Comprehensive Scoping Review of External Validation Studies on AI Pathology Models for Lung Cancer Diagnosis

This review identified 4,423 initial studies on AI models for lung cancer detection, supplemented by 440 additional sources. After removing duplicates, 22 studies met the inclusion criteria, focusing primarily on non-small cell lung cancer (NSCLC), including adenocarcinoma and squamous cell carcinoma. Only 10% of these studies led to external model validation. Most studies (16 of 22) were retrospective, with many employing varied datasets, although sample sizes ranged widely. Technical diversity was addressed through various imaging techniques. Quality assessments indicated biases, particularly in participant and image selection. Performance metrics, notably the area under the receiver operating characteristic curve (AUC), were reported in 17 studies, suggesting high diagnostic capability for models that subtyped lung cancers. However, there was significant heterogeneity in evaluation metrics and study designs, preventing a meta-analysis. Critical details on clinical application and external validation were often lacking.

Source link