Magnetic resonance imaging-based tool to predict breast cancer recurrence
Breast cancer is a leading cancer diagnosis among women with over 270,000 new cases every year in the United States. Breast cancer comprises a diverse collection of diseases with varying genomic signatures and treatment responses. Breast cancer treatment planning is often dictated by predictive factors such as patient age, tumor characteristics, and disease extent. Genomic biomarkers are an increasingly widely used tool that oncologists use to inform their treatment decisions. These biomarkers, such as Oncotype DX, perform subclassification of tumors and provide a risk score that oncologists can then use to provide a personalized treatment plan. However, this strategy is expensive and requires a physical sample of the tumor. There is a need for cost-effective, non-invasive methods to collect prognostic information about breast cancer tumors.
Duke inventors have developed a computational method that assesses MR images and provides a risk score for breast cancer patients. This technology is intended to be used by radiologists and oncologists to inform treatment decisions. Breast cancer patients typically have medical imaging completed prior to any treatment, which makes MRI information an underutilized source for treatment planning. The technology is trained by previously acquired MR images, then a radiologist or oncologist indicates the tumor location by quickly drawing a box within the software. The invention automatically segments the tumor and extracts its features using computer vision algorithms. The prediction of a recurrence risk is generated using the extracted features and a previously defined statistical/machine learning model. The generated score is then provided to the clinician. This technology has been studied with datasets totaling close to 1000 breast cancer patients.
- Provides a cost-effective, non-invasive method for collecting prognostic data for breast cancer patients using MR images
- Could improve patient outcomes of breast cancer patients through more personalized treatment strategies
- Can be easily used by radiologists after implementation into commercially available MRI hardware and workstations