A multi-tumor machine learning model to identify tertiary lymphoid structures (TLS) in histopathological H&E images as a potential clinical biomarker

Tertiary lymphoid structures (TLS) are ectopic lymphoid structures composed of B-cells, T-cells, and supportive cells that develop in non-lymphoid organs and are often found in tumors. The criticality and functions of TLS in an adaptive anti-tumor response are still being elucidated, but studies have shown associations between TLS and IO outcomes across multiple indications. These correlations are dependent on TLS maturity and localization within the tumor microenvironment (TME). Currently, identification of TLS in tumors by pathologists is not routine or standardized.

Objectives: Develop a machine-learning algorithm based on H&E images to score TLS as a clinical biomarker to: 1) Accurately and reproducibly identify TLS regions within the TME 2) Predict TLS subregions and maturity state 3) Extract TLS model-derived features.

This algorithm can be deployed in an exploratory manner to score TLS features in research and trial cohorts to assess its utility as a predictive biomarker and complement immune response measurements.