医用画像AIのシステム的考察

抄録

In recent years, modern artificial intelligence (AI) is one of the most attractive topics in many scientific fields and now developed for real world applications including medical ones. In fact, deep learning, a typical modern AI, for medical image diagnosis has achieved break-through performance that is comparable to human experts. However, deep learning has so-called black-box nature that is one of critical issues for medical applications. In this review, the-state-of-the-art AIs for medical image diagnosis are overviewed with some problems caused by the black-box nature. Then, systems approach will be introduced to investigate the problems and reveal the core mechanism of break-through and drawback. Future direction will be suggested for overcoming the problems.