Analysis and Research on Predicting the Motor Classification of Parkinson's Disease Based on Radiomics

Analysis and Research on Predicting the Motor Classification of Parkinson's Disease Based on Radiomics | 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 Analysis and Research on Predicting the Motor Classification of Parkinson's Disease Based on Radiomics Shu-fen Liu, Yuan-zhe Li, Yi Wang, Jian-long Zhuang, Lin-yi Li, and 7 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6223585/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 developed and compared single-sequence and multimodal imaging omics models for Parkinson's disease (PD) classification using 3.0T MRI scans (T1WI, T2-FLAIR) from 160 PD patients (82 tremor-type, 78 non-tremor-type) and 100 healthy controls. Regions of interest included the Hippocampus, Substantia Nigra, Red Nucleus, Thalamus, and Amygdala. Data were split into training/test sets (8:2), with the Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection and Support Vector Machine (SVM) for modeling, evaluated via Receiver Operating Characteristic (ROC) curves and area under curve (AUC). The single-sequence Hippocampal-T1WI model showed AUCs of control (training/test:0.940/0.834), non-tremor PD (training/test: 0.923/0.740), and tremor PD (training/test:0.914/0.524). The multimodal model achieved higher AUCs: control (training/test:0.966/0.877), non-tremor PD (training/test:0.952/0.861), and tremor PD (training: 0.942, test: 0.760), indicating improved predictive accuracy, demonstrating superior predictive accuracy. Multimodal imaging omics significantly enhanced PD diagnosis and differentiation compared to single-sequence models. Biological sciences/Neuroscience Health sciences/Biomarkers Health sciences/Diseases Parkinson's disease clinical classification radiomics magnetic resonance imaging Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Introduction Parkinson's disease (PD), the second most prevalent neurodegenerative disorder after Alzheimer's disease (AD), predominantly affects the elderly population with increasing incidence rates observed annually [ 1 ] . Its pathology involves substantia nigra dopaminergic neuron degeneration and Lewy body formation, leading to both motor (resting tremor, bradykinesia, gait abnormalities, rigidity) and non-motor symptoms (depression, sleep disorders, cognitive impairment) [ 2 ] . Clinical studies have further classified PD into tremor-dominant and non-tremor-dominant subtypes [ 3 ] . Tremor is defined as an involuntary, rhythmic, oscillatory movement of a body part [ 4 ] .Current evidence suggests the cerebello-thalamo-cortical pathway plays a central role in tremor-related disorders. In PD, tremor signals are amplified through the motor cortex-ventral intermediate thalamic nucleus (Vim) circuit, explaining the efficacy of Vim thalamotomy and deep brain stimulation [ 5 ] . Emerging research indicates PD tremor originates from the subthalamic nucleus, whose burst firing depends on motor cortex input, propagating through the cortex-subthalamic-pallidum-thalamus-cortical pathway [ 6 ] . The cerebellum may modulate tremor amplitude via the cerebello-thalamo-cortical circuit [ 7 ] . At the same time, Iron deposition in the red nucleus, as demonstrated by susceptibility-weighted imaging, correlates with PD tremor [ 8 ] . As a dopamine production cofactor, iron serves as a biomarker for surviving substantia nigra dopaminergic neurons. Additionally, tremor severity appears influenced by emotional and cognitive states, with stress or cognitive load potentially amplifying tremor through thalamic arousal pathways [ 9 ] . The hippocampus, crucial for cognitive function [ 10 ] , shows structural plasticity changes in PD. Kim B et al. [ 11 ] demonstrated hippocampal alterations in a 6-hydroxydopamine-induced PD rat model, supporting the hippocampus's involvement in PD pathology. Recent studies have demonstrated Artificial Intelligence (AI)'s significant contribution to PD diagnosis [ 12 ] . AI, primarily based on Machine Learning (ML), employs complex algorithms for decision-making, pattern recognition, and data analysis [ 13 ] .ML algorithms are divided into supervised learning and unsupervised learning. Supervised learning includes Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and Decision Tree (DT); unsupervised learning includes Hidden Markov Models (HMM) [ 14 ] . Radiomics, pioneered by Philippe Lambin et al. in 2012 [ 15 ] , has emerged as a promising field that utilizes AI/ML to extract imaging features and develop predictive models [ 16 ] .The selection of the best machine learning method for establishing radiomic methods is a crucial step. The choice of ML algorithms is critical in radiomics, with multiple studies indicating SVM's superior performance compared to other supervised classifiers [ 17 ] [ 18 ] [ 19 ] . Radiomics involves five key steps: image acquisition, segmentation, feature extraction, feature selection, and model construction and evaluation [ 20 ] . By quantifying tissue characteristics like heterogeneity and shape, radiomic features combined with clinical data can enhance disease diagnosis, prognosis, and prediction accuracy [ 21 , 22 ] . This approach has shown clinical utility across various diseases [ 23 – 25 ] , including PD.Prodoehl J [ 26 ] successfully differentiated PD, atypical Parkinsonism, essential tremor, and healthy controls using Diffusion Tensor Imaging with radiomic analysis of key regions (basal ganglia, red nucleus, dentate nucleus, cerebellum). Vitali P [ 27 ] achieved accurate classification of early PD, advanced PD, and healthy individuals through substantia nigra texture analysis. Additionally, radiomic studies have revealed significant differences in hippocampus, thalamus, and amygdala between PD patients and controls, correlating with cognitive performance [ 28 ] . While MRI-based radiomics offers promising diagnostic tools for PD, research on its application in PD tremor subtypes remains limited. This study aims to develop single- and multi-modality radiomic models to improve PD subtype prediction and classification, ultimately enhancing disease management strategies. Materials and Methods Patients This retrospective study enrolled 260 participants from January 2013 to December 2023, comprising 82 tremor-type PD, 78 non-tremor PD, and 100 control cases. Approved by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (Approval No. 2023 Fu Medical Affiliated No. 2 Ethics Review (286)), the study followed strict inclusion and exclusion criteria. Inclusion criteria required: (1) Standardized cranial MRI with T1-Weighted Imaging (T1WI) and T2-Weighted Fluid-Attenuated Inversion Recovery (T2-FLAIR) sequences using consistent scanning parameters; (2) PD diagnosis confirmed by two neurologists according to 2015 MDS diagnostic criteria [ 29 ] , with consensus resolution for discrepancies; (3) Complete imaging and clinical data; (4) Age 50–80 years. Exclusion criteria included: (1) Cognitive impairment from other causes, cerebrovascular diseases, or intracranial abnormalities; (2) Substandard image quality; (3) Incomplete clinical data. PD patients were classified into tremor group based on presence of resting tremor in head/neck or limbs according to the MDS Unified Parkinson's Disease Rating Scale (MDS-UPDRS), while those without resting tremor were assigned to non-tremor group. Controls were selected from individuals undergoing cranial MRI during the same period, confirmed PD-free and tremor-negative, meeting the same exclusion criteria. Magnetic Resonance Imaging All participants underwent 3.0T MRI (Dutch Philips Ingenia) using an 8-channel head coil. During scanning, subjects maintained a supine position with head stabilization using foam pads and noise-reducing earplugs, remaining conscious and relaxed. The imaging protocol included axial T1WI (repetition time (TR)/echo time (TE): 7.9/3.0 ms) and T2-FLAIR (TR/TE: 9000/100 ms) sequences. Image Acquisition and Delineation of Regions of Interest Figure 1 illustrates the radiomics workflow, which includes image acquisition and Regions of interest(ROI) delineation. Using ITK-SNAP software (v3.8.0; http://www.itk-snap.org ), we processed anonymized cranial MRI the Picture Archiving and Communication System (PACS) files exported from the Picture Archiving and Communication System (PACS). A trained specialist manually outlined bilateral ROIs (hippocampus, substantia nigra, red nucleus, thalamus, and amygdala) on both T1WI and T2-FLAIR sequences (Fig. 1 ). Radiomic Feature Extraction, Selection, and Model Building Participants were randomly allocated to training and test sets (8:2 ratio). Radiomic features were extracted from T1WI and T2-FLAIR ROIs using Pyradiomics (Python v3.6; https://pyradiomics.readthedocs.io ). Following feature standardization, we implemented a two-step feature selection: (1) minimal Redundancy Maximal Relevance (mRMR) for eliminating redundant features [ 30 ] , and (2) LASSO logistic regression for optimal feature subset selection through regularization parameter λ optimization. LASSO's penalty function compresses coefficients of irrelevant variables, reducing model complexity while addressing collinearity and overfitting. SVM classifier was employed for model construction. SVM operates as a maximum-margin linear classifier, transforming data separation into convex quadratic programming. Based on Vapnik-Chervonenkis Dimension theory and structural risk minimization principles, SVM optimally balances model complexity and learning capacity, particularly effective for small sample sizes. The SVM model identified key Parkinson's-related features from training data, enabling effective differentiation between PD subtypes and healthy controls. Statistics Statistical analyses were performed using SPSS and R software. Quantitative radiomic scores were generated from SVM-based models for both training and testing groups. We developed single-sequence and multimodal radiomic models using T1WI and T2-FLAIR sequences from five brain regions (hippocampus, substantia nigra, red nucleus, thalamus, and amygdala). Model performance was evaluated through Macro-avg, Micro-avg, and ROC curve analyses, with AUC, sensitivity, specificity, and accuracy as key metrics [ 31 , 32 ] . Micro-avg assesses overall performance across all categories, while Macro-avg evaluates category-specific performance before averaging. AUC values > 0.8 and 0.7–0.8 indicate good and moderate discriminative power, respectively. Higher sensitivity reflects better positive sample identification, specificity indicates superior negative sample recognition, and accuracy represents overall model performance Finally, the DeLong test by MedCalc ( www.medcalc.org ) was used to compare the differences between the ROC curves of different models, considering a p <0.05 as indicating a statistically significant difference. Results 3.1 Comparison of General Clinical Data Based on the inclusion and exclusion criteria, the study enrolled 260 participants: 82 tremor-type PD patients (38 male, 44 female), 78 non-tremor-type PD patients (41 male, 37 female), and 100 controls (39 male, 61 female). No significant gender differences were observed among groups ( p > 0.05) (Table 1 ). Table 1 The basic information of included people Control group non-tremor type PD tremor type PD p value Total number of people 100 78 82 - Gender (Male/Female) 39/61 41/37 38/44 0.192 Age 64.8 ± 8.4 67.6 ± 7.9 70.1 ± 7.0 - The data are presented as mean ± SD. p < 0.05 showed signifcant diference between different groups. 3.2 Based on Single-Sequence Radiomic Models 3.2.1 Radiomic Feature Selection Single-sequence radiomic models were developed using T1WI and T2-FLAIR sequences, focusing on bilateral regions including the Hippocampus, Substantia Nigra, Red Nucleus, Thalamus, and Amygdala. A total of 1197 radiomic features were initially extracted. Following feature selection with mRMR and LASSO, the optimal features for each region were identified: Hippocampus: 20 from T1WI and 12 from T2-FLAIR (Fig. 2 ); Substantia Nigra: 18 from T1WI and 19 from T2-FLAIR (Fig. 3 );Red Nucleus: 19 from T1WI and 16 from T2-FLAIR (Fig. 3 ); Thalamus: 16 from T1WI and 17 from T2-FLAIR (Fig. 4 ); Amygdala: 23 from T1WI and 10 from T2-FLAIR (Fig. 4 ). 