A wound app can draw a beautiful border around an ulcer. The harder question is what that border means for the person attached to it.
Our Mancunian colleagues have given us plenty to think about in Mind the Gap: A survey of chronic wound datasets, methods, and clinical translation, newly available online in Medical Image Analysis. Anum Fatima, Moi Hoon Yap and Connah Kendrick at Manchester Metropolitan University join Neil D. Reeves at Lancaster University and Michael Odling-Smee and David Bond at Aire Innovate to examine the distance between promising wound-image algorithms and practical clinical tools.
Twenty datasets. Fifty methods. Thirteen apps.
The survey analyses 20 chronic wound datasets—14 public or partially public and six private—alongside 50 deep learning methods and 13 mobile applications, encompassing commercial tools and research prototypes. Its scope includes diabetic foot, venous, arterial and pressure ulcers.
The central finding: the data used to develop and benchmark algorithms do not consistently match the capabilities that wound applications report. Gaps include standardised image acquisition, calibration for physical measurement, depth information, detailed tissue labels and representation of different skin tones.
Applications may report tissue classification, calibrated measurement or depth estimation, while the available public datasets provide insufficient support for developing and evaluating those capabilities. That mismatch deserves attention.
A good score needs a clinical destination
Our clinical reading is that the next question must go beyond whether an algorithm performs well on a familiar collection of photographs. Can a team trust its measurements across visits, cameras, care settings and patients? Can those measurements help someone recognise a concerning change and act in time?
For those of us working at the intersection of consumer electronics and medical devices, this is a useful reminder: putting an algorithm in a phone is one step. Building a dependable pathway from the photograph to an appropriate clinical response is the larger task.
Equity belongs in the dataset
The authors’ attention to skin-tone representation is particularly welcome. Our implication for developers and clinical partners is straightforward: ask who is represented in the development data, who appears in independent testing, and whether performance is reported across those groups. A reassuring overall score cannot answer every question about the patient in front of us.
What we would ask before bringing a tool into care
This review prompts several practical questions for a clinical demonstration or research collaboration: How is scale established? What happens when lighting or image quality changes? Has the tool been tested on patients and settings beyond its development dataset? How does an uncertain result reach a clinician? And what evidence connects its use to better care?
These are our implementation questions, rather than outcomes established by this survey. The paper is a review of datasets, methods and applications; it does not demonstrate that a particular app improves healing or prevents amputation. Nor does a shortage of public data establish that an individual commercial product lacks private validation.
From the Grand Challenge to the Foot Selfie
This paper connects directly to work we have shared with our Manchester colleagues and partners over the years. The common thread is making a useful photograph part of a useful care pathway.
Build the benchmark. Our coauthored Diabetic Foot Ulcer Grand Challenge 2021 evaluation tackled image classification and the shortage of public datasets. The later Diabetic Foot Ulcers Grand Challenge segmentation report extended that conversation to identifying wound boundaries. These challenges give researchers a shared testing ground; the new survey asks what else is needed for clinical translation.
Connect the image to a person. Our original Foot Selfie pilot explored a simple home system for taking and transmitting plantar photographs for clinical review. It demonstrated feasibility in a small study, while leaving the question of reduced morbidity for further research. We also highlighted the Manchester team’s early Foot Selfie-plus-AI work with Naser Ahmed, Moi Hoon Yap and Oracle.
Build with the people using it. Our more recent Foot Selfie co-creation work in Peru brought patients, caregivers and clinicians into the design process. Usable image capture, understandable messages and a practical route to care are part of the same translation problem.
For us, these are connected pieces of the same effort: representative data, careful evaluation, usable tools and a clinical team able to respond. The benchmark, the foot selfie and the person’s next clinical decision all need to meet.
Congratulations to our Manchester colleagues and the entire team for connecting the computing and clinical conversations. The destination we should keep in view is more hospital-free, activity-rich, dignity-filled days. Better wound images should help us get there.
Read the paper
Fatima A, Yap MH, Reeves ND, Odling-Smee M, Bond D, Kendrick C. Mind the Gap: A survey of chronic wound datasets, methods, and clinical translation. Medical Image Analysis. Published online October 7, 2026;115:104344. doi:10.1016/j.media.2026.104344. PubMed.
This commentary draws on the indexed abstract and publisher’s article summary.
Leave a Reply