Immunotherapy is a Nobel Prize winning approach to treat cancer patients, which has given spectacular responses in some previously untreatable cancer patients. However, many patients do not respond to immunotherapy and are only exposed to the toxicity of the drugs. Additionally, the cost of immunotherapy treatment is very high (~100kEuro per patient), which puts a dramatic financial burden on the healthcare system. In this context, there is an urgent need for identifying responders and non-reponders at an early stage, to guarantee an efficient and personalized cancer treatment with immunotherapy. The biomarker currently used in the clinic relies on assessment of tumor cells positive to a PD-L1 immunohistochemistry assay. Scoring PD-L1 suffers from subjective assessment made by pahtologists, from stain variability and from the fact that it mostly relies on visual estimation.
In this project, we will develop computational pathology models based on artificial intelligence, to assist quantification of PD-L1 positive tumor cells in lung cancer specimens, with the aim of supporting pathologists and oncologist in assessment of PD-L1 based biomarkers via a subjective, robust and reproducible quantification of PD-L1 positive cells.
We will develop a demonstrator that allows to run the built AI model/pipeline on digitized whole-slide images stained with a PD-L1 marker.