Related Experiment Video
Updated: Jul 1, 2026
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.0K
Application of Tuning-ensemble N-Best in Auto-Sklearn for Mammographic Radiomic Analysis for Breast Cancer Prediction
Faikah Awang Ismail1,2, Muhammad Khalis Abdul Karim2,3, Siti Izzatul Akma Zaidon3
1School of Biology, Faculty of Applied Sciences, Universiti Teknologi MARA, Cawangan Negeri Sembilan, Kampus Kuala Pilah, 72000, Malaysia.
Current Medical Imaging
|August 4, 2025
Summary
Auto-Sklearn automated machine learning improved breast tumor classification accuracy using mammographic radiomic features. This automated approach enhances diagnostic performance, offering a scalable and clinically relevant tool for early breast cancer detection.
Area of Science:
- Medical Imaging
- Machine Learning
- Oncology
Background:
- Breast cancer poses a significant global health challenge, with mammography as a primary screening tool.
- Mammography interpretation faces limitations due to inter-observer variability and difficulty distinguishing benign from malignant lesions.
- Automated machine learning (AutoML) offers potential to enhance diagnostic accuracy in breast cancer detection.
Purpose of the Study:
- To evaluate the efficacy of Auto-Sklearn, an AutoML framework, for classifying breast tumors using radiomic features from mammograms.
- To assess the performance of automated model selection, hyperparameter tuning, and ensemble construction in breast cancer classification.
- To compare the performance of Auto-Sklearn with conventional models in a mammographic context.
Main Methods:
- Radiomic features were extracted from 244 mammographic images after enhancement (CLAHE) and segmentation (ACM).
- Thirty-seven radiomic features (first-order, GLCM texture, shape) were extracted, standardized, and used to train Auto-Sklearn models.
- A dataset split of 80% training and 20% testing was employed, with a resampling strategy applied to address class imbalance.
Main Results:
- Auto-Sklearn achieved an initial accuracy of 55.10% on the test set, which improved to 76.16% after applying a resampling strategy.
- The Receiver Operating Curve Area Under Curve (ROC-AUC) improved from 0.660 to 0.840 with the resampling strategy, outperforming conventional models.
- The study demonstrated Auto-Sklearn's effectiveness in automating the classification of breast tumors based on radiomic features.
Conclusions:
- Auto-Sklearn effectively automates tumor classification, enhancing performance with handcrafted radiomic features.
- The framework presents a scalable, automated, and clinically relevant tool for breast cancer classification using mammographic radiomics.
- Future research should consider larger datasets and incorporate clinical metadata for further validation.
