Subtype-Specific Spatial Descriptors Of Tumor-Immune Microenvironment Are Prognostic Of Survival In Lung Adenocarcinoma

In-depth quantification of the tumor microenvironment cellular primitives is crucial in understanding and characterizing the prognosis cancer. In this work, we have developed an automated pipeline for lung cancer survival analysis by exploiting computationally derived patterns of tumor-immune cell interaction on H&E images. We argue that the integration of the phenotypic information of the tumor with hand-crafted features of the tumor-immune microenvironment improves survival analysis in lung cancer. For characterizing the tumor-immune microenvironment, we have utilized two types of features - graph-based and spatial heterogeneity-based descriptors. These features capture heterogeneity in spatial distributions of tumor and immune cells and quantify the interaction between them in specific lung adenocarcinoma subtypes. The prognostic role of these individual microenvironmental components is demonstrated through survival modeling. Our results on N=411 lung adenocarcinoma cases suggest that histologic subtype-specific heterogeneity descriptors are more prognostic of survival than subtype-agnostic features.