Identifying novel prognostic markers and genotype-phenotype associations in endometrioid endometrial carcinoma by computational analysis of histopathological images

Hematoxylin and eosin stained slides are routinely used for the diagnosis and grading of endometrioid endometrial carcinoma (EEC). These images present a high degree of cellular heterogeneity, which may contain clinically relevant information such as prognosis and is difficult to be quantified objectively by eyes. Besides traditional microscopic image assessment, a lot of effort has been put in molecular characterization of tumors. How molecular events manifest at tumor tissue level is not well understood. In this paper, we investigated whether quantitative morphological features extracted from histopathological images are associated with patient survival and somatic mutation of genes in EEC using the multi-modality data from The Cancer Genome Atlas. A computational image analysis pipeline was developed to extract image features that characterize the size, shape, staining, and density of cell nuclei. For prognosis prediction, we built a prognostic model based on the image features. In a validation set, the risk score predicted by our model was an independent prognostic factor for overall survival in a multivariate Cox proportional hazards model (hazard ratio with 95% confidence interval: 3.38 [1.55–7.37], p = 2.15e−3). To link tumor tissue morphology with somatic mutation, a two-sided Mann-Whitney U test was used to compare the distribution of each feature between mutated and nonmutated cases for frequently mutated genes. We found that TP53 and TTN were significantly associated with tissue morphological changes. These findings show the promising potential of computational histopathology image analysis in predicting patient survival and exploring genotype-phenotype associations.