3.2.2 Diagnostic efficacy of a single sequence radiomics model In the Hippocampal-T1WI radiomics model, the training set AUC values for the control, non-tremor PD, and tremor PD groups were 0.940, 0.923, and 0.914, respectively, while the test set AUC values were 0.834, 0.740, and 0.524. For the Hippocampal-T2-FLAIR model, the training set AUC values were 0.911, 0.844, and 0.754, with corresponding test set values of 0.830, 0.747, and 0.656 (Fig. 5 ). In the Substantia Nigra-T1WI model, the training set AUC values were 0.928, 0.856, and 0.905, while the test set values were 0.619, 0.559, and 0.562. For the Substantia Nigra-T2-FLAIR model, the training set AUC values were 0.895, 0.689, and 0.673, with test set values of 0.697, 0.639, and 0.566 (Fig. 5 ). For the Red Nucleus-T1WI radiomics model, the training set AUC values for the control, non-tremor PD, and tremor PD groups were 0.915, 0.659, and 0.716, respectively, with test set values of 0.547, 0.693, and 0.335. In the Red Nucleus-T2-FLAIR model, the training set AUC values were 0.920, 0.894, and 0.888, while the test set values were 0.748, 0.688, and 0.578 (Fig. 6 ). For the Thalamus-T1WI model, the training set AUC values were 0.914, 0.850, and 0.866, with test set values of 0.684, 0.684, and 0.651. In the Thalamus-T2-FLAIR model, the training set AUC values were 0.902, 0.872, and 0.873, while the test set values were 0.817, 0.844, and 0.556 (Fig. 6 ). For the Amygdala-T1WI model, the training set AUC values were 0.905, 0.867, and 0.871, with test set values of 0.728, 0.715, and 0.424. In the Amygdala-T2-FLAIR model, the training set AUC values were 0.886, 0.819, and 0.710, while the test set values were 0.775, 0.689, and 0.602 (Fig. 6 ) The AUC values, sensitivity, accuracy, and specificity of each individual sequence radiomics model are shown in Table 2 . Table 2 AUC values, sensitivity, specificity, and accuracy based in single sequence imaging omics model Brain Region Group AUC(A/B) SEN(A/B) SPE(A/B) ACC(A/B) T1WI Hippocampal Control 0.940/0.834 0.838/0.750 0.859/0.750 0.788/0.652 Non-tremor 0.923/0.740 0.774/0.500 0.925/0.750 0.814/0.471 Tremor 0.914/0.524 0.727/0.250 0.887/0.778 0.750/0.333 Substantia Nigra Control 0.928/0.619 0.950/0.550 0.820/0.625 0.768/0.478 Non-tremor 0.856/0.559 0.629/0.125 0.979/0.833 0.929/0.250 Tremor 0.905/0.562 0.818/0.500 0.908/0.639 0.806/0.381 Red Nucleus Control 0.915/0.547 0.925/0.450 0.766/0.448 0.712/0.360 Non-tremor 0.659/0.693 0.661/0.438 0.925/0.848 0.788/0.583 Tremor 0.716/0.335 0.652/0.154 0.937/0.722 0.827/0.167 Thalamus Control 0.914/0.684 0.900/0.650 0.734/0.625 0.679/0.520 Non-tremor 0.850/0.684 0.581/0.375 0.952/0.833 0.837/0.500 Tremor 0.866/0.651 0.727/0.438 0.923/0.778 0.814/0.467 Amygdala Control 0.905/0.728 0.913/0.750 0.758/0.531 0.702/0.500 Non-tremor 0.867/0.715 0.661/0.188 0.925/0.833 0.788/0.333 Tremor 0.871/0.424 0.621/0.125 0.923/0.694 0.788/0.154 T2- Hippocampal Control 0.911/0.830 0.889/0.800 0.781/0.750 0.717/0.667 FLAIR Non-tremor 0.844/0.747 0.613/0.500 0.911/0.861 0.745/0.615 Tremor 0.754/0.656 0.606/0.375 0.873/0.750 0.690/0.400 Substantia Nigra Control 0.895/0.697 0.888/0.600 0.781/0.656 0.717/0.522 Non-tremor 0.689/0.639 0.694/0.313 0.945/0.806 0.843/0.417 Tremor 0.673/0.566 0.727/0.313 0.930/0.667 0.828/0.294 Red Nucleus Control 0.920/0.748 0.938/0.800 0.789/0.563 0.735/0.533 Non-tremor 0.894/0.688 0.661/0.563 0.945/0.806 0.837/0.563 Tremor 0.888/0.578 0.667/0.188 0.908/0.917 0.772/0.500 Thalamus Control 0.902/0.817 0.838/0.800 0.917/0.844 0.882/0.762 Non-tremor 0.872/0.844 0.806/0.625 0.835/0.778 0.704/0.556 Tremor 0.873/0.556 0.766/0.375 0.958/0.806 0.857/0.462 Amygdala Control 0.886/0.775 0.850/0.800 0.805/0.625 0.731/0.571 Non-tremor 0.819/0.689 0.839/0.438 0.747/0.722 0.584/0.412 Tremor 0.710/0.602 0.318/0.063 0.965/0.833 0.808/0.143 * AUC, mean area under the curve; SEN, sensitivity; SPE, specificity; ACC, accuracy; A/B, Training set/Test set. 3.3 Based on multi sequence radiomics model A multimodal radiomics model was developed using T1WI and T2-FLAIR sequences from the Hippocampus, Substantia Nigra, Red Nucleus, Thalamus, and Amygdala. A total of 140 radiomics features were extracted, distributed as follows: Hippocampus (26 features: 16 from T1WI, 10 from T2-FLAIR), Substantia Nigra (34 features:19 from T1WI, 15 from T2-FLAIR), Red Nucleus (27 features: 15 from T1WI, 12 from T2-FLAIR), Thalamus (27 features: 13 from T1WI, 14 from T2-FLAIR), and Amygdala (26 features: 18 from T1WI, 8 from T2-FLAIR). Following feature selection using mRMR and LASSO, 18 optimal features were identified from the multimodal radiomics model. These comprised: 2 from Hippocampus-T1WI, 1 from Hippocampus-T2-FLAIR, 2 from Substantia Nigra-T1WI, 1 from Substantia Nigra-T2-FLAIR, 1 from Thalamus-T1WI, 7 from Thalamus-T2-FLAIR, 2 from Red Nucleus-T2-FLAIR, and 2 from Amygdala-T2-FLAIR (Fig. 7 ). The multimodal radiomics model demonstrated strong diagnostic performance, with training group AUC values of 0.966 (control), 0.952 (non-tremor PD), and 0.942 (tremor PD). Corresponding test group AUC values were 0.877, 0.861, and 0.760 respectively (Fig. 7 ). The results indicate that multimodal radiomics outperforms single image sequence analysis in differentiating between control, non-tremor PD, and tremor PD groups. Detailed performance metrics, including AUC, sensitivity, specificity, and accuracy, are presented in Table 3 . Table 3 AUC values, sensitivity, specificity, and accuracy in multimodal imaging omics models Group AUC SEN SPE ACC Control Training set 0.966 0.900 0.898 0.847 Test set 0.877 0.850 0.844 0.773 Non-tremor Training set 0.952 0.887 0.918 0.821 Test set 0.861 0.750 0.861 0.706 Tremor Training set 0.942 0.742 0.951 0.875 Test set 0.760 0.563 0.889 0.692 * AUC, mean area under the curve; SEN, sensitivity; SPE, specificity; ACC, accuracy; A/B, Training set/Test set. The DeLong test was employed to compare ROC curve differences across models. In the training set, significant differences were observed between the multimodal radiomics model and all single-sequence models (Hippocampus-T2-FLAIR, Substantia Nigra-T1WI/T2-FLAIR, Red Nucleus-T1WI/T2-FLAIR, Thalamus-T1WI/T2-FLAIR, and Amygdala-T1WI/T2-FLAIR). In the test set, significant differences were found between the multimodal model and single-sequence models using Substantia Nigra-T1WI/T2-FLAIR, Red Nucleus-T1WI/T2-FLAIR, and Thalamus-T1WI. Additionally, the Hippocampus-T1WI single-sequence model showed significant differences compared to Substantia Nigra-T1WI and Red Nucleus-T1WI models. Other comparisons were not statistically significant (Table 4 ). Table 4 p-values between different models by Delong test. Training set Test set T1WI T2-FLAIR T1WI T2-FLAIR Hippocampal vs Substantia Nigra 0.6092 0.5118 0.0379 0.0595 Hippocampal vs Red Nucleus 0.3050 0.6703 0.0045 0.0999 Hippocampal vs Thalamus 0.2763 0.6838 0.1548 0.8253 Hippocampal vs Amygdala 0.1331 0.2742 0.2750 0.5211 Substantia Nigra vs Red Nucleus 0.5447 0.2992 0.4787 0.4219 Substantia Nigra vs Thalamus 0.5168 0.7769 0.4427 0.0693 Substantia Nigra vs Amygdala 0.3426 0.7269 0.1423 0.2480 Red Nucleus vs Thalamus 0.9470 0.4058 0.1400 0.2549 Red Nucleus vs Amygdala 0.6744 0.2040 0.0581 0.5774 Thalamus vs Amygdala 0.7135 0.5036 0.6181 0.6382 Hippocampal vs Multimodality 0.1184 0.0017 0.5134 0.3667 Substantia Nigra vs Multimodality 0.0278 0.0003 0.0060 0.0075 Red Nucleus vs Multimodality 0.0133 0.0159 0.0009 0.0303 Thalamus vs Multimodality 0.0056 0.0004 0.0323 0.1922 Amygdala vs Multimodality 0.0025 p < 0.0001 0.0666 0.1876 * P < 0.05 has statistical significance Discussion PD demonstrates significant clinical heterogeneity, often leading to misdiagnosis or delayed detection due to the absence of definitive diagnostic methods for clinical subtypes. Early detection and intervention are crucial for effective PD management. While PD's hallmark is dopaminergic neuron degeneration in the substantia nigra and striatum, clinical symptoms typically emerge only after substantial neuronal loss, indicating a temporal disconnect between pathology and symptom onset. Clinically, PD manifests as tremor-dominant and non-tremor subtypes, with tremor pathology involving multiple brain regions. The basal ganglia-thalamus-cortical pathway, comprising the striatum, globus pallidus, subthalamic nucleus, and substantia nigra pars reticulata, plays a pivotal role in PD tremor generation through its connections with cortical, thalamic, and midbrain dopaminergic systems [ 33 ] . Recent studies have further implicated the hippocampus, thalamus, red nucleus, and amygdala in PD subtype heterogeneity [ 34 , 35 ] .While neuroimaging aids PD differential diagnosis, it cannot independently confirm PD. Radiomics has emerged as a promising diagnostic tool, enabling extraction of high-throughput quantitative features from MRI beyond visual analysis. Although radiomics has demonstrated diagnostic and prognostic value in various diseases, its application in PD subtype classification remains limited. Radiomic Features and Their Correlation with PD In radiomics, the fundamental characteristics that differentiate lesion from non-lesion images include gray-level contrast, texture uniformity, depth, and roughness. Concurrently, the approach to feature selection is pivotal in the development of radiomic models. In this investigation, employing LASSO, we identified 20 optimal radiomic features from 1197 for the Hippocampus-T1WI, 12 for the Hippocampus- T2-FLAIR, 18 for the Substantia Nigra-T1WI, 19 for the Substantia Nigra-T2-FLAIR, 19 for the Red Nucleus-T1WI, 16 for the Red Nucleus-T2-FLAIR, 16 for the Thalamus-T1WI, 17 for the Thalamus-T2-FLAIR, 23 for the Amygdala-T1WI, and 10 for the Amygdala-T2-FLAIR. Analogously, 18 optimal radiomic features were selected from 140 in multimodal radiomics. Predominantly, the selected features are high-order, characterized by wavelet filters. The wavelet filter processing algorithm stands out for its ability to discern varying resolutions and preserve signal details across these resolutions. Its capacity to magnify and reduce enables the transformation of signals to ROI while maintaining the resolution of that signal segment [ 36 ] . Moreover, wavelet filters enhance the information content of low-frequency signals and unveil deeper, high-throughput features beyond the discernment of the naked eye [ 37 ] . Additionally, their scalability renders them suitable for molecular biological signals and images exhibiting fractal or scale-invariant properties. Consequently, this study's LASSO-selected radiomic features are predominantly wavelet filters. The findings of this study indicate that the multimodal radiomic model exhibits superior predictive performance compared to the single-sequence radiomic model. The AUC values for the training and test groups are as follows: 0.966 and 0.877 for the healthy control group, 0.952 and 0.861 for the non-tremor type PD group, and 0.942 and 0.760 for the tremor type PD group, respectively. Within the realm of multimodal radiomics, LASSO was utilized to select 18 optimal radiomic features with the highest predictive efficacy. Notably, the feature with the greatest coefficient is wavelet_LLH_glrlm_GrayLevelNonUniformity_Hippocampus_T2-FLAIR, underscoring the significant role that variations in image gray levels play in predicting PD typing and differentiation. Among all feature classifications, GLRLM features constitute the largest proportion. These features characterize the coarseness and directional properties of image textures, such as the run length of textures at specific angles and directions, which are influenced by the voxel gray level distribution of the image. Changes in these distributions can reflect lesion heterogeneity, making GLRLM particularly sensitive for predicting PD typing and differentiation. Additionally, radiomic features derived from the substantia nigra account for the largest proportion, likely due to the pathological hallmark of PD: the degeneration and loss of dopaminergic neurons in the substantia nigra pars compacta. This observation is supported by Poston K [ 38 ] , who demonstrated a strong correlation between the substantia nigra and the severity of PD motor symptoms, further validating the results of this study. Kaya O [ 39 ] also highlighted significant shape differences in the subthalamic nucleus of PD patients compared to healthy controls. In this study, shape features represent the second-largest proportion after GLRLM, reinforcing the critical importance of shape-based characteristics in PD analysis. Comparison of Radiomics Performance Between Single Sequence and Multimodal Sequence Imaging This study utilizes T1WI and T2-FLAIR images from cranial MRI plain scans, with participants randomly allocated into training and test groups at an 8:2 ratio. Support SVM was employed to construct both single-sequence and multimodal radiomic models, from which the optimal model for predicting PD typing was selected. In the single-sequence radiomic models developed based on T1WI and T2-FLAIR images of the hippocampus, substantia nigra, red nucleus, thalamus, and amygdala, a quantitative assessment of each model's performance was conducted. It was observed that the single-sequence radiomic models based on Hippocampus-T1WI, Hippocampus-T2-FLAIR, Substantia Nigra-T1WI, Red Nucleus-T2-FLAIR, Thalamus-T1WI, Thalamus-T2-FLAIR, Amygdala-T1WI, and Amygdala-T2-FLAIR exhibited high efficacy in predicting PD types and distinguishing healthy controls within the training set. The Hippocampus-T1WI model demonstrated the highest performance (AUC values for the control group, non-tremor PD, and tremor PD were 0.940, 0.923, and 0.914, respectively), although the differences compared to other single-sequence models were not statistically significant (P > 0.05). However, the sensitivity, specificity, and accuracy of the test group were lower than those of the training group, potentially due to the limited sample size. Despite this, the clinical relevance of single-sequence radiomic models remains significant. For instance, Cao X [ 40 ] utilized SVM to develop a single-modality radiomic model for differentiating PD from healthy controls, achieving an AUC and accuracy of 100% in the training group and an AUC of 0.97 in the test group, underscoring the utility of single-sequence radiomics in PD differentiation. Similarly, SUN D [ 41 ] identified the left hippocampus as the region with the best radiomic features for distinguishing postural instability-gait difficulty type PD, tremor type PD, and healthy controls (training group AUC = 0.889, specificity = 80.0%, sensitivity = 88.2%, accuracy = 82.4%; test group AUC = 0.833, specificity = 83.3%, sensitivity = 75.0%, accuracy = 80.7%). Furthermore, Liu A et al. [ 42 ] constructed a single-sequence radiomic model based on chest CT using LASSO and multivariate logistic regression to differentiate benign from malignant pulmonary nodules, achieving AUC values of 0.836 and 0.809 in the test and training groups, respectively. This highlights the model's robust diagnostic capability in early lung cancer screening. In summary, single-sequence radiomics plays a critical role in the diagnosis and differential diagnosis of various diseases. This study further confirms that the single-sequence radiomic model based on Hippocampus-T1WI exhibits strong predictive performance in PD clinical typing and differentiation. This study further investigated the role of multimodal radiomics models in PD typing and differentiation. The results revealed that the AUC values for the control group in the training and test sets were 0.966 and 0.877, respectively. Similarly, the AUC values for the non-tremor PD group were 0.952 and 0.861 in the training and test sets, respectively, while for the tremor PD group, the AUC values were 0.942 and 0.760, respectively. Compared to single-sequence radiomics, multimodal radiomics demonstrated superior predictive