Related Experiment Video
Updated: Oct 3, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
3.0K
Deep learning approach based on superpixel segmentation assisted labeling for automatic pressure ulcer diagnosis.
Che Wei Chang1,2, Mesakh Christian3, Dun Hao Chang2,4
1Graduate Institute of Biomedical Electronics & Bioinformatics, National Taiwan University, Taipei, Taiwan.
Plos One
|February 17, 2022
Summary
Deep learning models can automatically diagnose pressure ulcers using superpixel-assisted image labeling. This approach improves wound healing assessment and monitoring, aiding clinical decision-making.
Area of Science:
- Medical Imaging Analysis
- Computational Pathology
- Artificial Intelligence in Healthcare
Background:
- Pressure ulcers pose significant challenges in diagnosis and management, exacerbated by limited access to care.
- Traditional machine learning (ML) methods for automatic diagnosis require extensive feature engineering, limiting clinical applicability.
- Deep learning (DL) offers automated feature extraction but necessitates large, expertly labeled datasets, which are difficult to obtain for pressure ulcers.
Purpose of the Study:
- To develop and evaluate a superpixel-assisted, region-based image labeling method for pressure ulcer tissue classification.
- To create datasets for wound and re-epithelialization segmentation to train deep learning models.
- To assess the performance of five popular deep learning models for pressure ulcer image analysis.
Main Methods:
- A superpixel-assisted, region-based method was employed for image labeling to facilitate tissue classification.
- Boundary-based techniques were used to generate datasets for wound and re-epithelialization segmentation.
- Five deep learning models (U-Net, DeeplabV3, PsPNet, FPN, Mask R-CNN) with a ResNet-101 encoder were trained on the prepared datasets (2836 images for classification, 2893 for segmentation).
Main Results:
- All five deep learning models demonstrated satisfactory performance in both tissue classification and wound/re-epithelialization segmentation tasks.
- The DeeplabV3 model achieved the highest performance, with precision, recall, and accuracy exceeding 0.99 for tissue classification and 0.98 for segmentation.
- The algorithm successfully integrated segmentation results with clinical data for wound healing detection, progress monitoring, size estimation, and debridement need assessment.
Conclusions:
- The proposed superpixel-assisted labeling method effectively supports the creation of datasets for deep learning-based pressure ulcer analysis.
- Deep learning models, particularly DeeplabV3, show high accuracy in classifying pressure ulcer tissues and segmenting wound areas.
- This automated approach holds significant potential for improving the clinical management of pressure ulcers through objective wound assessment and monitoring.
