Related Experiment Video
Updated: Aug 5, 2026
15:48
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
Deep Learning for Early Breast Cancer Detection on Contrast-Enhanced Breast MRI: A Multicenter Study
Na Young Jung1, Jihe Lim2, Hyug-Gi Kim3
1Department of Radiology, Eunpyeong St. Mary's Hospital, College of Medicine, The Catholic University of Korea, Seoul 03312, Republic of Korea.
Cancers
|July 28, 2026
Summary
A deep learning (DL) model aids radiologists in detecting small invasive breast cancers on MRI. While not improving sensitivity, the DL model enhanced radiologist precision in identifying these challenging lesions.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Accurate visual categorization of small enhancing lesions on contrast-enhanced breast MRI is challenging.
- Early detection of small invasive breast cancers is crucial for patient outcomes.
- Multi-institutional data is essential for developing robust AI models in medical imaging.
Purpose of the Study:
- To develop and evaluate a deep learning (DL) model for detecting small (≤2 cm) invasive breast cancers using multi-institutional contrast-enhanced breast MRI data.
- To assess the performance of the DL model in identifying enhancing breast masses.
- To investigate the potential of the DL model as an assistive tool for radiologists.
Main Methods:
- A retrospective study involving 1721 women with T1-stage invasive breast cancer across five hospitals.
- Development of a DL model using 14,917 labeled cancer images and 1443 labeled benign images from contrast-enhanced breast MRI.
- Evaluation of the DL model's performance using area under the precision-recall curve (AUPRC), sensitivity, precision, and F1 score, and comparison with radiologist performance.
Main Results:
- The DL model achieved an AUPRC of 0.42 for detecting small (≤2 cm) invasive breast cancers (sensitivity 83.2%, precision 33.2%, F1 score 0.47).
- For subcentimeter (≤1 cm) cancers, the DL model's performance was lower (sensitivity 75.3%, precision 20.8%, F1 score 0.33) compared to radiologists (mean F1 score 0.80).
- Radiologists using the DL model showed improved mean precision (83.2%) for subcentimeter cancers, with no significant change in sensitivity.
Conclusions:
- The DL model shows potential as an assistive tool to improve radiologist precision in detecting small invasive breast cancers on contrast-enhanced breast MRI.
- The DL model did not significantly improve radiologist sensitivity in detecting subcentimeter invasive breast cancers.
- Further research and validation are needed to optimize DL models for breast cancer detection in clinical practice.