Related Experiment Video
Updated: Jun 14, 2025
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
Deep Learning Applied to Diffusion-weighted Imaging for Differentiating Malignant from Benign Breast Tumors without
Mami Iima1, Ryosuke Mizuno1, Masako Kataoka1
1From the Department of Fundamental Development for Advanced Low Invasive Diagnostic Imaging, Nagoya University Graduate School of Medicine, 65 Tsurumai-cho, Showa-ku, Nagoya 466-8550, Japan (M.I.); Department of Diagnostic Imaging and Nuclear Medicine, Kyoto University Graduate School of Medicine, Kyoto, Japan (M.I., M.K., M.H., Y.N.); A.I. System Research, Kyoto, Japan (R.M.); Kyoto University Faculty of Medicine, Kyoto, Japan (K.T., T.Y.); Department of Diagnostic Radiology, Kyoto City Hospital, Kyoto, Japan (A.M.); Department of Diagnostic Radiology, Kansai Electric Power Hospital, Osaka, Japan (M.H.); e-Growth, Kyoto, Japan (K.I.); and Department of Breast Surgery, Kyoto University Graduate School of Medicine, Kyoto, Japan (M.T.).
Artificial intelligence (AI) models, specifically a small 2D convolutional neural network (CNN), demonstrated strong performance in distinguishing benign from malignant breast tumors using diffusion-weighted imaging (DWI). The AI models matched radiologist accuracy without requiring manual segmentation.
Area of Science:
- Radiology and Medical Imaging
- Artificial Intelligence in Medicine
- Oncology
Background:
- Accurate differentiation between benign and malignant breast tumors is crucial for effective patient management.
- Diffusion-weighted imaging (DWI) offers valuable information for characterizing breast lesions.
- Artificial intelligence (AI) holds promise for enhancing diagnostic accuracy in medical imaging.
Purpose of the Study:
- To evaluate and compare the performance of various AI models in differentiating benign from malignant breast tumors using DWI.
- To compare the diagnostic performance of AI models against radiologist assessments.
- To investigate the impact of data augmentation techniques on AI model performance.
Main Methods:
- Retrospective analysis of 3-T breast MRI data from 293 patients with 334 breast lesions.
- DWI acquisition with five b values (0, 200, 800, 1000, 1500 sec/mm²).
- Development and assessment of AI models including 2D CNN, ResNet-18, EfficientNet-B0, and 3D CNN, with comparative analysis against radiologist performance using ROC analysis.
Main Results:
- 2D CNN models outperformed 3D CNN models on the test dataset (AUC range: 0.83-0.88 vs 0.75-0.76).
- A small 2D CNN with specific data augmentations achieved an AUC of 0.88, comparable to radiologists (AUC: 0.86).
- No significant difference was found in specificity or sensitivity between the small 2D CNN and radiologists.
Conclusions:
- AI models, particularly a small 2D CNN, can effectively differentiate benign from malignant breast tumors on DWI.
- These AI models offer comparable diagnostic performance to radiologists without the need for manual segmentation.
- AI-powered DWI analysis represents a promising advancement in breast lesion characterization.
