Related Experiment Video
Updated: Aug 17, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Attention-based deep learning for breast lesions classification on contrast enhanced spectral mammography: a
Ning Mao1,2, Haicheng Zhang2, Yi Dai3
1Department of Radiology, Yantai Yuhuangding Hospital, Qingdao University, 264000, Yantai, Shandong, P. R. China.
British Journal of Cancer
|December 15, 2022
Summary
An attention-based deep learning model effectively distinguishes benign from malignant breast lesions on contrast-enhanced mammography (CESM). This AI tool also enhances the diagnostic accuracy of radiologists, improving patient care.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Distinguishing benign from malignant breast lesions is crucial for effective cancer treatment.
- Contrast-enhanced mammography (CESM) is an important imaging modality for breast lesion characterization.
Purpose of the Study:
- To develop and evaluate an attention-based deep learning model for improved classification of breast lesions on CESM images.
- To compare the performance of the developed model against conventional methods and human experts.
Main Methods:
- Utilized a multicentre cohort of 1239 preoperative CESM images with pathological diagnoses.
- Integrated the convolutional block attention module (CBAM) into DenseNet 121, Xception, and ResNet 50 architectures.
- Assessed model performance using ROC curves, accuracy, PPV, NPV, F1 score, PRC, and heat maps.
Main Results:
- The CBAM-based Xception model achieved an AUC of 0.970 on the external test set, outperforming other CNNs, radiomics, and radiologists.
- The model demonstrated high sensitivity (0.848) and perfect specificity (1.000), with an accuracy of 0.891.
- Heat maps and PRC confirmed the model's favorable predictive performance, and radiologists' performance improved with AI assistance.
Conclusions:
- Attention-based deep learning models applied to CESM images significantly aid in differentiating benign from malignant breast lesions.
- The integration of this AI technology enhances the diagnostic capabilities of specialized radiologists, potentially leading to better clinical outcomes.
