Adding Dimensionality Reduction analysis of Texture descriptors for Tourette’s Syndrome classification

Tourette Syndrome(TS) is a hereditary condition characterized by involuntary motor and vocal actions. Although there is no cure for TS, various prescription medications are typically provided to patients to help alleviate symptoms. Incorporating imaging data with traditional clinical assessment is widely reported to improve diagnosis. Incorporating imaging data analysis with traditional clinical assessment is widely reported to improve TS diagnosis. Besides, the combination of techniques from the Image Processing (IP) and Machine Learning (ML) areas can enhance the identification and categorization of texture patterns in brain regions, which is otherwise a time-consuming and challenging task to do manually. However, not every texture pattern is relevant for analysis, thus requiring an approach to reduce its dimensionality, and only retrieving the essential information. This paper proposes a dimensionality reduction analysis on texture features in the classification of TS. A total of 68 T1-weighted MR(T1-w) scans are used, with an equal distribution of TS and healthy participants. The proposed method includes the following steps: (i) parcellation of anatomical areas; (ii) texture feature extraction using 3D Gray-level Co-occurrence Matrix; (iii) dimensionality reduction by principal component analysis(PCA); and (iv) classification of PCA components using ML approaches. We found that reducing dimensionality considerably improves prediction of TS patients. Specifically, the identification of anatomical regions showed an improvement above 80% in accuracy when compared to baseline. We suggest that PCA should be incorporated as it increases the accuracy of classification when compared to the baseline and feature selection techniques.