Prostate MRI Analysis by Radiologists and Artificial Intelligence - Disease Identification and Guided Management

Aim: To assess whether artificial intelligence is non-inferior to radiologists in the diagnosis of clinically significant prostate cancer on MRI.

Objectives

Primary

1\. To compare the proportion of men who have clinically significant prostate cancer detected on MRI using AI ± targeted biopsy with radiologists ± targeted biopsy.

Secondary

1. To compare the proportion of men who have clinically insignificant prostate cancer detected on MRI using AI ± targeted biopsy with radiologists ± targeted biopsy.
2. To compare the proportion of men with non-suspicious MRIs for AI vs radiologists.
3. To compare the proportion of men with indeterminately scored MRI as reported by AI vs radiologists.
4. To compare the diagnostic test performance of AI vs radiologist.
5. To compare the additive value of AI when used together with a radiologist interpretation (summative of all identified lesions) compared to a radiologist alone.
6. To compare the additive value of AI when used together with a radiologist interpretation (where the radiologist can interact with the AI system by accepting or rejecting AI-identified lesions) compared to a radiologist alone.
7. To determine the frequency of AI failures.
8. To compare treatment eligibility decisions between AI and Radiologist.
9. To compare the cost-effectiveness of unblinded AI interpreted by the radiologist compared to radiologist alone for prostate cancer detection, and AI alone vs. radiologist alone, and a 3-arm analysis considering all three.

Design:

Prospective, international, within-patient, multi-centre, level-1 evidence trial in participants referred to hospital with a clinical suspicion of prostate cancer.

• AI (Lucida Pi) interpretation (Diagnostic test): AI algorithm that will interpretate the prostate MRI
• Radiologist interpretation (Diagnostic test): Radiologist will interpret the prostate MRI (as per standard of care)

Primary outcomes

• Proportion of men with clinically significant cancer [When biopsy results available, at an expected average of 30 days post-biopsy]: Proportion of men with clinically significant cancer detected (any pattern 4 disease on any core (i.e. Gleason Grade ≥ 3+4/Gleason grade group ≥2).

Secondary outcomes

• Proportion of men with clinically insignificant cancer [When biopsy results available, at an expected average of 30 days post-biopsy]: Proportion of men with clinically insignificant cancer detected (Gleason grade 3+3/Gleason grade group 1).
• Proportion of men with non-suspicious MRIs [When MRI results available, at an expected average of 30 days post-MRI]: Proportion of men with non-suspicious MRIs for AI vs Radiologists
• Proportion of MRIs with indeterminate scores. [When MRI results available, at an expected average of 30 days post-MRI]: Proportion of men with indeterminately scored MRI as reported by AI vs radiologists
• Agreement between AI and Radiologist in score of suspicion [When MRI results available, at an expected average of 30 days post-MRI]: Compare the proportion of MRIs with concordant scores between AI and Radiologist in score of suspicion
• Diagnostic test performance characteristics (AI versus Radiologist) [When biopsy results available, at an expected average of 30 days post-biopsy]: Test performance characteristics for AI and Radiologists, including sensitivity, specificity, area under the receive operating characteristic curve, positive predictive value and negative predictive value.
• Diagnostic test performance characteristics (AI plus Radiologist) [When biopsy results available, at an expected average of 30 days post-biopsy]: Test performance characteristics of AI in combination with Radiologist (summative of all identified lesions) compared to a radiologist alone, including sensitivity, specificity, area under the receive operating characteristic curve, positive predictive value and negative predictive value.
• Diagnostic test performance characteristics (AI-assisted Radiologist) [When biopsy results available, at an expected average of 30 days post-biopsy]: Test performance characteristics of AI in combination with Radiologist (where the radiologist can interact with the AI system by accepting or rejecting AI-identified lesions) compared to a radiologist alone, including sensitivity, specificity, area under the receive operating characteristic curve, positive predictive value and negative predictive value.
• Significant cancer detected by peri-lesional biopsies [When biopsy results available, at an expected average of 30 days post-biopsy]: Proportion of patients with significant cancer detected taking into account peri-lesional biopsies of AI and Radiologist declared lesions.
• Significant cancer detected by systematic biopsies [When biopsy results available, at an expected average of 30 days post-biopsy]: Proportion of patients with significant cancer detected by systematic biopsies
• Frequency of AI failures [When MRI results available, at an expected average of 30 days post-MRI]: Proportion of patients where AI was unable to interpret the MRI scan
• Treatment eligibility decisions [When biopsy results available, at an expected average of 30 days post-biopsy]: Proportion of patients where treatment eligibility changed between AI and Radiologist
• Cost-efffectiveness [At an expected average of 30 days post-intervention]: Cost-effectiveness of unblinded AI interpreted by the radiologist compared to radiologist alone in detecting significant prostate cancer, and AI alone vs. radiologist alone, and a 3-arm analysis considering all three.

Sex: Male, ages: 18 Years to any

Inclusion Criteria:

1. Men at least 18 years of age referred with clinical suspicion of prostate cancer
2. Serum PSA ≤ 20 ng/mL
3. Fit to undergo all procedures listed in the protocol
4. Able to provide written informed consent

Exclusion Criteria:

1. Prior prostate biopsy
2. Prior prostate MRI on a previous encounter*
3. Prior treatment for prostate cancer
4. Contraindication to MRI (e.g. claustrophobia, pacemaker)
5. Metalwork that would give rise to artefact on MRI (e.g. hip prosthesis, pelvic/spinal metalwork)
6. Contraindication to prostate biopsy
7. Unfit to undergo any procedures listed in protocol

   * An MRI on a previous encounter means a previous prostate MRI which has been seen by a doctor and has been used to inform patient management at the time of the original MRI.

• Veeru Kasivisvanathan, MBBS BSc FRCS MSc PGCert PhD, Study chair, Division of Surgery and Interventional Science, University College London, UK
• Doug Pendse, MB ChB MD (Res) MRCS FRCR, Study chair, Department of Radiology, Universiy College London Hospitals NHS Foundation Trust, UK