Predicting response to immunotherapy by Machine Learning of Quantitative Tissue Analysis
Currently, immunotherapy treatment decision making is based on quantification of Program Death Ligand 1 (PD-L1) or in clinical trials based on the percentage of tumour-infiltrating lymphocytes in pathology samples. Pathologists apply thresholds to assessments visually, which are image-dependent and operator-dependent. As a result, the critical decision of administering immunotherapy is made by applying very sensitive thresholds to a possibly inaccurate and subjective quantification of a largely variable stain. Therefore, there is an urgent need for reliable and more accurate biomarkers that can aid in the selection of cancer patients eligible for immunotherapy.
In this theme we will develop deep learning algorithms to quantify the geometry of the tumor, including amount and spatial distribution of tumor infiltrating lymphocytes, PD-L1 and CD8 positive cells in digitised histopathology images of breast, bladder, melanoma, colon and lung cancer patients treated with immunotherapy. The tool will be based on deep learning algorithms, trained using multiple stainings of retrospective cancer cohorts and clinical trials . The resulting tool will allow accurate, objective and reproducible quantification of imaging biomarkers for the prediction of treatment response both in a metastatic and in an early-stage scenario. The system will be validated and applied to ongoing prospective clinical trials of cancer patients treated with immunotherapy.
