NAIST Team Uses Dark-Field ML to Detect Cancer Cells

Researchers at Japan’s Nara Institute of Science and Technology (NAIST), with collaborators at Kindai University Faculty of Medicine and Hyogo Medical University, have developed a machine learning approach combining dark-field microscopy with spectral analysis to improve cancer cell detection in cytology tests. Published in Scientific Reports on 3 August 2026, the study demonstrated approximately 91% accuracy in differentiating mesothelioma from reactive mesothelial cells, both notoriously difficult to distinguish visually.

Traditional cytology relies on pathologist interpretation of stained cell samples under conventional microscopes, examining nuclear enlargement and abnormal cell shapes. However, cancerous and normal cells sometimes look identical at conventional scales, with differences occurring at nanometric scales in structures such as actin filaments that potentially alter how cells scatter light. The NAIST team used dark-field microscopy to capture light scattered by cells in the 420 to 720 nanometer range across mesothelioma and reactive mesothelial cell specimens, condensed the spectra using principal component analysis, and classified them using a support vector machine. The system achieved approximately 91% accuracy in patient-based validation and showed potential for distinguishing gastric, urothelial, lung and thoracic cancer cells.

Assistant Professor Yuka Tsuri, who led the NAIST study, positioned the findings as demonstrating that submicron-scale light scattering information is highly sensitive to cell-type differences compared with visual inspection or image analysis alone. The team envisions the technology as complementary to pathologist skill and judgment rather than replacement, with an integrated spectroscopic microscope acting as a diagnostic aid. The proof-of-concept adds light-scattering biophysical information to conventional cytology tests, opening new commercial opportunities across the growing digital pathology and cytology AI category.

The commercial signal is a rapidly maturing digital pathology and cytology AI market where multi-modal imaging (combining visual, spectral and biophysical data) is emerging as a key competitive differentiator. Integration with existing microscopy workflows accelerates commercial adoption. Expect Roche Tissue Diagnostics, Danaher (Leica Biosystems), Philips, Hologic (Genius platform), Paige AI, PathAI and Ibex Medical Analytics to sharpen positioning through 2027.

For the latest updates and in-depth insights into the world of Life Science, including breakthrough treatments, industry trends, and regulatory news, contact Adam Tiberius today