Related Experiment Video
Updated: Jan 1, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.5K
A similarity measure method fusing deep feature for mammogram retrieval
Zhiqiong Wang1,2,3, Junchang Xin4, Yukun Huang5
1College of Medicine and Biological Information Engineering, Northeastern University, China.
Journal of X-Ray Science and Technology
|December 24, 2019
Summary
This study introduces a new mammogram retrieval method that fuses deep learning features with content and location information. This approach significantly improves the accuracy of finding similar breast cancer cases for radiologists.
Area of Science:
- Radiology
- Medical Imaging
- Computer Science
Background:
- Breast cancer significantly impacts women's health, necessitating accurate diagnostic tools.
- Mammography is crucial for breast cancer screening and diagnosis.
- Efficient retrieval of similar past cases is vital for radiologists but time-consuming.
Purpose of the Study:
- To develop an advanced medical image retrieval system for mammograms.
- To enhance diagnostic accuracy by addressing the "semantic gap" in image retrieval.
- To improve the efficiency of referencing previous breast cancer diagnosis cases.
Main Methods:
- A novel similarity measure combining deep features (from CNN, SAE, DBN) with low-level content and location features.
- Image preprocessing and registration for feature extraction.
- Fusion of content, location, and deep similarity scores using optimized weights.
Main Results:
- The proposed method achieved a precision of 0.745, recall of 0.850, and a comprehensive evaluation index of 0.794.
- Fusing 60% Deep Belief Network (DBN) deep feature similarity with 40% low-level feature similarity yielded optimal results.
- The fused method demonstrated superior performance compared to content-based and location-based retrieval alone.
Conclusions:
- A new mammogram retrieval method effectively fuses deep, content, and location features.
- The developed system offers significant advantages in precision and recall for medical image retrieval.
- This approach enhances radiologists' ability to access relevant prior cases for improved diagnosis.
