The differential diagnosis value of radiomics-based machine learning in Parkinson’s disease: a systematic review and meta-analysis

Background In recent years, radiomics has been increasingly utilized for the differential diagnosis of Parkinson’s disease (PD). However, application in PD still lacks sufficient evidence-based support. To address this gap, we carried out a systematic review and meta-analysis to evaluate diagnostic value radiomics-based machine learning (ML) PD. Methods We systematically searched Embase, Cochrane, PubMed, Web Science databases as November 14, 2022. The quality assessment scale (RQS) was used included studies. outcome measures were c-index, which reflects overall accuracy model, well sensitivity specificity. During meta-analysis, discussed ML various atypical parkinsonism syndromes (APS). Results Twenty-eight articles with total 6,057 participants included. mean RQS score all 10.64, relative 29.56%. pooled sensitivity, specificity predicting 0.862 (95% CI: 0.833–0.891), 0.91 0.86–0.94), 0.93 0.87–0.96) training set, 0.871 0.853–0.890), 0.86 0.81–0.89), 0.87 0.83–0.91) validation respectively. Additionally, differentiating from APS 0.866 0.843–0.889), 0.84–0.88), 0.80 0.75–0.84) 0.879 0.854–0.903), 0.85–0.89), 0.82 0.77–0.86) Conclusion Radiomics-based can serve potential tool diagnosis. Moreover, it an excellent performance distinguishing APS. support vector (SVM) model exhibits robustness when number samples is relatively abundant. due diverse implementation process radiomics, expected that more large-scale, multi-class image data be develop intelligent tools broader applicability, promoting development prediction related fields. Systematic registration https://www.crd.york.ac.uk/PROSPERO/display_record.php?RecordID=383197 , identifier ID: CRD42022383197.