It’s Sunday morning here in Los Angeles. I’m walking down Ventura Boulevard, coffee in hand, and I found myself thinking: where is the review that pulls together all of the strategies for using artificial intelligence to analyze thermometric data in the diabetic foot? We’ve known for decades that temperature monitoring is one of the most powerful — and most underutilized — tools in our preventive arsenal. And we’ve watched AI eat the world in medical imaging. But who has systematically mapped how these two worlds are converging?
My wish was answered this morning by Wartakusumah, Yamada, Noguchi, and Oe out of Kanazawa University and Osaka Metropolitan University.
Their scoping review, just published in Diabetes Research and Clinical Practice, is exactly the paper I was hoping somebody would write. They systematically searched six databases — PubMed, MEDLINE, CINAHL, ScienceDirect, Scopus, and Google Scholar — and identified 60 original research articles and conference proceedings on AI-based foot thermography for diagnosing or monitoring diabetic adults.
Here’s the landscape they mapped. Most of the 60 studies focused on detecting increased temperature patterns — the inflammatory signature we’ve long associated with pre-ulcerative states and the Charcot foot. A smaller group examined decreased temperature patterns, pointing toward the vascular compromise of peripheral arterial disease. And a third cluster tackled severity classification of existing DFUs using thermal imaging data.
The AI performance numbers are striking: accuracy ranged from 61% to 100% across studies. That’s a huge spread, and it tells us something important. The algorithms themselves are maturing rapidly — convolutional neural networks, vision transformers, support vector machines, random forests — but the environments in which they’re being tested remain overwhelmingly controlled. Nearly half (46.7%) of the studies were conducted in controlled laboratory settings. Only 6.6% tested their systems in uncontrolled, real-world conditions. And perhaps most telling, 46.7% didn’t even report the testing environment at all.
This is the gap. This is always the gap. We can build beautiful algorithms in controlled settings, but the diabetic foot lives in the wild. It lives in shoes and socks. It lives in homes without climate control and in clinics with variable ambient temperatures. It lives on feet that have been walking and standing and bearing load in ways that change their thermal signature minute by minute. If we want AI-powered thermography to fulfill its promise as a scalable screening tool — and I believe it can — we need to test it where people actually live.
The clinical applications the authors identified are exactly right: clinical decision support, remote monitoring, and reducing clinician workload. Think about what that means in practice. A patient at home in rural Mississippi or downtown Mumbai points a smartphone thermal camera at their feet. An algorithm flags an asymmetry. A clinician gets an alert. An ulcer is prevented before it ever forms. That is the vision. And the technology is getting close. But close isn’t there yet.
Those of us who have worked in temperature monitoring for a long time — and I’ve been involved with this since our early randomized controlled trials in the 2000s — know that the 2.2°C (4°F) contralateral temperature difference remains a robust clinical signal. What AI brings to the table is the ability to detect subtler patterns, to integrate thermal data with other risk factors, and to do it at scale without requiring a trained clinician to interpret every image. That is transformative. But only if we do the hard work of validating these systems outside the lab.
Wartakusumah and colleagues have given us a clear map of where we are and where we need to go. The algorithms work. The hardware is becoming commodity. Smartphone-attached thermal cameras are already available for under $300. What we need now are large, multicenter, prospective studies in real-world settings — in homes, in community clinics, in low-resource environments — with diverse populations and standardized protocols.
Sunday morning wish: fulfilled. Now let’s get to work on the next chapter.
Wartakusumah R, Yamada A, Noguchi H, Oe M. Analysis of foot thermography images of diabetic patients using artificial intelligence: a scoping review. Diabetes Res Clin Pract. 2025;218:112446. doi: 10.1016/j.diabres.2025.112446
#ActAgainstAmputation #DiabeticFoot #AI #ArtificialIntelligence #MachineLearning #DeepLearning #Thermography #Thermometry #DFU #DiabeticFootUlcer #Prevention #Screening #RemoteMonitoring #Neuropathy #PAD #LimbPreservation #DigitalHealth #mHealth #SmartphoneThermography
Leave a Reply