Artificial intelligence is helping researchers predict how well rectal cancer patients will respond to treatment by analyzing standard biopsy samples. A new study led by scientists at University College London (UCL) and University College London Hospitals (UCLH) demonstrates that AI can detect immune cell patterns in tumor tissue that correlate with survival and recurrence risk—information not currently used in routine clinical decisions. The study, published in eBioMedicine, focused on the tumor microenvironment—the complex mix of immune cells, cancer cells, and surrounding tissue that influences how cancer grows and responds to therapy. While pathologists traditionally examine tissue slides under a microscope, the team used AI to rapidly analyze millions of images from routine biopsies, identifying key immune cell types and their spatial distribution. Dr. Charles-Antoine Collins-Fekete, a senior author from UCL’s Medical Physics & Biomedical Engineering department, explained that AI can extract valuable insights in minutes, far faster and more affordably than methods like whole-genome sequencing or spatial transcriptomics. “Pathology slides are already part of standard care, so they’re a rich, underused data source. AI can reveal hidden immune signatures that are hard to spot by eye but are critical to understanding treatment outcomes.” The researchers analyzed samples from over 900 patients, including those in the ARISTOTLE clinical trial. They found that higher levels of lymphocytes—immune cells that attack cancer—were linked to longer survival and lower risk of cancer returning. Conversely, elevated macrophages—immune cells that can support tumor growth—were associated with poorer outcomes. The AI system was trained on vast image datasets and tested on patient samples before and after chemoradiotherapy. Patients who showed an increase in tumor-infiltrating lymphocytes after treatment, indicating a stronger anti-tumor immune response, fared better. Those whose tumors remained immunologically “cold” were more likely to relapse. The study also revealed that genetic factors interact with immune profiles. For instance, patients with normal KRAS genes and high lymphocyte levels had better survival, while those with KRAS mutations and low lymphocytes did not. Similarly, high macrophage levels were especially dangerous in patients with TP53 gene mutations. Combining immune data with genetic information allowed the team to better predict patient risk. This could help guide treatment—intensifying therapy for high-risk patients and reducing exposure to chemoradiotherapy in low-risk individuals. The team also found that tumors with high mitotic activity—rapid cell division—tended to suppress immune responses, contributing to worse outcomes. To support clinical use, the researchers developed Octopath, a free online tool that allows doctors to upload pathology slides and receive automated immune analysis. However, they caution that larger, more diverse studies are needed to validate the findings before widespread adoption. Professor Maria Hawkins, a senior author and consultant clinical oncologist at UCLH, said the work represents a promising step toward personalized cancer care. “This is an early but exciting move toward using AI to refine cancer classification. In the future, AI could provide timely, actionable insights to help doctors and patients make better-informed treatment decisions.” The team plans to expand the research to include more detailed immune cell types and advanced analytical techniques, aiming to deepen understanding of how cancer evades the immune system and how therapies can be optimized.