Applications of Artificial Intelligence and Deep Learning in Early Cancer Detection

Early cancer diagnosis is a rapidly evolving area and a key global health challenge. Artificial intelligence consists of many elements from simple algorithms to complex networks, which can be applied to a wide range of potential problems. When these paths intersect, important opportunities arise. This review discusses the potential implications of this for early cancer diagnosis. The possible inferences made by an artificial intelligence system making a prediction pertinent to a mammographic finding are an important area of research and have so far received relatively little attention.

Early cancer diagnosis is a key national and international focus. In the United Kingdom, the national registry data suggest that cancer stage is closely correlated with 1-year cancer mortality. In 1-year survival is a surrogate for poor outcome, and at diagnosis there are many cancers presenting without cure in this timeframe when detected after symptoms. This underpins the national priority to improve early diagnosis rates to 75% by 2028 outlined in the long-term plan. Internationally, early diagnosis is recognised as a key priority by a number of organisations. It is widely accepted that early cancer can have improved outcomes and the morbidity and cost associated with advanced stage disease may be mitigated. This in turn leads to a number of initiatives to investigate routes to improve early detection. Screening is one of these. Randomised trials for lung, breast, liver, colon and rectal, and oesophagogastric cancers show promising results that are being incorporated into clinical guidelines and practice.

Screened individuals are usually of high risk and follow a pathway for confirmation. However incidence screening, incidence or interval with malignancy outside the screen detected range may provide challenge even for very high-risk populations such as those with a genetic mutation, small cell lung or a sub-set of high foul former smokers with lung. Any programme may have to consider well as which diagnostic pathway to offer those under consideration for screen entry. Patient selection and risk stratification are key challenges for any screening programme with the definition of high-risk individuals on a population level a current problem for most cancers. In the near term, AI algorithms and software solutions may have a role in the analysis to aid in and define what the heuristics are. More broadly, AI has the potential to directly facilitate cancer diagnosis by triggering investigation or referral in screened individuals in response to clinical parameters, in which site is one component. Digital mammography in wide use and forthcoming AI applications which can be imbedded being a prototype system demonstrating this. Other areas exist where the finding is less subjective where AI could be applied more widely, such as radiomics, and the PI-RADS.