[2E1-GS-13] AI application: Medical application (2)

[2E1-GS-13-02] THA auto templating by multimodal machine learning model using structured and unstructured data

〇Tomohisa Kobayashi¹, Naoya Kishimoto¹, Nobusuke Hanagasaki, Lu Xiangxun¹, Toshihiro Sasai², Takuya Sugasawa², Kazuhiko Kitano², Saori Kamiya², Masaaki Matsubara³ (1. IBM Japan, Ltd., 2. Johnson & Johnson K.K., 3. Nissan Tamagawa Hospital)

Keywords:AI, Multimodal Machine Learning, Image Analysis, THA, Templating

In total hip arthroplasty (THA), optimization of preoperative planning, which is called “Templating”, for each patient involves many considerations such as clinical images, measurements, bone tissue density, and muscle tension, so that it is difficult even for senior doctors. On the other hand, the optimization of templating leads to outcome improvement and reliable procurement of implants. Also, medical device manufacturers are shipping implants with different sizes and types more than actually needed in surgery, which squeezes profit. In this paper, we build prediction models using past THA case data (about 200 cases), which can predict the optimal implant size and the insertion position / angle / coordinate prediction on the X-ray for each patient, and developed a web application to visualize the prediction results for doctors. The prediction results exceed some doctor’s accuracy. And, about 40-61% potential shipment reduction was discovered.