Abstract Introduction: The understanding of morphological heterogeneity at the cellular level in High-Grade Serous Ovarian Cancer (HGSOC), particularly in relation to BRCA1 and BRCA2 mutations, is still an ongoing area of research. This study aims to elucidate the morphological features that distinguish wild-type (WT) tumors from those with BRCA1/BRCA2 mutations using deep learning models. Methods: A total of 300 H&E stained HGSOC slides were collected in a patient-wise manner and tumor areas were annotated by an expert pathologist. We used a convolutional neural network model with spatial attention pre-trained on pathological images to extract features from slides, and the features were clustered into 96 groups. Subsequently, a nuclei segmentation model was employed for precise cell segmentation, followed by measurement of cellular features including cell shape (area, eccentricity) and cell texture (dissimilarity, correlation). The assessment of cell texture evaluates the variation and uniformity in gray-level intensity across the cell. We performed independent t-tests on samples of 100 patches from each cluster, comparing cells of selected patches from the top-N clusters representing the highest proportion of each label. Results: H&E slides (n=300) include 69 (23%) BRCA1 mutants and 50 (16.7%) BRCA2 mutants. In Table1, negative t-statistics suggest that WT carcinoma cancer cells exhibit lower values for the measured features compared to BRCA mutants, whereas positive values denote higher values in WT. BRCA1 and BRCA2 mutant carcinoma cancer cells showed significantly larger areas and higher eccentricity than WT carcinoma cancer cells in all top-N clusters settings (N=3, 5, 10, 30). Furthermore, WT carcinoma cancer cells showed higher dissimilarity and reduced correlation compared to BRCA1 mutant carcinoma cells in four settings. Conclusions: Our observations in BRCA-mutated HGSOC cells enrich our understanding of tumor cell morphology with BRCA mutations. Table 1. Comparison of cell features between WT and BRCA mutants cell feature comparison top-3 clusters (t-statistics, p-value) top-5 clusters (t-statistics, p-value) top-10 clusters (t-statistics, p-value) top-30 clusters (t-statistics, p-value) area WT vs BRCA1 -37.41, <0.001 -50.2, <0.001 -37.53, <0.001 -85.95, <0.001 WT vs BRCA2 -42.14, <0.001 -50.61, <0.001 -16.57, <0.001 -51.98, <0.001 eccentricity WT vs BRCA1 -24.40, <0.001 -34.14, <0.001 -38.02, <0.001 -37.40, <0.001 WT vs BRCA2 -26.77, <0.001 -30.46, <0.001 -29.47, <0.001 -47.64, <0.001 dissimilarity WT vs BRCA1 55.17, <0.001 52.73, <0.001 103.57, <0.001 166.78, <0.001 WT vs BRCA2 -38.15, <0.001 -24.75, <0.001 0.2, <0.001 107.73, <0.001 correlation WT vs BRCA1 1.94, <0.001 -2.46, <0.001 -50.24, <0.001 -65.96, <0.001 WT vs BRCA2 87.62, <0.001 90.83, <0.001 52.6, <0.001 -48.90, <0.001 Citation Format: JaeHeon Lee, Hyunil Kim, Yongeun Lee, Yoon-La Choi, Kyungsoo Jung, Tae-Yeong Kwak, Sun Woo Kim, Hyeyoon Chang. Morphological feature discrepancies in wild-type vs. BRCA1/BRCA2 mutated high-grade serous ovarian cancer [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 4913.
Read full abstract