A Scottish research team has devised a novel way of detecting mesothelioma from among the hundreds of lung scans radiologists read every day: train a computer to do it for them.
A pilot program that marries diagnostics with cutting-edge data science has trained an AI program to spot the signs of mesothelioma as it processes computed tomography (CT) scans of patients, according to an April 13 release.
The prototype system is intended to speed the diagnosis of mesothelioma cases in Scotland, which has the world’s highest incidence of the disease due to the historical use of asbestos in construction and shipbuilding, the release states.
Training Days
The research team — which includes staff from Canon Medical Research, the University of Glasgow, Queen Elizabeth University Hospital, and the regional National Health System — trained the AI to spot mesothelioma tumors by showing it more than 100 CT scans with the cancer cells marked.
After that training process, the AI “was able to find and measure the tumor extremely accurately, without any human input,” the team stated in the release. No objective information was disclosed.
Measuring is Key
The team focused on malignant pleural mesothelioma due to its prominence in Scotland and because of the difficulty in accurately measuring it via CT scan. Properly measuring tumors is key to determining the best course of treatment. But in mesothelioma’s case, this process is hindered by its tendency to grow in a ring-like pattern around the surface of the lung, as compared to other lung cancers, which form a ball-like shape, the release noted.
Thus, the AI system can benefit mesothelioma patients in two ways: speeding their diagnosis and streamlining measurements. The latter could make clinical trials of new drugs less expensive, less time-consuming, and more accurate, the release states.
Canon’s principal scientist, Keith Goatman, said that the “speed and accuracy of the AI algorithm could have a wide-reaching impact on mesothelioma treatment. Accurate tumor volume measurements are much too time-consuming to perform by hand. Automating these measurements will open the way for clinical trials of new treatments, by detecting even small changes in the tumor size. Ultimately, it could be used routinely in hospitals to decide the best treatment for each individual.”
The prototype system is already undergoing further testing. The research team intends to publish their findings in a scientific journal, though the release did not specify which publication or when.
