AI model found to accurately spot lung cancer better than other methods

‘AI model can potentially save lives through early intervention in the future’

An artificial intelligence tool has been developed by a team of researchers to identify lung cancer more accurately than currently available methods.

The team of scientists, including those from the UK’s Royal Marsden NHS foundation trust, used an AI analysis of around 800 CT scans to look out for signs of cancer that are not easily found by the human eye.

Data from about 500 patients with large lung nodules was used to develop the AI algorithm, said the new study, that was published recently in the journal eBioMedicine and first reported by The Guardian.

The AI algorithm performed more effectively than current AI tools, said scientists, who said the advance can lead to better diagnosis of the malignant condition.

After feeding it the analysis of CT scans, the AI system was tested to determine if it can accurately spot and flag cancerous nodules.

Scientists used a unit of measure called area under the curve (AUC) to predict how efficient the model was at finding signs of cancer.

A unit of 1 AUC indicated the model was perfect, while 0.5 was as good as randomly guessing.

The AI model was found to predict the risk of cancer from analysing scans with an AUC of 0.87, scientists noted, adding that the new tool “could potentially save lives through early intervention in the future”.

“The model accurately segments and classifies large lung nodules, and may improve upon existing clinical models,” researchers wrote in the study.

Cancer is one of the leading causes of mortality across the world with nearly one in six deaths due to the condition, according to the World Health Organisation.

The most common cancers as of 2020 were breast and lung cancers – over 2.2 million cases of each type were reported that year.

While diagnosing lung cancers, nodules in the lung are seen as potential signs of the disease, with large ones of over 15mm having the highest risk of malignancy.

Clinicians have previously found that discovering incidental lung nodules on CT scans is common with most being benign, but some may represent early-stage cancers, providing an opportunity for early lung cancer diagnosis.

Researchers have found that in some cases flagging a high-risk nodule as benign may lead to delayed cancer diagnosis, while on the other hand, over analysing low-risk nodules in patients could expose patients to undue medical complications.

Scientists are hopeful that, with further evaluation, the new AI tool can be applied to clinician decision making with respect to large lung nodules in the future.