Automated Vetting and Protocolling of MRI Examinations

Background Vetting and assigning of protocols to MRI scan requests are usually carried out by radiologists and radiographers. This way is both time-consuming and prone to error. Recent improvements in Artificial Intelligence (Al) like Large Language Models (LLMs) are better at processing human language. It could be used to improve tasks like image reporting and assigning protocols in radiology. Approach Firstly, I will find out about the delays in the way this is currently done. I will use a survey to find out the perceptions of radiologists to both the current process and the proposed Al intervention. I will then use an LLM to undertake referral and protocol tasks. I will compare its performance to the performance of clinical specialists and established guidelines. I will present the explanation reports to radiologists and ask them if they can be understood and whether they can be trusted as clinically reliable. Impact My research could help reduce the workload and time radiologists spend dealing with scan requests. The findings will help design future research to consider application in real clinical settings.

Inclusion Criteria: * Patients who have had MRI examinations in the study sites * Patients with MRI referrals submitted through electronic referral systems * Patients whose MRI referrals were vetted and protocolled prior to examination Exclusion Criteria: * Patients who have had other imaging examination outside the MRI * Patients with paper referrals scanned in electronic referral systems * Patients for MRI examinations without vetting and protocolling