning to Identify Proteomic Signatures for Parkinson Disease Development and Progression. Neurology 2023, 101 (14), e1434–e1447. https://doi.org/10.1212/WNL.0000000000207725. Simuni, T.; Merchant, K.; Brumm, M. C.; Cho, H.; Caspell-Garcia, C.; Coffey, C. S.; Chahine, L. M.; Alcalay, R. N.; Nudelman, K.; Foroud, T.; Mollenhauer, B.; Siderowf, A.; Tanner, C.; Iwaki, H.; Sherer, T.; Marek, K.; Parkinson’s Progression Marker Initiative Authors. Longitudinal Clinical and Biomarker Characteristics of Non-Manifesting LRRK2 G2019S Carriers in the PPMI Cohort. NPJ Parkinsons Dis 2022, 8 (1), 140. https://doi.org/10.1038/s41531-022-00404-w. Koros, C.; Simitsi, A.-M.; Papagiannakis, N.; Bougea, A.; Prentakis, A.; Papadimitriou, D.; Pachi, I.; Beratis, I.; Stanitsa, E.; Angelopoulou, E.; Antonelou, R.; Bregianni, M.; Lourentzos, K.; Papageorgiou, S. G.; Bonakis, A.; Trapali, X. G.; Stamelou, M.; Stefanis, L. Serum Uric Acid as a Putative Biomarker in Prodromal Parkinson’s Disease: Longitudinal Data from the PPMI Study. J Parkinsons Dis 2023, 13 (5), 811–818. https://doi.org/10.3233/JPD-230007. Movement Disorder Society Task Force on Rating Scales for Parkinson’s Disease. The Unified Parkinson’s Disease Rating Scale (UPDRS): Status and Recommendations. Mov Disord 2003, 18 (7), 738–750. https://doi.org/10.1002/mds.10473. Goetz, C. G.; Poewe, W.; Rascol, O.; Sampaio, C.; Stebbins, G. T.; Counsell, C.; Giladi, N.; Holloway, R. G.; Moore, C. G.; Wenning, G. K.; Yahr, M. D.; Seidl, L.; Movement Disorder Society Task Force on Rating Scales for Parkinson’s Disease. Movement Disorder Society Task Force Report on the Hoehn and Yahr Staging Scale: Status and Recommendations. Mov Disord 2004, 19 (9), 1020–1028. https://doi.org/10.1002/mds.20213. Regnault, A.; Boroojerdi, B.; Meunier, J.; Bani, M.; Morel, T.; Cano, S. Does the MDS-UPDRS Provide the Precision to Assess Progression in Early Parkinson’s Disease? Learnings from the Parkinson’s Progression Marker Initiative Cohort. J Neurol 2019, 266 (8), 1927–1936. https:// doi.org/10.1007/s00415-019-09348-3. Deo, R. C. Machine Learning in Medicine. Circulation 2015, 132 (20), 1920–1930. https://doi.org/10.1161/CIRCULATIONAHA.115.001593. Goecks, J.; Jalili, V.; Heiser, L. M.; Gray, J. W. How Machine Learning Will Transform Biomedicine. Cell 2020, 181 (1), 92–101. https://doi.org/10.1016/j.cell.2020.03.022. May, M. Eight Ways Machine Learning Is Assisting Medicine. Nat Med 2021, 27 (1), 2–3. https://doi.org/10.1038/s41591-020-01197-2. Bae, Y. J.; Kim, J.-M.; Sohn, C.-H.; Choi, J.-H.; Choi, B. S.; Song, Y. S.; Nam, Y.; Cho, S. J.; Jeon, B.; Kim, J. H. Imaging the Substantia Nigra in Parkinson Disease and Other Parkinsonian Syndromes. Radiology 2021, 300 (2), 260–278. https://doi.org/10.1148/radiol.2021203341. Chu, Y.; Hirst, W. D.; Federoff, H. J.; Harms, A. S.; Stoessl, A. J.; Kordower, J. H. Nigrostriatal Tau Pathology in Parkinsonism and Parkinson’s Disease. Brain 2024, 147 (2), 444–457. https://doi.org/10.1093/brain/awad388. O’Callaghan, C.; Hornberger, M.; Balsters, J. H.; Halliday, G. M.; Lewis, S. J. G.; Shine, J. M. Cerebellar Atrophy in Parkinson’s Disease and Its Implication for Network Connectivity. Brain 2016, 139 (Pt 3), 845–855. https://doi.org/10.1093/brain/awv399. Li, T.; Le, W.; Jankovic, J. Linking the Cerebellum to Parkinson Disease: An Update. Nat Rev Neurol 2023, 19 (11), 645–654. https://doi.org/10.1038/s41582-023-00874-3. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5859401","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":414508956,"identity":"5adf69e1-8b8b-4b08-8e43-f80f6f5d7ac1","order_by":0,"name":"Yuhua Xu","email":"","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":false,"prefix":"","firstName":"Yuhua","middleName":"","lastName":"Xu","suffix":""},{"id":414508957,"identity":"abca2e01-ff82-4fc0-b21a-0cada356957b","order_by":1,"name":"Zhou Feifan","email":"","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":false,"prefix":"","firstName":"Zhou","middleName":"","lastName":"Feifan","suffix":""},{"id":414508958,"identity":"66457c62-f886-403c-9d9a-5f3a45c77ea7","order_by":2,"name":"Qianqian Gao","email":"","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":false,"prefix":"","firstName":"Qianqian","middleName":"","lastName":"Gao","suffix":""},{"id":414508960,"identity":"b4e7f189-c310-4ed5-ae91-8eb4cdc912f3","order_by":3,"name":"Yazhou Ma","email":"","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":false,"prefix":"","firstName":"Yazhou","middleName":"","lastName":"Ma","suffix":""},{"id":414508966,"identity":"bb3ab324-f3fd-4802-81f3-79e1c0cce5ca","order_by":4,"name":"Xin Chen","email":"","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":false,"prefix":"","firstName":"Xin","middleName":"","lastName":"Chen","suffix":""},{"id":414508969,"identity":"1763f3d3-aa39-4598-9f5f-35c049b231d1","order_by":5,"name":"Dong Li","email":"","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":false,"prefix":"","firstName":"Dong","middleName":"","lastName":"Li","suffix":""},{"id":414508972,"identity":"da9a5af3-51fe-4c5e-b193-bdbcba756ab3","order_by":6,"name":"Xuegan Lian","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAwElEQVRIiWNgGAWjYBACPhiDn5n54AOitLDBGJLtbMkGpGkxOM9jJkCcFon0aw8+tt2TMz7MYMbAUGMTTVgLz5lyw5ltxcZmhxnSHjAcS8ttIKiFvSdNmrctIXHbYYbjBowNh4nQwsyTJv0XqGVzM2ObBHFa2NuPSTMCtWxgZmYjUgvPGTbJnnMJxhKH2ZgNEojxC79E+jOJH2UJcvz95z8++FBjQ1gLAwMPUgQmEFYOAuwPiFM3CkbBKBgFIxcAAPzTN+GjzSQ9AAAAAElFTkSuQmCC","orcid":"","institution":"Third Affiliated Hospital,Soochow University","correspondingAuthor":true,"prefix":"","firstName":"Xuegan","middleName":"","lastName":"Lian","suffix":""}],"badges":[],"createdAt":"2025-01-19 12:23:14","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5859401/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5859401/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":76193192,"identity":"0f144197-430a-4508-8b5f-037a4424ffb8","added_by":"auto","created_at":"2025-02-13 09:56:09","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":292624,"visible":true,"origin":"","legend":"\u003cp\u003eOn the left is the original MRI; on the right is the MRI after N4 bias field correction.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/dd1f0cceb093390df3f5308d.png"},{"id":76193188,"identity":"a9312dda-25fc-4d55-a204-738a8dfa89b0","added_by":"auto","created_at":"2025-02-13 09:56:08","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":267518,"visible":true,"origin":"","legend":"\u003cp\u003eOn the left is before registration; on the right is after registration.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/bb5cfda242bf2a8b5fa4512a.png"},{"id":76193191,"identity":"0bd91dab-b289-4f4c-9570-27f94a8d3576","added_by":"auto","created_at":"2025-02-13 09:56:08","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":283869,"visible":true,"origin":"","legend":"\u003cp\u003eFlowchart of the research process\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/613b7023c3dcea1a88b616f0.png"},{"id":76193193,"identity":"6db745bc-912f-46e6-b92a-cc90ca543529","added_by":"auto","created_at":"2025-02-13 09:56:09","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":232769,"visible":true,"origin":"","legend":"\u003cp\u003eA. The relationship between the Mean Squared Error (MSE) and the regularization coefficient in the KNN model. Figure4B. The trend of feature coefficients selected by Lasso as the regularization coefficient changes. Figure4C. A bar chart of the importance of non-zero feature coefficients selected by Lasso. Figure4D. The ROC curve and Area Under the Curve (AUC) for KNN modeling based on the selected important features.\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/65eba45c7eb069c72c0b92cf.png"},{"id":76193237,"identity":"b2d27209-96f6-41ba-8a34-22179dfc9e1e","added_by":"auto","created_at":"2025-02-13 09:56:11","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":247906,"visible":true,"origin":"","legend":"\u003cp\u003eA. The relationship between the Mean Squared Error (MSE) and the regularization coefficient in the SVM model. Figure5B. The trend of feature coefficients selected by Lasso as the regularization coefficient changes. Figure5C. A bar chart of the importance of non-zero feature coefficients selected by Lasso. Figure5D. The ROC curve and Area Under the Curve (AUC) for SVM modeling based on the selected important features.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/ba260b661e32ea35718cb5f3.png"},{"id":76194926,"identity":"581b8c11-7060-4410-a3c5-d263049097fa","added_by":"auto","created_at":"2025-02-13 10:12:09","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":237900,"visible":true,"origin":"","legend":"\u003cp\u003eA. The relationship between the Mean Squared Error (MSE) and the regularization coefficient in the RF model. Figure6B. The trend of feature coefficients selected by Lasso as the regularization coefficient changes. Figure6C. A bar chart of the importance of non-zero feature coefficients selected by Lasso. Figure6D. The ROC curve and Area Under the Curve (AUC) for RF modeling based on the selected important features.\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/8035dee7495482e12f6ffe49.png"},{"id":76194930,"identity":"16b7a21a-aede-456b-b709-4563b9a940fb","added_by":"auto","created_at":"2025-02-13 10:12:11","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":285543,"visible":true,"origin":"","legend":"\u003cp\u003eA. The relationship between the Mean Squared Error (MSE) and the regularization coefficient in the LR model. Figure7B. The trend of feature coefficients selected by Lasso as the regularization coefficient changes. Figure7C. A bar chart of the importance of non-zero feature coefficients selected by Lasso. Figure7D. The ROC curve and Area Under the Curve (AUC) for LR modeling based on the selected important features.