• Text in Japanese

Using the perspective of a doctor, the development of a high precision AI-CAD system into the clinical workflow

Deep learning is one of the biggest advances in artificial intelligence (AI), and its effects in several different fields can already be felt. Perhaps no field anticipates more from deep learning than the field of medicine. The potential of deep learning in image recognition was demonstrated at Large Scale Visual Recognition Challenge (ILSVRC) 2012, and three years later was shown to surpass human ability. In response, the number of sessions devoted to deep learning at medical imaging conferences has dramatically increased. Topics include its application to image classification, image detection, image segmentation, and image generation. Key to this research development has been computer-aided diagnosis (CAD), a technology that assists doctors in imaging diagnostics. CAD is creating a paradigm shift in imaging diagnostics away from algorithms towards models based on large amounts of data. However, many issues remain before CAD becomes a mainstay in the clinic.

First is the insufficient amounts of data for the learning. In particular, a large-scale database with the lesion site labeled is required in medical imaging. In addition, although CAD is already being deployed for breast and colon cancer screenings, diagnostic imaging at actual clinical sites is not progressing at an acceptable rate. Further, CAD has not been incorporated effectively into clinical workflows.

Therefore, we are working with companies to develop the new AI technologies for imaging diagnostics. As next generation AI-CAD, these technologies include Explainable AI, and X-short learning AI for rare medical cases. Finally, by using data science approaches such as radiomics and radiogenomics to integrate imaging and non-imaging information, we are enhancing diagnostic accuracy.