Researchers at Kingston University are exploring how artificial intelligence could be trained to detect the early signs of oral cancer using a mobile phone app. Sarah Barman, PhD, and Paolo Remagnino, PhD, have secured £146,000 for the two-year Medical Research Council-funded project, which will see them work alongside experts from the University of Malaya and Cancer Research Malaysia.
The researchers will use a technology called deep learning, a form of machine learning that mimics the way the brain makes connections between pieces of information without being specifically programmed through artificial neural networks. If successful, the researchers say, it could help save both time and money in triaging patients, with potentially far-reaching benefits for the National Health Service in the long term.
“We’re basically training a system to detect abnormalities in the mouth that could be the early indications of oral cancer. Our challenge is to develop deep learning models that demonstrate a high accuracy and prediction of disease,” said Barman, a professor of computer vision with the Department of Computer Science at the university’s School of Computer Science and Mathematics.
“If we find this approach is reliable enough, artificial intelligence could be used for other forms of disease screening with a wide range of possible applications in the field of medical diagnostics. The idea that you could take an image on a mobile phone, then use artificial intelligence to quickly determine whether that patient needs referring or not, is really exciting,” said Barman.
Ensuring early diagnosis is a particular challenge in rural Asia, including parts of Malaysia, due to a lack of easy access to healthcare and specialist treatment. To combat this, Cancer Research Malaysia has developed a phone app called MeMoSA (Mobile Mouth Screening Anywhere).
The app can capture images of the oral cavity that can be interpreted remotely but is still reliant on the availability of oral medicine and surgical specialists to view the image and decide which are referable, which can be expensive and time consuming. The hope is that artificial intelligence can speed up this process.
“With the possibility of increasing survival rates for oral cancer patients, the incorporation of artificial intelligence within MeMoSA holds a lot of promise in ensuring that the efforts we make in early detection continue to break barriers across regions, particularly in south and southeast Asia where the disease is most prevalent,” said Sok Ching Cheong, PhD, lead scientist of the Head and Neck Cancer Team at Cancer Research Malaysia.
The Kingston University team will be training the system with thousands of images of normal mouths and examples of different categories of abnormality provided by experts from University Malaysia and Cancer Research Malaysia. They then will examine the accuracy of the system’s identification of referable and non-referable cases to see how closely it matches an expert clinician’s opinion.
“In Malaysia, they have already done studies with their app that shows you can take a photo of your mouth and send to an expert to judge if it’s likely to be malignant or referable,” said Remagnino, a professor with the Department of Computer Science.
“However, there is always a time delay when sending these images to a specialist remotely, as well as a reliance on their availability, which is where a system like this that can automatically triage patients could make a real difference,” said Remagnino.
As part of the project, the team also will be using a further type of artificial intelligence called a generative adversarial network (GAN) that can create realistic images based on what it has learned from images of people or animals through generating original works of art.
“We will be looking to see whether we can use the GAN to create unique images of mouths, both normal and with abnormalities, to help provide additional data for training the artificial intelligence system,” said Remagnino.