Related Experiment Video
Updated: Sep 27, 2025
13:44
Detection of Architectural Distortion in Prior Mammograms via Analysis of Oriented Patterns
Published on: August 30, 2013
43.1K
A novel wavelet decomposition and transformation convolutional neural network with data augmentation for breast
Olaide N Oyelade1,2, Absalom E Ezugwu3
1School of Mathematics, Statistics, and Computer Science, University of KwaZulu-Natal, King Edward Avenue, Pietermaritzburg Campus, Pietermaritzburg, 3201, KwaZulu-Natal, South Africa.
Scientific Reports
|April 9, 2022
Summary
This study introduces a novel wavelet-CNN-wavelet architecture for enhanced breast cancer detection in digital mammography. The method improves classification accuracy by effectively identifying and enhancing discriminative features using wavelet transforms.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Computer Vision
Background:
- Deep learning (DL) models, particularly convolutional neural networks (CNNs), show promise in breast cancer detection from digital mammograms.
- Existing methods often lack mechanisms to effectively discriminate and enhance relevant features for improved classification.
- Wavelet transform has been used for feature extraction, but not to restructure CNN architectures for enhanced feature detection.
Purpose of the Study:
- To propose a novel wavelet-CNN-wavelet architecture for improved breast cancer detection in digital mammography.
- To address limitations in current DL approaches by introducing a mechanism for feature enhancement and discrimination.
- To investigate the impact of restructuring CNNs with wavelet transforms on detecting discriminant features.
Main Methods:
- A hybrid approach combining seam carving and wavelet decomposition for image preprocessing to identify discriminative features.
- Development of a CNN-wavelet structure incorporating a new wavelet transformation function for feature extraction and map reduction.
- Utilized generative adversarial networks (GANs) to synthesize image samples, addressing potential training dataset insufficiency.
Main Results:
- The proposed wavelet-CNN-wavelet architecture demonstrated improved classification accuracy in detecting breast cancer abnormalities.
- The method resulted in lower loss function values, indicating enhanced model performance.
- Wavelet transform proved effective in restructuring CNN architectures for better feature detection in digital mammography.
Conclusions:
- The wavelet-CNN-wavelet architecture offers a significant advancement in breast cancer detection using digital mammography.
- Restructuring CNNs with wavelet transforms enhances the detection of subtle abnormalities indicative of breast cancer.
- This approach provides a more complete solution for feature enhancement and discrimination in medical image analysis.
