CCNY, MSKCC experts develop high-performance open-source AI for breast cancer detection

In a major breakthrough, a team of researchers from The City College of New York and Memorial Sloan Kettering Cancer Center (MSKCC) has developed a new AI model that can detect breast cancer in MRI images and pinpoint the location of tumors. The news appears in the journal Radiology: Artificial Intelligence. 

While AI methods have made significant strides in breast cancer detection, deep learning models often lack interpretability and are rarely openly available for external validation. This is particularly important for MRI, with heterogeneous imaging protocols, and small datasets. The CCNY-MSKCC team’s objective was to address these issues, by publicly releasing a model that has been trained to detect and localize breast, tested on data from two different clinical sites. 

The new model’s performance is comparable to that of specialized breast radiologists but is better than existing automated tools, said Lucas C. Parra, Professor of Biomedical Engineering at CCNY and co-head of the project. It was trained on the largest breast MRI dataset to date and has been released publicly to facilitate independent evaluation and foster future development. 

Since early detection is crucial for successful treatment and improving patient outcomes, the model establishes a new state-of-the-art method for detecting breast cancer, which is a leading cause of cancer-related deaths among women in the United States. 

Breast MRI is more sensitive in detecting cancer than conventional mammograms. Given recent recommendations to expand the use of breast MRI to radiologically dense breasts, the use of breast MRI in breast cancer screening is likely to expand.

So far, mammography is the primary screening tool for this cancer because it combines good sensitivity, easy access, and low cost. However, women with a high risk of developing breast cancer are recommended supplemental annual screening with Magnetic Resonance Imaging (MRI) due to its higher sensitivity. Breast MRI is also used in diagnostic contexts when a tumor is suspected based on clinical findings, mammography, or ultrasound. 

In addition to Parra, other members of the research team included [CCNY first]: Lukas Hirsch, Yu Huang, Beliz Kayis, and Hernan A. Makse (Benjamin Levich Institute). Elizabeth J. Sutton, Mary Hughes, and Danny Martinez were CCNY’s collaborators from MSKCC’s Department of Radiology.

Parra and Sutton, MD, Attending Radiologist at MSK’s Breast Center, co-lead the $4 million NIH-funded project, “Machine learning for risk-adjusted breast MRI screening.” The project is leveraging modern machine learning techniques to analyze medical images, an area of expertise for Parra. The goal is to detect breast cancer as early as possible while limiting the burden of screening in high-risk women.