We now know that asbestos is a killer, but in the last century it was used widely in industry, valued for its heat and fire-resistant properties. It could be found in the structure of all kinds of buildings and power stations as insulation on ships and even in the manufacture of all manner of household appliances. So, while buildings went up, ships were built, and industry boomed, no one realised that microscopic asbestos fibres were being inhaled into workers’ lungs.
Subsequently, since the 1970s, the western world saw a steady increase in the amount of asbestos-related disease, the most devastating of which has to be Malignant Pleural Mesothelioma (MPM), commonly known as ‘asbestos cancer’. While rare, when compared to other types of cancer, the prognosis for MPM is considerably bleaker, as it is incredibly difficult to diagnose in the early stages and there are no effective treatments.
MPM doesn’t grow like other cancers, presenting a number of challenges for radiologists. Unlike most types of tumour, which are roughly spherical in shape, MPM ‘fills’ the cavity between the lungs and diaphragm, called the ‘pleura’, which contains the fluids that allow the lungs to expand and contract without friction. Eventually the tumour starts to wrap itself around the lung, causing breathlessness, chest pain, persistent coughing and weight loss, among other symptoms. From this point, even with treatment, life expectancy is short.
But in Edinburgh, Scotland, something quite remarkable is happening. The lack of an effective early diagnosis from imaging and the challenge of staging and managing this most difficult of cancers, formed the basis of a proposal from Canon Medical Research Europe (CMRE) to the Scottish Cancer Innovation Challenge in mid-2018. Having previously investigated the rationale, clinical need, and theoretical feasibility of a method to automatically identify MPM tumours and their boundaries in CT images (a process known as ‘image segmentation’), CMRE and their partners, NHS Greater Glasgow and Clyde, were subsequently awarded a second phase of funding worth €180,000 towards development of a prototype algorithm which combines Artificial Intelligence (AI) and imaging technology that would put this theory into action – a project that could pave the way to saving time, money and lives in the fight against cancer.
Dr Sandy Weir, Technical Manager at Canon Medical Research’s Centre of Excellence in AI and his team of Data Scientists, led by Dr Keith Goatman, have been working with NHS Greater Glasgow and Clyde, Dr Kevin Blyth, Consultant Respiratory Physician & Professor at the University of Glasgow to develop a deep learning algorithm – or ‘convolutional neural network’ – to create an automatic RECIST measurement for MPM tumours that rapidly and accurately segments the tumour in chest CT. This knowledge is essential in treating patients with the disease.
The unusual non-spherical presentation of MPM means that automatically segmenting MPM tumour volumes is extremely complicated and challenging, even for the most experienced of radiologists. Sandy explains that the successful algorithm would automatically “segment and identify the tumour in the lung cavity, ideally at the very early stages” and is already seeing positive results. “When we compare results of automatic segmentation of large tumours against manual segmentation by an expert, we can see that our results compare very favourably. So, we know that our algorithm is showing extremely good promise. The next stage is an evaluation period where we’ll look at the effect of our algorithm on data from a small patient cohort where we have imaging before and after chemotherapy. This will allow us to compare the volumetric change and assess the correlation of our algorithm with expert mRECIST measurements.”
The WHO estimates that about 125m people in the world are currently exposed to asbestos at the workplace
MPM is now thankfully in decline in Europe. However, there are mesothelioma hotspots all over the world and it’s unfortunately on the increase in developing countries. The World Health Organisation estimates that about 125 million people in the world are currently exposed to asbestos at the workplace. But crucially, the AI prototype being developed by Sandy and his team has the potential to be transferrable to other types of tumour, “By starting with the most challenging diagnosis of mesothelioma, we hope to be able to use the techniques we develop to improve the performance and accuracy for other tumours which present in less challenging ways” he explains, so the knock-on effect of this research could reach far beyond just MPM.
The project team also hopes that an AI-based assessment tool could have a positive impact on the cost of cancer drugs, as clinical trials may become more efficient to stage more accurately using AI tools to determine whether new drugs are having a useful effect. It’s still early days, but this prototype could play a vital part in the future of cancer diagnosis and is certainly contributing to the growing body of evidence supporting the use of AI in global medical advancements.