In a pathology lab in Central Finland, a tissue sample that once took several days to analyze can now be evaluated in minutes. Researchers at the Faculty of Information Technology at the University of Jyväskylä have developed an artificial intelligence model that can rapidly assess colorectal cancer samples and predict whether a patient's cells have a functioning DNA repair mechanism—a crucial piece of information that influences both how the disease develops and how it should be treated.
The work addresses a genuine bottleneck in cancer diagnosis. Liisa Petäinen, who led the study, explains the current reality: pathologists manually examine magnified tissue samples under a microscope, a painstaking process that can consume days per sample. "Analyzing a cancer sample in a pathology laboratory—regarding, for example, the MMR mechanism—can take several days," Petäinen says, "whereas artificial intelligence can reduce the analysis time to minutes." Those minutes matter enormously. Faster diagnosis means patients can access treatment sooner, while hospitals reduce costs and free up pathologists to focus on other critical work.
The AI model's core task is to identify whether the MMR mechanism—the cell's own error-correction system that catches small mistakes during DNA replication—is working properly. When this mechanism fails, it can accelerate cancer development and fundamentally change treatment strategy. The researchers tested their model on approximately 1,300 colorectal cancer patients from Central Finland, then validated it using data from Oulu University Hospital and the United States, published in Computer Methods and Programs in Biomedicine.
What makes this approach distinctive is its flexibility. Typically, pathologists analyze tissue at twentyfold magnification, a high zoom level that requires precise identification of the tumor area beforehand. The researchers discovered their model also performed well at fivefold magnification—a much broader view that captures the surrounding tissue context. This matters because the study revealed something unexpected: tissue features in the areas surrounding the tumor, not just the tumor itself, can help predict whether the repair mechanism functions. Eventually, the entire tissue sample could be analyzed in a single step, eliminating the need to pinpoint the tumor area first.
The collaboration between University of Jyväskylä and the Central Finland Welfare Region proved essential. Tiina Jokela notes that Finland's infrastructure—high-quality biobanks, robust health registers, and a unified healthcare system—created an ideal laboratory for this work. "Central Finland offers a good pilot environment where research and clinical work can collaborate flexibly," Jokela explains. The Central Hospital Nova provided clinical data and real-world requirements, while the university contributed AI and data analytics expertise.
The implications extend beyond speed and cost. The AI model's consistent performance across different magnification levels and geographic datasets suggests it could be adapted and deployed in laboratories worldwide, potentially reaching patients in regions where pathologist shortages make current analysis impractical. As the researchers continue refining the approach with larger datasets—a validation step they emphasize as essential for any new diagnostic method—the path toward routine AI-assisted cancer analysis becomes clearer. The model doesn't replace human expertise; it amplifies it, turning days of painstaking work into minutes of precise analysis.