performance in PD typing and distinguishing healthy controls. Additionally, the Delong test indicated statistically significant differences between multimodal radiomics and single-sequence radiomic models based on Substantia Nigra-T1WI (p = 0.0278), Substantia Nigra-T2-FLAIR (p = 0.0003), Red Nucleus-T1WI (p = 0.0133), Red Nucleus-T2-FLAIR (p = 0.0159), Thalamus-T1WI (p = 0.0056), and Amygdala-T1WI (p = 0.0004). However, no statistically significant differences were observed when compared to other single-sequence radiomic models. Despite this, multimodal radiomics models have been validated to exhibit excellent predictive, differential, and diagnostic performance for various diseases. For instance, Wang K et al. [ 43 ] developed a multimodal radiomic model for glioma recurrence prediction based on 8F-FDG, 11C-methionine (11C-MET) PET, and MRI images, combined with clinical data. Using LASSO binary logistic regression analysis, they achieved an AUC of 0.932 in the training group and 0.910 in the test group, highlighting the superior diagnostic performance of multimodal models over single-modal ones. Similarly, Jiang Z et al. [ 44 ] constructed a multimodal radiomic model using multimodal MRI images and Random Forest (RF) to predict the treatment response of lung cancer brain metastasis to gamma knife therapy. The model achieved AUC values of 0.930 and 0.8532 in the training and test groups, respectively, outperforming single-modal radiomic models based on T1WI (AUC values of 0.722 and 0.656 in the training and test sets, respectively) and T2-FLAIR (AUC values of 0.805 and 0.704 in the training and test sets, respectively). In summary, multimodal radiomics demonstrates superior predictive and diagnostic capabilities compared to single-modal models in the context of disease assessment. Conclusion Our findings demonstrate that the Hippocampus-T2-FLAIR-based unimodal radiomics model shows superior performance in PD classification and differential diagnosis. Furthermore, the multimodal radiomics model, integrating multiple sequence images, outperforms single-sequence models in predictive accuracy for PD classification and differentiation. Declarations Ethics approval and consent to participate This study has been granted an exemption from requiring ethics approval by The Second Affiliated Hospital of Fujian Medical University. All methods were performed in accordance with the relevant guidelines and regulations. Consent for publication Written informed consent for publication of their details was obtained from the patient. Author contributions SL and CC conceived the study, designed the methodology, and conducted the primary data analysis. YL, YW and JZ contributed to the literature review, data collection, and interpretation of results. LL, XY, WH，BY, LY, XC and ML assisted in experimental design, provided critical feedback, and revised the manuscript. All authors participated in writing, editing, and approving the final version of the manuscript. Acknowledgements This work was supported by grants from the Natural Science Foundation of Fujian Province of China (NO.2023J01104), the Scientific Foundation of Quanzhou City for High Level Talents (NO.2023C005YR), Joint funds for the innovation of science and technology, Fujian province (NO.2023Y9255), and Fujian Health Science and Technology Plan Project (NO.2024GGA039) from Dr. Chunnuan Chen. Funding This work was supported by the Natural Science Foundation of Fujian Province of China under Grant number 2023J01104; the Scientific Foundation of Quanzhou City for High Level Talents under Grant number 2023C005YR; Joint funds for the innovation of science and technology, Fujian province under Grant number 2023Y9255; and Fujian Health Science and Technology Plan Project under Grant number 2024GGA039. Competing interests The authors report no conflict of interest. Availability of data and material Not applicable. References Zhu B, Kohn R, Patel A,et al. Demoralization and Quality of Life of Patients with Parkinson Disease. Psychother Psychosom. 2021. 90(6): 415–421. Guatteo E, Berretta N, Monda V,et al. Pathophysiological Features of Nigral Dopaminergic Neurons in Animal Models of Parkinson's Disease. Int J Mol Sci. 2022. 23(9): 4508. Mortimer JA, Borenstein AR, Nelson LM. Associations of welding and manganese exposure with Parkinson disease: review and meta-analysis. Neurology. 2012. 79(11): 1174–80. Bhatia KP, Bain P, Bajaj N,et al. Consensus Statement on the classification of tremors. from the task force on tremor of the International Parkinson and Movement Disorder Society. Mov Disord. 2018. 33(1): 75–87. Kremer NI, Pauwels R, Pozzi NG,et al. Deep Brain Stimulation for Tremor: Update on Long-Term Outcomes, Target Considerations and Future Directions. J Clin Med. 2021. 10(16): 3468. Huang CS, Wang GH, Chuang HH,et al. Conveyance of cortical pacing for parkinsonian tremor-like hyperkinetic behavior by subthalamic dysrhythmia. Cell Rep. 2021. 35(3): 109007. Zhong Y, Liu H, Liu G,et al. A review on pathology, mechanism, and therapy for cerebellum and tremor in Parkinson's disease. NPJ Parkinsons Dis. 2022. 8(1): 82. Guan X, Xuan M, Gu Q,et al. Influence of regional iron on the motor impairments of Parkinson's disease: A quantitative susceptibility mapping study. J Magn Reson Imaging. 2017. 45(5): 1335–1342. Dirkx MF, Zach H, van Nuland AJ,et al. Cognitive load amplifies Parkinson’s tremor through excitatory network influences onto the thalamus. Brain. 2020. 143(5): 1498–1511. Maurer AP, Nadel L. The Continuity of Context: A Role for the Hippocampus. Trends Cogn Sci. 2021. 25(3): 187–199. Kim B, Weerasinghe-Mudiyanselage P, Ang MJ,et al. Changes in the Neuronal Architecture of the Hippocampus in a 6-Hydroxydopamine-Lesioned Rat Model of Parkinson Disease. Int Neurourol J. 2022. 26(Suppl 2): S94-105. Birkenbihl C, Ahmad A, Massat NJ,et al. Artificial intelligence-based clustering and characterization of Parkinson’s disease trajectories. Sci Rep. 2023. 13(1): 2897. Gupta R, Kumari S, Senapati A,et al. New era of artificial intelligence and machine learning-based detection, diagnosis, and therapeutics in Parkinson’s disease. Ageing Res Rev. 2023. 90: 102013. Sarker IH. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Comput Sci. 2021. 2(3): 160. Lambin P, Rios-Velazquez E, Leijenaar R,et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012. 48(4): 441–6. DeJohn CR, Grant SR, Seshadri M. Application of Machine Learning Methods to Improve the Performance of Ultrasound in Head and Neck Oncology: A Literature Review. Cancers (Basel). 2022. 14(3): 665. Ozer ME, Sarica PO, Arga KY. New Machine Learning Applications to Accelerate Personalized Medicine in Breast Cancer: Rise of the Support Vector Machines. OMICS. 2020. 24(5): 241–246. Zheng Y, Zhou D, Liu H, Wen M. CT-based radiomics analysis of different machine learning models for differentiating benign and malignant parotid tumors. Eur Radiol. 2022. 32(10): 6953–6964. Li C, Chen H, Zhang B,et al. Radiomics Signature Based on Support Vector Machines for the Prediction of Pathological Complete Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer. Cancers (Basel). 2023. 15(21): 5134. Zhang YP, Zhang XY, Cheng YT,et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Mil Med Res. 2023. 10(1): 22. Mayerhoefer ME, Materka A, Langs G,et al. Introduction to Radiomics. J Nucl Med. 2020. 61(4): 488–495. Tunali I, Gillies RJ, Schabath MB. Application of Radiomics and Artificial Intelligence for Lung Cancer Precision Medicine. Cold Spring Harb Perspect Med. 2021. 11(8): a039537. Li S, Zhou B. A review of radiomics and genomics applications in cancers: the way towards precision medicine. Radiat Oncol. 2022. 17(1): 217. Conti A, Duggento A, Indovina I,et al. Radiomics in breast cancer classification and prediction. Semin Cancer Biol. 2021. 72: 238–250. Ai Y, Zhu H, Xie C, Jin X. Radiomics in cervical cancer: Current applications and future potential. Crit Rev Oncol Hematol. 2020. 152: 102985. Prodoehl J, Li H, Planetta PJ,et al. Diffusion tensor imaging of Parkinson's disease, atypical parkinsonism, and essential tremor. Mov Disord. 2013. 28(13): 1816–22. Vitali P, Pan MI, Palesi F,et al. Substantia Nigra Volumetry with 3-T MRI in De Novo and Advanced Parkinson Disease. Radiology. 2020. 296(2): 401–410. Betrouni N, Lopes R, Defebvre L,et al. Texture features of magnetic resonance images: A marker of slight cognitive deficits in Parkinson’s disease. Mov Disord. 2020. 35(3): 486–494. Postuma RB, Berg D, Stern M,et al. MDS clinical diagnostic criteria for Parkinson's disease. Mov Disord. 2015. 30(12): 1591–601. Pan X, Liu C, Feng T, Qi XS. A multi-objective based radiomics feature selection method for response prediction following radiotherapy. Phys Med Biol. 2023. 68(5). Guiot J, Vaidyanathan A, Deprez L,et al. A review in radiomics: Making personalized medicine a reality via routine imaging. Med Res Rev. 2022. 42(1): 426–440. Liu Z, Wang S, Dong D,et al. The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics. 2019. 9(5): 1303–1322. Lazarus M, Huang ZL, Lu J,et al. How do the basal ganglia regulate sleep-wake behavior. Trends Neurosci. 2012. 35(12): 723–32. Zur G, Lesman-Segev OH, Schlesinger I,et al. Tremor Relief and Structural Integrity after MRI-guided Focused US Thalamotomy in Tremor Disorders. Radiology. 2020. 294(3): 676–685. Hu J, Xiao C, Gong D,et al. Regional homogeneity analysis of major Parkinson’s disease subtypes based on functional magnetic resonance imaging. Neurosci Lett. 2019. 706: 81–87. Akbari H, Fei B. 3D ultrasound image segmentation using wavelet support vector machines. Med Phys. 2012. 39(6): 2972–84. Jing R, Wang J, Li J,et al. A wavelet features derived radiomics nomogram for prediction of malignant and benign early-stage lung nodules. Sci Rep. 2021. 11(1): 22330. Poston KL, Ua Cruadhlaoich M, Santoso LF,et al. Substantia Nigra Volume Dissociates Bradykinesia and Rigidity from Tremor in Parkinson's Disease: A 7 Tesla Imaging Study. J Parkinsons Dis. 2020. 10(2): 591–604. Kaya MO, Ozturk S, Ercan I,et al. Statistical Shape Analysis of Subthalamic Nucleus in Patients with Parkinson Disease. World Neurosurg. 2019. 126: e835-e841. Cao X, Wang X, Xue C,et al. A Radiomics Approach to Predicting Parkinson’s Disease by Incorporating Whole-Brain Functional Activity and Gray Matter Structure. Front Neurosci. 2020. 14: 751. Sun D, Wu X, Xia Y,et al. Differentiating Parkinson’s disease motor subtypes: A radiomics analysis based on deep gray nuclear lesion and white matter. Neurosci Lett. 2021. 760: 136083. Liu A, Wang Z, Yang Y,et al. Preoperative diagnosis of malignant pulmonary nodules in lung cancer screening with a radiomics nomogram. Cancer Commun (Lond). 2020. 40(1): 16–24. Wang K, Qiao Z, Zhao X,et al. Individualized discrimination of tumor recurrence from radiation necrosis in glioma patients using an integrated radiomics-based model. Eur J Nucl Med Mol Imaging. 2020. 47(6): 1400–1411. Jiang Z, Wang B, Han X,et al. Multimodality MRI-based radiomics approach to predict the posttreatment response of lung cancer brain metastases to gamma knife radiosurgery. Eur Radiol. 2022. 32(4): 2266–2276. 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-6223585","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Article","associatedPublications":[],"authors":[{"id":453029053,"identity":"10022263-a043-408d-b64c-1b2108991942","order_by":0,"name":"Shu-fen Liu","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Shu-fen","middleName":"","lastName":"Liu","suffix":""},{"id":453029054,"identity":"b3e22c93-ecc0-4a18-8c16-3c0e9b3c2990","order_by":1,"name":"Yuan-zhe Li","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yuan-zhe","middleName":"","lastName":"Li","suffix":""},{"id":453029055,"identity":"7a3c45e0-0ce0-45ed-b19a-5757e2ee22ed","order_by":2,"name":"Yi Wang","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yi","middleName":"","lastName":"Wang","suffix":""},{"id":453029057,"identity":"328f3b67-bc01-43de-a345-f9855f7712e0","order_by":3,"name":"Jian-long Zhuang","email":"","orcid":"","institution":"Quanzhou Women’s and Children’s Hospital","correspondingAuthor":false,"prefix":"","firstName":"Jian-long","middleName":"","lastName":"Zhuang","suffix":""},{"id":453029058,"identity":"7b39003a-dfb3-4ece-b548-1487d907f800","order_by":4,"name":"Lin-yi Li","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Lin-yi","middleName":"","lastName":"Li","suffix":""},{"id":453029059,"identity":"58a7c598-b6ed-4793-bac2-ba982ad790b9","order_by":5,"name":"Xiao-fang Ye","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xiao-fang","middleName":"","lastName":"Ye","suffix":""},{"id":453029060,"identity":"216cb88d-904c-4252-b5ea-2396e62ef0fd","order_by":6,"name":"Wan-li Huang","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Wan-li","middleName":"","lastName":"Huang","suffix":""},{"id":453029062,"identity":"d2ef7106-3695-417e-8201-304d669cc923","order_by":7,"name":"Bin-bin Yu","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Bin-bin","middleName":"","lastName":"Yu","suffix":""},{"id":453029063,"identity":"bcd805aa-7366-446d-949d-5452a0b968ed","order_by":8,"name":"Li-chao Ye","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Li-chao","middleName":"","lastName":"Ye","suffix":""},{"id":453029065,"identity":"2e52f479-81b1-4d5d-8fcf-47041a43cd6e","order_by":9,"name":"Xiang-rong Chen","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xiang-rong","middleName":"","lastName":"Chen","suffix":""},{"id":453029066,"identity":"9d981255-292b-4924-afa5-f812be2d8f8a","order_by":10,"name":"Mi-mi Li","email":"","orcid":"","institution":"Fujian Medical University","correspondingAuthor":false,"prefix":"","firstName":"Mi-mi","middleName":"","lastName":"Li","suffix":""},{"id":453029071,"identity":"5b10e902-73e3-4275-9d81-6d6fcd38efd4","order_by":11,"name":"Chunnuan Chen","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAABCklEQVRIiWNgGAWjYDACCSBmbIByEips7PgZmBsOEK/lwZm0ZMkGRhK0MD5sO8y44QDcBOyAf3bzsQc/dxzOM5dIfvYgsS2N2fj4wcbDvG13GPjbuxOwWnLnWLph75nDxZYz0swNEs7Z8JmdSWwAannGIHHm7AZsWgwkcsykGdsOJ264kWAmkVCWxmx2AKzlMFAqF4eW/G9QLenfJBLYDjNu7n9ISEsOG1RLDtAWkPclCNgicSPNTLK3LT1xw5k3ZRIJwECWuPGw4eCcc4d5cPmFf0byM4mfbdaJG46nb5P8AYrK/uTDH96UHZbjb+/FqgU7YOJhYOAhXjkIMP4gTf0oGAWjYBQMbwAAP0twNfcpq7UAAAAASUVORK5CYII=","orcid":"","institution":"Fujian Medical University","correspondingAuthor":true,"prefix":"","firstName":"Chunnuan","middleName":"","lastName":"Chen","suffix":""}],"badges":[],"createdAt":"2025-03-14 04:38:08","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6223585/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6223585/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82581870,"identity":"56b6b43b-894d-4513-9a9b-5bcf3d159ced","added_by":"auto","created_at":"2025-05-13 06:39:36","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":382053,"visible":true,"origin":"","legend":"\u003cp\u003eROI delineation with color coding: right hippocampus (sky blue), left hippocampus (pink), right amygdala (purple), left amygdala (orange), right substantia nigra (dark blue), left substantia nigra (green), right red nucleus (gray), left red nucleus (skin tone), right thalamus (red), and left thalamus (yellow).