首页|Detection and Diagnosis for Pressure Injury by Using SE-Swin Cascade R-CNN

Detection and Diagnosis for Pressure Injury by Using SE-Swin Cascade R-CNN

扫码查看
原文链接
  • NETL
  • NSTL
  • IEEE
Pressure injuries, a chronic disease with high incidence and costly treatment, are categorized by the National Pressure Injury Advisory Panel from stage 1 to 4 based on skin invasion severity. Variations in injury range and depth can lead to incorrect staging. A computer-aided diagnosis system using a convolutional neural network (CNN) architecture has proven reliable in object detection and classification. This study proposes a cascade R-CNN with a squeeze-and-excitation shifted windows transformer (SE-Swin transformer) model for detecting and classifying pressure injury stages. The system includes data augmentation, feature extraction, and stage classification. Using 883 images, the system achieves a mean average precision (mAP) of 81.3% in detection and accuracy of 87.1%, sensitivity of 85.7%, and positive predictive value (PPV) of 86.6% in stage classification.

InjuriesFeature extractionTransformersConvolutional neural networksData augmentationCostsComputational modelingHospitalsComputer architectureAccuracyPressure measurementMedical diagnosis

Yao-Sian Huang、Chiao-Min Chen、Yi-Sin Liu、Shou-Chuan Sun、Mei-Chu Chen、Shu-Chen Chang、Ruey-Feng Chang

展开 >

Department of Computer Science and Information Engineering, National Changhua University of Education, Changhua, Taiwan

Department of Mathematics, National Changhua University of Education, Changhua, Taiwan

National Taiwan University, Taipei, Taiwan

Nursing Department, Changhua Christian Hospital, Changhua, Taiwan

Erlin Christian Hospital, Changhua Christian Hospital, Changhua, Taiwan

National Taiwan University and MOST Joint Research Center for AI Technology and All Vista Healthcare, Taipei, Taiwan|Artificial Intelligence Development Center, Changhua Christian Hospital, Changhua, Taiwan

展开 >

2025

10.1109/MMUL.2024.3524427

IEEE multimedia

IEEE multimedia

ISSN：
年,卷(期)：2025.32(1)
  • 20