Artificial intelligence (AI) can be used to support clinicians with reviewing chest x-ray (CXR) images from patients with suspected lung cancer following referral from primary care. The intended use of AI in the clinical pathway is to read and flag higher risk CXR images so that clinicians can prioritise patients for urgent computerised tomography (CT).
The evidence for the use of AI supported clinical review of CXRs for patients with suspected lung cancer is emerging. No published evidence on the clinical effectiveness, cost-effectiveness, or safety of the AI use case was identified. No studies were identified that captured patient or staff views on the use of AI in this setting.
Interim analysis (n=41 reaching diagnosis stage, n=27 reaching treatment stage) from an ongoing service evaluation in NHS Grampian shows that use of AI alongside an adjusted clinical pathway shows promise in reducing time from CXR report to CT, reducing time to treatment, and increasing the identification of patients with treatable lung cancers.
AI does not have a universally agreed definition.1 A type of AI, known as machine learning, uses algorithms (that is, a sets of rules) to automatically learn from data, and find patterns or relationships within the data. An advanced form of AI, often referred to as deep learning, is a sub-type of machine learning that attempts to simulate the neural structures of the human brain to learn from larger volumes of different types of information than in machine learning only.
The use of AI to support clinical review of CXR in patients with suspected lung cancer is being explored in two NHSScotland health boards.The AI technologies being used are annalise.ai (Annalise AI, Sydney, Australia) and qXR (Qure.ai, Mumbai, India). Other AI tools with similar capabilities exist and fourteen similar technologies were reviewed in a recent health technology evaluation by the National Institute for Health and Care Excellence (NICE). NHS Grampian is using the Annalise Enterprise CXR AI module (annalise.ai) in a Scottish Government funded service evaluation in adults over 18 years old. The module was trained on more than 520,000 CXR studies, including over 820,000 individual de-identified CXR images. Using a machine learning model, annalise.ai scans for 124 potential issues in each CXR image, 34 of which are deemed priority findings. If the AI picks up any issues with a CXR image, it acts as an automated triage system and highlights that the patient should be urgently reviewed by a clinician.
In NHS Greater Glasgow and Clyde (GGC), a pragmatic mixed-methods research study, funded by the manufacturer and Scottish Government, uses deep learning AI technology to highlight abnormalities on CXR images. The qXR (Qure.ai) module was trained on 4.4 million CXR images worldwide, validated on a set of more than 93,000 CXR images and has a processing time per CXR of under 20 seconds. The qXR module analyses a CXR image and highlights whether the scan should be reviewed by a clinician.
The use of AI to support clinician review of CXRs is the first application of the technology in a clinical setting in Scotland. By assisting clinicians with prioritising referrals for urgent CT scans, the use of AI may speed up time to diagnosis and lead to earlier treatment and improved health outcomes for patients with suspected lung cancer.
