LesionIQ skin cancer diagnostic

A photograph is taken of a skin lesion using an interface designed to be as user-friendly as possible.
The system is a demonstration of TinyML – machine learning running on lower-powered, non-connected devices

With this in mind, Tess, a PhD candidate in the School of Mathematical and Computer Sciences at Heriot-Watt University in Edinburgh, has created an early detection system. She’s developed a set of AI tools designed to diagnose skin cancer and other skin conditions using a Raspberry Pi 3 Model B computer attached to a small camera. The aim has been to produce a low-cost device that can be used by patients living in remote areas of the world. It allows skin conditions to be monitored from home without the need for an internet connection.

Making a diagnosis

Tess’s system uses machine learning to analyse images of skin lesions. “Machine learning has overcome most of the challenges faced in traditional methods of skin lesion classification by analysing many images at once, more accurately than human dermatologists,” she says. “Machine learning can then identify patterns that the human eye cannot and can therefore be more accurate and objective.”

Tess has been demonstrating the system at Heriot-Watt’s advanced health and care technologies suite

To assess a skin complaint, patients equipped with one of Tess’s devices would be asked to take a photograph of the affected area. Since the skin lesion image datasets are stored on the Raspberry Pi computer, her program is able to analyse the photograph in real-time before comparing it to thousands of images stored in the dataset. The idea is that the information would be shared with a GP and a decision made over potential treatment. “Current legislation [in the UK] requires a medical professional to validate the outputted diagnosis,” Tess says.

Global impact

Tess envisages LesionIQ being used in rural Scotland and she is in talks with NHS Scotland to gain ethical approval. But it could also become a vital tool for remote communities across the world, particularly given Raspberry Pi devices are so widely used and affordable. “I am often asked why I chose not to deploy my AI model on a smartphone, and this is because smartphones are costly and not widely available/used in developing countries where access to the Internet is also limited,” she explains.

Tess has been working with academics at London South Bank University, Edinburgh Napier University, and the Foundation for Research and Technology – Hellas in Greece

The device certainly looks promising. Currently, it’s proving to be 85% accurate in diagnosing skin cancer and Tess says this can be improved. “I am working with the small amount of publicly available skin lesion datasets available and I am working to create and access new datasets soon which are larger and more diverse,” she reveals. This is proving to be the greatest challenge, however.

“There is a lack of diverse skin lesion data available and the landscape of clinical AI is still in its infancy,” Tess adds. “My future plans are to create/source more diverse data and conduct a study to test this device in a real-world setting.” She hopes the device will be well on the path to being used by real-life patients before 2030.

Find more Raspberry Pi projects in Raspberry Pi Official Magazine

This article appears in issue 169 of Raspberry Pi Official Magazine, which you can access online. You can also subscribe to the print version of our magazine. Not only do we deliver worldwide, but those who sign up to the six- or twelve-month print subscription will receive a FREE Raspberry Pi Pico 2 W!

You can find Raspberry Pi Official Magazine on Facebook, X, Threads, LinkedIn, and Mastodon. You can also contact the team via email: magazine@raspberrypi.com

The post LesionIQ skin cancer diagnostic appeared first on Raspberry Pi.