Related Experiment Video
Updated: Dec 29, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.5K
Deep Learning Computer-Aided Diagnosis for Breast Lesion in Digital Mammogram.
Mugahed A Al-Antari1,2, Mohammed A Al-Masni1, Tae-Seong Kim3
1Department of Biomedical Engineering, College of Electronics and Information, Kyung Hee University, Yongin, Republic of Korea.
Advances in Experimental Medicine and Biology
|February 8, 2020
Summary
This study presents an integrated deep learning system for breast lesion diagnosis using mammograms. The system achieves high accuracy in detecting, segmenting, and classifying lesions, aiding radiologists in diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Accurate diagnosis of breast lesions from mammograms is crucial for effective cancer treatment.
- Computer-aided diagnosis (CAD) systems aim to assist physicians by automating key tasks in medical image analysis.
- Existing CAD systems often lack integration across detection, segmentation, and classification stages.
Purpose of the Study:
- To develop and evaluate an integrated deep learning-based CAD system for breast lesion diagnosis.
- To assess the system's performance in detection, segmentation, and classification of breast lesions from digital X-ray mammograms.
- To demonstrate the system's potential to assist radiologists in improving diagnostic accuracy.
Main Methods:
- An integrated CAD system was developed using deep learning techniques.
- Lesion detection was performed using the You-Only-Look-Once (YOLO) algorithm.
- Lesion segmentation was achieved with a full resolution convolutional network (FrCN).
- Classification of lesions as benign or malignant utilized regular feedforward CNN, ResNet-50, and InceptionResNet-V2 models.
- The INbreast database was used for evaluation with fivefold cross-validation.
Main Results:
- The YOLO-based detection achieved 97.27% accuracy, 93.93% MCC, and 98.02% F1-score.
- FrCN segmentation yielded 92.97% overall accuracy, 85.93% MCC, and 92.69% Dice score.
- Classification accuracies were 88.74% (CNN), 92.56% (ResNet-50), and 95.32% (InceptionResNet-V2).
- The integrated system demonstrated superior performance compared to conventional deep learning methods.
Conclusions:
- The developed integrated CAD system effectively performs detection, segmentation, and classification of breast lesions.
- The system shows significant potential to enhance the diagnostic capabilities of radiologists.
- Deep learning approaches offer a promising avenue for advancing computer-aided diagnosis in mammography.