\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/690e79bb5b3160182fee0137.png"},{"id":76193897,"identity":"89d2d633-20e1-4d2f-8195-fdbd5d9b82a4","added_by":"auto","created_at":"2025-02-13 10:04:09","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":251229,"visible":true,"origin":"","legend":"\u003cp\u003eA. The relationship between the Mean Squared Error (MSE) and the regularization coefficient in the NB model. Figure8B. The trend of feature coefficients selected by Lasso as the regularization coefficient changes. Figure8C. A bar chart of the importance of non-zero feature coefficients selected by Lasso. Figure8D. The ROC curve and Area Under the Curve (AUC) for NB modeling based on the selected important features.\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/3f9ecc1709a39dfa16293316.png"},{"id":95818662,"identity":"07a7089c-b496-4a63-8c0a-0a96ed1bbf62","added_by":"auto","created_at":"2025-11-13 10:27:09","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3308090,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5859401/v1/e72e76ae-ea87-460d-8eb2-2c6a279dbac5.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"A Machine Learning Approach for Early Parkinson's Disease Diagnosis Based on Brain Texture Features from T1-Weighted Imaging","fulltext":[{"header":"Introduction","content":"\u003cp\u003eParkinson's disease is a neurodegenerative disorder that not only profoundly impacts patients' quality of life but also imposes a substantial burden on China's economy and healthcare system, especially with the onset of an aging population\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. The disease was firstly reported by Jame Parkinson in 1817, so it was then named after him.\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. The clinical features of Parkinson's disease include both motor and non-motor symptoms. The primary motor symptoms consist of the following four: resting tremor, rigidity, bradykinesia, and postural instability\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e.Non-motor symptoms include olfactory dysfunction, cognitive impairment, depression, anxiety, autonomic dysfunction, and others\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. Over time, the symptoms of Parkinson's disease gradually worsen, and patients may also experience on-off phenomena, dyskinesia, and other complications\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e. The exact cause of Parkinson's disease is not yet fully understood. Possible contributing factors include genetics, environmental factors, age, and others. Currently, there is no cure for Parkinson's disease; treatment focuses on controlling symptoms to improve the patient's quality of life\u003csup\u003e\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u003c/sup\u003e. These treatment approaches include pharmacological therapy (such as levodopa), surgical interventions (such as deep brain stimulation), and rehabilitation exercises as adjunctive therapies.\u003csup\u003e\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e,\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u003c/sup\u003e. As mentioned above, Parkinson's disease patients experience numerous and severe symptoms, which significantly impact their quality of life and cause great distress to both the patients and their families\u003csup\u003e\u003cspan additionalcitationids=\"CR10\" citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Therefore, early diagnosis is especially important. It enables doctors to start treatment as soon as possible, slowing the progression of symptoms and providing the greatest benefit to the patients.\u003c/p\u003e \u003cp\u003eTo diagnose Parkinson's disease, clinicians typically need to gather a detailed medical history of the patient, observe their symptoms, and integrate the findings with additional auxiliary tests for a comprehensive diagnosis\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. The tools frequently used in diagnosis include clinical rating scales such as the UPDRS, MoCA, MMSE, and PDQ-39, as well as imaging examinations like cranial ultrasound, MRI, and SPECT\u003csup\u003e\u003cspan additionalcitationids=\"CR14 CR15\" citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. However, these tests are highly dependent on the clinician's experience and are quite subjective. Additionally, the early symptoms of Parkinson's disease are often subtle, which greatly increases the difficulty of early diagnosis. In recent years, there has been an increasing effort to find a more objective and accurate method. As a result, there has been growing exploration of clinical biomarkers for Parkinson's disease\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMRI is a commonly used imaging technique in clinical practice \u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e. Although Parkinson's disease cannot be directly diagnosed through MRI, studies have shown that analysis of MRI texture features can facilitate the early detection of Parkinson's disease\u003csup\u003e\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e,\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e\u003c/sup\u003e. Texture features refer to the spatial arrangement of pixels and the distribution pattern of gray levels in an image. By extracting and analyzing these texture features, computers can understand the surface structure or surface properties of objects in the image, such as smoothness, roughness, periodicity, and so on\u003csup\u003e\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e\u003c/sup\u003e. Machine learning is an important branch of artificial intelligence (AI) that focuses on how computers can automatically learn from experience and make decisions or predictions based on data and algorithms\u003csup\u003e\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e,\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u003c/sup\u003e. In recent years, the application of machine learning in medical diagnostics has been increasingly widespread \u003csup\u003e\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e,\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e\u003c/sup\u003e. This study aims to develop a machine learning model that analyzes texture features from MRI T1-weighted sequences for the early diagnosis of Parkinson's disease. This approach offers an objective diagnostic method and holds significant clinical relevance.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003ch3\u003eData Acquisition\u003c/h3\u003e\n\u003cp\u003eThe PPMI (Parkinson\u0026apos;s Progression Markers Initiative) data is a global, long-term, prospective clinical research project initiated by the Michael J. Fox Foundation (MJFF)\u003csup\u003e26\u003c/sup\u003e. It aims to provide data for the discovery of early biomarkers and the monitoring of disease progression in Parkinson\u0026apos;s disease (PD). The goal of this research initiative is to help scientists better understand the pathophysiology of Parkinson\u0026apos;s disease by collecting extensive data, thereby advancing early diagnosis and personalized treatment\u003csup\u003e27,28\u003c/sup\u003e. We applied for and were granted access to the PPMI data(https://www.ppmi-info.org). In the PPMI database, 67 patients of Parkinson\u0026apos;s disease (PD) and 22 healthy controls were selected from those enrolled between January 1, 2021, and October 1, 2024. The participants met the following criteria: 1) Underwent MRI T1-weighted imaging; 2) Had relatively complete registration information; 3) Showed no apparent abnormalities on MRI images; 4) Had high-quality MRI images with no artifacts. We downloaded the MRI T1 DICOM files for these participants and converted them to Nii format for subsequent preprocessing. Additionally, we collected basic demographic and clinical information, including sex, age, Moca score, H-Y stage and MDS-UPDRS scores.\u003c/p\u003e\n\u003ch3\u003eImage Preprocessing\u003c/h3\u003e\n\u003cp\u003eTo reduce image quality errors caused by scanner artifacts or other factors, we performed N4 bias field correction on all MRI T1-weighted images(Figure 1). N4 bias field correction is a widely used technique in medical image processing, primarily aimed at correcting the inhomogeneities in MRI images, commonly referred to as bias field effects. These effects are typically caused by non-uniformities in the magnetic field distribution, resulting in uneven intensity distributions in the image, which can subsequently affect the accuracy of image segmentation and subsequent analyses\u003csup\u003e29\u0026ndash;31\u003c/sup\u003e.\u003c/p\u003e\n\u003cp\u003eSince each subject\u0026apos;s brain MRI is unique, in order to more easily study the relationship between different brain regions and Parkinson\u0026apos;s patients, we registered each subject\u0026apos;s MRI to the MNI152_1mm template. MNI152 is a standard brain template based on the Montreal Neurological Institute (MNI) space, created by spatially registering, averaging, and normalizing brain imaging data from 152 healthy adults. The MNI152 template can be considered an \u0026ldquo;average brain\u0026rdquo;, representing the approximate average morphology of all the samples. This template can be downloaded from the ICBM official website (https:// www.bic.mni.mcgill.ca/ ServicesAtlases/ ICBM152NLin2009)\u003csup\u003e32,33\u003c/sup\u003e, where we selected the 1*1*1 mm resolution template. We registered each subject\u0026apos;s MRI to the MNI152_1mm template using the Elastix module in 3D-Slicer version 5.6.2. At the same time, we used the AAL116_1mm template to label the MNI152_1mm template.