MRI-Based Machine Learning Approach Versus Radiologist MRI Reading for the Detection of Prostate Cancer, The PRIMER Trial

Not Applicable

Not yet recruiting

• Conditions: Prostate Carcinoma

• Registration Number: NCT07162194
• Lead Sponsor: University of Southern California

Brief Summary

This clinical trial studies how well a magnetic resonance imaging (MRI)-based machine learning approach (i.e., artificial intelligence \[AI\]) works as compared to radiologist MRI readings in detecting prostate cancer. One of the current methods used to help diagnose possible prostate cancer is performing a prostate MRI. An MRI uses a magnetic field to take pictures of the body. The MRI images are examined by a radiologist. If a suspicious area is seen in the MRI, the radiologist assigns it a PIRADS score. This stands for Prostate Imaging Reporting and Data System. The PIRADS score is used to report how likely it is that a suspicious area in the prostate is cancer. The AI system has been developed also to be able to analyze prostate MRI images and detect suspicious areas in the prostate that may be cancer. The AI system's ability to diagnose aggressive prostate cancer may be similar to detection performed by experienced radiologists using the standard PIRADS system of analyzing prostate MRI.

Detailed Description

PRIMARY OBJECTIVE:

I. To determine the non-inferiority of targeted biopsy according to Green Learning (GL) AI over Prostate Imaging Reporting \& Data System (PIRADS).

SECONDARY OBJECTIVES:

I. To determine the patient-level diagnostic performance of GL AI, Deep Learning (DL) AI and PIRADS for clinically significant prostate cancer (CSPCa) detection.

II. To assess Targeted biopsy core characteristics. III. To evaluate the predictors for patient-level CSPCa detection. IV. To assess the spatial correlation of CSPCa distribution on radical prostatectomy (RP) specimens and region of interest (ROI) generated by GL AI and PIRADS.

OUTLINE: Patients undergoing prostate biopsy per standard of care (SOC) are assigned to Group 1. Patients who underwent a prostate biopsy followed by a radical prostatectomy within 6 months, as well as patients only undergoing a radical prostatectomy are assigned to Group 2.

GROUP 1: Patients are randomized to 1 of 6 arms.

ARM I: Patients undergo MRI/transrectal ultrasound (TRUS) followed by a targeted prostate biopsy using PIRADS on study. Patients then undergo a 2nd MRI/TRUS followed by a targeted prostate biopsy based on GL AI predictions. Patients then undergo a 3rd MRI/TRUS followed by a targeted biopsy based on DL AI predictions. Finally, patients undergo up to 12 additional prostate biopsies per SOC.

ARM II: Patients undergo MRI/TRUS followed by a targeted prostate biopsy using PIRADS. Patients then undergo a 2nd MRI/TRUS followed by a targeted prostate biopsy based on DL AI predictions. Patients then undergo a 3rd MRI/TRUS followed by a targeted biopsy based on GL AI predictions. Finally, patients undergo up to 12 additional prostate biopsies per SOC.

ARM III: Patients undergo MRI/TRUS followed by a targeted prostate biopsy using GL AI predictions. Patients then undergo a 2nd MRI/TRUS followed by a targeted prostate biopsy using PIRADS. Patients then undergo a 3rd MRI/TRUS followed by a targeted biopsy based on DL AI predictions. Finally, patients undergo up to 12 additional prostate biopsies per SOC.

ARM IV: Patients undergo MRI/TRUS followed by a targeted prostate biopsy using GL AI predictions. Patients then undergo a 2nd MRI/TRUS followed by a targeted prostate biopsy based on DL AI predictions. Patients then undergo a 3rd MRI/TRUS followed by a targeted biopsy using PIRADS. Finally, patients undergo up to 12 additional prostate biopsies per SOC.

ARM V: Patients undergo MRI/TRUS followed by a targeted prostate biopsy using DL AI predictions. Patients then undergo a 2nd MRI/TRUS followed by a targeted prostate biopsy based on GL AI predictions. Patients then undergo a 3rd MRI/TRUS followed by a targeted biopsy using PIRADS. Finally, patients undergo up to 12 additional prostate biopsies per SOC.

ARM VI: Patients undergo MRI/TRUS followed by a targeted prostate biopsy using DL AI predictions. Patients then undergo a 2nd MRI/TRUS followed by a targeted prostate biopsy using PIRADS. Patients then undergo a 3rd MRI/TRUS followed by a targeted biopsy based on GL AI predictions. Finally, patients undergo up to 12 additional prostate biopsies per SOC.

GROUP 2: Patients have their removed prostate evaluated using a special mold on study. Prostate tissue is mapped and compared with the prostate cancer prediction on MRI generated by radiologists and AI reports.