\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/561ac83f3c2f7b36c1091b70.png"},{"id":82582969,"identity":"e01393eb-d3d8-41be-b11a-63fe17a8f391","added_by":"auto","created_at":"2025-05-13 06:47:36","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":324759,"visible":true,"origin":"","legend":"\u003cp\u003eThe most significant imaging features and corresponding coefficients of T1WI in Hippocampal(A); MSE path diagram(B); LASSO coefficient path(C). The most significant imaging features of T2-FLAIR in Hippocampal and their corresponding coefficient maps(D); MSE path diagram(E); LASSO coefficient path(F).\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/c6666a3c534d4161789e09b8.png"},{"id":82581871,"identity":"c26bef9d-a183-469b-b23f-10413a6bdbb9","added_by":"auto","created_at":"2025-05-13 06:39:36","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":604531,"visible":true,"origin":"","legend":"\u003cp\u003eKey image features and their coefficients for T1WI in the Substantia Nigra (A), along with the MSE path (B) and LASSO coefficient path (C). Similarly, key image features and their coefficients for T2-FLAIR in the Substantia Nigra (D), the MSE path (E) and LASSO coefficient path (F) are shown. The same details are provided for T1WI in the Red Nucleus (G, H, I) and T2-FLAIR in the Red Nucleus (J, K, L).\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/32c8c531a4466d9ec149a444.png"},{"id":82582968,"identity":"49a12959-4679-4d36-aecb-c076d0f9f571","added_by":"auto","created_at":"2025-05-13 06:47:36","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":562195,"visible":true,"origin":"","legend":"\u003cp\u003eKey image features and their coefficientsfor T1WI in the Thalamus (A), along with the MSE path (B) and LASSO coefficient path (C). Similarly, key image features and their coefficients for T2-FLAIR in the Thalamus (D), the MSE path (E) and LASSO coefficient path (F) are shown. The same details are provided for T1WI in the Amygdala (G, H, I) and T2-FLAIR in the Amygdala (J, K, L).\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/beb2c20303b156fb50fffb3b.png"},{"id":82581874,"identity":"bfbd38f5-d8d9-4edf-99a2-ed25c4491cc3","added_by":"auto","created_at":"2025-05-13 06:39:36","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":332567,"visible":true,"origin":"","legend":"\u003cp\u003eThe ROC curves of the single-sequence radiomics model for the Hippocampus and Substantia Nigra in training and test sets. (A) Hippocampus-T1WI training set; (B) Hippocampus-T1WI test set; (C) Hippocampus-T2-FLAIR training set; (D) Hippocampus-T2-FLAIR test set; (E) Substantia Nigra-T1WI training set; (F) Substantia Nigra-T1WI test set; (G) Substantia Nigra-T2-FLAIR training set; (H) Substantia Nigra-T2-FLAIR test set. Class 0: Control group; Class 1: Non-tremor group; Class 2: Tremor group.\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/9215bf28a27b7aa4ba52c00b.png"},{"id":82584744,"identity":"6f142731-9b49-4dad-8b3c-53567bebc0b8","added_by":"auto","created_at":"2025-05-13 06:55:36","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":411839,"visible":true,"origin":"","legend":"\u003cp\u003eROC curves of the single-sequence radiomics model for the Red Nucleus, Thalamus, and Amygdala in training and test sets. (A) Red Nucleus-T1WI training set; (B) Red Nucleus-T1WI test set; (C) Red Nucleus-T2-FLAIR training set; (D) Red Nucleus-T2-FLAIR test set; (E) Thalamus-T1WI training set; (F) Thalamus-T1WI test set; (G) Thalamus-T2-FLAIR training set; (H) Thalamus-T2-FLAIR test set; (I) Amygdala-T1WI training set; (J) Amygdala-T1WI test set; (K) Amygdala-T2-FLAIR training set; (L) Amygdala-T2-FLAIR test set. Class 0: Control group; Class 1: Non-tremor group; Class 2: Tremor group\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/211862cfef7d8ee41615f3f5.png"},{"id":82581876,"identity":"26f76365-ccc1-4d23-ab3c-9513a9685381","added_by":"auto","created_at":"2025-05-13 06:39:36","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":391434,"visible":true,"origin":"","legend":"\u003cp\u003e(A) Multimodal radiomics features with corresponding coefficient maps; (B) MSE trajectory; (C) LASSO coefficient path. (D-E) ROC curves for the multimodal radiomics model in training and test sets, respectively. Class 0: Control group; Class 1: Non-tremor group; Class 2: Tremor group\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/06766f80d53a1e657648947f.png"},{"id":97892928,"identity":"e32df573-907f-48a7-bc00-f107534c4a89","added_by":"auto","created_at":"2025-12-10 15:24:17","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3423205,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6223585/v1/a8cb4059-33b9-4a99-a552-142b1255f4ee.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Analysis and Research on Predicting the Motor Classification of Parkinson's Disease Based on Radiomics","fulltext":[{"header":"Introduction","content":"\u003cp\u003eParkinson's disease (PD), the second most prevalent neurodegenerative disorder after Alzheimer's disease (AD), predominantly affects the elderly population with increasing incidence rates observed annually\u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. Its pathology involves substantia nigra dopaminergic neuron degeneration and Lewy body formation, leading to both motor (resting tremor, bradykinesia, gait abnormalities, rigidity) and non-motor symptoms (depression, sleep disorders, cognitive impairment)\u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]\u003c/sup\u003e. Clinical studies have further classified PD into tremor-dominant and non-tremor-dominant subtypes\u003csup\u003e[\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eTremor is defined as an involuntary, rhythmic, oscillatory movement of a body part\u003csup\u003e[\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]\u003c/sup\u003e.Current evidence suggests the cerebello-thalamo-cortical pathway plays a central role in tremor-related disorders. In PD, tremor signals are amplified through the motor cortex-ventral intermediate thalamic nucleus (Vim) circuit, explaining the efficacy of Vim thalamotomy and deep brain stimulation\u003csup\u003e[\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]\u003c/sup\u003e. Emerging research indicates PD tremor originates from the subthalamic nucleus, whose burst firing depends on motor cortex input, propagating through the cortex-subthalamic-pallidum-thalamus-cortical pathway\u003csup\u003e[\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e]\u003c/sup\u003e. The cerebellum may modulate tremor amplitude via the cerebello-thalamo-cortical circuit\u003csup\u003e[\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]\u003c/sup\u003e. At the same time, Iron deposition in the red nucleus, as demonstrated by susceptibility-weighted imaging, correlates with PD tremor\u003csup\u003e[\u003cspan citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e]\u003c/sup\u003e. As a dopamine production cofactor, iron serves as a biomarker for surviving substantia nigra dopaminergic neurons. Additionally, tremor severity appears influenced by emotional and cognitive states, with stress or cognitive load potentially amplifying tremor through thalamic arousal pathways\u003csup\u003e[\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e]\u003c/sup\u003e. The hippocampus, crucial for cognitive function\u003csup\u003e[\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e, shows structural plasticity changes in PD.\u003c/p\u003e \u003cp\u003eKim B et al.\u003csup\u003e[\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]\u003c/sup\u003e demonstrated hippocampal alterations in a 6-hydroxydopamine-induced PD rat model, supporting the hippocampus's involvement in PD pathology.\u003c/p\u003e \u003cp\u003eRecent studies have demonstrated Artificial Intelligence (AI)'s significant contribution to PD diagnosis\u003csup\u003e[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e]\u003c/sup\u003e. AI, primarily based on Machine Learning (ML), employs complex algorithms for decision-making, pattern recognition, and data analysis\u003csup\u003e[\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]\u003c/sup\u003e.ML algorithms are divided into supervised learning and unsupervised learning. Supervised learning includes Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), and Decision Tree (DT); unsupervised learning includes Hidden Markov Models (HMM)\u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e]\u003c/sup\u003e. Radiomics, pioneered by Philippe Lambin et al. in 2012\u003csup\u003e[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e, has emerged as a promising field that utilizes AI/ML to extract imaging features and develop predictive models\u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e .The selection of the best machine learning method for establishing radiomic methods is a crucial step. The choice of ML algorithms is critical in radiomics, with multiple studies indicating SVM's superior performance compared to other supervised classifiers \u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e] [\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e] [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eRadiomics involves five key steps: image acquisition, segmentation, feature extraction, feature selection, and model construction and evaluation\u003csup\u003e[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]\u003c/sup\u003e. By quantifying tissue characteristics like heterogeneity and shape, radiomic features combined with clinical data can enhance disease diagnosis, prognosis, and prediction accuracy\u003csup\u003e[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e, \u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e]\u003c/sup\u003e. This approach has shown clinical utility across various diseases \u003csup\u003e[\u003cspan additionalcitationids=\"CR24\" citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e]\u003c/sup\u003e, including PD.Prodoehl J\u003csup\u003e[\u003cspan citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e]\u003c/sup\u003e successfully differentiated PD, atypical Parkinsonism, essential tremor, and healthy controls using Diffusion Tensor Imaging with radiomic analysis of key regions (basal ganglia, red nucleus, dentate nucleus, cerebellum). Vitali P\u003csup\u003e[\u003cspan citationid=\"CR27\" class=\"CitationRef\"\u003e27\u003c/span\u003e]\u003c/sup\u003e achieved accurate classification of early PD, advanced PD, and healthy individuals through substantia nigra texture analysis. Additionally, radiomic studies have revealed significant differences in hippocampus, thalamus, and amygdala between PD patients and controls, correlating with cognitive performance\u003csup\u003e[\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eWhile MRI-based radiomics offers promising diagnostic tools for PD, research on its application in PD tremor subtypes remains limited. This study aims to develop single- and multi-modality radiomic models to improve PD subtype prediction and classification, ultimately enhancing disease management strategies.\u003c/p\u003e"},{"header":"Materials and Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003ePatients\u003c/h2\u003e \u003cp\u003eThis retrospective study enrolled 260 participants from January 2013 to December 2023, comprising 82 tremor-type PD, 78 non-tremor PD, and 100 control cases. Approved by the Ethics Committee of the Second Affiliated Hospital of Fujian Medical University (Approval No. 2023 Fu Medical Affiliated No. 2 Ethics Review (286)), the study followed strict inclusion and exclusion criteria.\u003c/p\u003e \u003cp\u003eInclusion criteria required: (1) Standardized cranial MRI with T1-Weighted Imaging (T1WI) and T2-Weighted Fluid-Attenuated Inversion Recovery (T2-FLAIR) sequences using consistent scanning parameters; (2) PD diagnosis confirmed by two neurologists according to 2015 MDS diagnostic criteria\u003csup\u003e[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]\u003c/sup\u003e, with consensus resolution for discrepancies; (3) Complete imaging and clinical data; (4) Age 50\u0026ndash;80 years. Exclusion criteria included: (1) Cognitive impairment from other causes, cerebrovascular diseases, or intracranial abnormalities; (2) Substandard image quality; (3) Incomplete clinical data.\u003c/p\u003e \u003cp\u003ePD patients were classified into tremor group based on presence of resting tremor in head/neck or limbs according to the MDS Unified Parkinson's Disease Rating Scale (MDS-UPDRS), while those without resting tremor were assigned to non-tremor group. Controls were selected from individuals undergoing cranial MRI during the same period, confirmed PD-free and tremor-negative, meeting the same exclusion criteria.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMagnetic Resonance Imaging\u003c/h3\u003e\n\u003cp\u003eAll participants underwent 3.0T MRI (Dutch Philips Ingenia) using an 8-channel head coil. During scanning, subjects maintained a supine position with head stabilization using foam pads and noise-reducing earplugs, remaining conscious and relaxed. The imaging protocol included axial T1WI (repetition time (TR)/echo time (TE): 7.9/3.0 ms) and T2-FLAIR (TR/TE: 9000/100 ms) sequences.