I like where this is heading
AI Skin Cancer Diagnosis with Edge Computing Tools
Insights | 11-09-2025 | By
Robin Mitchell
The project to develop the technology was led by PhD researcher Tess Watt (Heriot-Watt University/PA).
Key Things to Know:
- AI in medicine shows promise for early disease detection, but clinical integration still faces ethical, technical, and operational barriers.
- Edge AI tools, such as low-cost diagnostic systems, reduce reliance on cloud infrastructure and protect patient privacy.
- Skin cancer diagnosis is a leading use case for embedded AI, highlighting the importance of early intervention in improving survival rates.
- Clinical validation, regulatory approval, and diverse training datasets remain essential for advancing trustworthy AI healthcare solutions.
Artificial intelligence is already proving valuable in a range of medical applications, from improving diagnostic imaging to supporting early detection of disease. However, the clinical integration of AI is still in its early stages. While development continues to accelerate, significant technical, ethical, and operational barriers remain.
What are the core challenges preventing widespread adoption? Why is medical data so difficult to access, and how do privacy concerns shape system design? More importantly, how close are we to real-world AI tools that can assist — not replace — medical professionals?
The Challenges of AI in Medicine
I've said it before, and I'll keep saying it: AI needs to enter medicine and fast.
Why? Because the potential benefits are staggering, if not obscene. From spotting anomalies in radiology scans to predicting genetic disorders before symptoms emerge, AI brings accuracy, speed, and the holy grail of healthcare, early detection. And we know that early detection is the difference between manageable treatment and life-altering disease. AI could take that to a whole new level.
But like most things in tech, it's not that simple.
For all its promise,
AI in medicine faces serious roadblocks. The first and most obvious one is data. AI thrives on data, thousands or even millions of examples to learn from. However, medical data is hard to find, as it is locked behind layers of privacy legislation, ethical standards, and institutional red tape. And rightly so, as many would not want their MRI scans showing up in a training set without consent.
Then comes the problem of testing. It's one thing to develop a promising diagnostic model in a lab; it's a very different thing to run it in a hospital where real patients are on the line. There's no sandbox mode when you're dealing with human lives, making real-world validation ethically challenging, legally risky, and painfully slow.
And for all the benefits that AI presents, it will make mistakes. Especially in its early stages, even a 1% error rate can have huge consequences when you're dealing with diagnoses, prescriptions, or surgical recommendations. The stakes are high, and medicine is not an industry where "move fast and break things" is tolerated.
Then there's the human factor: resistance. Many medical professionals are sceptical of AI technology, with some fearing it will be replaced, while others simply don't trust black-box systems when making life-or-death decisions.
Combining data scarcity, ethical complexity, operational risk, and cultural resistance to
medical AI, it's no surprise that AI hasn't taken over hospitals yet.
Edge AI and Skin Cancer: A Glimpse at the Future of Remote Diagnosis
If there was ever a compelling case for embedded AI in healthcare, then the development of a new AI system capable of running in remote regions of the world, identifying skin cancer early on, would certainly be it.
At Heriot-Watt University in Scotland, PhD researcher Tess Watt has developed a low-cost, offline
diagnostic tool that uses artificial intelligence to identify potential skin cancers. Running on a Raspberry Pi requires no cloud backend, no internet connection, and no waiting for lab results. This development is an excellent example of AI at the edge, and it could potentially change the way rural communities access healthcare.
Why Early Detection with Edge AI Matters
The study highlights that skin cancer remains one of the most common cancers worldwide, with early intervention being critical for survival rates. By embedding AI directly into portable systems, researchers can bypass many of the infrastructure limitations that traditionally delay diagnosis. As the MDPI research notes, such solutions are not only affordable but also scalable, meaning they could serve as an accessible first-line diagnostic option in low-resource environments.
To use the device, a patient uses a basic camera module connected to a Raspberry Pi to take a photo of a skin lesion. The device then runs local inference using an image classification model trained on thousands of known dermatological cases. In seconds, it outputs a preliminary diagnosis. That result can be passed to a local healthcare provider, if one is available, for further review and treatment planning.
