AI-assisted gastric cancer diagnosis using a multimodal hyperspectral imaging system

Optical imaging techniques have been widely used in biomedical applications due to their high resolution, compact design, safety, and cost-effectiveness compared to other medical imaging methods. These methods hold significant diagnostic potential by capturing variations in tissue optical properties arising from genetic and phenotypic alterations. This presentation will introduce recently developed optical imaging techniques, including hyperspectral imaging technique with structured illumination and laser speckle imaging. Using a multimodal hyperspectral imaging system, excised tissues obtained via endoscopic submucosal dissection were analyzed to quantify intrinsic tissue properties such as scattering, absorption, and reflectance across the visible spectrum. In addition, artificial intelligence-based image analysis methods were developed to enhance diagnostic accuracy. In particular, a vision transformer model was trained to distinguish between normal tissue, adenomas, and gastric cancer based on complex intrinsic optical properties, enabling more precise disease classification.