Related Experiment Video
Updated: Jul 8, 2025
Machine Learning Algorithms for Early Detection of Bone Metastases in an Experimental Rat Model
Published on: August 16, 2020
A Deep Learning Decision Support Tool to Improve Risk Stratification and Reduce Unnecessary Biopsies in BI-RADS 4
Chika F Ezeana1, Tiancheng He1, Tejal A Patel1
1From the Department of Systems Medicine and Bioengineering, Houston Methodist Neal Cancer Center, Houston Methodist Hospital, Houston, Tex (C.F.E., T.H., L.W., S.T.C.W.); Houston Methodist Neal Cancer Center, Houston Methodist Hospital, Houston, Tex (J.E., J.C.C.); Departments of General Oncology (T.A.P.), Health Services Research (Y.C.T.S., B.K., I.W.P.), and Radiology (D.S., W.T.Y.), University of Texas MD Anderson Cancer Center, Houston, Tex; University of Texas Health Science Center, San Antonio, Tex (V.K., M.E., E.B., P.M.O., K.A.K.); University of the Incarnate Word School of Osteopathic Medicine, San Antonio, Tex (H.S.); Huntsman Cancer Institute, University of Utah, Salt Lake City, Utah (A.L.C., K.K.); and Department of Radiology, Houston Methodist Hospital, Weill Cornell Medicine, 6670 Bertner Ave, Houston, TX 77030 (S.T.C.W.).
The intelligent-augmented breast cancer risk calculator (iBRISK) accurately predicts malignancy in BI-RADS 4 mammography lesions. This AI tool can reduce unnecessary biopsies and associated costs, improving breast cancer screening efficiency.
Area of Science:
- Radiology and Oncology
- Artificial Intelligence in Medicine
- Medical Decision Support Systems
Background:
- Breast cancer diagnosis relies on accurate risk assessment for BI-RADS category 4 lesions.
- Overbiopsy of benign lesions leads to increased healthcare costs and patient anxiety.
- Existing risk calculators may lack precision in stratifying malignancy probability.
Purpose of the Study:
- To evaluate the performance of the intelligent-augmented breast cancer risk calculator (iBRISK) on a multicenter patient dataset.
- To assess iBRISK's accuracy in predicting malignancy for BI-RADS 4 lesions.
- To determine the potential cost savings and reduction in unnecessary biopsies facilitated by iBRISK.
Main Methods:
- iBRISK, developed using deep learning on clinical and mammographic data, was tested on an independent multicenter dataset.
- Data from 4209 women with BI-RADS 4 lesions were analyzed.
- iBRISK's precision in risk stratification and probability of malignancy (POM) estimation was evaluated, alongside its performance as a continuous predictor.
Main Results:
- The iBRISK model achieved 89.5% accuracy and an AUC of 0.93, with 100% sensitivity and 81% specificity.
- Low and high probability of malignancy (POM) groups showed distinct malignancy rates (0.16% vs. 85.9%).
- As a continuous predictor, iBRISK yielded an AUC of 0.97, with potential cost savings exceeding $420 million.
Conclusions:
- iBRISK demonstrates high sensitivity for malignancy prediction in BI-RADS 4 lesions.
- The tool can potentially obviate biopsies in up to 50% of patients with low or moderate POM.
- iBRISK offers a promising approach to reduce biopsy-associated costs and improve diagnostic accuracy in breast cancer screening.
