Validation of histopathology-based deep learning algorithms for detection of actionable non-small cell lung cancer biomarkers.

Non-small cell lung cancer (NSCLC) patient management relies on molecular analysis to determine eligibility for targeted therapy. Furthermore, neoadjuvant immunotherapy is primarily suitable in the absence of specific genomic alterations. However, significant challenges remain, including suboptimal molecular testing and patients being assigned to non-optimal treatment strategies. Here, we present AI classifiers for the identification of EGFR, ALK, BRAF and MET alterations directly from hematoxylin and eosin (H&E)-stained tissue using CanvOI 1.1, a digital pathology foundation model. Their performance was evaluated on an independent validation dataset of 968 NSCLC samples. The classifiers achieved AUCs of 0.87 for EGFR, 0.96 for ALK, 0.88 for BRAF and 0.83 for MET. Moreover, they demonstrated high accuracy in identifying cases lacking alterations. Our results highlight the potential of deep-learning tools for the detection of NSCLC biomarkers and specifically the identification of tumors without EGFR or ALK driver alterations, supporting more informed clinical decision-making.