Using artificial intelligence to detect diabetic foot ulcers early
Diabetic foot ulcers can begin with a small blister, crack, pressure mark or area of discolouration. Reduced sensation may mean a person does not feel an injury, while poor circulation and elevated blood glucose can slow healing. Early recognition gives clinicians more time to reduce pressure, treat infection and prevent a wound from becoming limb-threatening.
Artificial intelligence (AI) is being developed to support this process by analysing foot photographs, identifying changes over time and helping care teams prioritise assessments. In Australia, the value of these tools will depend on how well they fit everyday clinical practice, protect personal information and reach people in metropolitan, regional and remote communities.
Why early detection matters
A diabetic foot ulcer can progress quickly when pressure continues during walking. A person may keep working, commuting or caring for family without realising that a wound is worsening. Common risk factors include peripheral neuropathy, vascular disease, previous ulceration, foot deformity and footwear that rubs against vulnerable areas.
Regular inspection is therefore central to prevention. A patient or carer may photograph the sole, heel and sides of each foot at home, while a nurse, podiatrist or general practitioner checks circulation, sensation, skin integrity and signs of infection. AI-assisted screening can help flag a concerning image, but it should support—not replace—a hands-on assessment.
How image-based screening works
A smartphone or clinic camera can capture images of the foot under reasonably consistent lighting. Machine-learning software may assess redness, swelling, skin breakdown, callus, colour changes or wound dimensions. Some systems compare current images with earlier records to detect subtle changes that may be difficult to notice during a busy appointment.
The most useful tools are designed around clinical decisions rather than impressive technical performance alone. A high-risk alert should lead to a clear pathway, such as same-day review, podiatry referral or vascular assessment. The system should also show its limitations, identify poor-quality images and allow clinicians to override an alert when the clinical picture differs from the algorithm’s interpretation.
Turning a promising tool into routine care
Introducing AI requires more than purchasing software. Health services need agreed protocols for taking images, documenting consent, responding to alerts, escalating suspected infection and recording outcomes. Staff must know who reviews flagged images, how quickly they respond and what happens when a patient cannot upload a usable photograph.
This is where implementation science can help research findings become dependable clinical practice. A Brisbane hospital, community health service or general practice may need to adapt the workflow to staffing levels, existing electronic records, podiatry capacity and local referral arrangements.
Evaluation should measure patient outcomes as well as algorithm accuracy. Useful indicators include time from alert to review, healing rates, hospital admissions, avoidable amputations, false alarms and whether people continue using the system. Feedback from patients, carers, Aboriginal health workers, nurses and podiatrists can reveal practical barriers that a laboratory trial may miss.
Reaching people across Australian communities
Access will look different in Brisbane, Cairns, regional Queensland and remote communities. A metropolitan clinic may have podiatrists and vascular specialists nearby, while a patient in a rural town may need a telehealth appointment or outreach visit. Store-and-forward images could help clinicians review wounds between scheduled appointments, provided there is a reliable escalation pathway for urgent cases.
Australia’s everyday habits also matter. Walking barefoot at home, wearing thongs in summer, working outdoors, swimming at local beaches or using poorly fitting work boots can expose feet to injury. Hot weather may increase sweating and skin irritation, while long travel distances can delay treatment. Education should focus on practical foot protection, daily checks and prompt reporting of changes rather than relying on technology alone.
Culturally safe care is essential. Aboriginal and Torres Strait Islander communities experience significant health inequities, and digital screening must be developed with community-controlled health organisations rather than imposed as a purely technical solution. Language, trust, internet access, device ownership and the role of family or carers should all be considered in the design.
Protecting privacy and clinical safety
Foot images are health information, especially when linked with a name, medical record or location. Australian providers need processes consistent with the Privacy Act 1988 and the Australian Privacy Principles, as well as relevant Queensland Health policies and organisational governance requirements. Patients should understand what is collected, why it is needed, where it is stored, who can access it and whether it will be used to improve the algorithm.
Services should also consider the regulatory status of an AI product. Software intended to support diagnosis or clinical decisions may fall within the Therapeutic Goods Administration’s regulatory framework for medical devices. Procurement teams should ask about validation in Australian populations, cybersecurity, data hosting, model updates, audit logs and arrangements if the supplier ceases operation.
Remote care can strengthen follow-up when it is connected to clinical support. For example, telehealth support demonstrates how virtual services can be integrated into broader care pathways. A similar principle applies to diabetic foot monitoring: digital contact should complement local examination, not create a substitute for urgent in-person treatment.
Practical steps for health services
A cautious implementation can begin with a limited pilot involving clinicians, patients and carers. The service can compare AI alerts with expert assessment, test image quality in real conditions and identify whether the tool improves time to treatment. Australian market considerations are important because licensing costs, device compatibility, broadband coverage and integration with platforms such as My Health Record may affect long-term sustainability.
Clear communication also protects trust. Patients should be told that an algorithm may miss a problem or generate a false alert, and they should know which symptoms require immediate care. Increasing pain, spreading redness, warmth, swelling, discharge, fever, blackened tissue or a rapidly changing wound warrants prompt clinical attention.
- Establish a daily or regular foot-check routine for people at increased risk.
- Use standardised photography guidance, including lighting, distance and image angles.
- Create a documented pathway for reviewing alerts and escalating urgent findings.
- Validate performance across different skin tones, ages, devices and Australian settings.
- Review privacy, cybersecurity, consent and TGA-related obligations before procurement.
AI can make early warning more visible, particularly when it connects a patient’s home observations with a responsive clinical team. Its success should be judged by fewer severe infections, faster access to treatment, better self-management and reduced preventable hospitalisation—not by the novelty of the software.
Health services, researchers, technology developers and community organisations can work together to test safe, equitable diabetic foot screening in real Australian settings. Building that evidence through partnerships will help ensure that an image captured at home or in a local clinic leads to timely, human-centred care.