\u003c/p\u003e\n\u003ch3\u003eImage Acquisition and Delineation of Regions of Interest\u003c/h3\u003e\n\u003cp\u003eFigure \u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e illustrates the radiomics workflow, which includes image acquisition and Regions of interest(ROI) delineation. Using ITK-SNAP software (v3.8.0; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttp://www.itk-snap.org\u003c/span\u003e\u003cspan address=\"http://www.itk-snap.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e), we processed anonymized cranial MRI the Picture Archiving and Communication System (PACS) files exported from the Picture Archiving and Communication System (PACS). A trained specialist manually outlined bilateral ROIs (hippocampus, substantia nigra, red nucleus, thalamus, and amygdala) on both T1WI and T2-FLAIR sequences (Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e\n\u003ch3\u003eRadiomic Feature Extraction, Selection, and Model Building\u003c/h3\u003e\n\u003cp\u003eParticipants were randomly allocated to training and test sets (8:2 ratio). Radiomic features were extracted from T1WI and T2-FLAIR ROIs using Pyradiomics (Python v3.6; \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://pyradiomics.readthedocs.io\u003c/span\u003e\u003cspan address=\"https://pyradiomics.readthedocs.io\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e). Following feature standardization, we implemented a two-step feature selection: (1) minimal Redundancy Maximal Relevance (mRMR) for eliminating redundant features\u003csup\u003e[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e]\u003c/sup\u003e, and (2) LASSO logistic regression for optimal feature subset selection through regularization parameter λ optimization. LASSO's penalty function compresses coefficients of irrelevant variables, reducing model complexity while addressing collinearity and overfitting.\u003c/p\u003e \u003cp\u003eSVM classifier was employed for model construction. SVM operates as a maximum-margin linear classifier, transforming data separation into convex quadratic programming. Based on Vapnik-Chervonenkis Dimension theory and structural risk minimization principles, SVM optimally balances model complexity and learning capacity, particularly effective for small sample sizes. The SVM model identified key Parkinson's-related features from training data, enabling effective differentiation between PD subtypes and healthy controls.\u003c/p\u003e\n\u003ch3\u003eStatistics\u003c/h3\u003e\n\u003cp\u003eStatistical analyses were performed using SPSS and R software. Quantitative radiomic scores were generated from SVM-based models for both training and testing groups. We developed single-sequence and multimodal radiomic models using T1WI and T2-FLAIR sequences from five brain regions (hippocampus, substantia nigra, red nucleus, thalamus, and amygdala). Model performance was evaluated through Macro-avg, Micro-avg, and ROC curve analyses, with AUC, sensitivity, specificity, and accuracy as key metrics\u003csup\u003e[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e, \u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMicro-avg assesses overall performance across all categories, while Macro-avg evaluates category-specific performance before averaging. AUC values\u0026thinsp;\u0026gt;\u0026thinsp;0.8 and 0.7\u0026ndash;0.8 indicate good and moderate discriminative power, respectively. Higher sensitivity reflects better positive sample identification, specificity indicates superior negative sample recognition, and accuracy represents overall model performance\u003c/p\u003e \u003cp\u003eFinally, the DeLong test by MedCalc (\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ewww.medcalc.org\u003c/a\u003e\u003c/span\u003e\u003cspan address=\"http://www.medcalc.org\" targettype=\"URL\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e) was used to compare the differences between the ROC curves of different models, considering a \u003cem\u003ep\u003c/em\u003e\u0026lt;0.05 as indicating a statistically significant difference.\u003c/p\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003e3.1 Comparison of General Clinical Data\u003c/h2\u003e \u003cp\u003eBased on the inclusion and exclusion criteria, the study enrolled 260 participants: 82 tremor-type PD patients (38 male, 44 female), 78 non-tremor-type PD patients (41 male, 37 female), and 100 controls (39 male, 61 female). No significant gender differences were observed among groups (\u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026gt;\u0026thinsp;0.05) (Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eThe basic information of included people\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"5\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eControl group\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003enon-tremor type PD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003etremor type PD\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003e\u003cem\u003ep\u003c/em\u003e value\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTotal number of people\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e100\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e78\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e82\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGender (Male/Female)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e39/61\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e41/37\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e38/44\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.192\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e64.8\u0026thinsp;\u0026plusmn;\u0026thinsp;8.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e67.6\u0026thinsp;\u0026plusmn;\u0026thinsp;7.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e70.1\u0026thinsp;\u0026plusmn;\u0026thinsp;7.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e-\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eThe data are presented as mean\u0026thinsp;\u0026plusmn;\u0026thinsp;SD. \u003cem\u003ep\u003c/em\u003e\u0026thinsp;\u0026lt;\u0026thinsp;0.05 showed signifcant diference between different groups.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003e3.2 Based on Single-Sequence Radiomic Models\u003c/h3\u003e\n\u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003e3.2.1 Radiomic Feature Selection\u003c/h2\u003e \u003cp\u003eSingle-sequence radiomic models were developed using T1WI and T2-FLAIR sequences, focusing on bilateral regions including the Hippocampus, Substantia Nigra, Red Nucleus, Thalamus, and Amygdala. A total of 1197 radiomic features were initially extracted. Following feature selection with mRMR and LASSO, the optimal features for each region were identified: Hippocampus: 20 from T1WI and 12 from T2-FLAIR (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e); Substantia Nigra: 18 from T1WI and 19 from T2-FLAIR (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e);Red Nucleus: 19 from T1WI and 16 from T2-FLAIR (Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e); Thalamus: 16 from T1WI and 17 from T2-FLAIR (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e); Amygdala: 23 from T1WI and 10 from T2-FLAIR (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003e3.2.2 Diagnostic efficacy of a single sequence radiomics model\u003c/h2\u003e \u003cp\u003eIn the Hippocampal-T1WI radiomics model, the training set AUC values for the control, non-tremor PD, and tremor PD groups were 0.940, 0.923, and 0.914, respectively, while the test set AUC values were 0.834, 0.740, and 0.524. For the Hippocampal-T2-FLAIR model, the training set AUC values were 0.911, 0.844, and 0.754, with corresponding test set values of 0.830, 0.747, and 0.656 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eIn the Substantia Nigra-T1WI model, the training set AUC values were 0.928, 0.856, and 0.905, while the test set values were 0.619, 0.559, and 0.562. For the Substantia Nigra-T2-FLAIR model, the training set AUC values were 0.895, 0.689, and 0.673, with test set values of 0.697, 0.639, and 0.566 (Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eFor the Red Nucleus-T1WI radiomics model, the training set AUC values for the control, non-tremor PD, and tremor PD groups were 0.915, 0.659, and 0.716, respectively, with test set values of 0.547, 0.693, and 0.335. In the Red Nucleus-T2-FLAIR model, the training set AUC values were 0.920, 0.894, and 0.888, while the test set values were 0.748, 0.688, and 0.578 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). For the Thalamus-T1WI model, the training set AUC values were 0.914, 0.850, and 0.866, with test set values of 0.684, 0.684, and 0.651. In the Thalamus-T2-FLAIR model, the training set AUC values were 0.902, 0.872, and 0.873, while the test set values were 0.817, 0.844, and 0.556 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e). For the Amygdala-T1WI model, the training set AUC values were 0.905, 0.867, and 0.871, with test set values of 0.728, 0.715, and 0.424. In the Amygdala-T2-FLAIR model, the training set AUC values were 0.886, 0.819, and 0.710, while the test set values were 0.775, 0.689, and 0.602 (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003e)\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe AUC values, sensitivity, accuracy, and specificity of each individual sequence radiomics model are shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAUC values, sensitivity, specificity, and accuracy based in single sequence imaging omics model\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003eBrain Region\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eAUC(A/B)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eSEN(A/B)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eSPE(A/B)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eACC(A/B)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" morerows=\"14\" nameend=\"c2\" namest=\"c1\" rowspan=\"15\"\u003e \u003cp\u003eT1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eHippocampal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.940/0.834\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.838/0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.859/0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.788/0.652\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.923/0.740\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.774/0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.925/0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.814/0.471\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.914/0.524\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.727/0.250\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.887/0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.750/0.333\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eSubstantia\u003c/p\u003e \u003cp\u003eNigra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.928/0.619\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.950/0.550\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.820/0.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.768/0.478\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.856/0.559\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.629/0.125\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.979/0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.929/0.250\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.905/0.562\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.818/0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.908/0.639\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.806/0.381\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eRed Nucleus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.915/0.547\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.925/0.450\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.766/0.448\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.712/0.360\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.659/0.693\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.661/0.438\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.925/0.848\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.788/0.583\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.716/0.335\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.652/0.154\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.937/0.722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.827/0.167\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eThalamus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.914/0.684\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.900/0.650\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.734/0.