\u003c/p\u003e\n\u003ch3\u003eExtraction of Texture Features\u003c/h3\u003e\n\u003cp\u003eBased on prior research, we selected brain regions closely associated with the pathophysiology of Parkinson\u0026rsquo;s disease, including the thalamus\u003csup\u003e34\u0026ndash;37\u003c/sup\u003e, hippocampus\u003csup\u003e38\u0026ndash;40\u003c/sup\u003e, caudate nucleus\u003csup\u003e41\u003c/sup\u003e, amygdala\u003csup\u003e42,43\u003c/sup\u003e,\u0026nbsp;Globus pallidus\u0026nbsp;and putamen\u003csup\u003e44\u003c/sup\u003e. For each of these regions, corresponding masks were constructed using the RestPlus data analysis package in Matlab 2020. Specifically, the AAL116 anatomical atlas (provided by the Montreal Neurological Institute, MNI) was used as a basis, and the Image Calculator function in RestPlus was employed to extract the relevant regions, which were then used as masks. Additionally, AAL116 was selected as the functional region label and co-registered with the MNI152 standard brain template. Texture features were subsequently extracted from the pre-registered MRI images. These features included first-order statistical features, second-order texture features, and higher-order texture features derived from wavelet transform\u003csup\u003e45\u0026ndash;47\u003c/sup\u003e. The texture feature extraction process was implemented using the Radiomics function in Python, ensuring both efficiency and precision in the data processing.\u003c/p\u003e\n\u003ch3\u003eFeature Selection and Machine Learning Classification\u003c/h3\u003e\n\u003cp\u003eBefore performing machine learning modeling and prediction, we conducted an initial feature selection process on the extracted texture features.\u0026nbsp;First, we applied\u0026nbsp;Student\u0026apos;s t-test\u0026nbsp;to assess the significance of each feature. Based on the absolute values of the t-statistics, we selected the top 10% of features. Next, we employed Lasso regression with 5-fold cross-validation to determine the optimal regularization parameter, and then selected the non-zero coefficient features as the most important ones. Finally, the selected features were used for machine learning modeling in Python 3.7 version. The models chosen for this study included K-Nearest Neighbors (KNN)\u003csup\u003e48\u003c/sup\u003e, Support Vector Machine (SVM)\u003csup\u003e49\u003c/sup\u003e, Random Forest (RF)\u003csup\u003e50\u003c/sup\u003e, Logistic Regression (LR)\u003csup\u003e51\u003c/sup\u003e, and Naive Bayes (NB)\u003csup\u003e52\u003c/sup\u003e.\u0026nbsp;To stabilize the results, the SMOTE library in Python was used to balance the data .We compared and analyzed the results of different models, plotted the ROC curves, and visualized the important features selected by Lasso. Additionally, the models were evaluated using performance metrics such as the area under the ROC curve (AUC), accuracy, precision,sensitivity, specificity and F1 score.\u003c/p\u003e\n\u003ch3\u003eDeveloping a Simplified Model\u003c/h3\u003e\n\u003cp\u003eIn the modeling process, the initial features we used included all relevant features of the thalamus, hippocampus, caudate nucleus, amygdala, and putamen. To simplify the diagnostic process, we selected the top ten features with the highest absolute weight coefficients from the LASSO regression analysis to construct \u0026ldquo;simplified models\u0026rdquo;. We then compared those performance with that of the full models.\u003c/p\u003e"},{"header":"Results","content":"\u003ch3\u003eBaseline characteristics\u003c/h3\u003e\n\u003cp\u003eIn the image preprocessing stage, low-quality images with artifacts were excluded. A total of 34 patients of Parkinson\u0026apos;s disease (PD) and 19 healthy controls (HC) were selected for the study, and their MRI T1-weighted images were analyzed. Additionally, clinical data including sex, age, Moca score, H-Y stage, and MDS-UPDRS scores were collected for all participants. The statistical analysis revealed a significant difference in the UPDRS II scores between PD patients and healthy controls (P \u0026lt; 0.001), with the detailed results presented in Table 1.\u003c/p\u003e\n\u003cp\u003eTabel 1. Basic information\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\u003cbr\u003e\u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eFeature\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eMean\u0026plusmn;Std\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eMean\u0026plusmn;Std\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eP\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eSex\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eM:20\u003c/p\u003e\n \u003cp\u003eF:14\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eM: 9\u003c/p\u003e\n \u003cp\u003eF:10\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e0.606\u003c/p\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eAge\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e63.264\u0026plusmn;10.924\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e64.947\u0026plusmn;7.524\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e0.554\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eMoca\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e26.382\u0026plusmn;2.243\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e27.158\u0026plusmn;1.537\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e0.187\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eH-Y\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e1.853\u0026plusmn;0.558\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eUPDRSⅠ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e2.029\u0026plusmn;2.611\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e1.000\u0026plusmn;1.491\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e0.121\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eUPDRSⅡ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e8.735\u0026plusmn;5.539\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e0.632\u0026plusmn;1.065\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e＜0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eUPDRSⅢ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e23.206\u0026plusmn;10.959\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003eUPDRSⅣ\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e1.118\u0026plusmn;2.772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 138px;\"\u003e\n \u003cp\u003e-\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003eModel evaluation\u003c/h3\u003e\n\u003cp\u003eWe compared the performance of five machine learning models: KNN, SVM, RF, LR, and NB, using evaluation metrics including AUC, accuracy, precision, sensitivity, specificity and F1 score.\u0026nbsp;The results showed that, except for the LR model with an AUC of 0.84 and the NB model with an AUC of 0.88, all other models had an AUC greater than 0.90, with the KNN model performing the best, achieving an AUC of 0.95. The RF model excelled across all metrics, with an accuracy of 0.889, precision of 0.923, sensitivity of 0.923, specificity of 0.800, and F1 score of 0.923, demonstrating superior performance (Table 2).\u003c/p\u003e\n\u003cp\u003eTable 2.Model Evaluation Results\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.846\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.917\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.880\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.870\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.889\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.923\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.923\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.800\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.923\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.778\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.923\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.857\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.889\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;\u003c/p\u003e\n\u003ch3\u003eResults presentation of each model\u003c/h3\u003e\n\u003cp\u003eIn the KNN algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure4A) was employed to determine the optimal regularization parameter, which was found to be \u0026lambda; = 0.01105.\u0026nbsp;At this optimal \u0026lambda; value, 30 non-zero coefficient features were further selected (Figure4B), which represent the most important features, and their visualization is shown in Figure4C. These important features include:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eAmygdala-related features:\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-LLL_firstorder_10Percentile_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glcm_Correlation_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_firstorder_Kurtosis_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_DependenceVariance_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glcm_Idmn_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_firstorder_Median_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Mean_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LLH_firstorder_Skewness_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_glszm_SmallAreaEmphasis_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glszm_GrayLevelNonUniformity_AMYG\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eThalamus-related features:\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glrlm_GrayLevelNonUniformity_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_gldm_SmallDependenceLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_LowGrayLevelZoneEmphasis_THA\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003ePutamen-related features:\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_DependenceNonUniformity_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_firstorder_Skewness_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_SizeZoneNonUniformityNormalized_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glcm_Correlation_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis_PUT\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eHippocampus-related