Computer-aided diagnostics helps to accurately determine different expression levels of claudin-18.2 in gastric cancer

Introduction Determination of claudin-18.2 expression by immunohistochemistry (IHC) is prerequisite for targeted treatment of gastric cancers (GC) with zolbetuximab. Precise assessment of IHC expression categories, however, may be challenging and prone to interobserver variability. Computer-aided diagnosis has a high potential of improving diagnostic accuracy and reproducibility. We established a computer-aided analysis tool for claudin-18.2 positivity scoring. Methods Analysis steps included the identification of tumour tissue on haematoxylin-3,3′-diaminobenzidine-stained tissue microarray (TMA) slides, cell segmentation, and membranous staining intensity estimation of claudin-18.2 (clone 43-14A). We analysed 2248 cores from 417 primary resected GC with detailed pathological data available. Results In 51.6% (1159/2248) of TMA cores, no stained tumour cells were detected. Among cases with claudin-18.2 expression, predominantly 1+ and 2+ cells, and a minority of 3+ stained cells were found, and 2+ to 3+ staining was unevenly distributed. Utilizing the SPOTLIGHT claudin-18.2 positivity threshold we identified 12% (187/1555) positive cores corresponding to 2.5% (9/365) positive cases. Lower staining intensities in tumour centre cores point to intratumoural heterogeneity. Conclusion Computer-aided diagnostics helps to accurately measure claudin-18.2 expression levels, allowing to precisely determine claudin-18.2 status in GC patients. Previously uncaptured categorization of staining intensities may enhance the understanding of claudin-18.2 threshold for patient stratification.