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.679/0.520\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.850/0.684\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.581/0.375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.952/0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.837/0.500\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.866/0.651\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.727/0.438\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.923/0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.814/0.467\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eAmygdala\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.905/0.728\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.913/0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.758/0.531\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.702/0.500\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.867/0.715\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.661/0.188\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.925/0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.788/0.333\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.871/0.424\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.621/0.125\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.923/0.694\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.788/0.154\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eT2-\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eHippocampal\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.911/0.830\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.889/0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.781/0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.717/0.667\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e \u003cp\u003eFLAIR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.844/0.747\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.613/0.500\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.911/0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.745/0.615\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.754/0.656\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.606/0.375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.873/0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.690/0.400\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eSubstantia\u003c/p\u003e \u003cp\u003eNigra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.895/0.697\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.888/0.600\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.781/0.656\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.717/0.522\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.689/0.639\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.694/0.313\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.945/0.806\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.843/0.417\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.673/0.566\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.727/0.313\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.930/0.667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.828/0.294\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eRed Nucleus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.920/0.748\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.938/0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.789/0.563\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.735/0.533\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.894/0.688\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.661/0.563\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.945/0.806\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.837/0.563\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.888/0.578\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.667/0.188\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.908/0.917\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.772/0.500\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eThalamus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.902/0.817\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.838/0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.917/0.844\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.882/0.762\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.872/0.844\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.806/0.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.835/0.778\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.704/0.556\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.873/0.556\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.766/0.375\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.958/0.806\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.857/0.462\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\" morerows=\"2\" rowspan=\"3\"\u003e \u003cp\u003eAmygdala\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.886/0.775\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.850/0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.805/0.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.731/0.571\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.819/0.689\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.839/0.438\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.747/0.722\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.584/0.412\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c2\" namest=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.710/0.602\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.318/0.063\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c7\"\u003e \u003cp\u003e0.965/0.833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c8\"\u003e \u003cp\u003e0.808/0.143\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e* AUC, mean area under the curve; SEN, sensitivity; SPE, specificity; ACC, accuracy; A/B, Training set/Test set.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec13\" class=\"Section2\"\u003e \u003ch2\u003e3.3 Based on multi sequence radiomics model\u003c/h2\u003e \u003cp\u003eA multimodal radiomics model was developed using T1WI and T2-FLAIR sequences from the Hippocampus, Substantia Nigra, Red Nucleus, Thalamus, and Amygdala. A total of 140 radiomics features were extracted, distributed as follows: Hippocampus (26 features: 16 from T1WI, 10 from T2-FLAIR), Substantia Nigra (34 features:19 from T1WI, 15 from T2-FLAIR), Red Nucleus (27 features: 15 from T1WI, 12 from T2-FLAIR), Thalamus (27 features: 13 from T1WI, 14 from T2-FLAIR), and Amygdala (26 features: 18 from T1WI, 8 from T2-FLAIR).\u003c/p\u003e \u003cp\u003eFollowing feature selection using mRMR and LASSO, 18 optimal features were identified from the multimodal radiomics model. These comprised: 2 from Hippocampus-T1WI, 1 from Hippocampus-T2-FLAIR, 2 from Substantia Nigra-T1WI, 1 from Substantia Nigra-T2-FLAIR, 1 from Thalamus-T1WI, 7 from Thalamus-T2-FLAIR, 2 from Red Nucleus-T2-FLAIR, and 2 from Amygdala-T2-FLAIR (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e).\u003c/p\u003e \u003cp\u003eThe multimodal radiomics model demonstrated strong diagnostic performance, with training group AUC values of 0.966 (control), 0.952 (non-tremor PD), and 0.942 (tremor PD). Corresponding test group AUC values were 0.877, 0.861, and 0.760 respectively (Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e). The results indicate that multimodal radiomics outperforms single image sequence analysis in differentiating between control, non-tremor PD, and tremor PD groups. Detailed performance metrics, including AUC, sensitivity, specificity, and accuracy, are presented in Table\u0026nbsp;\u003cspan refid=\"Tab3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab3\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 3\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eAUC values, sensitivity, specificity, and accuracy in multimodal imaging omics models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGroup\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSEN\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eSPE\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eACC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eControl\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.966\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.900\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.898\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.847\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.877\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.850\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.844\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.773\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNon-tremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.952\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.887\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.918\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.821\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.861\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.706\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTremor\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.942\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.742\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.951\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.875\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eTest set\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0.760\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.563\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.889\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.692\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003e* AUC, mean area under the curve; SEN, sensitivity; SPE, specificity; ACC, accuracy; A/B, Training set/Test set.\u003c/p\u003e \u003cp\u003eThe DeLong test was employed to compare ROC curve differences across models. In the training set, significant differences were observed between the multimodal radiomics model and all single-sequence models (Hippocampus-T2-FLAIR, Substantia Nigra-T1WI/T2-FLAIR, Red Nucleus-T1WI/T2-FLAIR, Thalamus-T1WI/T2-FLAIR, and Amygdala-T1WI/T2-FLAIR). In the test set, significant differences were found between the multimodal model and single-sequence models using Substantia Nigra-T1WI/T2-FLAIR, Red Nucleus-T1WI/T2-FLAIR, and Thalamus-T1WI. Additionally, the Hippocampus-T1WI single-sequence model showed significant differences compared to Substantia Nigra-T1WI and Red Nucleus-T1WI models. Other comparisons were not statistically significant (Table\u0026nbsp;\u003cspan refid=\"Tab4\" class=\"InternalRef\"\u003e4\u003c/span\u003e).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab4\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 4\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ep-values between different models by Delong test.\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/th\u003e \u003cth align=\"left\" colspan=\"3\" nameend=\"c4\" namest=\"c2\"\u003e \u003cp\u003eTraining set\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colspan=\"2\" nameend=\"c6\" namest=\"c5\"\u003e \u003cp\u003eTest set\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e\u0026nbsp;\u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003eT1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c4\" namest=\"c3\"\u003e \u003cp\u003eT2-FLAIR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003eT1WI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003eT2-FLAIR\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHippocampal vs Substantia Nigra\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.6092\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.5118\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0379\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0595\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHippocampal vs Red Nucleus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.3050\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.6703\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0045\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0999\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHippocampal vs Thalamus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.2763\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.6838\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.1548\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.8253\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHippocampal vs Amygdala\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.1331\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.2742\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.2750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.5211\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubstantia Nigra vs Red Nucleus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.5447\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.2992\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.4787\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.4219\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubstantia Nigra vs Thalamus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.5168\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.7769\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.4427\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0693\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubstantia Nigra vs Amygdala\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.3426\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.7269\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.1423\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.2480\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed Nucleus vs Thalamus\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.9470\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.4058\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.1400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.2549\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed Nucleus vs Amygdala\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.6744\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.2040\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0581\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.5774\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThalamus vs Amygdala\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.7135\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.5036\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.6181\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.6382\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHippocampal vs Multimodality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.1184\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.0017\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.5134\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.3667\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSubstantia Nigra vs Multimodality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.0278\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.0003\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0060\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0075\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRed Nucleus vs Multimodality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.0133\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.0159\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0009\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.0303\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eThalamus vs Multimodality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.0056\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003e0.0004\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0323\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.1922\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAmygdala vs Multimodality\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colspan=\"2\" nameend=\"c3\" namest=\"c2\"\u003e \u003cp\u003e0.0025\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c4\"\u003e \u003cp\u003ep\u0026thinsp;\u0026lt;\u0026thinsp;0.0001\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c5\"\u003e \u003cp\u003e0.0666\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c6\"\u003e \u003cp\u003e0.1876\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003e* P\u0026thinsp;\u0026lt;\u0026thinsp;0.05 has statistical significance\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003ePD demonstrates significant clinical heterogeneity, often leading to misdiagnosis or delayed detection due to the absence of definitive diagnostic methods for clinical subtypes. Early detection and intervention are crucial for effective PD management. While PD's hallmark is dopaminergic neuron degeneration in the substantia nigra and striatum, clinical symptoms typically emerge only after substantial neuronal loss, indicating a temporal disconnect between pathology and symptom onset.