This technology is an AI-powered software tool for the robust and automated quantification of breast density from digital mammograms, providing improved performance in breast cancer risk assessment.
The current clinical standard for breast density assessment, visual BI-RADS grading, is highly subjective and does not provide the quantitative, continuous measure needed for refined risk stratification. Existing automated analysis tools are expensive, inaccessible for research use, and function as "black boxes" by not providing dense tissue segmentation maps, rendering their results unverifiable. Most research-based alternatives are not publicly available and have not undergone independent validation using large, racially diverse, multi-institutional datasets, which are required to demonstrate robust clinical performance.
This technology, termed Deep-LIBRA, combines deep learning with radiomic machine learning to automatically quantify breast density from digital mammography images through a three-stage process. First, convolutional neural networks segment the breast region by removing the background and the pectoralis muscle. Next, the algorithm divides breast tissue into localized regions and extracts quantitative radiomic features. Finally, machine learning classifiers differentiate dense from non-dense tissue to generate both percent density measurements and spatial segmentation maps.
This technology has been trained on multi-institutional datasets and independently validated for breast cancer risk prediction.
Licensing Contact: Joan Martinez
