A new artificial intelligence tool called SMMILe is revolutionizing the way cancer is analyzed in tissue samples, offering the potential to personalize treatment by providing detailed insights into tumor composition and distribution. Developed by researchers at the University of Cambridge, the machine learning algorithm can detect cancer cells in biopsy and surgical tissue slides, identify the locations of tumor lesions, and estimate the proportions of regions with varying levels of aggressiveness—all without requiring time-intensive, region-by-region annotations from pathologists. Traditionally, training AI models for digital pathology has relied on high-quality, meticulously labeled slides, which are expensive and labor-intensive to produce. SMMILe overcomes this limitation by being trained on large datasets that only include simple, patient-level diagnostic labels such as cancer type or grade. Despite this minimal input, the algorithm delivers highly detailed spatial maps of tissue samples, revealing the structure and heterogeneity of tumors. In a study published in Nature Cancer, the team tested SMMILe on 3,850 whole-slide images across six cancer types: lung, kidney, ovarian, breast, stomach, and prostate. The model performed on par with, and in some cases outperformed, nine other state-of-the-art AI tools in classifying entire slides. More impressively, it significantly surpassed them in estimating the spatial distribution and proportions of different tumor subtypes and grades. Dr. Zeyu Gao, who led the development of SMMILe at the Early Cancer Institute, explained that cancer is often not uniform—different parts of a single tumor can vary in aggressiveness and biological behavior. “Our model doesn’t just say ‘yes, there’s cancer,’ it maps out these differences,” he said. “This could one day help doctors tailor treatments more precisely to each patient’s unique tumor profile.” Dr. Mireia Crispin-Ortuzar, joint senior author and Co-Lead of the Cancer Research UK Cambridge Centre Integrated Cancer Medicine Virtual Institute, compared the tool to a “sonar” for tissue images. “We often know a tumor exists, but not how it’s organized within the tissue,” she said. “Our method uses affordable, widely available data to generate detailed maps—without the high cost of specialized imaging or annotation.” The researchers plan to extend SMMILe’s capabilities to predict molecular biomarkers—biological indicators that reveal how a tumor behaves at a deeper level. This could unlock new insights into cancer development and progression, paving the way for more accurate, personalized treatment strategies. Dr. Dani Skirrow, Research Information Manager at Cancer Research UK, praised the findings, calling it a step toward faster, more precise cancer diagnosis. “This technology could help doctors quickly understand the full picture of a patient’s cancer, enabling better treatment decisions,” she said. “While further clinical testing is needed, the early results are promising.” The research was funded by Cancer Research UK and GE HealthCare. The University of Cambridge and Addenbrooke’s Charitable Trust are also advancing cancer care through the upcoming Cambridge Cancer Research Hospital, set to open on the Cambridge Biomedical Campus, combining clinical excellence with cutting-edge research to transform cancer diagnosis and treatment.