\u003c/p\u003e \u003cp\u003eClinically, PD manifests as tremor-dominant and non-tremor subtypes, with tremor pathology involving multiple brain regions. The basal ganglia-thalamus-cortical pathway, comprising the striatum, globus pallidus, subthalamic nucleus, and substantia nigra pars reticulata, plays a pivotal role in PD tremor generation through its connections with cortical, thalamic, and midbrain dopaminergic systems\u003csup\u003e[\u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e]\u003c/sup\u003e. Recent studies have further implicated the hippocampus, thalamus, red nucleus, and amygdala in PD subtype heterogeneity\u003csup\u003e[\u003cspan citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e, \u003cspan citationid=\"CR35\" class=\"CitationRef\"\u003e35\u003c/span\u003e]\u003c/sup\u003e.While neuroimaging aids PD differential diagnosis, it cannot independently confirm PD. Radiomics has emerged as a promising diagnostic tool, enabling extraction of high-throughput quantitative features from MRI beyond visual analysis. Although radiomics has demonstrated diagnostic and prognostic value in various diseases, its application in PD subtype classification remains limited.\u003c/p\u003e \u003cdiv id=\"Sec15\" class=\"Section2\"\u003e \u003ch2\u003eRadiomic Features and Their Correlation with PD\u003c/h2\u003e \u003cp\u003eIn radiomics, the fundamental characteristics that differentiate lesion from non-lesion images include gray-level contrast, texture uniformity, depth, and roughness. Concurrently, the approach to feature selection is pivotal in the development of radiomic models. In this investigation, employing LASSO, we identified 20 optimal radiomic features from 1197 for the Hippocampus-T1WI, 12 for the Hippocampus- T2-FLAIR, 18 for the Substantia Nigra-T1WI, 19 for the Substantia Nigra-T2-FLAIR, 19 for the Red Nucleus-T1WI, 16 for the Red Nucleus-T2-FLAIR, 16 for the Thalamus-T1WI, 17 for the Thalamus-T2-FLAIR, 23 for the Amygdala-T1WI, and 10 for the Amygdala-T2-FLAIR. Analogously, 18 optimal radiomic features were selected from 140 in multimodal radiomics. Predominantly, the selected features are high-order, characterized by wavelet filters. The wavelet filter processing algorithm stands out for its ability to discern varying resolutions and preserve signal details across these resolutions. Its capacity to magnify and reduce enables the transformation of signals to ROI while maintaining the resolution of that signal segment\u003csup\u003e[\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]\u003c/sup\u003e. Moreover, wavelet filters enhance the information content of low-frequency signals and unveil deeper, high-throughput features beyond the discernment of the naked eye\u003csup\u003e[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e]\u003c/sup\u003e. Additionally, their scalability renders them suitable for molecular biological signals and images exhibiting fractal or scale-invariant properties. Consequently, this study's LASSO-selected radiomic features are predominantly wavelet filters.\u003c/p\u003e \u003cp\u003eThe findings of this study indicate that the multimodal radiomic model exhibits superior predictive performance compared to the single-sequence radiomic model. The AUC values for the training and test groups are as follows: 0.966 and 0.877 for the healthy control group, 0.952 and 0.861 for the non-tremor type PD group, and 0.942 and 0.760 for the tremor type PD group, respectively. Within the realm of multimodal radiomics, LASSO was utilized to select 18 optimal radiomic features with the highest predictive efficacy. Notably, the feature with the greatest coefficient is wavelet_LLH_glrlm_GrayLevelNonUniformity_Hippocampus_T2-FLAIR, underscoring the significant role that variations in image gray levels play in predicting PD typing and differentiation. Among all feature classifications, GLRLM features constitute the largest proportion. These features characterize the coarseness and directional properties of image textures, such as the run length of textures at specific angles and directions, which are influenced by the voxel gray level distribution of the image. Changes in these distributions can reflect lesion heterogeneity, making GLRLM particularly sensitive for predicting PD typing and differentiation. Additionally, radiomic features derived from the substantia nigra account for the largest proportion, likely due to the pathological hallmark of PD: the degeneration and loss of dopaminergic neurons in the substantia nigra pars compacta. This observation is supported by Poston K\u003csup\u003e[\u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]\u003c/sup\u003e, who demonstrated a strong correlation between the substantia nigra and the severity of PD motor symptoms, further validating the results of this study. Kaya O\u003csup\u003e[\u003cspan citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e]\u003c/sup\u003e also highlighted significant shape differences in the subthalamic nucleus of PD patients compared to healthy controls. In this study, shape features represent the second-largest proportion after GLRLM, reinforcing the critical importance of shape-based characteristics in PD analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec16\" class=\"Section2\"\u003e \u003ch2\u003eComparison of Radiomics Performance Between Single Sequence and Multimodal Sequence Imaging\u003c/h2\u003e \u003cp\u003eThis study utilizes T1WI and T2-FLAIR images from cranial MRI plain scans, with participants randomly allocated into training and test groups at an 8:2 ratio. Support SVM was employed to construct both single-sequence and multimodal radiomic models, from which the optimal model for predicting PD typing was selected. In the single-sequence radiomic models developed based on T1WI and T2-FLAIR images of the hippocampus, substantia nigra, red nucleus, thalamus, and amygdala, a quantitative assessment of each model's performance was conducted. It was observed that the single-sequence radiomic models based on Hippocampus-T1WI, Hippocampus-T2-FLAIR, Substantia Nigra-T1WI, Red Nucleus-T2-FLAIR, Thalamus-T1WI, Thalamus-T2-FLAIR, Amygdala-T1WI, and Amygdala-T2-FLAIR exhibited high efficacy in predicting PD types and distinguishing healthy controls within the training set. The Hippocampus-T1WI model demonstrated the highest performance (AUC values for the control group, non-tremor PD, and tremor PD were 0.940, 0.923, and 0.914, respectively), although the differences compared to other single-sequence models were not statistically significant (P\u0026thinsp;\u0026gt;\u0026thinsp;0.05). However, the sensitivity, specificity, and accuracy of the test group were lower than those of the training group, potentially due to the limited sample size. Despite this, the clinical relevance of single-sequence radiomic models remains significant. For instance, Cao X\u003csup\u003e[\u003cspan citationid=\"CR40\" class=\"CitationRef\"\u003e40\u003c/span\u003e]\u003c/sup\u003e utilized SVM to develop a single-modality radiomic model for differentiating PD from healthy controls, achieving an AUC and accuracy of 100% in the training group and an AUC of 0.97 in the test group, underscoring the utility of single-sequence radiomics in PD differentiation. Similarly, SUN D\u003csup\u003e[\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e]\u003c/sup\u003e identified the left hippocampus as the region with the best radiomic features for distinguishing postural instability-gait difficulty type PD, tremor type PD, and healthy controls (training group AUC\u0026thinsp;=\u0026thinsp;0.889, specificity\u0026thinsp;=\u0026thinsp;80.0%, sensitivity\u0026thinsp;=\u0026thinsp;88.2%, accuracy\u0026thinsp;=\u0026thinsp;82.4%; test group AUC\u0026thinsp;=\u0026thinsp;0.833, specificity\u0026thinsp;=\u0026thinsp;83.3%, sensitivity\u0026thinsp;=\u0026thinsp;75.0%, accuracy\u0026thinsp;=\u0026thinsp;80.7%). Furthermore, Liu A et al.\u003csup\u003e[\u003cspan citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e]\u003c/sup\u003e constructed a single-sequence radiomic model based on chest CT using LASSO and multivariate logistic regression to differentiate benign from malignant pulmonary nodules, achieving AUC values of 0.836 and 0.809 in the test and training groups, respectively. This highlights the model's robust diagnostic capability in early lung cancer screening. In summary, single-sequence radiomics plays a critical role in the diagnosis and differential diagnosis of various diseases. This study further confirms that the single-sequence radiomic model based on Hippocampus-T1WI exhibits strong predictive performance in PD clinical typing and differentiation.\u003c/p\u003e \u003cp\u003eThis study further investigated the role of multimodal radiomics models in PD typing and differentiation. The results revealed that the AUC values for the control group in the training and test sets were 0.966 and 0.877, respectively. Similarly, the AUC values for the non-tremor PD group were 0.952 and 0.861 in the training and test sets, respectively, while for the tremor PD group, the AUC values were 0.942 and 0.760, respectively. Compared to single-sequence radiomics, multimodal radiomics demonstrated superior predictive performance in PD typing and distinguishing healthy controls. Additionally, the Delong test indicated statistically significant differences between multimodal radiomics and single-sequence radiomic models based on Substantia Nigra-T1WI (p\u0026thinsp;=\u0026thinsp;0.0278), Substantia Nigra-T2-FLAIR (p\u0026thinsp;=\u0026thinsp;0.0003), Red Nucleus-T1WI (p\u0026thinsp;=\u0026thinsp;0.0133), Red Nucleus-T2-FLAIR (p\u0026thinsp;=\u0026thinsp;0.0159), Thalamus-T1WI (p\u0026thinsp;=\u0026thinsp;0.0056), and Amygdala-T1WI (p\u0026thinsp;=\u0026thinsp;0.0004). However, no statistically significant differences were observed when compared to other single-sequence radiomic models. Despite this, multimodal radiomics models have been validated to exhibit excellent predictive, differential, and diagnostic performance for various diseases. For instance, Wang K et al.\u003csup\u003e[\u003cspan citationid=\"CR43\" class=\"CitationRef\"\u003e43\u003c/span\u003e]\u003c/sup\u003e developed a multimodal radiomic model for glioma recurrence prediction based on 8F-FDG, 11C-methionine (11C-MET) PET, and MRI images, combined with clinical data. Using LASSO binary logistic regression analysis, they achieved an AUC of 0.932 in the training group and 0.910 in the test group, highlighting the superior diagnostic performance of multimodal models over single-modal ones. Similarly, Jiang Z et al.\u003csup\u003e[\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]\u003c/sup\u003e constructed a multimodal radiomic model using multimodal MRI images and Random Forest (RF) to predict the treatment response of lung cancer brain metastasis to gamma knife therapy. The model achieved AUC values of 0.930 and 0.8532 in the training and test groups, respectively, outperforming single-modal radiomic models based on T1WI (AUC values of 0.722 and 0.656 in the training and test sets, respectively) and T2-FLAIR (AUC values of 0.805 and 0.704 in the training and test sets, respectively).\u003c/p\u003e \u003cp\u003eIn summary, multimodal radiomics demonstrates superior predictive and diagnostic capabilities compared to single-modal models in the context of disease assessment.\u003c/p\u003e \u003c/div\u003e"},{"header":"Conclusion","content":"\u003cp\u003eOur findings demonstrate that the Hippocampus-T2-FLAIR-based unimodal radiomics model shows superior performance in PD classification and differential diagnosis. Furthermore, the multimodal radiomics model, integrating multiple sequence images, outperforms single-sequence models in predictive accuracy for PD classification and differentiation.\u003c/p\u003e"},{"header":"Declarations","content":"\u003ch2\u003eEthics approval and consent to participate\u003c/h2\u003e\n\u003cp\u003eThis study has been granted an exemption from requiring ethics approval by The Second Affiliated Hospital of Fujian Medical University. All methods were performed in accordance with the relevant guidelines and regulations.\u003c/p\u003e\n\u003ch2\u003eConsent for publication\u003c/h2\u003e\n\u003cp\u003eWritten informed consent for publication of their details was obtained from the patient.