August 30, 2022
Keio University School of Medicine
A multi-institutional collaborative research group, led by Senior Assistant Professor Satoshi Hayashida and Professor Yuko Kitagawa of the Department of Surgery (General and Gastroenterological), Keio University School of Medicine, has jointly developed an artificial intelligence (AI)-based image diagnosis system using deep learning technology with Fixstars Corporation. This system, designed for breast ultrasound examinations, has been shown to be capable of determining with high accuracy whether patients undergoing breast cancer screening should receive further detailed examinations.
The AI, developed using a deep learning technique called Convolutional Neural Network (CNN), was designed to perform diagnoses based on the BI-RADS assessment criteria, an international standard for breast ultrasound imaging, and to determine whether the target ultrasound image contains lesions requiring further examination.
When this AI was used to diagnose 3,166 breast ultrasound images (containing 3,656 lesions) different from the training data, it was found to be capable of diagnosis with a sensitivity of 91.2% and a specificity of 90.7%. Considering that the passing criteria for "physicians certified for performing and interpreting breast cancer screening ultrasonography" by the Central Committee of the Japan Association of Breast Cancer Screening are a sensitivity of 80% and a specificity of 80%, these results demonstrate a diagnostic accuracy that significantly surpasses these standards.
Furthermore, when the results of breast ultrasound image diagnoses by 20 physicians, including 10 board-certified surgeons, were compared with the diagnostic results from the AI, the AI demonstrated statistically significantly superior accuracy.
In the future, this AI system is expected to be used as a diagnostic aid for physicians in breast ultrasound screening during breast cancer screenings and comprehensive health checkups. By preventing missed diagnoses and overdiagnosis, it is anticipated to contribute to improving accuracy and correcting disparities in medical technology, such as resolving differences in quality between facilities.
The results of this research were published in the online edition of the international scientific journal Cancer Science on August 3, 2022 (U.S. time).
Please see below for the full press release.