Artificial intelligence-assisted screening for NTRK fusion-positive salivary gland tumors: A novel digital pathology workflow.

NTRK fusion is a promising therapeutic target for salivary gland cancer (SGC). However, the diagnostic complexity of the histological SGC subtype and low NTRK fusion-positive SGC incidence (5%), except for secretory carcinoma (90%), limit targeted therapy. Despite recommended pan-TRK immunohistochemistry (IHC) screening, evaluation remains unstandardized. Artificial intelligence (AI) has now enabled the prediction of genetic and molecular information. We aimed to evaluate the usefulness of the AI model, a histopathological image retrieval system-Luigi-Oral-in detecting NTRK fusion-positive cases. All 273 primary salivary gland tumors surgically obtained between 2005 and 2023 were retrospectively and blindly assessed using fluorescence in situ hybridization (FISH) for ETV6 rearrangements, pan-TRK IHC (EPR17341), and differential diagnosis with Luigi-Oral for hematoxylin-eosin staining queries in December 2024 (https://luigi-pathology.com/). Luigi-Oral defined inclusion of secretory carcinoma among the top five diagnoses as positive differential indication. Overall, 273 salivary gland tumors were identified, including 104 benign and 169 malignant tumors, with 7 samples pathologically diagnosed as secretory carcinomas. FISH, pan-TRK IHC, and Luigi-Oral identified 7, 7, and 24 positive cases, respectively. All the FISH-positive cases were pathologically diagnosed as secretory carcinoma. Luigi-Oral exhibited a sensitivity, specificity, false-positive rate, and false-negative rate of 85.7%, 93.2%, 6.8%, and 14.3% for detecting FISH-positive cases, respectively. The false-positive rate was 13.3% in acinic cell carcinoma, histologically similar to secretory carcinoma. Luigi-Oral might be a useful alternative to IHC for screening NTRK fusion-positive rare SGC cases, and advances in digital pathology can facilitate implementation of an AI model in the NTRK screening workflow.