\u003c/p\u003e\n\u003ch2\u003eAuthor contributions\u003c/h2\u003e\n\u003cp\u003eSL and CC conceived the study, designed the methodology, and conducted the primary data analysis. YL, YW and JZ contributed to the literature review, data collection, and interpretation of results. LL, XY, WH，BY, LY, XC and ML assisted in experimental design, provided critical feedback, and revised the manuscript. All authors participated in writing, editing, and approving the final version of the manuscript.\u003c/p\u003e\n\u003ch2\u003eAcknowledgements\u003c/h2\u003e\n\u003cp\u003eThis work was supported by grants from the Natural Science Foundation of Fujian Province of China (NO.2023J01104), the Scientific Foundation of Quanzhou City for High Level Talents (NO.2023C005YR), Joint funds for the innovation of science and technology, Fujian province (NO.2023Y9255), and Fujian Health Science and Technology Plan Project (NO.2024GGA039) from Dr. Chunnuan Chen.\u003c/p\u003e\n\u003ch2\u003eFunding\u003c/h2\u003e\n\u003cp\u003eThis work was supported by\u0026nbsp;the Natural Science Foundation of Fujian Province of China under Grant number 2023J01104; the Scientific Foundation of Quanzhou City for High Level Talents under Grant number 2023C005YR; Joint funds for the innovation of science and technology, Fujian province under Grant number 2023Y9255; and Fujian Health Science and Technology Plan Project under Grant number 2024GGA039.\u003c/p\u003e\n\u003ch2\u003eCompeting interests\u003c/h2\u003e\n\u003cp\u003eThe authors report no conflict of interest.\u0026nbsp;\u003c/p\u003e\n\u003ch2\u003eAvailability of data and material\u003c/h2\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eZhu B, Kohn R, Patel A,et al. Demoralization and Quality of Life of Patients with Parkinson Disease. Psychother Psychosom. 2021. 90(6): 415\u0026ndash;421.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuatteo E, Berretta N, Monda V,et al. Pathophysiological Features of Nigral Dopaminergic Neurons in Animal Models of Parkinson's Disease. Int J Mol Sci. 2022. 23(9): 4508.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMortimer JA, Borenstein AR, Nelson LM. Associations of welding and manganese exposure with Parkinson disease: review and meta-analysis. Neurology. 2012. 79(11): 1174\u0026ndash;80.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBhatia KP, Bain P, Bajaj N,et al. Consensus Statement on the classification of tremors. from the task force on tremor of the International Parkinson and Movement Disorder Society. Mov Disord. 2018. 33(1): 75\u0026ndash;87.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKremer NI, Pauwels R, Pozzi NG,et al. Deep Brain Stimulation for Tremor: Update on Long-Term Outcomes, Target Considerations and Future Directions. J Clin Med. 2021. 10(16): 3468.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHuang CS, Wang GH, Chuang HH,et al. Conveyance of cortical pacing for parkinsonian tremor-like hyperkinetic behavior by subthalamic dysrhythmia. Cell Rep. 2021. 35(3): 109007.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhong Y, Liu H, Liu G,et al. A review on pathology, mechanism, and therapy for cerebellum and tremor in Parkinson's disease. NPJ Parkinsons Dis. 2022. 8(1): 82.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuan X, Xuan M, Gu Q,et al. Influence of regional iron on the motor impairments of Parkinson's disease: A quantitative susceptibility mapping study. J Magn Reson Imaging. 2017. 45(5): 1335\u0026ndash;1342.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDirkx MF, Zach H, van Nuland AJ,et al. Cognitive load amplifies Parkinson\u0026rsquo;s tremor through excitatory network influences onto the thalamus. Brain. 2020. 143(5): 1498\u0026ndash;1511.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaurer AP, Nadel L. The Continuity of Context: A Role for the Hippocampus. Trends Cogn Sci. 2021. 25(3): 187\u0026ndash;199.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKim B, Weerasinghe-Mudiyanselage P, Ang MJ,et al. Changes in the Neuronal Architecture of the Hippocampus in a 6-Hydroxydopamine-Lesioned Rat Model of Parkinson Disease. Int Neurourol J. 2022. 26(Suppl 2): S94-105.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBirkenbihl C, Ahmad A, Massat NJ,et al. Artificial intelligence-based clustering and characterization of Parkinson\u0026rsquo;s disease trajectories. Sci Rep. 2023. 13(1): 2897.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGupta R, Kumari S, Senapati A,et al. New era of artificial intelligence and machine learning-based detection, diagnosis, and therapeutics in Parkinson\u0026rsquo;s disease. Ageing Res Rev. 2023. 90: 102013.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSarker IH. Machine Learning: Algorithms, Real-World Applications and Research Directions. SN Comput Sci. 2021. 2(3): 160.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLambin P, Rios-Velazquez E, Leijenaar R,et al. Radiomics: extracting more information from medical images using advanced feature analysis. Eur J Cancer. 2012. 48(4): 441\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeJohn CR, Grant SR, Seshadri M. Application of Machine Learning Methods to Improve the Performance of Ultrasound in Head and Neck Oncology: A Literature Review. Cancers (Basel). 2022. 14(3): 665.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOzer ME, Sarica PO, Arga KY. New Machine Learning Applications to Accelerate Personalized Medicine in Breast Cancer: Rise of the Support Vector Machines. OMICS. 2020. 24(5): 241\u0026ndash;246.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZheng Y, Zhou D, Liu H, Wen M. CT-based radiomics analysis of different machine learning models for differentiating benign and malignant parotid tumors. Eur Radiol. 2022. 32(10): 6953\u0026ndash;6964.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi C, Chen H, Zhang B,et al. Radiomics Signature Based on Support Vector Machines for the Prediction of Pathological Complete Response to Neoadjuvant Chemoradiotherapy in Locally Advanced Rectal Cancer. Cancers (Basel). 2023. 15(21): 5134.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang YP, Zhang XY, Cheng YT,et al. Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling. Mil Med Res. 2023. 10(1): 22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMayerhoefer ME, Materka A, Langs G,et al. Introduction to Radiomics. J Nucl Med. 2020. 61(4): 488\u0026ndash;495.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eTunali I, Gillies RJ, Schabath MB. Application of Radiomics and Artificial Intelligence for Lung Cancer Precision Medicine. Cold Spring Harb Perspect Med. 2021. 11(8): a039537.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi S, Zhou B. A review of radiomics and genomics applications in cancers: the way towards precision medicine. Radiat Oncol. 2022. 17(1): 217.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eConti A, Duggento A, Indovina I,et al. Radiomics in breast cancer classification and prediction. Semin Cancer Biol. 2021. 72: 238\u0026ndash;250.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAi Y, Zhu H, Xie C, Jin X. Radiomics in cervical cancer: Current applications and future potential. Crit Rev Oncol Hematol. 2020. 152: 102985.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eProdoehl J, Li H, Planetta PJ,et al. Diffusion tensor imaging of Parkinson's disease, atypical parkinsonism, and essential tremor. Mov Disord. 2013. 28(13): 1816\u0026ndash;22.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVitali P, Pan MI, Palesi F,et al. Substantia Nigra Volumetry with 3-T MRI in De Novo and Advanced Parkinson Disease. Radiology. 2020. 296(2): 401\u0026ndash;410.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBetrouni N, Lopes R, Defebvre L,et al. Texture features of magnetic resonance images: A marker of slight cognitive deficits in Parkinson\u0026rsquo;s disease. Mov Disord. 2020. 35(3): 486\u0026ndash;494.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePostuma RB, Berg D, Stern M,et al. MDS clinical diagnostic criteria for Parkinson's disease. Mov Disord. 2015. 30(12): 1591\u0026ndash;601.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePan X, Liu C, Feng T, Qi XS. A multi-objective based radiomics feature selection method for response prediction following radiotherapy. Phys Med Biol. 2023. 68(5).\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGuiot J, Vaidyanathan A, Deprez L,et al. A review in radiomics: Making personalized medicine a reality via routine imaging. Med Res Rev. 2022. 42(1): 426\u0026ndash;440.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu Z, Wang S, Dong D,et al. The Applications of Radiomics in Precision Diagnosis and Treatment of Oncology: Opportunities and Challenges. Theranostics. 2019. 9(5): 1303\u0026ndash;1322.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLazarus M, Huang ZL, Lu J,et al. How do the basal ganglia regulate sleep-wake behavior. Trends Neurosci. 2012. 35(12): 723\u0026ndash;32.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZur G, Lesman-Segev OH, Schlesinger I,et al. Tremor Relief and Structural Integrity after MRI-guided Focused US Thalamotomy in Tremor Disorders. Radiology. 2020. 294(3): 676\u0026ndash;685.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHu J, Xiao C, Gong D,et al. Regional homogeneity analysis of major Parkinson\u0026rsquo;s disease subtypes based on functional magnetic resonance imaging. Neurosci Lett. 2019. 706: 81\u0026ndash;87.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAkbari H, Fei B. 3D ultrasound image segmentation using wavelet support vector machines. Med Phys. 2012. 39(6): 2972\u0026ndash;84.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJing R, Wang J, Li J,et al. A wavelet features derived radiomics nomogram for prediction of malignant and benign early-stage lung nodules. Sci Rep. 2021. 11(1): 22330.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePoston KL, Ua Cruadhlaoich M, Santoso LF,et al. Substantia Nigra Volume Dissociates Bradykinesia and Rigidity from Tremor in Parkinson's Disease: A 7 Tesla Imaging Study. J Parkinsons Dis. 2020. 10(2): 591\u0026ndash;604.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKaya MO, Ozturk S, Ercan I,et al. Statistical Shape Analysis of Subthalamic Nucleus in Patients with Parkinson Disease. World Neurosurg. 2019. 126: e835-e841.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCao X, Wang X, Xue C,et al. A Radiomics Approach to Predicting Parkinson\u0026rsquo;s Disease by Incorporating Whole-Brain Functional Activity and Gray Matter Structure. Front Neurosci. 2020. 14: 751.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSun D, Wu X, Xia Y,et al. Differentiating Parkinson\u0026rsquo;s disease motor subtypes: A radiomics analysis based on deep gray nuclear lesion and white matter. Neurosci Lett. 2021. 760: 136083.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLiu A, Wang Z, Yang Y,et al. Preoperative diagnosis of malignant pulmonary nodules in lung cancer screening with a radiomics nomogram. Cancer Commun (Lond). 2020. 40(1): 16\u0026ndash;24.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang K, Qiao Z, Zhao X,et al. Individualized discrimination of tumor recurrence from radiation necrosis in glioma patients using an integrated radiomics-based model. Eur J Nucl Med Mol Imaging. 2020. 47(6): 1400\u0026ndash;1411.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eJiang Z, Wang B, Han X,et al. Multimodality MRI-based radiomics approach to predict the posttreatment response of lung cancer brain metastases to gamma knife radiosurgery. Eur Radiol. 2022. 32(4): 2266\u0026ndash;2276.\u003c/span\u003e\u003c/li\u003e\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, clinical classification, radiomics, magnetic resonance imaging","lastPublishedDoi":"10.21203/rs.3.rs-6223585/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6223585/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eThis study developed and compared single-sequence and multimodal imaging omics models for Parkinson's disease (PD) classification using 3.0T MRI scans (T1WI, T2-FLAIR) from 160 PD patients (82 tremor-type, 78 non-tremor-type) and 100 healthy controls. Regions of interest included the Hippocampus, Substantia Nigra, Red Nucleus, Thalamus, and Amygdala. Data were split into training/test sets (8:2), with the Least Absolute Shrinkage and Selection Operator (LASSO) for feature selection and Support Vector Machine (SVM) for modeling, evaluated via Receiver Operating Characteristic (ROC) curves and area under curve (AUC). The single-sequence Hippocampal-T1WI model showed AUCs of control (training/test:0.940/0.834), non-tremor PD (training/test: 0.923/0.740), and tremor PD (training/test:0.914/0.524). The multimodal model achieved higher AUCs: control (training/test:0.966/0.877), non-tremor PD (training/test:0.952/0.861), and tremor PD (training: 0.942, test: 0.760), indicating improved predictive accuracy, demonstrating superior predictive accuracy. Multimodal imaging omics significantly enhanced PD diagnosis and differentiation compared to single-sequence models.\u003c/p\u003e","manuscriptTitle":"Analysis and Research on Predicting the Motor Classification of Parkinson's Disease Based on Radiomics","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-13 06:39:31","doi":"10.21203/rs.3.rs-6223585/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":"3df2a441-0270-4206-944a-34c1aa488feb","owner":[],"postedDate":"May 13th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[{"id":48176881,"name":"Biological sciences/Neuroscience"},{"id":48176882,"name":"Health sciences/Biomarkers"},{"id":48176883,"name":"Health sciences/Diseases"}],"tags":[],"updatedAt":"2025-12-08T11:39:05+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-13 06:39:31","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6223585","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6223585","identity":"rs-6223585","version":["v1"]},"buildId":"8U1c8b4HqxoKbykW_rLl7","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}