AI may help detect deep vein thrombosis, study shows

Machine-learning algorithms could potentially diagnose deep vein thrombosis (DVT) as accurately as radiologists, a study has found.

The finding could cut long waiting lists for treatment, and avoid patients being prescribed drugs for DVT when they don’t have it. DVT is a blood clot most commonly found in the leg, which can lead to fatal pulmonary embolism (PE). In about a third to a half of patients, the condition leads to long-term symptoms and disability.

Many people who experience symptoms, however, are eventually found not to have the condition, leading to anxiety for patients and a heavier workload for NHS doctors. “Currently, many patients do not have a definitive diagnosis within 24 hours of a suspected DVT, and so many patients end up receiving painful injections of what can often be an unnecessary anticoagulant, with potential side-effects,” said Dr Nicola Curry, consultant haematologist at Oxford University Hospitals NHS Foundation Trust and study lead.

Researchers at the University of Oxford, Imperial College and the University of Sheffield collaborated with ThinkSono, a technology company, to train a machine-learning algorithm, AutoDVT, to distinguish patients who had DVT from those who did not. The results, published in the journal Digital Medicine, showed that the AI algorithm accurately diagnosed DVT when compared to the gold standard ultrasound scan. The researchers calculated that using the algorithm could potentially save health services $150 per examination.

“Traditionally, DVT diagnoses need a specialist ultrasound scan performed by a trained radiographer, and we have found that the preliminary data using the AI algorithm coupled to a hand-held ultrasound machine shows promising results,” Dr Curry said.

AI may enable GPs and nurses to diagnose the condition

The research team will now begin a blinded clinical study, comparing the accuracy of AutoDVT with standard care to determine the sensitivity of the algorithm for picking up DVT cases.

“The AI algorithm can not only be trained to analyse ultrasound images to discriminate the presence or absence of a blood clot – it can also direct the user using the ultrasound wand to the right locations along the femoral vein, so that even a non-specialist user can acquire the right images,” said Christopher Deane from the Oxford Haemophilia and Thrombosis Centre, and a member of the research team.

If the study is successful, the technology may enable non-specialist healthcare professionals, such as GPs and nurses, to diagnose and treat DVT quickly. It may also allow non-specialists to collect images and send them to an expert, making it possible to diagnose patients who cannot reach a specialist.