In a groundbreaking partnership, researchers from The City College of New York (CCNY) and Memorial Sloan Kettering Cancer Center (MSKCC) have developed a powerful new AI model that detects and localizes breast cancer in MRI scans accurately. The AI model was published in Radiology: Artificial Intelligence and combining interpretability and deep learning, is a completely open-source architecture that has advanced state of the art in both accuracy and transparency.
Unlike traditional “black-box” AI models, this AI system visually localizes the suspected tumor, which allows radiologists to glean essential diagnostic information from the AI system and ideally allows it to be used more seamlessly in existing clinical workflows.
Built on the Largest Breast MRI Dataset to Date
Breast MRI offers superior sensitivity over mammography, especially for women with dense breast tissue, but has been underutilized due to lack of robust AI tools. This new model is trained on thousands of annotated breast MRIs from two leading institutions, ensuring high generalizability across scanner types, imaging protocols, and demographics.
The system uses convolutional neural networks (CNNs) and combines classification and localization in a unified framework an approach key to enhancing precision medicine via accurate tumor mapping.
Open-Source AI Rivals Experts in Breast Cancer Detection
An innovative AI model created by CCNY and MSKCC has successfully matched breast radiologists and thoroughly outperformed commercial AI products on a metric of accuracy.
This model produces accurate tumor localization, interpretable outputs to provide confidence to the clinician, and integrates seamlessly into MRI workflows. This has the potential to minimize radiologist workload, increase diagnostic throughput, and allow for earlier, more confident cancer diagnosis.
NIH-Funded Project Promotes Transparency and Collaboration
This project is unique in that it is fully open-source everything about the model architecture, training code, and the datasets are released to the public.
The work is being funded by a $4 million NIH grant, and the key focus of the project is to leverage AI to enable risk-adjusted breast MRI screening to avoid unnecessary MRI exams. There are plans to use AI on longitudinal imaging for the purpose of better monitoring and treatment planning in the future by prioritizing reproducibility and collaboration worldwide.
Conclusion
This breakthrough model is more than a technical milestone it’s a call to action for open, collaborative innovation in medical AI. By offering transparency, robustness, and expert-level accuracy, the CCNY and MSKCC project stands to transform oncologic imaging, radiologist support, and ultimately, patient care.
About The Author
