Professor Manabu Fujimoto, Associate Professor Yasuhiro Fujisawa and others of the University of Tsukuba Medical and Medical Sciences have collaborated with Kyocera Communication Systems Co., Ltd. to develop a skin tumor artificial intelligence (AI) diagnostic assistance system with extremely high diagnostic accuracy of 90% or more. Successful.
In skin diseases, "visual inspection" is the most important diagnostic method, but skin cancer called malignant melanoma is difficult to distinguish because it looks very similar to a "mole".Although differential research using diagnostic imaging technology has been conducted for a long time, it is generally said that learning using at least 1,000 images for each category is required for AI image identification, and it is necessary to identify skin tumors. More than 14,000 images are required to build the system for this.The group has shown that skin tumors can be identified even by AI learning using about 6,000 clinical photographs, making full use of high-quality data after the diagnosis is confirmed and the image analysis know-how accumulated by Kyocera Communication Systems. rice field.
AI診断補助システムと日本皮膚科学会認定皮膚科専門医13名とで、画像診断テストを行ったところ、専門医の良悪性の識別率85.3%±3.7%に対し、AIの識別率は92.4%±2.1%と有意に高かった。また、良悪性の識別より難しい14種類の詳細な診断の正答率についても、皮膚科専門医が59.7%±7.1%、AIの正答率は74.5%±4.6%であり、こちらもAI診断補助システムのほうが優れていた。
The goal of this skin tumor AI diagnostic aid system is to use it in the actual clinical setting within a few years after sufficient performance evaluation.