Deep transfer learning approach to predict tumor mutation burden (TMB) and delineate spatial heterogeneity of TMB within tumors from whole slide images

Purpose Tumor Mutation Burden (TMB) is an emerging prognostic and/or predictive biomarker across many cancers. Tissue-based sequencing approaches have been used to determine TMB status. However, the clinical utility of these approaches is limited, due to time, cost, and tissue availability constraints, and could provide inconsistent TMB status driven by spatial heterogeneity. We present deep transfer learning approach utilizing whole slide images (WSIs) for predicting patient-level TMB status and spatial TMB heterogeneity within tumors.

Experimental Design In experiments to predict patient-level TMB status, we used WSIs from TCGA Bladder (BLCA) and Lung (LUAD) cancer cohorts. To investigate spatial TMB heterogeneity and its prognostic utility, we used TCGA BLCA cohort.

Results In experiments of patient-level TMB predictions, our proposed method achieved overall best Area-Under-ROC scores of 0.752 (95% CI, 0.683-0.802) and 0.742 (95%, 0.682-0.794) for TCGA BLCA and LUAD cohorts, respectively, compared to the state-of-the-art methods. In experiments to delineate spatial TMB heterogeneity within tumors, we predicted TMB status for each tile representing small tumor region in WSIs. We calculated Shannon entropy using predicted TMB status of tiles to determine high or low spatial TMB heterogeneity for each patient. Kaplan Meier analysis showed that the patient subgroup having patient-level TMB high with low spatial TMB heterogeneity within tumors have good prognosis (log-rank test P<0.05).

Conclusions Our computational pipeline using WSIs can accurately predict patient-level TMB status and have potential to provide rapid and cost-effective TMB testing. Spatial analysis of TMB illuminates clinical significance of TMB heterogeneity as a prognostic biomarker. Code Availability: https://github.com/hwanglab/tcga_tmb_prediction