The role of artificial intelligence in improving the diagnosis of prostate cancer

Research question: What is role of artificial intelligence (AI) in identifying prostate cancer on MRI scans? Background: In the past five years, MRI of the prostate +/- targeted biopsy has been recommended as the gold standard for prostate cancer diagnosis. However, there is high variability in performance of radiologists for detecting cancer on MRI. Further, many patients are not even getting an MRI. A barrier to adoption includes the availability of expert radiologists with a 40% shortfall projected by 2027. AI has performed similarly to expert radiologists on retrospective MRI datasets but there are no prospective studies evaluating AI in prostate MRI cancer detection. A formal prioritisation exercise by patients highlighted that this is a priority area for research. This research aims to: Assess the performance of AI alone and in addition to radiologists in detecting prostate cancer on MRI in an actual patient pathway, identifying barriers to routine NHS implementation Define key standards for AI solutions in prostate cancer detection Evaluate the cost-effectiveness of AI Create an image databank available to other researchers for development of future AI solutions Methods: WP1-PARADIGM-trial: This is a within-patient multicentre 500-person non-inferiority diagnostic yield clinical trial. Men with suspected prostate cancer will undergo a prostate MRI reported separately by AI and a radiologist. Targeted biopsies of suspicious lesions on either report will be carried out. This will determine how good AI is compared to and in addition to radiologists in detecting cancer. Post trial, MRIs will be combined with images from our other MRI clinical trials conducted over the last decade to create a dataset of more than 3000 high quality images which we will make available to other researchers for development of future AI. WP2-NHS-perspective cost-effectiveness analysis determining cost per diagnosis of significant cancer detected and insignificant cancer avoided comparing AI alone and in addition to radiologist. Inputs from WP1 and WP3 will update the decision model. WP3-Semi-structured interviews will be carried out with patients, clinicians, AI developers and policymakers. Views will be explored on acceptability of an AI alone pathway, how clinicians should best interact with AI and barriers to implementation. WP4-A target product profile for AI in prostate cancer detection will be carried out. This will define the key standards required for AI in prostate cancer detection to be safely used. This involves performing a systematic review followed by interviews in WP3 to identify the key characteristics followed by Delphi consensus meetings to agree on standards. Timelines: Year 1-3: WP1-PARADIGM-trial. Years 2-4 WP3. Years 1-5: WP2 & WP4. The implication of this work is to support the radiology workflow thus allowing every man who needs a prostate MRI to have a rapid high-quality scan result, no matter their background, geography or access to particular hospitals. This will increase cancer detection in a cost-effective manner and increase global adoption of prostate MRI. Dissemination of the research findings will be co-delivered with the PPI advisory group via social media channels, national prostate cancer charities, international conferences and peer-reviewed publications.