Systematic review and meta-analysis of AI in lung cancer metastasis imaging for diagnosis and prognosis.

Abstract

Lung cancer (LC) remains the leading cause of cancer-related mortality, and distant metastases (DMs) are common. Imaging-based AI research has largely focused on primary tumors, with metastatic lesions insufficiently investigated. This study provides the first metastasis-focused quantitative synthesis of AI performance for imaging-based evaluation of DMs in LC across classification and outcome-prediction tasks, with 59 of 64 eligible reports in the meta-analysis. Task-specific meta-analysis showed pooled sensitivity, specificity, and AUC (best-performing model) of 0.88, 0.87, and 0.91 for molecular-level prediction; 0.87, 0.90, and 0.93 for metastasis differentiation; 0.89, 0.94, and 0.90 for tumor type classification; and 0.82, 0.86, and 0.89 for outcome prediction, respectively. Heterogeneity between studies was substantial (I² > 90% for key analyses). Subgroup analyses showed favorable performance across study settings. The findings support AI for imaging-based evaluation of DMs in LC, highlighting the need for robust and interpretable models for translation.