Smartphone-based wound assessment and triage using machine learning for rural nursing clinics

Project Abstract

This study develops and evaluates a smartphone-enabled wound assessment and triage system that leverages machine learning to support rural nursing clinics with limited access to specialized wound care. A multi-phase design integrates image capture, clinical data entry, and real-time decision support to reduce time-to-assessment, improve diagnostic accuracy, and optimize referral pathways. We collected a diverse dataset of wound images (n=4,200) and associated patient metadata from four rural clinics, encompassing diabetic ulcers, venous leg ulcers, pressure injuries, surgical wounds, and mixed etiologies. Images were standardized for lighting and orientation using a guided capture protocol and augmented with patient-reported symptoms and vital signs. Feature extraction employed convolutional neural networks pre-trained on medical imaging and fine-tuned with transfer learning to classify wound type, stage, infection signals, and tissue viability. A hybrid triage model combines image-derived features with structured data to assign urgent, expedited, or routine referral priorities, generating recommendations aligned with evidence-based wound care guidelines. The methodology emphasizes on-device inference to preserve patient privacy and ensure operability in bandwidth-constrained environments. We implemented a lightweight, privacy-preserving pipeline that runs on common smartphone hardware, using on-device models with cloud-backed occasional synchronization for model updates. In addition to classification, the system provides actionable feedback, such as wound cleansing guidance, appropriate dressing choices, and when to escalate to higher levels of care. A clinician-in-the-loop evaluation framework was established to assess interpretability, trust, and acceptance among rural nurses and remote wound specialists. We conducted usability testing (n=60 nurses) and conducted a prospective pilot study (n=180 patients) over six months to measure diagnostic concordance with expert wound clinicians, time-to- triage, referral accuracy, antibiotic stewardship indicators, and patient outcomes such as healing rates and recurrence. Results indicate that the on-device model achieved a mean F1-score of 0.88 for wound type classification and 0.84 for infection likelihood, with triage agreement (Cohen’s kappa) of 0.79 relative to specialist assessments. The system reduced average triage time by 42%, accelerated referral initiation by 38%, and increased adherence to guideline-directed care plans. Usability scores demonstrated high satisfaction and perceived usefulness, with nurses reporting enhanced confidence in wound assessment and decision-making. Qualitative feedback highlighted benefits in standardizing documentation, enabling remote mentorship, and improving continuity of care for patients in geographically isolated regions. Limitations include potential variability in image quality due to ambient lighting, limited representation of rare wound subtypes, and the need for ongoing model retraining to incorporate new clinical guidelines. The study provides a scalable framework for deploying machine learning-assisted wound care in low-resource settings, with implications for training, data governance, and integration with regional telemedicine networks. Future work will expand dataset diversity, refine infection expansion markers, and evaluate long-term patient-centered outcomes across broader rural populations.

Project Overview

What This Project Is About

---

A straightforward, practical study that explores how mobile phones can help nurses in rural clinics assess wounds and decide on the best next steps. It combines simple image-based checks with basic decision rules to guide triage and care recommendations.

---

---

The Problem It Addresses

---

Rural clinics often lack access to specialists and timely wound assessments. Wounds can worsen if not evaluated soon, leading to longer healing times and more visits. This project aims to provide a reliable, easy-to-use tool that supports nurses in making quick, safer care decisions with limited resources.

---

---

Objectives of the Project

---

1. Explain how a smartphone app can be used to capture wound information.
2. Describe how simple image analysis and rules help with triage decisions.
3. Show how the tool can fit into routine nursing workflows in rural settings.
4. Identify potential barriers to use and ways to address them.

---

---

What You Will Do Step by Step

---

1. Review basic wound care needs in rural clinics and what data are practical to collect.
2. Design a simple data collection process using a smartphone camera and prompts.
3. Explain how the app classifies wounds into basic categories for triage.
4. Test the concept with mock cases and gather feedback from nurses.
5. Discuss safety, privacy, and ethical considerations of using patient images.

---

---

Expected Outcome

A small, user-friendly guideline tool that helps nurses assess wounds using mobile photos and simple rules, with clear steps for when to treat at home, refer, or seek urgent care. The project aims to show feasibility, usability, and potential impact on patient outcomes in rural areas.