A research group from Nara Institute of Science and Technology, Kindai University, and Osaka Institute of Technology has developed a method to diagnose cancer cells with high accuracy by using AI to analyze the "light scattering spectrum" generated when light is shone on cells.
In cytology, a pathologist, a specialist in cancer diagnosis, examines the size and shape of cells and nuclei, as well as their arrangement, under a microscope to determine the presence and type of cancer cells. However, cancer cells, which closely resemble normal cells in morphology, are difficult to distinguish solely through microscopic observation, and the diagnosis requires the experience and skill of the physician.
Cancerous cells undergo various changes not only in shape but also within the cell itself. However, many of these changes cannot be observed under a microscope and have not been fully utilized in diagnosis until now. Therefore, the research group focused on light scattering spectra, which reflect changes in the ultrastructure inside cells.
By irradiating cells with light, capturing the scattered light, and analyzing its spectrum using machine learning, researchers demonstrated that even morphologically similar cells can be classified with high accuracy from normal cells and cancer cells based on their spectral characteristics. The effectiveness of this method has been confirmed for various cancer cells, including malignant mesothelioma, which is considered extremely difficult to distinguish even for pathologists, as well as lung adenocarcinoma, small cell lung cancer, gastric adenocarcinoma, and urothelial carcinoma.
Furthermore, the system demonstrated high classification performance even with cytological specimens prepared at different medical institutions, confirming its relative reliance on facility differences. This suggests its potential for application to a wide range of cytological diagnoses.
This research is unprecedented and has been patented. Moving forward, we aim to verify its effectiveness using cells collected from a larger number of patients, further enhance machine learning, and realize a cell diagnostic support system that is easy to use in actual medical settings. If put into practical use, it is expected to support diagnoses by pathologists, leading to improved diagnostic accuracy and efficiency, and ultimately resulting in higher quality cancer diagnoses.
Paper information: [Scientific Reports] Accurate Tumor Cell Identification via AI analysis of White Light Scattering Spectra in Cytology
