Identification of topological features in renal tumor microenvironment associated with patient survival

Abstract

Motivation: As a highly heterogeneous disease, the progression of tumor is not only achieved by unlimited growth of the tumor cells, but also supported, stimulated, and nurtured by the microenvironment around it. However, traditional qualitative and/or semi-quantitative parameters obtained by pathologist's visual examination have very limited capability to capture this interaction between tumor and its microenvironment. With the advent of digital pathology, computerized image analysis may provide a better tumor characterization and give new insights into this problem.

Results: We propose a novel bioimage informatics pipeline for automatically characterizing the topological organization of different cell patterns in the tumor microenvironment. We apply this pipeline to the only publicly available large histopathology image dataset for a cohort of 190 patients with papillary renal cell carcinoma obtained from The Cancer Genome Atlas project. Experimental results show that the proposed topological features can successfully stratify early- and middle-stage patients with distinct survival, and show superior performance to traditional clinical features and cellular morphological and intensity features. The proposed features not only provide new insights into the topological organizations of cancers, but also can be integrated with genomic data in future studies to develop new integrative biomarkers.

Availability and implementation: https://github.com/chengjun583/KIRP-topological-features.

Contact: 1271992826@qq.com or kunhuang@iu.edu.

Supplementary information: Supplementary data are available at Bioinformatics online.

Fig. 1.

Overview of our workflow. (A) Learning nucleus patterns in an unsupervised manner. (B) Generating bag of edge histogram features and identifying survival-related edge patterns (Color version of this figure is available atBioinformatics online.)

Fig. 2.

Illustration of the three main steps involved in our feature extraction workflow. (A) Nucleus segmentation. (B) Nucleus pattern learning using stacked sparse autoencoder to learn high-level features followed by clustering. Nucleus patterns are indicated by different colors. There are eight nucleus patterns. (C) Delaunay triangle edge patterns showed in different colors. Edge patterns are defined in terms of their end nodes. There are 36 edge patterns since we have eight nucleus patterns. The H&E image is converted to a grayscale image to highlight colors (Color version of this figure is available atBioinformatics online.)

Fig. 3.

The proposed BOEH features provide better prognosis prediction than clinical variables. (A–C) Kaplan–Meier curves stratified by tumor stage, tumor subtype, and predicted risk index of lasso-Cox model built on BOEH features, respectively. (D) ROC curves that predict the binary outcome of 5-year survival using predicted risk index of lasso-Cox model built on BOEH features, tumor stage, and tumor subtype, respectively. For extracting BOEH features, the number of nucleus patterns is set to 64 (Color version of this figure is available atBioinformatics online.)

Fig. 4.

Examples of the learned nucleus patterns forming edge types that are strongly associated with survival. The number of nucleus clusters is set to 64. The number in the upper-left corner of each image is the cluster index. Each image consists of 10 × 10 nucleus patches from the same cluster (Color version of this figure is available atBioinformatics online.)