Related Experiment Video
Updated: Aug 3, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.0K
DeepMiCa: Automatic segmentation and classification of breast MIcroCAlcifications from mammograms
Alessia Gerbasi1, Greta Clementi1, Fabio Corsi2
1Department of Electrical, Computer and Biomedical Engineering, University of Pavia, Pavia, Italy.
Computer Methods and Programs in Biomedicine
|April 8, 2023
Summary
DeepMiCa is a novel AI system for detecting and classifying breast microcalcifications from mammograms. This automated pipeline offers a visual explanation, aiding clinicians in diagnosis and potentially reducing unnecessary biopsies.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Breast cancer is the most common cancer globally, with early detection crucial for survival.
- Microcalcifications are the earliest sign of breast cancer, but their classification as benign or malignant remains challenging.
- Current diagnostic methods often require invasive biopsy procedures to confirm malignancy.
Purpose of the Study:
- To develop DeepMiCa, a fully automated, visually explainable deep-learning pipeline for analyzing microcalcifications in mammograms.
- To provide a reliable decision support system to aid clinicians in diagnosing borderline difficult cases.
- To improve the accuracy and efficiency of microcalcification detection and classification.
Main Methods:
- The DeepMiCa pipeline involves three stages: raw scan preprocessing, automatic patch-based semantic segmentation using a UNet network with a custom loss function, and lesion classification via deep transfer learning.
- A custom loss function was designed to effectively handle extremely small lesions during segmentation.
- Explainable AI methods are integrated to generate visual interpretation maps of classification results.
Main Results:
- The segmentation and classification algorithms achieved an area under the ROC curve of 0.95 and 0.89, respectively.
- The system demonstrates high accuracy in detecting and classifying microcalcifications.
- DeepMiCa operates efficiently without requiring high-performance computational resources and provides visual explanations.
Conclusions:
- A novel, fully automated pipeline for breast microcalcification detection and classification has been developed.
- DeepMiCa has the potential to serve as a valuable second opinion tool, enabling clinicians to rapidly visualize and inspect critical imaging features.
- The proposed system may help reduce misclassification rates and the number of unnecessary biopsies in clinical practice.
