Head-to-Head Comparison of 2 Artificial Intelligence Tools for Detecting Lymph Node Metastases in Whole-Slide Pathology Images Within and Beyond Their Intended Use

The increasing diagnostic workload in pathology, driven by rising cancer incidences, highlights the need for scalable, cost effective solutions. Artificial intelligence (AI) has shown promise in supporting lymph node (LN) metastasis detection, a key prognostic factor in cancer staging. However, the current Conformité Européene In Vitro Diagnostics--certified AI tools are often limited to specific tumor types, reducing their cost efficiency and clinical use. This study evaluates the performance of 2 Conformité Européene In Vitro Diagnostics-certified AI tools—Visiopharm Metastasis Detection App (VMD) and DeepPath LYDIA (DPL)—for multipurpose LN metastasis detection across 6 tumor types, both within and beyond their intended use. We retrospectively analyzed whole-slide images from 455 patients with LN metastases from melanoma, colorectal, head and neck, lung, vulvar, and breast cancer. Both sentinel and nonsentinel LNs were included, with expert pathologists establishing the reference standard, according to clinical practice. Sensitivity was calculated per case and stratified based on metastasis size. False-positive alerts (FPAs) were assessed in 1012 tumor-negative slides. Both applications demonstrated excellent sensitivity for macrometastases across tumor types. DPL showed slightly higher sensitivity for micrometastases and isolated tumor cells compared with VMD, particularly in lung cancer and melanoma. FPA rates were substantial for both tools, with VMD generally producing more alerts, especially in lung and breast cancer. Our findings suggest that a single AI tool may be suitable for LN metastasis detection across multiple tumor types, even beyond its intended use. However, high FPA rates—particularly in lung cancer (inside intended use for DPL)—may limit practical use. Prospective studies are needed to confirm workflow efficiency gains and define optimal implementation strategies. These results support a broader, pragmatic approach to AI validation and regulatory approval, potentially improving the business case for AI adoption in pathology laboratories.