Related Experiment Video
Updated: Oct 25, 2025
05:41
Author Spotlight: Integrating Ultrasound Imaging with Biochemical Markers for Thyroid Disease Diagnosis
Published on: February 9, 2024
812
Dilated Semantic Segmentation for Breast Ultrasonic Lesion Detection Using Parallel Feature Fusion
Rizwana Irfan1, Abdulwahab Ali Almazroi1, Hafiz Tayyab Rauf2
1Department of Information Technology, College of Computing and Information Technology at Khulais, University of Jeddah, Jeddah 21959, Saudi Arabia.
Diagnostics (Basel, Switzerland)
|August 7, 2021
Summary
Early breast cancer detection is crucial for reducing mortality. This study introduces a novel Di-CNN and DenseNet201 fusion method for ultrasonic image analysis, achieving 98.9% accuracy in breast cancer diagnosis.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Oncology
- Biomedical Engineering
Background:
- Breast cancer poses a significant global health challenge, with rising mortality rates in developing nations.
- Early detection of breast cancer is paramount for improving patient outcomes and reducing death rates.
- Ultrasonic imaging offers a cost-effective and sensitive modality for breast lesion diagnosis.
Purpose of the Study:
- To develop and evaluate an automated system for segmenting and classifying breast lesions from ultrasonic images.
- To enhance the accuracy of breast cancer diagnosis through the fusion of deep learning features.
- To investigate the efficacy of a combined Dilated Semantic Segmentation Network (Di-CNN) and DenseNet201 approach.
Main Methods:
- Ultrasonic breast lesion images were segmented using a Dilated Semantic Segmentation Network (Di-CNN) integrated with morphological erosion.
- Feature extraction was performed using the DenseNet201 deep neural network with transfer learning.
- A novel 24-layer CNN was proposed for transfer learning-based feature extraction, validated for target intensity.
- Feature vectors from DenseNet201 and the 24-layer CNN were fused in parallel for nodule classification using a Support Vector Machine (SVM).
Main Results:
- The Support Vector Machine (SVM) classifier achieved 90.11% accuracy with CNN-activated features and 98.45% with DenseNet201-activated features.
- The fused feature vector approach using SVM demonstrated superior performance, reaching an accuracy of 98.9%.
- The proposed algorithm significantly outperformed recent methods in breast cancer diagnosis rates.
Conclusions:
- The fusion of DenseNet201 and a 24-layer CNN with SVM offers a highly accurate method for breast cancer diagnosis from ultrasonic images.
- The developed Di-CNN based segmentation combined with advanced feature fusion significantly improves diagnostic performance.
- This approach holds promise for enhancing early breast cancer detection and improving patient survival rates, particularly in resource-limited settings.
