How is AI used to help detect the risk of cancer through analysing medical images?

An AI system is trained on existing data, known as ​‘training data’. For example, training data may be images of already-diagnosed cancerous skin lesions. The system uses patterns in the training data to identify specific features of these images that relate to cancerous features. The trained AI system checks new data by testing it against features identified through the training data – in this example, new pictures of skin lesions. The system flags any images where it identifies cancerous features. 

AI systems can have considerable benefits around cancer detection and prognosis. The potential to diagnose cancer earlier will lead to better outcomes for many patients. Over time, AI could also make image interpretation more reliable and determine outcomes for patients with high precision, in a way that humans alone could not. 

AI can also help predict the risk of some cancers returning, and can analyse genetic data to help identify the cause of DNA mutations that are associated with cancer.

But there are risks involved with the use of AI for cancer detection. Studies have found that gaps in AI training data in healthcare – such as missing data from particular ethnicities and skin-types – can lead to different levels of accuracy for different demographic groups.

Clinicians are also concerned about the effects of uneven geographic access to AI diagnostic tools – known as a ​‘postcode lottery’ – and the risk that this uneven access could exacerbate existing health inequalities.