features:\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_Imc1_HIP\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformityNormalized_HIP\u003c/li\u003e\n \u003cli\u003eoriginal_glcm_InverseVariance_HIP\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_firstorder_10Percentile_HIP\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eGlobus pallidus-related features:\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_ClusterShade_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eCaudate nucleus-related features:\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-LHH_glcm_Idmn_CAU\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eIn the SVM algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure5A) was used to determine the optimal regularization parameter, which was found to be \u0026lambda; = 0.01650. At this optimal \u0026lambda; value, 30 non-zero coefficient features were further selected (Figure5B), which represent the most important features, and their visualization is shown in Figure5C. These important features include:\u003c/p\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003ePutamen-related features\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_DependenceNonUniformity_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_DependenceNonUniformityNormalized_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_firstorder_Skewness_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_SizeZoneNonUniformityNormalized_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis_PUT\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003eThalamus-related features\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_DependenceVariance_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_Energy_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glcm_MaximumProbability_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_TotalEnergy_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_SmallDependenceEmphasis_THA\u003c/li\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_THA\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003eGlobus pallidus-related features\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_glszm_SmallAreaEmphasis_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_ClusterShade_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glszm_ZoneEntropy_PAL\u003c/li\u003e\n \u003cli\u003eoriginal_firstorder_Median_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_glszm_SizeZoneNonUniformityNormalized_PAL\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003eHippocampus-related features\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_firstorder_Mean_HIP\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glcm_Imc1_HIP\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformityNormalized_HIP\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_glcm_ClusterShade_HIP\u003c/li\u003e\n \u003cli\u003ewavelet-LHH_firstorder_Skewness_HIP\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003eAmygdala-related features\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHH_firstorder_Skewness_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Kurtosis_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_firstorder_Median_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_glcm_Idn_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LLH_firstorder_Skewness_AMYG\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eIn the RF algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure6A) was used to determine the optimal regularization parameter, which was found to be \u0026lambda; = 0.01444. At this optimal \u0026lambda; value, 26 non-zero coefficient features were further selected (Figure6B), representing the most important features, and their visualization is shown in Figure6C. These important features include:\u003c/p\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eThalamus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SizeZoneNonUniformity_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLH_glcm_InverseVariance_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLH_glcm_ClusterShade_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eGlobus pallidus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_ClusterShade_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_glszm_SizeZoneNonUniformityNormalized_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003ePutamen-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_firstorder_Skewness_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_Idn_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_firstorder_Kurtosis_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_Energy_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_glszm_SmallAreaLowGrayLevelEmphasis_PUT\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eHippocampus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glcm_Imc1_HIP\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_glcm_ClusterShade_HIP\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eCaudate nucleus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis_CAU\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_firstorder_Kurtosis_CAU\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eAmygdala-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_Minimum_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glcm_Idmn_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Kurtosis_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_Maximum_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_glszm_SizeZoneNonUniformityNormalized_AMYG\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eIn the LR algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting,5-fold cross-validation (Figure7A) was used to determine the optimal regularization parameter, which was found to be \u0026lambda; = 0.01263. At this optimal \u0026lambda; value, 32 non-zero coefficient features were further selected ( Figure7B), representing the most important features, and their visualization is shown in Figure7C. These important features include:\u003c/p\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eHippocampus-related features\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_firstorder_10Percentile_HIP\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003ePutamen-related features\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_firstorder_Skewness_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_glszm_SmallAreaLowGrayLevelEmphasis_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_SizeZoneNonUniformityNormalized_PUT\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eThalamus-related features\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SizeZoneNonUniformity_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_Minimum_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glcm_MaximumProbability_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SmallAreaHighGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glszm_LargeAreaHighGrayLevelEmphasis_THA\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eGlobus pallidus-related features\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-LHL_glszm_SmallAreaEmphasis_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_ClusterShade_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glcm_Correlation_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_firstorder_RobustMeanAbsoluteDeviation_PAL\u003c/li\u003e\n \u003cli\u003eoriginal_glrlm_RunEntropy_PAL\u003c/li\u003e\n \u003cli\u003eoriginal_firstorder_Median_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glrlm_RunEntropy_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glrlm_LongRunEmphasis_PAL\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eAmygdala-related features\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-LHL_glcm_Idmn_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_gldm_DependenceNonUniformityNormalized_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Kurtosis_AMYG\u003c/li\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_glszm_ZoneEntropy_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_firstorder_Median_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Mean_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_gldm_DependenceNonUniformityNormalized_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_gldm_LowGrayLevelEmphasis_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_glrlm_LowGrayLevelRunEmphasis_AMYG\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul\u003e\n \u003cli\u003e\u003cstrong\u003eCaudate nucleus-related features\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis_CAU\u003c/li\u003e\n\u003c/ol\u003e\n\u003cp\u003eIn the NB algorithm, we initially performed a T-test to preliminarily select 670 features. Subsequently, Lasso regression was applied for fitting, 5-fold cross-validation (Figure8A) was used to determine the optimal regularization parameter, which was found to be \u0026lambda; = 0.02154. At this optimal \u0026lambda; value, 26 non-zero coefficient features were further selected (Figure8B), representing the most important features, and their visualization is shown in Figure8C. These important features include:\u003c/p\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003ePutamen-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_Idn_PUT\u003c/li\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_firstorder_Skewness_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_firstorder_Kurtosis_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HLL_gldm_LargeDependenceHighGrayLevelEmphasis_PUT\u003c/li\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PUT\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eThalamus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-3-0-mm-3D_glszm_SmallAreaLowGrayLevelEmphasis_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_glcm_MaximumProbability_THA\u003c/li\u003e\n \u003cli\u003ewavelet-LLH_glcm_ClusterShade_THA\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eAmygdala-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-2-0-mm-3D_glcm_Idmn_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_firstorder_Skewness_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LLL_firstorder_Minimum_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_glcm_Idmn_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Kurtosis_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-HLH_firstorder_Median_AMYG\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_gldm_LowGrayLevelEmphasis_AMYG\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eGlobus pallidus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_gldm_SmallDependenceLowGrayLevelEmphasis_PAL\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_glcm_ClusterShade_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-HHH_glszm_SmallAreaEmphasis_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-LHL_firstorder_Median_PAL\u003c/li\u003e\n \u003cli\u003ewavelet-HHL_glcm_Imc1_PAL\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eHippocampus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003elog-sigma-4-0-mm-3D_glszm_GrayLevelNonUniformityNormalized_HIP\u003c/li\u003e\n \u003cli\u003eoriginal_glcm_InverseVariance_HIP\u003c/li\u003e\n \u003cli\u003elog-sigma-5-0-mm-3D_firstorder_10Percentile_HIP\u003c/li\u003e\n \u003cli\u003ewavelet-LHH_firstorder_Skewness_HIP\u003c/li\u003e\n\u003c/ol\u003e\n\u003cul start=\"12\"\u003e\n \u003cli\u003e\u003cstrong\u003eCaudate nucleus-related features\u003c/strong\u003e\u003c/li\u003e\n\u003c/ul\u003e\n\u003col\u003e\n \u003cli\u003ewavelet-HHL_gldm_SmallDependenceLowGrayLevelEmphasis_CAU\u003c/li\u003e\n \u003cli\u003ewavelet-LHH_gldm_LargeDependenceHighGrayLevelEmphasis_CAU\u003c/li\u003e\n\u003c/ol\u003e\n\u003ch3\u003eSimplified Models Evaluation\u003c/h3\u003e\n\u003cp\u003eWe selected the top ten features with the highest absolute weight coefficients, as determined by t-tests and Lasso regression analysis for each model, to build simplified models.The results are presented in Table 3. The findings indicate that both the RF and SVM models achieved AUC values exceeding 0.8, while the AUC values of the remaining models also reached 0.7. The RF model, which performed well in the full models, continued to perform excellently in the simplified models, achieving a prediction accuracy of 0.778, still making it the best-performing model.\u003c/p\u003e\n\u003cp\u003eTable3.Simple Model Evaluation\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003eAUC\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eSensitivity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003eSpecificity\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003eF1 score\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eKNN\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.722\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.818\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.783\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eSVM\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.84\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.667\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.692\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.818\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.429\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.750\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eRF\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.778\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.769\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.571\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.833\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eLR\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.73\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.722\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.714\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.909\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.429\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.800\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 77px;\"\u003e\n \u003cp\u003eNB\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 68px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.722\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 85px;\"\u003e\n \u003cp\u003e0.846\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.786\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 87px;\"\u003e\n \u003cp\u003e0.500\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 79px;\"\u003e\n \u003cp\u003e0.815\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e"},{"header":"Discussion","content":"\u003cp\u003eThe Parkinson's Progression Markers Initiative (PPMI) database is a global, multi-center clinical research project designed to collect and share data and biomarkers related to Parkinson's disease (PD), with the aim of advancing the understanding of the disease, improving early diagnosis, monitoring disease progression, and developing potential therapies\u003csup\u003e\u003cspan citationid=\"CR53\" class=\"CitationRef\"\u003e53\u003c/span\u003e\u003c/sup\u003e. Although numerous researchers have focused on identifying early clinical biomarkers for PD\u003csup\u003e\u003cspan additionalcitationids=\"CR55\" citationid=\"CR54\" class=\"CitationRef\"\u003e54\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR56\" class=\"CitationRef\"\u003e56\u003c/span\u003e\u003c/sup\u003e, this area still faces significant challenges and gaps. In an effort to explore more objective and accessible early diagnostic methods, we utilized only MRI T1-weighted imaging in this study. This approach not only reduces the reliance on clinicians' personal experience but also eliminates the subjective bias introduced by scales (such as UPDRS, MoCA, H-Y, etc.)\u003csup\u003e\u003cspan additionalcitationids=\"CR58\" citationid=\"CR57\" class=\"CitationRef\"\u003e57\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR59\" class=\"CitationRef\"\u003e59\u003c/span\u003e\u003c/sup\u003e. As a more objective clinical biomarker, imaging-based methods provide a powerful tool for early disease differentiation, making it an ideal choice.In this study, we first performed N4 bias field correction on the MRI images to minimize errors introduced by equipment variability. To facilitate feature extraction, we registered all subject MRI scans to a standard brain space (MNI space). Based on the MNI standard brain space, we extracted several key brain regions (such as the thalamus, hippocampus, caudate nucleus, putamen, amygdala, and globus pallidus) as regions of interest (ROIs). These ROIs were then paired with each registered subject's MRI scan, and texture features were extracted using the AAL116 label and analyzed with the radiomics library in Python. To stabilize the results, the SMOTE library in Python was also used to balance the data .This method has not been widely discussed in the existing literature and offers certain innovation and practicality. Through this approach, we not only effectively extract diagnostically valuable features from an imaging perspective but also provide a portable and comprehensive technological pathway for the early diagnosis and monitoring of Parkinson's disease.\u003c/p\u003e \u003cp\u003eThe application of machine learning in the medical field has been widely validated across various subdomains\u003csup\u003e\u003cspan additionalcitationids=\"CR61\" citationid=\"CR60\" class=\"CitationRef\"\u003e60\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR62\" class=\"CitationRef\"\u003e62\u003c/span\u003e\u003c/sup\u003e, with its efficiency and accuracy being recognized by numerous researchers. This study aims to leverage machine learning models in conjunction with MRI T1-weighted imaging features for the early diagnosis of PD and has achieved promising diagnostic results. Prior to model training, we performed feature selection using T-tests and Lasso regression analysis, identifying key features associated with PD, which were then visualized.Experimental results indicate that, except for the LR model with an AUC of 0.84 and the NB model with an AUC of 0.88, all other models had an AUC greater than 0.90,with the KNN model performing the best, achieving an AUC of 0.95. Furthermore, the RF model outperformed other models across all evaluation metrics, with an accuracy of 0.889, precision of 0.923, sensitivity of 0.923, specificity of 0.800,and F1 score of 0.923, all of which were the highest among the models tested. These results suggest that the RF model exhibits significant advantages and reliability in the early diagnosis of PD.In the RF model, important features selected through Lasso regression primarily focused on the texture characteristics of multiple brain regions, including 6 thalamic features, 4 globus pallidus-related features, 6 putamen-related features, 2 hippocampal features, 2 caudate nucleus features, and 6 amygdala-related features. These features played a crucial role in the diagnostic process and provide valuable insights for future research. Through the extraction and analysis of these features, this study offers new approaches for the early diagnosis of Parkinson's disease and provides essential technical support and theoretical foundations for researchers in the field. Moreover, the simplified models, constructed using the top ten features with the highest absolute weight coefficients from each model, also demonstrated strong predictive capabilities, effectively streamlining our prediction process.