Inferring multicellular interactions in tumors from standard pathology slides (med.stanford.edu)

Stanford Medicine researchers have introduced CANVAS, an AI platform capable of predicting multicellular interactions within tumors from standard pathology slides, such as H&E stains. This innovative approach transforms the traditional, time-consuming analysis of tumor slides into a detailed understanding of cellular neighborhoods—complex ecosystems formed by cancer cells, immune cells, and stromal cells. The study, published in Cell, outlines how the platform identified ten distinct cellular neighborhoods in non-small cell lung cancer, including one linked to poorer patient outcomes and reduced effectiveness of immunotherapy. The significance of CANVAS lies in its ability to leverage existing pathology resources, significantly reducing costs and analysis time while providing insights into tumor ecosystems. By correlating these neighborhoods with patient prognosis, CANVAS demonstrates the potential to inform clinical decision-making more accurately than current methods. This advancement is further enhanced by previous AI tools developed in the lab, such as MUSK, which train on extensive medical image data to recognize key patterns. The researchers aim to validate CANVAS in clinical trials, potentially revolutionizing the way oncologists assess and approach cancer treatments by offering a new lens through which to view tumor biology.