Related Experiment Video
Updated: Mar 2, 2026
15:48
Tracking the Mammary Architectural Features and Detecting Breast Cancer with Magnetic Resonance Diffusion Tensor Imaging
Published on: December 15, 2014
23.3K
EMD-DWT based transform domain feature reduction approach for quantitative multi-class classification of breast
Sharmin R Ara1, Syed Khairul Bashar1, Farzana Alam2
1Department of Electrical and Electronic Engineering, Bangladesh University of Engineering and Technology, Dhaka 1000, Bangladesh.
Ultrasonics
|May 13, 2017
Summary
This study introduces a new method to reduce ultrasound features for better breast tumor classification. The new approach improves accuracy in distinguishing benign from malignant tumors and categorizing them by severity.
Area of Science:
- Medical Imaging and Diagnostics
- Biomedical Engineering
- Machine Learning in Healthcare
Background:
- Classifying breast tumors using a large number of ultrasound features can decrease classifier performance.
- Effective feature reduction is crucial for accurate quantitative classification of breast tumors.
Purpose of the Study:
- To propose an effective feature reduction approach in the transform domain for improved multi-class classification of breast tumors.
- To enhance the accuracy of classifying tumors into benign-malignant and BI-RADS categories (≤3, 4, 5).
Main Methods:
- Utilized feature transformation methods like empirical mode decomposition (EMD) and discrete wavelet transform (DWT).
- Applied filter- or wrapper-based subset selection to extract non-redundant transform domain features from bi-modal ultrasound data.
- Classified 201 breast tumors using the reduced feature set and conventional classifiers, validated with histopathology and radiologist BI-RADS scores.
Main Results:
- The transform domain reduced feature set improved sensitivity, specificity, and accuracy by 5.35%, 3.45%, and 3.98% respectively for benign-malignant classification compared to the original feature set.
- Achieved improvements in likelihood of malignancy (3.49%, 9.07%, 3.06%) and inadmissible error probability (4.48%) for BI-RADS categorization (≤3, 4, 5).
- Demonstrated the efficacy of the proposed method through comparative analysis of conventional classifiers.
Conclusions:
- Constructing a transform domain reduced feature set by extracting complementary information from bi-modal features enhances breast tumor classification.
- Integrating qualitative bi-modal BI-RADS scores further improves quantitative classification accuracy.
- The proposed approach can help reduce unnecessary biopsies and minimize the risk of misdiagnosis.
