An algorithm developed by UK tech start-up company behold.ai has been found to competently identify potential cancers in the lungs. The work was presented at the British Thoracic Oncology Group (BTOG) annual conference in Dublin on 29 January 2020.
Working with a team at the University Hospitals of Leicester NHS Trust, the software was run on the chest X-rays of 1,513 patients who had a GP direct referral for a chest X-ray over a two-week period in June and July 2019. The algorithm works by delivering an ‘instant triage’ of each X-ray within seconds.
Although all 1,513 examinations were analysed using the algorithm, the team was specifically interested in cases that were referred by a radiologist or reporting radiographer for a CT scan directly after X-ray, to confirm or rule out cancer.
The hospital radiology team suspected cancer in 39 out of the 1,513 X-rays and referred those 39 patients for a CT scan of the chest. Of the 39 suspected cases, 11 were eventually confirmed histologically to be cancer (through a tissue sample tested in the laboratory).
The algorithm correctly identified the presence and location of 10 out of the 11 histologically confirmed cancers. Therefore had the technology been used to triage cancer patients it could potentially save time to diagnosis on CT scans in most of these patients.
“The AI project is in its early days, but the initial outcomes are very promising. The next steps would be to run clinical trials to create a sound evidence base to demonstrate that it is both safe and beneficial in clinical practice,” said Dr Indrajeet Das, a consultant radiologist at University Hospitals of Leicester NHS Trust, who worked on the project.
The National Optimal Lung Cancer Pathway (NOLCP) recommends that all routine or urgent chest X-rays referred by the GP for suspected lung cancer should be reported within 24 hours. According to the latest Radiology Review by the Care Quality Commission, NHS hospitals are faced with increasing numbers of chest X-rays and scans and some trusts have reported backlogs in reporting the results(1). An algorithm which could identify the abnormal radiographs rapidly would quickly enable medical staff to prioritise the most urgent cases first.
“We are working with a number of hospitals in the UK, USA and India,” said Dr Thomas Naunton-Morgan, behold.ai’s Medical Director. “Radiologists, like those in the team we worked with in Leicester are keen to report the abnormal scans first, shortening the turnaround time for the patients that need the most urgent care. Therefore the potential for our algorithm to support this care pathway is a very exciting prospect.”