WoundTrack.
Connected care.
A clearer healing journey.
An AI-assisted chronic wound monitoring platform for home healthcare. Connecting thoughtful documentation, longitudinal insight, and clinical expertise.
Under development · Academic research concept
Research models explore wound categories and boundaries to support review.
01 / Project
What is WoundTrack?
An end-to-end chronic wound monitoring and clinical collaboration platform. The project connects wound images with clinical context and a history that healthcare professionals can review over time.
- Capture
- Standardized wound documentation.
- Analyze
- AI-assisted image analysis under research.
- Track
- Longitudinal monitoring over time.
- Consult
- Clinical collaboration with physicians and specialists.
Intended for patients and relatives, home healthcare nurses, physicians, and specialists.
02 / Our Goal
To make chronic wound monitoring more organized, objective, secure, and accessible.
- Standardized documentation
- AI-supported clinical decision-making
- Longitudinal wound history
- Physician and specialist consultation
AI supports clinical decision-making; it does not replace healthcare professionals.
03 / The Problem
Challenges in Current Wound Monitoring
Chronic wounds may need follow-up over weeks or months. Across visits, useful images and observations can become difficult to retrieve, compare, and share.
Fragmented documentation
Photos may be left on personal phones or shared informally instead of being stored in an organized longitudinal record.
Non-standardized images
Lighting, angle, distance, and zoom can vary between captures, making comparison difficult.
Subjective assessment
Wound observations can depend on individual judgment and may be hard to compare consistently.
Delayed consultation
Relevant visual information may not reach the appropriate specialist in time for review.
Limited longitudinal tracking
Previous images and clinical context can be difficult to retrieve and review together.
Unsuitable sharing channels
Sensitive wound images may pass through informal communication channels without appropriate access controls or traceability.
04 / Proposed workflow
One connected care journey.
From the first image to the next review. Select a step to explore the proposed pathway.
The patient or home care team begins a structured record for professional review.
Capture consistencyA calibration marker could support physical measurements. Image quality and capture conditions need careful evaluation.
Clinical responsibilityAI outputs support review. They do not independently diagnose or prescribe treatment.
05 / AI Research
AI-Assisted Wound Analysis
Current areas under consideration. These research directions are not implemented clinical capabilities or guaranteed final features.
Wound classification
Explore identifying wound categories such as diabetic foot ulcers and pressure injuries.
Stage / grade assessment
Investigate wound-specific frameworks. Pressure injuries and diabetic foot ulcers do not share a single staging system.
Wound segmentation
Candidate models may outline wound boundaries at pixel level. U-Net, DeepLabV3+, and SegFormer are under consideration.
Risk flagging
Explore prioritization cues for physician review; AI would not diagnose infection or decide emergency treatment.
Candidate model families include U-Net, DeepLabV3+, SegFormer, and EfficientNet / ResNet where appropriate. Model selection is still under evaluation.
06 / Privacy by design
Shared intelligence.
Local patient data.
Exploring collaborative medical AI without bringing raw patient images into a central dataset.
- 01 Federated learning
- Each institution trains locally. Model updates contribute to a shared model.
- 02 Privacy-preserving methods
- Research methods aim to reduce information leakage during collaborative training.
- 03 Homomorphic encryption
- Investigate whether selected operations can be performed on encrypted model information.
Patient images stay within the institution in this proposed architecture.
Collaborative model aggregation
Hospital A trains on its own images. Only model updates flow toward the shared model; raw patient images remain local.
Research consideration: federated learning alone does not guarantee privacy. Updates may leak information; privacy protection, accuracy, and computational cost must be evaluated together.
07 / Research agenda
Good research starts
with better questions.
Four questions guide the investigation. Methods and scope will evolve as datasets are evaluated.
Investigation in progressRESEARCH QUESTION
Can federated learning achieve performance comparable to centralized learning for chronic wound image analysis?
Compare federated and centralized training under consistent evaluation conditions.
08 / Research foundations
Better evidence starts with data.
Candidate sources under evaluation. Dataset selection has not been finalized.
| Dataset | Possible use | Status |
|---|---|---|
| DFUC2020 / DFUC2022 | Diabetic foot ulcer datasets being evaluated for detection and segmentation-related research. | Under evaluation |
| Pressure Ulcer Stages | Potential dataset sources for exploring pressure injury stage classification. | Under evaluation |
| Medetec Wound Database | Possible supplementary source for exploratory or educational use; not assumed to be a standardized benchmark. | Under evaluation |
Evaluation criteriaAnnotation quality, image count, segmentation masks, class and stage / grade labels, accessibility, licensing, patient-level metadata, and longitudinal availability.
09 / People
Our Team
Bilkent University · Department of Computer Engineering
CS491 / CS492 Senior Design Project
Student team
- Aybegüm Fatma Çelebi
- Nehir Konaş
- Aslı Kanber
- Ceyda Yıldırım
- Beren Irgat
- Supervisor
- Assist. Prof. Sinem Sav
- Innovation Expert
- Dr. Deniz Katırcıoğlu Öztürk
- Current stage
- Defining scope, evaluating datasets and AI methods, designing the architecture, and preparing project documentation.