Predicting Parkinson's disease and its progression based on radiomics in T1-weight images and α‑synuclein in cerebrospinal fluid.

This study aimed to develop a radiomics model that can predict Parkinson's disease (PD) and its progression of using T1-weighted images (T1WI), and to evaluate the prediction performance of a multimodal model incorporating clinical factors. We selected all participants from the Parkinson's Progression Markers Initiative (PPMI) (n = 205) database and Qilu Hospital of Shandong University (Qingdao) (n = 60). The patients were tracked over 4-5 years via Hoehn-Yahr Scale (HYS) and categorized into stable PD (SPD: HYS unchanged) and progression PD (PPD: HYS increase). Participants from the PPMI database were randomly divided into a training dataset and an internal testing dataset to construct radiomics models, which were validated by the Qilu Hospital database as an independent testing dataset. Only participants from the PPMI database had cerebrospinal fluid (CSF) α‑synuclein (α-syn) data, which were used to establish a combined model. The radiomics-based classifier for healthy controls (HC) and PD achieved areas under the receiver operating characteristic curves (AUROCs) of 0.787 and 0.746 in the internal and independent testing datasets, respectively, while the AUROCs were 0.857 and 0.802 in predicting SPD and PPD, respectively. Moreover, integrating CSF total α-syn in the combined model enhanced the predictive performance, with AUROC of 0.890, sensitivity of 0.846 and specificity of 0.857 for HC vs. PD, and AUROC of 0.939, sensitivity of 0.917 and specificity of 0.933 for SPD vs. PPD in the internal testing dataset. The current research presented evidence that radiomics utilizing conventional T1WI can predict the clinical stages of PD and that the efficacy of the multimodal model can be boosted by combining radiomics with CSF total α‑syn.