Can artificial intelligence improve the safety and effectiveness of chest x-rays to identify lung cancers?

A2002395

Lung cancer is the leading cause of cancer mortality in Yorkshire. It is a key driver of health inequalities with the highest incidence linked to older age and social deprivation.  Each year, around 150,000 chest x-rays are performed in Leeds, with specialists reviewing and providing written interpretations for each one. Notably, seven out of ten lung cancers are initially identified through these chest x-rays. However, up to one in four lung cancers present may go undetected by chest x-rays, with nearly half of these missed cases being visible in retrospect.

Artificial intelligence (AI) programs hold the potential to match or even surpass human performance in interpreting chest x-ray images, thereby enhancing the safety of the chest x-ray reporting process. However, the evidence regarding the accuracy of available products is of varying quality. The studies are conducted in groups not representative of the population of Leeds, and do not allow a comparison of products. 

This project, a collaboration with the School of Computing at the University of Leeds, aims to develop a "benchmarking study" to conduct independent evaluations of AI programs. The study will utilise approximately 70,000 chest x-rays from previous research in Leeds. The benchmarking study will assess how well AI performs compared to humans and other AI programs and will also identify any biases within AI programs that may discriminate based on age, gender, smoking status, or race.

The goal is to use the benchmarking study as part of an ongoing program to test AI algorithms, benefiting both Leeds and the wider NHS.

Lead Researcher	Dr Bobby Bhartia
Co-Researchers	Dr Nishant Ravikumar Dr Steven Bradley
Host Organisation/CSU	Radiology Leeds Teaching Hospitals NHS Trust
Grant Amount	£10,249.95
Start Date	12/01/2023
Estimated Duration	32 months
Status	In Progress
Impact Areas	Innovation and Health Technologies Health Inequalities – Cancer Care
Tags/key notes	Cancer and neoplasms