Abstract 2492: Diagnostic algorithm for salivary gland oncocytic lesions

Abstract Aim: Diagnosing salivary gland lesions with oncocytic differentiation is challenging as it encompassing 21 reactive and neoplastic lesions. With the extensive differentiation, morphological overlapping is possible especially that some tumors can be purely oncocytic. We propose a diagnostic algorithm based on analyzing oncocytic detection in the salivary gland lesion in AS’ registry (501 out of 6383 SGL) to propose a differential diagnosis based on morphology, immunohistochemistry, and molecular findings. Methods: We developed a weighted probabilistic scoring algorithm incorporating dimensional reduction and hierarchical clustering analysis of morphological features, immunohistochemical profiles, and molecular labelling that represents the greatest weighted values. If molecular information is missing, the algorithm computes the values corresponding to morphology and immunohistochemical vectors respectively. The algorithm uses supervised machine learning with backpropagation for continuous refinement of classification boundaries. Results: Salivary gland lesions with oncocytic differentiation comprise malignant neoplasms (e.g., acinic cell carcinoma, salivary duct carcinoma, myoepithelial carcinoma), benign neoplasms (e.g., pleomorphic adenoma, Warthin tumor, oncocytoma), cystic lesions and reactive process (e.g. multifocal oncocytic metaplasia), with different extent of oncocytic differentiation, either extensive, focal, or wall-lining, in which oncocytes are positive for MIA. Principal component analysis revealed three distinct oncocytic architectural patterns with eigenvalues >1.0: extensive (λ1=2.84), focal (λ2=1.92), and wall-lining (λ3=1.37). Hierarchical clustering demonstrated discrete entity groupings with a cophenetic correlation coefficient of 0.89. The algorithm achieved a primary diagnostic accuracy of 92% (95% CI: 88.3-95.7%), while maintaining differential diagnostic sensitivity of 85% at a lower boundary condition of 0.3. Nevertheless, the algorithm is designed to propose a diagnosis for well-investigated cases only. Otherwise, the differentials are provided. A user-friendly interface will be made available at https://www.biopticka.cz/en/publications/tools/Oncocytic/ Conclusions: The algorithm demonstrates acceptable performance metrics in differential diagnosis generation. Although oncocytic differentiation can theoretically occur in any lesion, the inclusion of the salivary oncocytic lesions was confined to real data with adequate molecular investigation. Coverage could be expanded if reliable inputs show oncocytic differentiation, for example, in secretory carcinoma, polymorphous adenocarcinoma (and its cribriform subtype), microsecretory carcinoma, or palisading adenocarcinoma. Citation Format: Bacem Khalele, Alena Skalova. Diagnostic algorithm for salivary gland oncocytic lesions. [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 2492.