\u003c/p\u003e \u003cp\u003eThis study has some limitations. Firstly, although we employed the MNI (Montreal Neurological Institute) standard brain space registration method, which significantly enhanced the efficiency of the study and improved the accuracy of extracting specific brain regions, this approach inevitably led to the loss of some feature information. For example, the volume of brain regions in MRI scans has been considered an important feature in previous studies; however, this information was not fully preserved during the MNI registration process. Secondly, this study did not consider the extraction of features from the substantia nigra, primarily because this region is difficult to extract directly in standard brain space\u003csup\u003e\u003cspan citationid=\"CR63\" class=\"CitationRef\"\u003e63\u003c/span\u003e,\u003cspan citationid=\"CR64\" class=\"CitationRef\"\u003e64\u003c/span\u003e\u003c/sup\u003e. However, the substantia nigra is a critical brain area closely associated with Parkinson\u0026rsquo;s disease, and omitting this region from feature extraction may impact the diagnostic accuracy of the model. Additionally, recent studies have increasingly focused on the relationship between the cerebellum and Parkinson\u0026rsquo;s disease\u003csup\u003e\u003cspan citationid=\"CR65\" class=\"CitationRef\"\u003e65\u003c/span\u003e,\u003cspan citationid=\"CR66\" class=\"CitationRef\"\u003e66\u003c/span\u003e\u003c/sup\u003e. While we previously analyzed cerebellar texture features, the results were not as expected, and therefore, the cerebellum was not further explored in this study. Finally, due to the relatively small sample size and the fact that data were only obtained from the PPMI database, future studies should aim to increase the sample size and conduct multi-center external validation to enhance the generalizability and reliability of the findings.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003e \u003col\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThis study presents an early Parkinson's disease diagnostic model based on brain texture features and machine learning methods. By extracting texture features from T1-weighted images and combining them with various machine learning algorithms such as KNN, SVM, RF, LR, and NB, we successfully differentiated Parkinson's disease patients from healthy controls. The experimental results show that the proposed method effectively identifies early signs of Parkinson's disease, providing strong support for early diagnosis.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eA comparison of several machine learning algorithms (KNN, SVM, RF, LR, and NB) revealed that both the KNN and RF models demonstrated superior performance across various metrics, including AUC, accuracy, precision, sensitivity, and F1 score. These results suggest that the KNN and RF algorithms have a significant advantage in using brain texture features for the early diagnosis of Parkinson's disease, with high clinical applicability.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eThis study shows that brain texture features, as a non-invasive and reliable type of imaging data, have broad potential for early diagnosis of neurodegenerative diseases. Particularly in the early stages of Parkinson's disease, texture features can reveal changes in the brain's microstructure, providing scientific evidence for early intervention and treatment of the disease.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003cspan\u003e \u003cli\u003e \u003cp\u003eEarly diagnosis is key to improving the quality of life for Parkinson's disease patients. The method demonstrated in this study, which combines brain texture features from T1-weighted images with machine learning algorithms, effectively distinguishes Parkinson's disease patients from healthy controls, suggesting its broad clinical application potential. Early detection and accurate diagnosis can facilitate early intervention and personalized treatment for Parkinson's disease, ultimately improving patient outcomes.\u003c/p\u003e \u003c/li\u003e \u003c/span\u003e \u003c/ol\u003e \u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eAcknowledgments：\u003c/h2\u003e\n\u003cp\u003eThe authors would like to thank the investigators of the original studies for providing the summary statistics of the PPMI database.\u003c/p\u003e\n\u003ch2\u003eDeclaration of interest statement：\u003c/h2\u003e\n\u003cp\u003eThe authors report no conflict of interest.\u003c/p\u003e\n\u003ch2\u003eFunding ：\u003c/h2\u003e\n\u003cp\u003eNone.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eConsent for publication ：\u003c/h2\u003e\n\u003cp\u003eNo consent for publication is needed.\u0026nbsp;\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eX.Y.H. was primarily responsible for conducting the main experiments and drafting the manuscript.Z.F.F., G.Q.Q., and L.D. assisted in the experimental procedures and contributed to the creation of figures and tables.M.Y.Z., C.X., and L.X.G. provided valuable input on the experimental design and offered constructive feedback on the manuscript revision.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n\u003cli\u003eLi, G.; Ma, J.; Cui, S.; He, Y.; Xiao, Q.; Liu, J.; Chen, S. Parkinson\u0026rsquo;s Disease in China: A Forty-Year Growing Track of Bedside Work. \u003cem\u003eTransl Neurodegener\u003c/em\u003e 2019, \u003cem\u003e8\u003c/em\u003e, 22. https://doi.org/10.1186/ s40035-019-0162-z.\u003c/li\u003e\n\u003cli\u003eParkinson, J. An Essay on the Shaking Palsy. \u003cem\u003eJNP\u003c/em\u003e 2002, \u003cem\u003e14\u003c/em\u003e (2), 223\u0026ndash;236. https:// doi.org/ 10.1176/jnp.14.2.223.\u003c/li\u003e\n\u003cli\u003eJankovic, J. Parkinson\u0026rsquo;s Disease: Clinical Features and Diagnosis. \u003cem\u003eJ Neurol Neurosurg Psychiatry\u003c/em\u003e 2008, \u003cem\u003e79\u003c/em\u003e (4), 368\u0026ndash;376. https: //doi.org /10.1136/jnnp.2007.131045.\u003c/li\u003e\n\u003cli\u003eSun, J.; Lai, Z.; Ma, J.; Gao, L.; Chen, M.; Chen, J.; Fang, J.; Fan, Y.; Bao, Y.; Zhang, D.; Chan, P.; Yang, Q.; Ye, C.; Wu, T.; Ma, T. Quantitative Evaluation of Iron Content in Idiopathic Rapid Eye Movement Sleep Behavior Disorder. \u003cem\u003eMov Disord\u003c/em\u003e 2020, \u003cem\u003e35\u003c/em\u003e (3), 478\u0026ndash;485. https:// doi.org/10.1002/mds.27929.\u003c/li\u003e\n\u003cli\u003eSantos-Garc\u0026iacute;a, D.; Fuente-Fern\u0026aacute;ndez, R. de la. Impact of Non-Motor Symptoms on Health-Related and Perceived Quality of Life in Parkinson\u0026rsquo;s Disease. \u003cem\u003eJournal of the Neurological Sciences\u003c/em\u003e 2013, \u003cem\u003e332\u003c/em\u003e (1), 136\u0026ndash;140. https://doi.org/10.1016/j.jns.2013.07.005.\u003c/li\u003e\n\u003cli\u003eArmstrong, M. J.; Okun, M. S. Diagnosis and Treatment of Parkinson Disease: A Review. \u003cem\u003eJAMA\u003c/em\u003e 2020, \u003cem\u003e323\u003c/em\u003e (6), 548\u0026ndash;560. https://doi.org/10.1001/jama.2019.22360.\u003c/li\u003e\n\u003cli\u003eWhone, A. L.; Watts, R. L.; Stoessl, A. J.; Davis, M.; Reske, S.; Nahmias, C.; Lang, A. E.; Rascol, O.; Ribeiro, M. J.; Remy, P.; Poewe, W. H.; Hauser, R. A.; Brooks, D. J.; REAL-PET Study Group. Slower Progression of Parkinson\u0026rsquo;s Disease with Ropinirole versus Levodopa: The REAL-PET Study. \u003cem\u003eAnn Neurol\u003c/em\u003e 2003, \u003cem\u003e54\u003c/em\u003e (1), 93\u0026ndash;101. https://doi.org/10.1002/ana.10609.\u003c/li\u003e\n\u003cli\u003eChen, W.; Xiao, Q.; Shao, M.; Feng, T.; Liu, W.-G.; Luo, X.-G.; Chen, X.-C.; Xie, A.-M.; Liu, C.-F.; Liu, Z.-G.; Liu, Y.-M.; Wang, J.; Chen, S.-D. Prevalence of Wearing-off and Dyskinesia among the Patients with Parkinson\u0026rsquo;s Disease on Levodopa Therapy: A Multi-Center Registry Survey in Mainland China. \u003cem\u003eTransl Neurodegener\u003c/em\u003e 2014, \u003cem\u003e3\u003c/em\u003e (1), 26. https://doi.org/10.1186/ 2047-9158-3-26.\u003c/li\u003e\n\u003cli\u003eWoo, J.; Lau, E.; Ziea, E.; Chan, D. K. Y. Prevalence of Parkinson\u0026rsquo;s Disease in a Chinese Population. \u003cem\u003eActa Neurol Scand\u003c/em\u003e 2004, \u003cem\u003e109\u003c/em\u003e (3), 228\u0026ndash;231. https://doi.org/10.1034/ j.1600-0404.2003.00220.x.\u003c/li\u003e\n\u003cli\u003eChen, R. C.; Chang, S. F.; Su, C. L.; Chen, T. H.; Yen, M. F.; Wu, H. M.; Chen, Z. Y.; Liou, H. H. Prevalence, Incidence, and Mortality of PD: A Door-to-Door Survey in Ilan County, Taiwan. \u003cem\u003eNeurology\u003c/em\u003e 2001, \u003cem\u003e57\u003c/em\u003e (9), 1679\u0026ndash;1686. https://doi.org/10.1212/wnl.57.9.1679.\u003c/li\u003e\n\u003cli\u003eYang, J.-X.; Chen, L. Economic Burden Analysis of Parkinson\u0026rsquo;s Disease Patients in China. \u003cem\u003eParkinsons Dis\u003c/em\u003e 2017, \u003cem\u003e2017\u003c/em\u003e, 8762939. https://doi.org/10.1155/2017/8762939.\u003c/li\u003e\n\u003cli\u003eGelb, D. J.; Oliver, E.; Gilman, S. Diagnostic Criteria for Parkinson Disease. \u003cem\u003eArchives of Neurology\u003c/em\u003e 1999, \u003cem\u003e56\u003c/em\u003e (1), 33\u0026ndash;39. https://doi.org/10.1001/archneur.56.1.33.\u003c/li\u003e\n\u003cli\u003eMakkos, A.; Kov\u0026aacute;cs, M.; Aschermann, Z.; Harmat, M.; Janszky, J.; Kar\u0026aacute;di, K.; Kov\u0026aacute;cs, N. Are the MDS-UPDRS-Based Composite Scores Clinically Applicable? \u003cem\u003eMov Disord\u003c/em\u003e 2018, \u003cem\u003e33\u003c/em\u003e (5), 835\u0026ndash;839. https://doi.org/10.1002/mds.27303.\u003c/li\u003e\n\u003cli\u003eGallagher, D. A.; Goetz, C. G.; Stebbins, G.; Lees, A. J.; Schrag, A. Validation of the MDS-UPDRS Part I for Nonmotor Symptoms in Parkinson\u0026rsquo;s Disease. \u003cem\u003eMov Disord\u003c/em\u003e 2012, \u003cem\u003e27\u003c/em\u003e (1), 79\u0026ndash;83. https://doi.org/10.1002/mds.23939.\u003c/li\u003e\n\u003cli\u003eDalrymple-Alford, J. C.; MacAskill, M. R.; Nakas, C. T.; Livingston, L.; Graham, C.; Crucian, G. P.; Melzer, T. R.; Kirwan, J.; Keenan, R.; Wells, S.; Porter, R. J.; Watts, R.; Anderson, T. J. The MoCA: Well-Suited Screen for Cognitive Impairment in Parkinson Disease. \u003cem\u003eNeurology\u003c/em\u003e 2010, \u003cem\u003e75\u003c/em\u003e (19), 1717\u0026ndash;1725. https://doi.org/10.1212/WNL.0b013e3181fc29c9.