Importantly, the diagnostic model tested by Watt and colleagues was trained using publicly available dermatological image datasets, ensuring reproducibility and transparency in its development. According to the MDPI paper, leveraging open datasets is vital for increasing trust in AI medical tools, particularly when addressing regulatory requirements and ethical concerns around patient data use.
In this design, there is no upload, no data leakage, and no latency.
Privacy Advantages of Offline Diagnostics
Data privacy remains one of the central barriers to medical AI adoption. By running entirely offline, edge devices mitigate risks associated with patient confidentiality breaches. The MDPI research stresses that reducing reliance on external servers is particularly important in maintaining compliance with data protection standards, an issue that continues to challenge cloud-based healthcare platforms.
Now, the device is not perfect, with Watts' team reporting about 85% diagnostic accuracy, but this is still impressive for an early-stage, edge-deployed model. This device is the sort of prototype that signals what's truly possible with medical AI and edge devices. With improved datasets, better onboard models, and regulatory clearance, a system like this could provide life-saving diagnoses where dermatologists are miles or even continents away.
Improving Accuracy and Addressing Bias
The paper also identifies that while current models reach around 85% diagnostic accuracy, further improvements depend on the diversity of training data. Skin tones, lesion types, and imaging conditions all influence model robustness. Addressing these gaps will be essential to ensure that AI-assisted tools serve global populations equitably rather than reinforcing existing biases in healthcare systems.
Of course, there's still a long road ahead. Clinical validation, regulatory approval, and scaling this for actual deployment take time. But as Watt points out, the key is starting now, because the longer we wait, the more people go diagnosed.
As the authors of the MDPI study underline, clinical validation will ultimately determine whether these devices can move beyond prototype stage. Integrating AI-driven diagnostics into established healthcare pathways requires collaboration between technologists, clinicians, and regulators. If these partnerships succeed, edge AI could become a cornerstone of community-level screening programmes, offering earlier intervention and reducing the burden on overstretched hospital systems.
Are Edge AI Devices the Future of Medical Care?
Ask anyone working in AI and healthcare what the biggest barrier to adoption is, and chances are they'll give you one word: privacy.
Right now, most medical AI tools rely on the cloud. That means sensitive patient data (images, vitals, records) is uploaded, stored, and analysed on remote servers, sometimes halfway across the world. Understandably, this makes patients and healthcare providers nervous. It opens the door to data breaches, surveillance concerns, and regulatory headaches that slow everything down.
But what if the AI didn't need the cloud?
That's the promise of Edge AI in medicine;
diagnostic tools that run locally, on-site, without any internet connection required. No upload or third-party processing, just instant analysis, done on a small, inexpensive device sitting in a clinic, an ambulance, or even a patient's home.
However, it's not just about privacy either, it's also about access and equality. A device that doesn't rely on cloud infrastructure doesn't require broadband. It can work just as well in a remote Scottish island as it can in a city hospital, and this drastically lowers the barrier to care, especially in under-served regions where seeing a specialist is either cost-prohibitive or logistically impossible.
Furthermore,
healthcare systems worldwide are overwhelmed. If we can take even 10% of early-stage diagnoses and push them to affordable AI tools running at the edge, we free up real doctors to focus on the cases that need them most.
Now, edge AI medical devices are still in their infancy. The Raspberry Pi-based cancer
detection prototype is an impressive start, but it's a long way from being an off-the-shelf product ready for clinical deployment. Accuracy, robustness, and regulatory approval are all hurdles that must be cleared.
AI devices are undeniably the future of medical care, not in some abstract, hand-wavy way, but in a boots-on-the-ground, "let's get this in the hands of patients" kind of way.
We're not there yet. But we're pointed in the right direction.
AI Skin Cancer Diagnosis Tools
www.electropages.com
www.linkedin.com
️ with IP announcement's
% confident brainchip will succeed even though it has taken years longer than I realised.
