Classification of Tourette's Syndrome Using Cortico-Striatal-Thalamic-Cortical Circuit Segmentation and Convolutional Neural Networks: A Machine Learning Study (Preprint)

Le résumé fourni par la source

BACKGROUND Tourette’s Syndrome (TS) is an inheritable neurological disorder characterized by repetitive, involuntary movements called “tics”. Predominantly affecting males, TS typically emerges in adolescence, with a lack of definitive diagnostic tests often causing delays in diagnosis. This study utilizes magnetic resonance (MR) imaging to explore brain patterns linked to TS, focusing on the cortico-striatal-thalamic-cortical (CSTC) circuit. This neural pathway, involved in motor control, emotion regulation, and cognition, is believed to play a key role in TS symptoms. Computational methods, including Machine Learning (ML), are used to analyze these images and deepen our understanding of TS. OBJECTIVE We aim to assess whether CSTC-based segmentation improves TS classification over whole brain analysis. The study employs Freesurfer and Slant segmentation methods, training VGG16, VGG19, and ResNet50 models. METHODS The study follows four steps: (1) Dataset Organization: 68 T1-weighted MR volumes; (2) Preprocessing and Segmentation: Data enhancement and CSTC related brain region segmentation using Freesurfer and Slant; (3) Data Augmentation: Increasing dataset size with 6 degrees of freedom; (4) Classification: Comparison of whole brain and CSTC based CNN classification. RESULTS Results show that CSTC based segmentation outperforms whole brain methods (pcorr < 0.001). Using Freesurfer with VGG16 achieves 82% accuracy, while Slant with VGG16 achieves 80.3% accuracy. Thus, CSTC based segmentation shows promise for advancing TS diagnosis. CONCLUSIONS VGG16’s superior performance suggests its balance of depth and parameter count was well suited to our dataset without overfitting, while rigid data augmentation was crucial for increasing sample variability and improving generalization CLINICALTRIAL This study was not registered as a clinical trial.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.