DOI 10.1080/01616412.2026.2676869
A machine learning approach for early Parkinson's disease diagnosis based on brain texture features from T1-weighted imaging.
Аннотация
OBJECTIVE
To investigate the diagnostic value of subcortical texture features from T1-weighted MRI combined with machine learning for early Parkinson's disease (PD), and to develop a simplified model for clinical application.
METHODS
Retrospective analysis of T1-weighted imaging data from the PPMI database was performed. A primary cohort (133 PD patients, 47 healthy controls [HC]) and an independent validation cohort (33 PD patients, 22 HC) were enrolled after quality control. MRI images were preprocessed with N4 bias field correction and registered to the MNI152_1mm template. Texture features were extracted from six subcortical regions (thalamus, hippocampus, caudate nucleus, amygdala, globus pallidus, putamen) via Python and Matlab. Features were selected by t-test and Lasso regression (5-fold cross-validation), with SMOTE for data balancing. Five machine learning models (KNN, SVM, RF, LR, NB) were constructed, and simplified models were built with the top 10 weighted features. Model performance was assessed by AUC, accuracy, precision, sensitivity, specificity and F1 score.
RESULTS
In the primary cohort, the RF model exhibited the best performance (AUC = 0.96, accuracy = 88.9%, precision = 92.9%, sensitivity = 92.9%, specificity = 75.0%, F1 = 92.9%). It maintained good efficacy in the validation cohort (AUC = 0.84), and the simplified RF model still achieved an AUC of 0.823. All simplified models except NB had an AUC > 0.80.
CONCLUSIONS
Subcortical texture features from T1-weighted MRI are valuable biomarkers for early PD. The RF model shows robust discriminative ability between PD patients and HC, and its simplified version has favorable clinical applicability, providing a non-invasive auxiliary tool for early PD diagnosis.
Подготовка и перевод публикации
Обработка статьи · этапы, ошибки и управление
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Тип публикации указан по данным источника. Уровень доказательности автоматически не присваивается.