Breast cancer is a global health crisis with millions of people diagnosed each year and hundreds of thousands dying from their disease. Screening is crucial because many breast cancers are curable when detected early. Mammography screening for breast cancers has reduced the mortality rate by 40%. However, mammography often misses cancers in dense breast tissue, and it can be difficult to differentiate benign and cancerous lesions. Even when mammography is supplemented with breast ultrasound, follow-up biopsies show that 70% of lumps tested are benign. Dr. Lisa Mullen from the Johns Hopkins School of Medicine, in collaboration with Drs. David Porter and Susan Harvey, is hoping to improve the accuracy and sensitivity of breast cancer screening. Using innovative artificial intelligence (AI) technologies, the team is developing new diagnostic tools to better read screening images, resulting in more accurate detection of cancer. They are developing machine learning algorithms that process real-world data from 100,000 digital mammographs and ultrasound images known to have been cancerous or benign. They are using an AI platform that combines data from several images to improve detection while reducing false positives. This method, known as upstream data fusion, has proved successful for the Department of Defense in detecting military targets. A pilot study applying these techniques to breast imaging data already shows promisingly improved accuracy. In this project, the team is aiming to optimize the platform and achieve up to 20% greater sensitivity and specificity than current screening methods.
Porter DW, Walton WC, Harvey SC, Mullen LA, Tsui BMW, Kim S, Peyton KS. Breast cancer detection/diagnosis with upstream data fusion and machine learning. Proc. SPIE. 2020.