Radiomics on routine T1-weighted MRI can delineate Parkinson's disease from multiple system atrophy and progressive supranuclear palsy

Objectives This study aimed to explore the feasibility of radiomics features extracted from T1-weighted MRI images to differentiate Parkinson's disease (PD) from atypical parkinsonian syndromes (APS). Methods Radiomics features were computed from T1 images of 65 patients with PD, 61 patients with APS (31: progressive supranuclear palsy and 30: multiple system atrophy), and 75 healthy controls (HC). These features were extracted from 19 regions of interest primarily from subcortical structures, cerebellum, and brainstem. Separate random forest classifiers were applied to classify different groups based on a reduced set of most important radiomics features for each classification as determined by the random forest-based recursive feature elimination by cross-validation method. Results The PD vs HC classifier illustrated an accuracy of 70%, while the PD vs APS classifier demonstrated a superior test accuracy of 92%. Moreover, a 3-way PD/MSA/PSP classifier performed with 96% accuracy. While first-order and texture-based differences like Gray Level Co-occurrence Matrix (GLCM) and Gray Level Difference Matrix for the substantia nigra pars compacta and thalamus were highly discriminative for PD vs HC, textural features mainly GLCM of the ventral diencephalon were highlighted for APS vs HC, and features extracted from the ventral diencephalon and nucleus accumbens were highlighted for the classification of PD and APS. Conclusions This study establishes the utility of radiomics to differentiate PD from APS using routine T1-weighted images. This may aid in the clinical diagnosis of PD and APS which may often be indistinguishable in early stages of disease.

Automated summary

This study demonstrates the feasibility of using radiomics features extracted from T1-weighted MRI images to differentiate Parkinson's disease from atypical parkinsonian syndromes. The results showed high accuracy rates for classifiers distinguishing PD from HC, PD from APS, and a 3-way classifier for PD/MSA/PSP, highlighting the potential of radiomics in aiding clinical diagnosis in early stages of disease.