[Paper Review] Pressure Ulcer Categorisation using Deep Learning: A Clinical Trial to Evaluate Model Performance

Paul Fergus, Carl Chalmers+3 more|arXiv (Cornell University)|Mar 7, 2022

Pressure Ulcer Prevention and Management4 citations

TL;DR

This study evaluates a deep learning-based mobile system using Faster R-CNN to automatically categorize pressure ulcers (I–IV, DTI, unstageable) from smartphone images in a real-world NHS clinical trial. With 216 validated images, the model achieved a mean F1-score of 0.6786 at a 0.75 confidence threshold, demonstrating promising but not yet clinically sufficient performance for standardised reporting in community care.

ABSTRACT

Pressure ulcers are a challenge for patients and healthcare professionals. In the UK, 700,000 people are affected by pressure ulcers each year. Treating them costs the National Health Service {\\pounds}3.8 million every day. Their etiology is complex and multifactorial. However, evidence has shown a strong link between old age, disease-related sedentary lifestyles and unhealthy eating habits. Pressure ulcers are caused by direct skin contact with a bed or chair without frequent position changes. Urinary and faecal incontinence, diabetes, and injuries that restrict body position and nutrition are also known risk factors. Guidelines and treatments exist but their implementation and success vary across different healthcare settings. This is primarily because healthcare practitioners have a) minimal experience in dealing with pressure ulcers, and b) a general lack of understanding of pressure ulcer treatments. Poorly managed, pressure ulcers lead to severe pain, poor quality of life, and significant healthcare costs. In this paper, we report the findings of a clinical trial conducted by Mersey Care NHS Foundation Trust that evaluated the performance of a faster region-based convolutional neural network and mobile platform that categorised and documented pressure ulcers. The neural network classifies category I, II, III, and IV pressure ulcers, deep tissue injuries, and unstageable pressure ulcers. Photographs of pressure ulcers taken by district nurses are transmitted over 4/5G communications to an inferencing server for classification. Classified images are stored and reviewed to assess the model's predictions and relevance as a tool for clinical decision making and standardised reporting. The results from the study generated a mean average Precision=0.6796, Recall=0.6997, F1-Score=0.6786 with 45 false positives using an @.75 confidence score threshold.

Motivation & Objective

• To address clinical variability and inconsistent categorization of pressure ulcers by healthcare professionals, particularly district nurses in community settings.
• To develop and evaluate an AI-powered mobile system that standardizes pressure ulcer classification using deep learning for real-time clinical decision support.
• To assess the feasibility and performance of a Faster R-CNN model trained on real-world, low-quality clinical images in a primary care environment.
• To identify barriers to model performance, such as image quality and class imbalance, and propose solutions for future improvement.
• To advocate for open sharing of high-quality pressure ulcer images and standardized photography protocols to enhance AI model accuracy.

Proposed method

• A Faster Region-based Convolutional Neural Network (Faster R-CNN) was trained to classify pressure ulcers into six categories: I, II, III, IV, deep tissue injury (DTI), and unstageable.
• District nurses captured pressure ulcer images using mobile phones and transmitted them over 4G/5G to a server at Liverpool John Moores University for inference.
• Images were preprocessed by cropping to focus on the wound, improving model performance by reducing environmental variance.
• The system used transfer learning with a pre-trained backbone, fine-tuned on a dataset of 1,016 images collected over eight months, reduced to 216 after quality filtering.
• Performance was evaluated using standard object detection metrics: mean Average Precision (mAP), mean Recall (mAR), and mean F1-score (mAF1).
• A confidence threshold of 0.75 was applied to balance precision and recall, with false positives analyzed for clinical relevance.

Experimental results

Research questions

• RQ1 Can a deep learning model accurately classify pressure ulcers across multiple stages and types using real-world, low-quality smartphone images in a community healthcare setting?
• RQ2 How does image cropping and preprocessing affect the performance of a Faster R-CNN model in pressure ulcer categorization?
• RQ3 To what extent does the current model’s performance meet clinical standards for use in routine patient care and reporting?
• RQ4 What are the main challenges in acquiring sufficient, high-quality, and balanced training data for pressure ulcer classification models?
• RQ5 How can standardized photography protocols and data sharing improve future model performance and clinical adoption?

Key findings

• The model achieved a mean F1-score of 0.6786, precision of 0.6796, and recall of 0.6997 at a 0.75 confidence threshold, indicating moderate but not yet clinically sufficient performance.
• Cropping the images significantly improved model performance, increasing mAP from 0.3639 to 0.6796, highlighting the importance of focus on the wound.
• The system generated 45 false positives, indicating room for improvement in distinguishing ulcers from similar skin conditions or artifacts.
• Despite training on low-quality internet images, the model produced reasonably good results, suggesting potential for improvement with higher-quality data.
• The study identified a critical need for at least 9,000 high-quality, balanced images (1,500 per class) to achieve performance in the 90th percentile, which was not met in this trial.
• The results underscore the importance of standardized photography and open data sharing to advance AI in pressure ulcer management.