AI-based prediction of actionable genomic alterations from H&E-stained slides

Precision oncology based on the identification of actionable genomic alterations (AGAs) has been transformative in the care of solid tumors, yet optimal treatment selection is often hindered by tissue exhaustion and slow turnaround times. Artificial intelligence (AI) addresses these barriers by predicting AGAs directly from routine hematoxylin and eosin (H&E)-stained slides. Recent landmark studies, including the prospective EAGLE and ANNAR trials, have demonstrated that deep learning models can identify alterations such as EGFR in non-small cell lung cancer and FGFR in urothelial cancer across diverse global cohorts. By providing near-instant, non-destructive results, AI-based prediction can triage patients for molecular testing or clinical trials at the time of initial diagnosis. This perspective evaluates the maturing evidence base and outlines a pragmatic pathway for integrating these tools into clinical practice. Ultimately, these models offer the opportunity to extract critical molecular information from slides generated in every laboratory worldwide and to significantly improve timely access to the rapidly evolving therapeutic landscape of solid tumors.