Researchers at the University of Jyväskylä have developed a new artificial intelligence (AI) tool demonstrating strong performance in identifying colorectal cancer. The AI-based model, developed in collaboration with the University of Turku’s Institute of Biomedicine, the University of Helsinki, and Nova Hospital of Central Finland, is designed for the automatic analysis of tissue samples.
The AI tool identifies key tissue categories relevant to colorectal cancer identification with impressive accuracy. The tool’s accuracy is 96.74%, outperforming all previous models in the classification of tissue microscopy samples.
“Based on our study, the developed model is able to identify all tissue categories relevant for cancer identification, with an accuracy of 96.74%,” said Fabi Prezja, Doctoral Student at the University of Jyväskylä.
The system works by analyzing digital microscopy slides prepared from a patient’s intestine sample, highlighting areas containing different tissue categories. This automated approach has the potential to significantly cut down on the time pathologists spend manually reviewing slides.
This innovative tool could significantly ease the workload of histopathologists, potentially leading to quicker diagnoses, prognoses, and greater clinical insights. The research team has made this AI tool freely available for further research and development.
“The free availability aims to accelerate future advances by encouraging scientists, developers and researchers worldwide to continue developing the tool and finding new applications for it,” Prezja explained.
However, the research team cautions that the introduction of AI tools into clinical practice must be approached with care, and stresses the need for rigorous validation. Before AI solutions become standard practice, their results must be consistently validated to ensure they align with clinical and regulatory standards.
Source: University of Jyväskylä
Journal Reference: Prezja, F., et al. (2025). Improving performance in colorectal cancer histology decomposition using deep and ensemble machine learning. Heliyon. doi.org/10.1016/j.heliyon.2024.e37561
