British researchers, together with the Dutch Research Institute for Diagnostic Accuracy, investigated the effectiveness of a Korean AI tool for analysing lung scans for the presence of lung cancer. Most scans do not show clearly identifiable tumours at first glance, making the screening process very labour-intensive for radiologists. An AI-driven solution could provide relief here.
There have long been calls for a preventive lung cancer screening programme similar to those we have for breast and colon cancer. However, the amount of work and lack of radiologists, apart from the cost, stand in the way of such programmes. In the UK, the added value of a screening programme has already been demonstrated. There, within the Lung Cancer Screening (UKLS) study, so-called low-dose CT scans are made that lead to lung cancer, specifically in people at high risk, being detected earlier, even before the first symptoms appear.
AI tool detects lung cancer earlier
To ease the workload for radiologists, an AI tool developed by South Korea's Coreline Soft was recently tested. Data from the UKLS trial were used for that study. The AI tool successfully identified scans without significant lung nodules - which represent the majority of cases - even in high-risk individuals. This allows radiologists to focus their expertise on cases requiring further analysis, improving efficiency while maintaining accuracy in detecting lung cancer.
The potential of AI-driven solutions in the field of (cancer) diagnostics has been explored for some time. Also in relation to possibly setting up a (nationwide) population screening. ‘However, with AI we might in the future be able to carry out population screening for lung cancer in an effective and affordable way,’ Mathias Prokop, professor of Radiology at Radboudumc, said back in 2023.
No diagnoses missed
A key finding of the UK study is that all confirmed cases of lung cancer were part of the scans flagged by the AI for further investigation. This ensures that no diagnoses were missed while significantly reducing the number of scans that had to be manually reviewed. The success of the study was made possible by the UKLS study's high quality radiology reporting and long-term follow-up data, which provided a reliable dataset for AI validation.
‘Implementing low-dose CT screening for lung cancer is very useful, but comes with logistical and financial challenges. Our research suggests that AI could play a crucial role in making screening programmes more efficient while maintaining diagnostic confidence,’ said Professor John Field, Professor of Molecular Oncology at the University of Liverpool.
‘This is the first breast AI validation study conducted in a real-world sequential screening programme for lung cancer, with histologically proven lung cancer outcomes and a follow-up of more than 5 years for disease-free survival. Therefore, a milestone for further AI validation in terms of methodology and accuracy with results that can be translated into medical implementation,’ added Professor Matthijs Oudkerk, emeritus professor of radiology at the University of Groningen, Chief Scientific Officer of the Institute for Diagnostic Accuracy.
