Abstract 6175: Deep learning-based survival prediction for hepatocellular carcinoma using whole slide images

Abstract There is an increasing need for classifiers to effectively stratify prognosis in patients with hepatocellular carcinoma (HCC). Deriving interpretable prognostic features using deep-learning (DL) -based prognostic models from pathology images remains a challenge. In this study, we developed a computational approach based on DL to predict the overall survival (OS) of HCC patients from formalin-fixed paraffin-embedded histopathology whole slide images. A DL model for classifying good and poor prognosis was developed on a total of 460 patients with 2 independent cohorts from Seoul St. Mary's Hospital. The primary outcome was OS. The training set (SSM1, n = 292) included HCC patients that had undergone surgical resection. The classifier was then independently tested on the second HCC dataset (SSM2, n = 168). The area under the curve (AUC) for the receiver operating characteristic (ROC) curves was 0. 850 and 0. 781 for SSM1 and SSM2 datasets, respectively. In addition, we have tried to establish DL model for prediction of OS based on clinicopathological features, including Edmondson-Steiner grade, pT stage, vascular invasion, Ki-67 index, and cytokeratin 19 immunohistochemical expression. Random forest model was applied to classify the good and poor prognosis groups and The AUROCs for the random forest was 0. 705 for the SSM1 dataset, which are inferior to that of DL prediction. These findings suggest that the DL techniques can be applied to refine the prognostic prediction of HCC patients. Citation Format: Sung Hak Lee, Hyun-Jong Jang. Deep learning-based survival prediction for hepatocellular carcinoma using whole slide images 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 (6Suppl): Abstract nr 6175.