Artificial intelligence models for psychiatric diagnostics have a high risk for bias

A systematic review has identified high risk for bias in neuroimaging-based artificial intelligence (AI) models for psychiatric diagnostics, indicating that improvements are needed before clinical application.
The review included 517 studies of 555 models, among which 83.1% had a high risk for bias according to the Prediction Model Risk of Bias Assessment Tool.
Risk for bias in the included models was found to be particularly high in the analysis domain, including insufficient sample size, poor examination of model performance, and lack of handling data complexity.