Related Experiment Video
Updated: May 31, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
CDCG-UNet: Chaotic Optimization Assisted Brain Tumor Segmentation Based on Dilated Channel Gate Attention U-Net Model
K Bhagyalaxmi1, B Dwarakanath2
1Department of Computer Science and Engineering, Faculty of Engineering and Technology, SRM Institute of Science and Technology, Ramapuram, Chennai, 600089, India. bk8019@srmist.edu.in.
Neuroinformatics
|January 22, 2025
Summary
A novel Chaotic Dilated Channel Gate attention U-Net (CDCG-UNet) model improves brain tumor segmentation accuracy in MRI scans. This advanced technique enhances early diagnosis by effectively identifying tumor regions with high precision.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Brain tumors represent a significant cause of cancer-related mortality in both children and adults.
- Late diagnosis and high costs of detection devices are major challenges in brain tumor identification.
- Existing machine learning (ML) approaches often suffer from low accuracy, high loss, and substantial computational costs.
Purpose of the Study:
- To propose a novel U-Net model for accurate brain tumor segmentation in magnetic resonance images (MRI).
- To address the limitations of existing methods by enhancing accuracy and reducing computational complexity.
- To effectively segment brain tumor subregions, including whole tumor (WT), tumor core (TC), and enhancing tumor (ET).
Main Methods:
- Utilized a Probabilistic Hybrid Wiener filter (PHWF) for pre-processing MRI scans to reduce noise and improve image quality.
- Employed a 3D Convolutional Vision Transformer (3D-VT) for efficient feature extraction, reducing model complexity.
- Implemented a Chaotic Dilated Channel Gate attention U-Net (CDCG-UNet) model, optimized with the Chaotic Harris Shrinking Spiral optimization algorithm (CHSOA), for tumor segmentation.
Main Results:
- The CDCG-UNet model achieved high dice scores across multiple datasets: BRATS 2021 (0.972 for ET, 0.987 for CT, 0.98 for WT), BRATS 2020 (98.87% for ET, 98.67% for CT, 99.1% for WT), and BRATS 2023 (98.42% for ET, 98.08% for CT, 99.3% for WT).
- Demonstrated superior performance in segmenting various brain tumor components compared to existing methods.
- The pre-processing and feature extraction steps effectively enhanced the segmentation accuracy.
Conclusions:
- The proposed CDCG-UNet model offers a highly accurate and efficient solution for brain tumor segmentation in MRI.
- This approach has the potential to improve early diagnosis and treatment planning for brain tumor patients.
- The integration of advanced filtering, feature extraction, and optimized deep learning architectures shows significant promise in medical image analysis.
