Diagnostic performance of PET-based artificial intelligence for differentiating Parkinson's disease from normal controls or atypical parkinsonism: a systematic review and meta-analysis.

PURPOSE: This study aims to evaluate the diagnostic performance of PET-based artificial intelligence (AI) for differentiating Parkinson's disease (PD) from normal controls (NC) or atypical parkinsonism (AP).

METHODS

A systematic literature search was conducted in PubMed, Embase, Web of Science, and the Cochrane Library for studies published up to July 2, 2025. Studies were included if they investigated PET-based AI for differentiating PD from NC or AP. Quality assessments were performed using the PROBAST-AI tool, and a bivariate random-effects model was implemented to calculate pooled sensitivity, specificity, and area under the curve (AUC).

RESULTS

Among 29 included studies, PET-based AI demonstrated a pooled sensitivity of 0.93 (95% CI 0.89-0.96) and specificity of 0.92 (95% CI 0.86-0.95) for PD vs. NC, with an AUC of 0.97 (95% CI 0.95-0.98). For PD vs. AP, pooled sensitivity was 0.92 (95% CI 0.88-0.95), specificity was 0.86 (95% CI 0.78-0.91), and AUC was 0.95 (95% CI 0.93-0.97). Subgroup analyses revealed dopaminergic tracers ([18F]FDOPA, [11C]CFT, [18F]AV-133) outperformed [18F]FDG in sensitivity for PD vs. NC, while [18F]FDG showed superior AUC versus dopaminergic tracers for PD vs. AP.

CONCLUSIONS

PET-based AI models exhibit high diagnostic performance in differentiating PD from NC or AP, indicating significant potential for clinical application. However, limitations such as study heterogeneity and small sample sizes highlight the need for larger, multi-center trials to validate these findings and improve the clinical utility of AI models in practice.