\u003c/li\u003e\n\u003cli\u003eSeifert, K. D.; Wiener, J. I. The Impact of DaTscan on the Diagnosis and Management of Movement Disorders: A Retrospective Study. \u003cem\u003eAm J Neurodegener Dis\u003c/em\u003e 2013, \u003cem\u003e2\u003c/em\u003e (1), 29\u0026ndash;34.\u003c/li\u003e\n\u003cli\u003eChen, Y.; Gao, C.; Sun, Q.; Pan, H.; Huang, P.; Ding, J.; Chen, S. MicroRNA-4639 Is a Regulator of DJ-1 Expression and a Potential Early Diagnostic Marker for Parkinson\u0026rsquo;s Disease. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e 2017, \u003cem\u003e9\u003c/em\u003e, 232. https://doi.org/10.3389/fnagi.2017.00232.\u003c/li\u003e\n\u003cli\u003eChougar, L.; Faucher, A.; Faouzi, J.; Lejeune, F.-X.; Gama Lobo, G.; Jovanovic, C.; Cormier, F.; Dupont, G.; Vidailhet, M.; Corvol, J.-C.; Colliot, O.; Leh\u0026eacute;ricy, S.; Grabli, D.; Degos, B. Contribution of MRI for the Early Diagnosis of Parkinsonism in Patients with Diagnostic Uncertainty. \u003cem\u003eMovement Disorders\u003c/em\u003e 2024, \u003cem\u003e39\u003c/em\u003e (5), 825\u0026ndash;835. https://doi.org/10.1002/ mds.29760.\u003c/li\u003e\n\u003cli\u003eChen, Y.; Qi, Y.; Li, T.; Lin, A.; Ni, Y.; Pu, R.; Sun, B. A More Objective PD Diagnostic Model: Integrating Texture Feature Markers of Cerebellar Gray Matter and White Matter through Machine Learning. \u003cem\u003eFront Aging Neurosci\u003c/em\u003e 2024, \u003cem\u003e16\u003c/em\u003e, 1393841. https://doi.org/10.3389/ fnagi.2024.1393841.\u003c/li\u003e\n\u003cli\u003eRyman, S. G.; Poston, K. L. MRI Biomarkers of Motor and Non-Motor Symptoms in Parkinson\u0026rsquo;s Disease. \u003cem\u003eParkinsonism Relat Disord\u003c/em\u003e 2020, \u003cem\u003e73\u003c/em\u003e, 85\u0026ndash;93. https://doi.org/10.1016/ j.parkreldis.2019.10.002.\u003c/li\u003e\n\u003cli\u003eCastellano, G.; Bonilha, L.; Li, L. M.; Cendes, F. Texture Analysis of Medical Images. \u003cem\u003eClin Radiol\u003c/em\u003e 2004, \u003cem\u003e59\u003c/em\u003e (12), 1061\u0026ndash;1069. https://doi.org/10.1016/j.crad.2004.07.008.\u003c/li\u003e\n\u003cli\u003eManickam, P.; Mariappan, S. A.; Murugesan, S. M.; Hansda, S.; Kaushik, A.; Shinde, R.; Thipperudraswamy, S. P. Artificial Intelligence (AI) and Internet of Medical Things (IoMT) Assisted Biomedical Systems for Intelligent Healthcare. \u003cem\u003eBiosensors (Basel)\u003c/em\u003e 2022, \u003cem\u003e12\u003c/em\u003e (8), 562. https://doi.org/10.3390/bios12080562.\u003c/li\u003e\n\u003cli\u003eGreener, J. G.; Kandathil, S. M.; Moffat, L.; Jones, D. T. A Guide to Machine Learning for Biologists.\u003cem\u003eNat Rev Mol Cell Biol\u003c/em\u003e 2022, \u003cem\u003e23\u003c/em\u003e (1), 40\u0026ndash;55. https://doi.org/10.1038/ s41580-021-00407-0.\u003c/li\u003e\n\u003cli\u003eDeo, R. C. Machine Learning in Medicine. \u003cem\u003eCirculation\u003c/em\u003e 2015, \u003cem\u003e132\u003c/em\u003e (20), 1920\u0026ndash;1930. https:// doi.org/10.1161/CIRCULATIONAHA.115.001593.\u003c/li\u003e\n\u003cli\u003eChoi, R. Y.; Coyner, A. S.; Kalpathy-Cramer, J.; Chiang, M. F.; Campbell, J. P. Introduction to Machine Learning, Neural Networks, and Deep Learning. \u003cem\u003eTransl Vis Sci Technol\u003c/em\u003e 2020, \u003cem\u003e9\u003c/em\u003e (2), 14. https://doi.org/10.1167/tvst.9.2.14.\u003c/li\u003e\n\u003cli\u003eThe Parkinson Progression Marker Initiative (PPMI). \u003cem\u003eProgress in neurobiology\u003c/em\u003e 2011, \u003cem\u003e95\u003c/em\u003e (4). https://doi.org/10.1016/j.pneurobio.2011.09.005.\u003c/li\u003e\n\u003cli\u003eA, S.; L, C.-M.; De, L.; Cm, F.; Y, M.; Pa, U.; H, N.; Rn, A.; Lm, C.; T, F.; D, G.; K, K.; K, M.; B, M.; Kl, P.; J, S.; T, S.; Cm, T.; D, W.; A, V.; Sh, C.; R, K.; C, C.-G.; Cs, C.; M, F.; Lma, O.; Sj, H.; T, S.; K, M.; C, S. Assessment of Heterogeneity among Participants in the Parkinson\u0026rsquo;s Progression Markers Initiative Cohort Using \u0026alpha;-Synuclein Seed Amplification: A Cross-Sectional Study. \u003cem\u003eThe Lancet. Neurology\u003c/em\u003e 2023, \u003cem\u003e22\u003c/em\u003e (5). https://doi.org/10.1016/S1474-4422(23)00109-6.\u003c/li\u003e\n\u003cli\u003eK, T.; H, S.-T.; S, K.; L, Z.; M, M.; P, S.-F.; R, T. High-Throughput CSF Proteomics and Machine Learning to Identify Proteomic Signatures for Parkinson Disease Development and Progression. \u003cem\u003eNeurology\u003c/em\u003e 2023, \u003cem\u003e101\u003c/em\u003e (14). https://doi.org/10.1212/WNL.0000000000207725.\u003c/li\u003e\n\u003cli\u003eSled, J. G.; Zijdenbos, A. P.; Evans, A. C. A Nonparametric Method for Automatic Correction of Intensity Nonuniformity in MRI Data. \u003cem\u003eIEEE Trans Med Imaging\u003c/em\u003e 1998, \u003cem\u003e17\u003c/em\u003e (1), 87\u0026ndash;97. https://doi.org/10.1109/42.668698.\u003c/li\u003e\n\u003cli\u003eVovk, U.; Pernus, F.; Likar, B. A Review of Methods for Correction of Intensity Inhomogeneity in MRI. \u003cem\u003eIEEE Trans Med Imaging\u003c/em\u003e 2007, \u003cem\u003e26\u003c/em\u003e (3), 405\u0026ndash;421. https://doi.org/10.1109/ TMI.2006.891486.\u003c/li\u003e\n\u003cli\u003eBelaroussi, B.; Milles, J.; Carme, S.; Zhu, Y. M.; Benoit-Cattin, H. Intensity Non-Uniformity Correction in MRI: Existing Methods and Their Validation. \u003cem\u003eMed Image Anal\u003c/em\u003e 2006, \u003cem\u003e10\u003c/em\u003e (2), 234\u0026ndash;246. https://doi.org/10.1016/j.media.2005.09.004.\u003c/li\u003e\n\u003cli\u003eVS Fonov, AC Evans, K Botteron, CR Almli, RC McKinstry, DL Collins and BDCG, Unbiased Average Age-Appropriate Atlases for Pediatric Studies, NeuroImage,Volume 54, Issue 1, January 2011, ISSN 1053\u0026ndash;8119.\u003c/li\u003e\n\u003cli\u003eVS Fonov, AC Evans, RC McKinstry, CR Almli and DL Collins, Unbiased Nonlinear Average Age-Appropriate Brain Templates from Birth to Adulthood, NeuroImage, Volume 47, Supplement 1, July 2009, Page S102 Organization for Human Brain Mapping 2009 Annual Meeting.\u003c/li\u003e\n\u003cli\u003eYe, Z.; Zhang, G.; Zhang, Y.; Li, S.; Liu, N.; Zhou, X.; Xiao, W.; M\u0026uuml;nte, T. F. The Role of the Subthalamic Nucleus in Sequential Working Memory in De Novo Parkinson\u0026rsquo;s Disease. \u003cem\u003eMov Disord\u003c/em\u003e 2021, \u003cem\u003e36\u003c/em\u003e (1), 87\u0026ndash;95. https://doi.org/10.1002/mds.28344.\u003c/li\u003e\n\u003cli\u003evan Nuland, A. J. M.; den Ouden, H. E. M.; Zach, H.; Dirkx, M. F. M.; van Asten, J. J. A.; Scheenen, T. W. J.; Toni, I.; Cools, R.; Helmich, R. C. GABAergic Changes in the Thalamocortical Circuit in Parkinson\u0026rsquo;s Disease. \u003cem\u003eHum Brain Mapp\u003c/em\u003e 2020, \u003cem\u003e41\u003c/em\u003e (4), 1017\u0026ndash;1029. https://doi.org/10.1002/hbm.24857.\u003c/li\u003e\n\u003cli\u003eLauro, P. M.; Lee, S.; Akbar, U.; Asaad, W. F. Subthalamic-Cortical Network Reorganization during Parkinson\u0026rsquo;s Tremor. \u003cem\u003eJ Neurosci\u003c/em\u003e 2021, \u003cem\u003e41\u003c/em\u003e (47), 9844\u0026ndash;9858. https://doi.org/10.1523/ JNEUROSCI.0854-21.2021.\u003c/li\u003e\n\u003cli\u003eDirkx, M. F.; den Ouden, H. E. M.; Aarts, E.; Timmer, M. H. M.; Bloem, B. R.; Toni, I.; Helmich, R. C. Dopamine Controls Parkinson\u0026rsquo;s Tremor by Inhibiting the Cerebellar Thalamus. \u003cem\u003eBrain\u003c/em\u003e 2017, \u003cem\u003e140\u003c/em\u003e (3), 721\u0026ndash;734. https://doi.org/10.1093/brain/aww331.\u003c/li\u003e\n\u003cli\u003eBelloso-Iguerategui, A.; Zamarbide, M.; Merino-Galan, L.; Rodr\u0026iacute;guez-Chinchilla, T.; Gago, B.; Santamaria, E.; Fern\u0026aacute;ndez-Irigoyen, J.; Cotman, C. W.; Prieto, G. A.; Quiroga-Varela, A.; Rodr\u0026iacute;guez-Oroz, M. C. Hippocampal Synaptic Failure Is an Early Event in Experimental Parkinsonism with Subtle Cognitive Deficit. \u003cem\u003eBrain\u003c/em\u003e 2023, \u003cem\u003e146\u003c/em\u003e (12), 4949\u0026ndash;4963. https:// doi.org/10.1093/brain/awad227.\u003c/li\u003e\n\u003cli\u003eLe, L.; M, K.; H, P.; T, I. Hippocampal Functional Connectivity in Parkinson\u0026rsquo;s Disease. \u003cem\u003eNeuro-degenerative diseases\u003c/em\u003e 2022, \u003cem\u003e22\u003c/em\u003e (1). https://doi.org/10.1159/000526377.\u003c/li\u003e\n\u003cli\u003eCamicioli, R.; Moore, M. M.; Kinney, A.; Corbridge, E.; Glassberg, K.; Kaye, J. A. Parkinson\u0026rsquo;s Disease Is Associated with Hippocampal Atrophy. \u003cem\u003eMov Disord\u003c/em\u003e 2003, \u003cem\u003e18\u003c/em\u003e (7), 784\u0026ndash;790. https://doi.org/10.1002/mds.10444.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eElevated caudate connectivity in cognitively normal Parkinson\u0026rsquo;s disease patients - PubMed\u003c/em\u003e. https://pubmed.ncbi.nlm.nih.gov/33087833/ (accessed 2024-12-04).\u003c/li\u003e\n\u003cli\u003eZhang, L.; Zhang, P.; Dong, Q.; Zhao, Z.; Zheng, W.; Zhang, J.; Hu, X.; Yao, Z.; Hu, B. Fine-Grained Features Characterize Hippocampal and Amygdaloid Change Pattern in Parkinson\u0026rsquo;s Disease and Discriminate Cognitive-Deficit Subtype. \u003cem\u003eCNS Neurosci Ther\u003c/em\u003e 2024, \u003cem\u003e30\u003c/em\u003e (1), e14480. https://doi.org/10.1111/cns.14480.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eCommon and distinct roles of amygdala subregional functional connectivity in non-motor symptoms of Parkinson\u0026rsquo;s disease - PubMed\u003c/em\u003e. https://pubmed.ncbi.nlm.nih.gov/36806219/ (accessed 2024-12-04).\u003c/li\u003e\n\u003cli\u003eSigirli, D.; Ozdemir, S. T.; Erer, S.; Sahin, I.; Ercan, I.; Ozpar, R.; Orun, M. O.; Hakyemez, B. Statistical Shape Analysis of Putamen in Early-Onset Parkinson\u0026rsquo;s Disease. \u003cem\u003eClin Neurol Neurosurg\u003c/em\u003e 2021, \u003cem\u003e209\u003c/em\u003e, 106936. https://doi.org/10.1016/j.clineuro.2021.106936.