Researchers in Japan have developed a machine learning system capable of identifying cancer cells by analyzing how they scatter light. This innovative method seeks to enhance cytological testing, which traditionally relies on a pathologist’s ability to visually distinguish cancerous cells from normal ones.
Machine Learning Enhances Cytological Testing
The study, led by Assistant Professor Yuka Tsuri from the Nara Institute of Science and Technology (NAIST), was published on August 3, 2026, in the journal Scientific Reports. Co-authors include a team of scientists from both NAIST and the Kindai University Faculty of Medicine. They focused on how subtle changes in nanometric structures of cells could affect light scattering, potentially helping to differentiate cancerous cells from healthy cells that may appear similar under traditional microscopy.
Cytological tests are widely used for early cancer detection due to their minimally invasive nature. However, the accuracy of these tests largely depends on the skill of the examining pathologist. In some instances, the differences between cancerous and non-cancerous cells are so minor that they may evade detection. The researchers aimed to determine whether a machine learning approach could identify these minute optical distinctions.
Utilizing dark-field microscopy, which captures light scattering rather than light passing through cells, the team examined specimens containing cancerous mesothelioma cells alongside benign reactive mesothelial cells. They analyzed white light scattering across wavelengths from 420 to 720 nanometers. The resulting spectral data was then processed through a machine learning pipeline, which included principal component analysis to condense the data before classification using a support vector machine algorithm.
The findings revealed that the system achieved approximately 91% accuracy in differentiating between the mesothelioma and reactive mesothelial cells during patient-based validation. The methodology demonstrated promise in identifying other types of cancer, including gastric and urothelial cancers, with varying levels of success based on tumor types.
Dr. Tsuri emphasized the potential of the light scattering spectrum to provide detailed insights that surpass traditional visual analysis, enabling more precise diagnostics. The team envisions integrating this spectroscopic technology into existing cytology practices as a supplementary tool for pathologists, potentially increasing diagnostic accuracy.
Currently, the researchers are focused on optimizing the optical settings and refining the operational protocol, alongside further enhancing the machine learning models to improve their diagnostic capabilities.
Why It Matters
This development marks a significant advancement in cancer diagnostics, leveraging technology to address limitations in conventional methods. The integration of machine learning with optical analysis could empower medical professionals to make more accurate diagnoses, ultimately aiding in earlier detection and treatment of various cancers.
