Radiologist uses AI to better diagnose lung condition

05 November 2018 - Consultant Simon Walsh has developed an algorithm which can accurately diagnose idiopathic pulmonary fibrosis

Simon Walsh

A radiologist at King’s College Hospital has developed an algorithm that could dramatically improve the diagnosis rate of a lung condition which kills up to 6,000 people in the UK each year.

Fibrotic lung disease causes scarring and stiffness in the lungs, which makes it difficult to breathe. Idiopathic pulmonary fibrosis (IPF) is the most common and deadly fibrotic lung disease.

More than 5,000 new cases of IPF are diagnosed in the UK each year and this figure is increasing by 5% annually. About 30,000 people in the UK are currently living with IPF.

Patients are diagnosed with a CT scan of their lungs, which needs to be interpreted by a radiologist. However, consultant radiologist Dr Simon Walsh says accurately interpreting the CT can be difficult and usually needs to be carried out by specialists with extensive experience of IPF.

“Making an early accurate diagnosis is crucial so patients can be given prompt, appropriate treatment,” said Dr Walsh. “However, many hospitals do not have radiologists who specialise in this condition, and even in the hands of experts, CT interpretation can be subjective. Half of all patients with this condition are misdiagnosed at least once.”

Patients with IPF are offered a drug which on average slows the progression of the disease by about 20%. Drugs used to treat other fibrotic lung diseases can be harmful in patients with IPF, so misdiagnosis has significant consequences for patients with IPF.

To improve diagnosis rates, particularly for patients who do not live close to hospitals with specialist radiologists in this field, Dr Walsh applied artificial intelligence technology, known as deep learning to develop a computer algorithm that analyses patients’ CT scans. The result is a system that can interpret CT scans in patients suspected of having IPF as accurately as an expert chest radiologist.

He said: “The algorithm uses deep learning software to interpret the CT scans. It’s a diagnostic tool that can be used to assist in the diagnosis of patients - it will not replace the work being done by radiologists. I developed this algorithm because it will allow for more accurate diagnoses of IPF for all patients, including those living in rural areas, where it may be difficult to access specialist services.”

Dr Walsh tested the algorithm against 150 cases of fibrotic lung disease and the algorithm provided the same level of performance as 91 world leading specialist radiologists. The results of this research have now been published in The Lancet Respiratory Medicine and were presented as a late-breaking abstract at the European Respiratory Society Congress in Paris this month. The next stage of testing will involve using the algorithm in a working clinical environment to see how it impacts patient care in practice.

Dr Walsh added: “The benefits of this software are that it is quick, cheap and accessible and can be used alongside current hospital software as a diagnostic support tool. In principle, it will mean better outcomes for patients with this condition, which can be incredibly difficult to diagnose. If the next stage of testing is successful, it could be used within the NHS within a few years.”


For further information please contact:
Mary McConnell
Senior Communications Officer
mary.mcconnell2@nhs.net
Extension: +44 (0)20 3299 4139