\u003c/li\u003e\n\u003cli\u003eJiang, Z.; Yin, J.; Han, P.; Chen, N.; Kang, Q.; Qiu, Y.; Li, Y.; Lao, Q.; Sun, M.; Yang, D.; Huang, S.; Qiu, J.; Li, K. Wavelet Transformation Can Enhance Computed Tomography Texture Features: A Multicenter Radiomics Study for Grade Assessment of COVID-19 Pulmonary Lesions. \u003cem\u003eQuant Imaging Med Surg\u003c/em\u003e 2022, \u003cem\u003e12\u003c/em\u003e (10), 4758\u0026ndash;4770. https://doi.org/10.21037/qims-22-252.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eWavelet transforms in separation science for denoising and peak overlap detection\u003c/em\u003e. https://pubmed.ncbi.nlm.nih.gov/32108426/ (accessed 2024-12-04).\u003c/li\u003e\n\u003cli\u003eChaddad, A.; Daniel, P.; Niazi, T. Radiomics Evaluation of Histological Heterogeneity Using Multiscale Textures Derived From 3D Wavelet Transformation of Multispectral Images. \u003cem\u003eFront Oncol\u003c/em\u003e 2018, \u003cem\u003e8\u003c/em\u003e, 96. https://doi.org/10.3389/fonc.2018.00096.\u003c/li\u003e\n\u003cli\u003eWang, J.; Geng, X. Large Margin Weighted K-Nearest Neighbors Label Distribution Learning for Classification. \u003cem\u003eIEEE Trans Neural Netw Learn Syst\u003c/em\u003e 2024, \u003cem\u003e35\u003c/em\u003e (11), 16720\u0026ndash;16732. https://doi.org/10.1109/TNNLS.2023.3297261.\u003c/li\u003e\n\u003cli\u003eWang, H.; Shao, Y.; Zhou, S.; Zhang, C.; Xiu, N. Support Vector Machine Classifier via L0/1 Soft-Margin Loss. \u003cem\u003eIEEE Trans Pattern Anal Mach Intell\u003c/em\u003e 2022, \u003cem\u003e44\u003c/em\u003e (10), 7253\u0026ndash;7265. https://doi.org/10.1109/TPAMI.2021.3092177.\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eA review on longitudinal data analysis with random forest \u003c/em\u003e. https://pubmed.ncbi.nlm.nih.gov/36653905/ (accessed 2024-12-04).\u003c/li\u003e\n\u003cli\u003e\u003cem\u003eLogistic Regression in Medical Research - \u003c/em\u003ehttps://pubmed.ncbi.nlm.nih.gov/33449558/ (accessed 2024-12-04).\u003c/li\u003e\n\u003cli\u003eSugahara, S.; Ueno, M. Exact Learning Augmented Naive Bayes Classifier. \u003cem\u003eEntropy (Basel)\u003c/em\u003e 2021, \u003cem\u003e23\u003c/em\u003e (12), 1703. https://doi.org/10.3390/e23121703.\u003c/li\u003e\n\u003cli\u003eParkinson Progression Marker Initiative. The Parkinson Progression Marker Initiative (PPMI). \u003cem\u003eProg Neurobiol\u003c/em\u003e 2011, \u003cem\u003e95\u003c/em\u003e (4), 629\u0026ndash;635. https://doi.org/10.1016/j.pneurobio.2011.09.005.\u003c/li\u003e\n\u003cli\u003eTsukita, K.; Sakamaki-Tsukita, H.; Kaiser, S.; Zhang, L.; Messa, M.; Serrano-Fernandez, P.; Takahashi, R. High-Throughput CSF Proteomics and Machine Learning to Identify Proteomic Signatures for Parkinson Disease Development and Progression. \u003cem\u003eNeurology\u003c/em\u003e 2023, \u003cem\u003e101\u003c/em\u003e (14), e1434\u0026ndash;e1447. https://doi.org/10.1212/WNL.0000000000207725.\u003c/li\u003e\n\u003cli\u003eSimuni, T.; Merchant, K.; Brumm, M. C.; Cho, H.; Caspell-Garcia, C.; Coffey, C. S.; Chahine, L. M.; Alcalay, R. N.; Nudelman, K.; Foroud, T.; Mollenhauer, B.; Siderowf, A.; Tanner, C.; Iwaki, H.; Sherer, T.; Marek, K.; Parkinson\u0026rsquo;s Progression Marker Initiative Authors. Longitudinal Clinical and Biomarker Characteristics of Non-Manifesting LRRK2 G2019S Carriers in the PPMI Cohort. \u003cem\u003eNPJ Parkinsons Dis\u003c/em\u003e 2022, \u003cem\u003e8\u003c/em\u003e (1), 140. https://doi.org/10.1038/s41531-022-00404-w.\u003c/li\u003e\n\u003cli\u003eKoros, C.; Simitsi, A.-M.; Papagiannakis, N.; Bougea, A.; Prentakis, A.; Papadimitriou, D.; Pachi, I.; Beratis, I.; Stanitsa, E.; Angelopoulou, E.; Antonelou, R.; Bregianni, M.; Lourentzos, K.; Papageorgiou, S. G.; Bonakis, A.; Trapali, X. G.; Stamelou, M.; Stefanis, L. Serum Uric Acid as a Putative Biomarker in Prodromal Parkinson\u0026rsquo;s Disease: Longitudinal Data from the PPMI Study. \u003cem\u003eJ Parkinsons Dis\u003c/em\u003e 2023, \u003cem\u003e13\u003c/em\u003e (5), 811\u0026ndash;818. https://doi.org/10.3233/JPD-230007.\u003c/li\u003e\n\u003cli\u003eMovement Disorder Society Task Force on Rating Scales for Parkinson\u0026rsquo;s Disease. The Unified Parkinson\u0026rsquo;s Disease Rating Scale (UPDRS): Status and Recommendations. \u003cem\u003eMov Disord\u003c/em\u003e 2003, \u003cem\u003e18\u003c/em\u003e (7), 738\u0026ndash;750. https://doi.org/10.1002/mds.10473.\u003c/li\u003e\n\u003cli\u003eGoetz, C. G.; Poewe, W.; Rascol, O.; Sampaio, C.; Stebbins, G. T.; Counsell, C.; Giladi, N.; Holloway, R. G.; Moore, C. G.; Wenning, G. K.; Yahr, M. D.; Seidl, L.; Movement Disorder Society Task Force on Rating Scales for Parkinson\u0026rsquo;s Disease. Movement Disorder Society Task Force Report on the Hoehn and Yahr Staging Scale: Status and Recommendations. \u003cem\u003eMov Disord\u003c/em\u003e 2004, \u003cem\u003e19\u003c/em\u003e (9), 1020\u0026ndash;1028. https://doi.org/10.1002/mds.20213.\u003c/li\u003e\n\u003cli\u003eRegnault, A.; Boroojerdi, B.; Meunier, J.; Bani, M.; Morel, T.; Cano, S. Does the MDS-UPDRS Provide the Precision to Assess Progression in Early Parkinson\u0026rsquo;s Disease? Learnings from the Parkinson\u0026rsquo;s Progression Marker Initiative Cohort. \u003cem\u003eJ Neurol\u003c/em\u003e 2019, \u003cem\u003e266\u003c/em\u003e (8), 1927\u0026ndash;1936. https:// doi.org/10.1007/s00415-019-09348-3.\u003c/li\u003e\n\u003cli\u003eDeo, R. C. Machine Learning in Medicine. \u003cem\u003eCirculation\u003c/em\u003e 2015, \u003cem\u003e132\u003c/em\u003e (20), 1920\u0026ndash;1930. https://doi.org/10.1161/CIRCULATIONAHA.115.001593.\u003c/li\u003e\n\u003cli\u003eGoecks, J.; Jalili, V.; Heiser, L. M.; Gray, J. W. How Machine Learning Will Transform Biomedicine. \u003cem\u003eCell\u003c/em\u003e 2020, \u003cem\u003e181\u003c/em\u003e (1), 92\u0026ndash;101. https://doi.org/10.1016/j.cell.2020.03.022.\u003c/li\u003e\n\u003cli\u003eMay, M. Eight Ways Machine Learning Is Assisting Medicine. \u003cem\u003eNat Med\u003c/em\u003e 2021, \u003cem\u003e27\u003c/em\u003e (1), 2\u0026ndash;3. https://doi.org/10.1038/s41591-020-01197-2.\u003c/li\u003e\n\u003cli\u003eBae, Y. J.; Kim, J.-M.; Sohn, C.-H.; Choi, J.-H.; Choi, B. S.; Song, Y. S.; Nam, Y.; Cho, S. J.; Jeon, B.; Kim, J. H. Imaging the Substantia Nigra in Parkinson Disease and Other Parkinsonian Syndromes. \u003cem\u003eRadiology\u003c/em\u003e 2021, \u003cem\u003e300\u003c/em\u003e (2), 260\u0026ndash;278. https://doi.org/10.1148/radiol.2021203341.\u003c/li\u003e\n\u003cli\u003eChu, Y.; Hirst, W. D.; Federoff, H. J.; Harms, A. S.; Stoessl, A. J.; Kordower, J. H. Nigrostriatal Tau Pathology in Parkinsonism and Parkinson\u0026rsquo;s Disease. \u003cem\u003eBrain\u003c/em\u003e 2024, \u003cem\u003e147\u003c/em\u003e (2), 444\u0026ndash;457. https://doi.org/10.1093/brain/awad388.\u003c/li\u003e\n\u003cli\u003eO\u0026rsquo;Callaghan, C.; Hornberger, M.; Balsters, J. H.; Halliday, G. M.; Lewis, S. J. G.; Shine, J. M. Cerebellar Atrophy in Parkinson\u0026rsquo;s Disease and Its Implication for Network Connectivity. \u003cem\u003eBrain\u003c/em\u003e 2016, \u003cem\u003e139\u003c/em\u003e (Pt 3), 845\u0026ndash;855. https://doi.org/10.1093/brain/awv399.\u003c/li\u003e\n\u003cli\u003eLi, T.; Le, W.; Jankovic, J. Linking the Cerebellum to Parkinson Disease: An Update. \u003cem\u003eNat Rev Neurol\u003c/em\u003e 2023, \u003cem\u003e19\u003c/em\u003e (11), 645\u0026ndash;654. https://doi.org/10.1038/s41582-023-00874-3.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"Parkinson's disease, PPMI database, Machine learning, MRI T1-weighted imaging, Texture analysis","lastPublishedDoi":"10.21203/rs.3.rs-5859401/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5859401/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study presents an early diagnostic method for Parkinson\u0026rsquo;s disease using T1-weighted imaging texture features combined with machine learning models. T1-weighted imaging data from the PPMI database were preprocessed to extract texture features from various brain regions,including the thalamus, hippocampus, caudate nucleus, amygdala, globus pallidus and putamen. The Random Forest (RF) model demonstrated excellent performance in distinguishing Parkinson\u0026rsquo;s patients from healthy controls, achieving an AUC of 0.90, accuracy of 88.9%, precision of 92.3%, sensitivity of 92.3%, specificity of 80.0%, and an F1 score of 92.3%. A simplified RF model also exhibited strong performance with a prediction accuracy of 77.8%. This method effectively leverages brain texture features to assist in the early diagnosis of Parkinson\u0026rsquo;s disease.\u003c/p\u003e","manuscriptTitle":"A Machine Learning Approach for Early Parkinson's Disease Diagnosis Based on Brain Texture Features from T1-Weighted Imaging","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-02-13 09:56:03","doi":"10.21203/rs.3.rs-5859401/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"15b7ecb0-208e-4c87-9850-5c3d75a8fce1","owner":[],"postedDate":"February 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":44201435,"name":"Biological sciences/Computational biology and bioinformatics/Computational neuroscience"},{"id":44201436,"name":"Biological sciences/Computational biology and bioinformatics/Databases"},{"id":44201437,"name":"Biological sciences/Computational biology and bioinformatics/Predictive medicine"},{"id":44201438,"name":"Biological sciences/Computational biology and bioinformatics"},{"id":44201439,"name":"Biological sciences/Neuroscience"},{"id":44201440,"name":"Health sciences/Anatomy"},{"id":44201441,"name":"Health sciences/Biomarkers"},{"id":44201442,"name":"Health sciences/Neurology"},{"id":44201443,"name":"Health sciences/Pathogenesis"}],"tags":[],"updatedAt":"2025-11-07T00:53:30+00:00","versionOfRecord":[],"versionCreatedAt":"2025-02-13 09:56:03","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5859401","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5859401","identity":"rs-5859401","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}