Related Experiment Video
Updated: Sep 6, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.0K
End-to-End Calcification Distribution Pattern Recognition for Mammograms: An Interpretable Approach with GNN
Melissa Min-Szu Yao1,2, Hao Du3,4, Mikael Hartman3,4,5
1Department of Radiology, Wan Fang Hospital, Taipei Medical University, Taipei 116, Taiwan.
Diagnostics (Basel, Switzerland)
|June 24, 2022
Summary
This study introduces an interpretable artificial intelligence (AI) model using graph convolution to automatically detect and classify breast cancer calcification patterns in mammograms, showing promising results superior to baseline models.
Area of Science:
- Medical Imaging Analysis
- Artificial Intelligence in Oncology
- Machine Learning for Breast Cancer Detection
Background:
- Mammographic calcifications are key indicators of breast cancer.
- Accurate classification of calcification patterns is crucial for diagnosis.
- Existing methods may lack interpretability and efficiency in pattern classification.
Purpose of the Study:
- To develop a novel, interpretable artificial intelligence (AI) model for automatic detection and classification of mammographic calcification patterns.
- To utilize a unique graph convolution approach for enhanced pattern recognition.
- To improve the accuracy and interpretability of AI-driven mammogram analysis for breast cancer detection.
Main Methods:
- Development of a graph-convolutional-network-based AI model.
- Training and validation using 581 mammographic images from 292 breast cancer patients with classified calcification patterns (diffuse, segmental, regional, grouped, linear).
- Evaluation of model performance using precision, recall, F1 score, accuracy, and area under the ROC curve, with interpretable visualization methods.
Main Results:
- The AI model achieved a precision of 0.522, sensitivity of 0.643, specificity of 0.847, F1 score of 0.559, accuracy of 64.3%, and AUC of 0.745.
- Performance was superior to all baseline models, particularly in predicting linear and diffuse patterns, and grouped and regional patterns.
- The model's predictions were interpretable through visualization, highlighting important calcification nodes.
Conclusions:
- The proposed deep neural network framework offers an effective AI solution for automatic detection and classification of calcification distribution patterns in mammograms.
- The model demonstrates high potential for assisting in the diagnosis of breast cancers.
- Further clinical validation and data augmentation are recommended to enhance the AI model's performance.
