Related Experiment Video
Updated: Aug 30, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Connected-SegNets: A Deep Learning Model for Breast Tumor Segmentation from X-ray Images
Mohammad Alkhaleefah1, Tan-Hsu Tan1, Chuan-Hsun Chang2
1Department of Electrical Engineering, National Taipei University of Technology, Taipei 10608, Taiwan.
Cancers
|August 26, 2022
Summary
This study introduces Connected-SegNets, a novel deep learning model for enhanced breast tumor segmentation in X-ray images. The model achieves superior performance using intersection over union loss and advanced data augmentation techniques.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Accurate breast tumor segmentation is crucial for effective diagnosis and treatment planning in mammography.
- Existing deep learning models face challenges with noise and data variability in medical images.
Purpose of the Study:
- To propose Connected-SegNets, a novel deep learning architecture for improved breast tumor segmentation in X-ray images.
- To enhance model robustness and accuracy through architectural modifications and optimized loss functions.
Main Methods:
- Developed Connected-SegNets by integrating two SegNet architectures with skip connections.
- Implemented intersection over union (IoU) loss to improve noise robustness.
- Applied contrast limit adaptive histogram equalization (CLAHE) for preprocessing and rotation/flipping for data augmentation.
Main Results:
- Connected-SegNets outperformed state-of-the-art methods on INbreast, CBIS-DDSM, and a private dataset.
- Achieved a maximum Dice score of 96.34% on INbreast and 92.86% on CBIS-DDSM.
- Attained the highest IoU scores, reaching 91.21% on INbreast and 87.34% on CBIS-DDSM.
Conclusions:
- Connected-SegNets demonstrate significant potential for accurate and robust breast tumor segmentation.
- The proposed model offers a promising advancement in AI-driven medical image analysis for breast cancer detection.
