Location
Innovation Lab Atrium
Start Date
4-3-2025 2:00 PM
End Date
4-3-2025 3:30 PM
Presentation Date
3 April 2025, 2:00pm - 3:30pm
Meeting Name
2025 - Miners Solving for Tomorrow Research Conference
Department(s)
Mechanical and Aerospace Engineering
Document Type
Poster
Document Version
Final Version
File Type
event
Language(s)
English
Rights
© 2025 The Authors, All rights reserved
Included in
Apr 3rd, 2:00 PM
Apr 3rd, 3:30 PM
Image-Based Stage Prediction of Pressure Injuries Using Deep Learning
Innovation Lab Atrium
Comments
Advisor: M. C. (Ming Chuan) Leu
Co-advisor: Fateme Fayyazbakhsh
Abstract:
This study focuses on enhancing the detection and staging of pressure injuries through the application of deep learning, specifically using the YOLOv8 object detection model. A high-quality, publicly available dataset was prepared to evaluate various YOLOv8 variants --YOLOv8n, YOLOv8s, YOLOv8m, YOLOv8l, and YOLOv8x --alongside five optimizers: Adam, AdamW, NAdam, RAdam, and Stochastic Gradient Descent. A simulation-based research approach, informed by CONSORT and STROBE guidelines, was followed for rigorous dataset preparation and model evaluation. Among all configurations, the YOLOv8s model combined with the AdamW optimizer and proper hyperparameter tuning delivered the best results, achieving a mean average precision (mAP) of 86.1% at IoU ≥ 0.5 and a recall of 82.31%. The model showed notable improvements in identifying difficult cases, such as Stage 2 pressure injuries, and achieved high accuracy rates across various stages, including 0.90 for Deep Tissue Injury and 0.91 for Unstageable injuries. An ensemble approach utilizing all YOLOv8 variants also demonstrated robust performance on unseen data, further strengthening the case for its clinical applicability. These findings highlight the promise of YOLOv8-based models in delivering accurate and reliable pressure injury staging, supporting more effective clinical decision-making.