Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Early Explorations of Lightweight Models for Wound Segmentation on Mobile Devices

Vanessa Borst, Timo Dittus, Konstantin Müller, Samuel Kounev

TL;DR

This work addresses the challenge of delivering objective wound assessment from smartphone images by evaluating lightweight segmentation architectures for mobile deployment. It systematically compares ENet, UNet, UNeXt, and TopFormer on a merged CFU dataset derived from FuSeg and DFUC 2022, exploring both scratch training and pretraining on large datasets, with a real-world Flutter app demonstration. Key findings show that lightweight models, especially TopFormer, achieve competitive segmentation metrics (e.g., $IoU$ and $DSC$ in the high70s to low80s with pretraining) and can operate on consumer devices, though edge contours could still be improved. The study demonstrates the practicality of on-device wound segmentation for home-based monitoring and outlines future work to broaden architecture coverage and dataset size, enhancing robustness and clinical relevance.

Abstract

The aging population poses numerous challenges to healthcare, including the increase in chronic wounds in the elderly. The current approach to wound assessment by therapists based on photographic documentation is subjective, highlighting the need for computer-aided wound recognition from smartphone photos. This offers objective and convenient therapy monitoring, while being accessible to patients from their home at any time. However, despite research in mobile image segmentation, there is a lack of focus on mobile wound segmentation. To address this gap, we conduct initial research on three lightweight architectures to investigate their suitability for smartphone-based wound segmentation. Using public datasets and UNet as a baseline, our results are promising, with both ENet and TopFormer, as well as the larger UNeXt variant, showing comparable performance to UNet. Furthermore, we deploy the models into a smartphone app for visual assessment of live segmentation, where results demonstrate the effectiveness of TopFormer in distinguishing wounds from wound-coloured objects. While our study highlights the potential of transformer models for mobile wound segmentation, future work should aim to further improve the mask contours.

Early Explorations of Lightweight Models for Wound Segmentation on Mobile Devices

TL;DR

This work addresses the challenge of delivering objective wound assessment from smartphone images by evaluating lightweight segmentation architectures for mobile deployment. It systematically compares ENet, UNet, UNeXt, and TopFormer on a merged CFU dataset derived from FuSeg and DFUC 2022, exploring both scratch training and pretraining on large datasets, with a real-world Flutter app demonstration. Key findings show that lightweight models, especially TopFormer, achieve competitive segmentation metrics (e.g., and in the high70s to low80s with pretraining) and can operate on consumer devices, though edge contours could still be improved. The study demonstrates the practicality of on-device wound segmentation for home-based monitoring and outlines future work to broaden architecture coverage and dataset size, enhancing robustness and clinical relevance.

Abstract

The aging population poses numerous challenges to healthcare, including the increase in chronic wounds in the elderly. The current approach to wound assessment by therapists based on photographic documentation is subjective, highlighting the need for computer-aided wound recognition from smartphone photos. This offers objective and convenient therapy monitoring, while being accessible to patients from their home at any time. However, despite research in mobile image segmentation, there is a lack of focus on mobile wound segmentation. To address this gap, we conduct initial research on three lightweight architectures to investigate their suitability for smartphone-based wound segmentation. Using public datasets and UNet as a baseline, our results are promising, with both ENet and TopFormer, as well as the larger UNeXt variant, showing comparable performance to UNet. Furthermore, we deploy the models into a smartphone app for visual assessment of live segmentation, where results demonstrate the effectiveness of TopFormer in distinguishing wounds from wound-coloured objects. While our study highlights the potential of transformer models for mobile wound segmentation, future work should aim to further improve the mask contours.
Paper Structure (15 sections, 2 figures, 2 tables)

This paper contains 15 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Approach
  3. Task Definition and Requirements
  4. Model Selection
  5. Experimental Setup
  6. Dataset
  7. Evaluation Protocol
  8. Implementation Details
  9. Evaluation
  10. Results
  11. Real-World Deployment
  12. Conclusion and Future Work
  15. Appendix

Figures (2)

  • Figure 1: Live segmentation of different model variants in our mobile app, arranged in descending order of parameter count.
  • Figure 2: Examples of identical or highly similar image pairs from both datasets.