Related Experiment Video
Updated: May 22, 2025
09:11
Use of MRI-ultrasound Fusion to Achieve Targeted Prostate Biopsy
Published on: April 9, 2019
21.3K
Mixture of Expert-Based SoftMax-Weighted Box Fusion for Robust Lesion Detection in Ultrasound Imaging
Se-Yeol Rhyou1, Minyung Yu1, Jae-Chern Yoo1
1Department of Electrical and Computer Engineering, College of Information and Communication Engineering, Sungkyunkwan University, Suwon 440-746, Republic of Korea.
Diagnostics (Basel, Switzerland)
|March 13, 2025
Summary
CSM-FusionNet enhances ultrasound (US) imaging for hepatocellular carcinoma (HCC) detection by reducing noise and improving lesion identification. This novel framework significantly boosts diagnostic accuracy and aids clinical decision-making.
Area of Science:
- Medical Imaging
- Artificial Intelligence in Medicine
- Hepatocellular Carcinoma Research
Background:
- Ultrasound (US) imaging is vital for early hepatocellular carcinoma (HCC) detection.
- Challenges in US imaging include speckle noise, low contrast, and varied lesion appearances, impacting diagnostic accuracy.
- Accurate HCC detection is crucial for timely treatment and improved patient outcomes.
Purpose of the Study:
- To introduce CSM-FusionNet, a novel framework designed to overcome limitations in US-based HCC lesion detection.
- To enhance the accuracy and reliability of lesion detection in ultrasound images.
- To improve diagnostic performance for hepatocellular carcinoma using advanced image processing and machine learning.
Main Methods:
- Developed CSM-FusionNet, integrating clustering, SoftMax-weighted Box Fusion (SM-WBF), and padding techniques.
- Applied image preprocessing including intensity adjustment, histogram equalization, and filtering to raw US images.
- Utilized data augmentation and trained 10 YOLOv8 networks, employing mAP@0.5 for SM-WBF weights and DBSCAN for bounding box clustering.
Main Results:
- Achieved a significant accuracy improvement from 82.48% to 97.58%, with sensitivity reaching 100%.
- Lesion detection accuracy increased from 56.11% to 95.56% following the application of clustering and SM-WBF.
- The framework effectively reduced redundant detections and enhanced feature representation for improved classification.
Conclusions:
- CSM-FusionNet shows significant potential to enhance diagnostic reliability in US-based lesion detection.
- The framework aids in precise clinical decision-making for hepatocellular carcinoma cases.
- This approach offers a promising advancement for improving the accuracy of ultrasound diagnostics in oncology.
