Related Experiment Video
Updated: Dec 30, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.5K
Cascaded multi-scale convolutional encoder-decoders for breast mass segmentation in high-resolution mammograms
Summary
This study presents a novel deep learning method for accurate breast mass segmentation in mammograms. The approach effectively handles diverse mass characteristics and improves generalizability for better cancer detection.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Accurate breast mass segmentation is crucial for mammogram interpretation.
- Challenges include class imbalance and diverse mass characteristics.
Purpose of the Study:
- To develop an effective deep learning model for high-resolution mammogram mass segmentation.
- To address limitations of existing methods like class imbalance and limited receptive fields.
Main Methods:
- A multi-scale cascade of deep convolutional encoder-decoders was employed.
- Auto-context was used for integrating multi-scale information and long-range spatial context.
- End-to-end training with simultaneous segmentation refinement was performed.
- Transfer learning from DDSM-CBIS to INbreast datasets was utilized.
Main Results:
- The proposed method achieved promising results in mass segmentation on high-resolution INbreast images.
- Demonstrated superior model generalizability compared to standard encoder-decoder strategies.
- Effectively managed class imbalance and diverse mass features.
Conclusions:
- The developed multi-scale cascade deep learning model offers an effective solution for breast mass segmentation.
- The approach shows significant potential for improving automated mammogram analysis and cancer diagnosis.
- The method's generalizability suggests broad applicability in clinical settings.
