Related Experiment Video
Updated: May 28, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.6K
A novel approach to skin disease segmentation using a visual selective state spatial model with integrated spatial
Yu Bai1,2, Hai Zhou1,2, Hongjie Zhu1,2
1College of Electronic and Information, Southwest Minzu University, Chengdu, 610225, China.
Scientific Reports
|February 9, 2025
Summary
We developed SSR-UNet, an efficient deep learning model for skin lesion segmentation. It achieves high accuracy in identifying skin lesions, improving diagnostic capabilities for dermatologists.
Area of Science:
- Medical Imaging
- Computer Vision
- Artificial Intelligence
Background:
- Accurate skin lesion segmentation is vital for clinical diagnosis and treatment.
- Automated segmentation aids dermatologists in early detection and monitoring of skin diseases.
- Existing methods face challenges with feature extraction and computational complexity.
Purpose of the Study:
- Introduce a novel U-shaped segmentation architecture, SSR-UNet.
- Improve efficiency and accuracy in skin lesion segmentation.
- Address limitations of Convolutional Neural Networks (CNNs) and Transformers.
Main Methods:
- Developed SSR-UNet utilizing a Residual Space State Block and bidirectional scanning.
- Implemented a spatially-constrained loss function to enhance gradient stability.
- Evaluated the model on ISIC2017 and ISIC2018 skin lesion segmentation benchmarks.
Main Results:
- SSR-UNet achieved high accuracy on ISIC2017 (Mean Intersection Over Union: 80.98%, Classification Accuracy: 96.50%, Specificity: 98.04%).
- On ISIC2018, SSR-UNet demonstrated strong performance (Mean Intersection Over Union: 82.17%, Dice Coefficient: 90.21%, Classification Accuracy: 95.34%, Sensitivity: 88.49%).
- Outperformed existing models in key segmentation metrics.
Conclusions:
- SSR-UNet offers an efficient and accurate solution for skin lesion segmentation.
- The model effectively balances computational load and feature extraction capabilities.
- Demonstrates excellent performance for dermatological image analysis.