All patients may also undergo digital rectal exam (DRE) on study.

After completion of study intervention, patients are followed up at 10 days and at 3 months.

Study Timeline

• Status Verified: 2025-08
• Study First Submitted: 2025-08-29
• Study First Submitted QC: 2025-08-29
• Last Update Submitted: 2025-08-29
• Study First Posted: 2025-09-09
• Last Update Posted: 2025-09-09
• Study Start: 2025-10-15
• Primary Completion: 2027-10-15
• Study Completion: 2028-10-15

Recruitment & Eligibility

• Status: NOT_YET_RECRUITING
• Sex: Male
• Target Recruitment: 130

Inclusion Criteria

• PROSTATE BIOPSY COHORT: Patients undergoing transperineal MRI/TRUS fusion prostate biopsy (PBx) as per standard of care
• PROSTATE BIOPSY COHORT: Patients who underwent or are undergoing 3T multiparametric MRI (T2W, diffusion weighted imaging [DWI], apparent diffusion coefficient [ADC], and dynamic contrast-enhanced [DCE]) within 90 days prior to biopsy
• PROSTATE BIOPSY COHORT: Patients who provided with consent for the study
• RADICAL PROSTATECTOMY COHORT: Patients undergoing radical prostatectomy without neo-adjuvant hormonal therapy for primary treatment of prostate cancer as per standard of care
• RADICAL PROSTATECTOMY COHORT: Patients who underwent or are undergoing 3T multiparametric MRI (T2W, DWI, ADC, and DCE) within 180 days prior to radical prostatectomy
• RADICAL PROSTATECTOMY COHORT: Patients who provided with consent for the study

Exclusion Criteria

• PROSTATE BIOPSY COHORT: Patients with a history of prostate cancer
• PROSTATE BIOPSY COHORT: Patients with a history of surgical treatment on benign prostate hyperplasia
• PROSTATE BIOPSY COHORT: Patients undergoing saturation prostate biopsy
• PROSTATE BIOPSY COHORT: Patients under 20 years old
• PROSTATE BIOPSY COHORT: Patients with previous PBx history
• PROSTATE BIOPSY COHORT: MRI which was not interpreted by PIRADS
• PROSTATE BIOPSY COHORT: MRI with significant artifact
• RADICAL PROSTATECTOMY COHORT: Patients with a history of surgical treatment on benign prostate hyperplasia
• RADICAL PROSTATECTOMY COHORT: Patients under 20 years old
• RADICAL PROSTATECTOMY COHORT: Patients without pre-treatment MRI
• RADICAL PROSTATECTOMY COHORT: MRI which was not interpreted by PIRADS
• RADICAL PROSTATECTOMY COHORT: MRI with significant artifact

Study & Design

• Study Type: INTERVENTIONAL
• Study Design: PARALLEL

Primary Outcome Measures

Name	Time	Method
Clinically-significant prostate cancer (CSPCa) detection rate on Green Learning (GL) artificial intelligence (AI)-targeted and Prostate Imaging-Reporting and Data System (PIRADS)-targeted biopsies	Up to 3 months	Detection rates of PIRADS and GL AI-targeted biopsy will be evaluated per index region of interest (ROI), respectively.
CSPCa detection rate on GL AI-targeted and Deep Learning (DL) AI-targeted biopsies	Up to 3 months	Detection rates of GL AI and DL AI targeted biopsy will be evaluated per index ROI, respectively.

Secondary Outcome Measures

Name	Time	Method
Patient-level diagnostic performance of GL AI and PIRADS for CSPCa detection	Up to 3 months	Sensitivity, specificity, accuracy, positive predictive value, and negative predictive value will be compared between PIRADS, GL AI, and DL AI predictions by McNemar test. The performance will be calculated according to the definitions below. Additionally, decision curve analysis will be conducted.
Targeted biopsy core characteristics	Up to 3 months	Will include prostate cancer subtypes, benign elements, lesion locations, cancer core length (mm), cancer core involvement (%), and Gleason Grade Group. Will be compared between PIRADS, GL AI, and DL AI by Chi-square test or Wilcoxon rank sum test.
Predictors for patient-level CSPCa detection	Up to 3 months	Independent predictors for patient-level CSPCa detection will be assessed by logistic regression. Predictors include age, race, ethnicity, prostate-specific antigen (PSA), PSA density (PSA/prostate volume measured on magnetic resonance imaging \[MRI\]), digital rectal examination (DRE) abnormality, PIRADS score, GL AI prediction score, and DL AI prediction score. Classifier will be created using the strong predictors for CSPCa, and its discriminant performance will be assessed by the receiver operating characteristic (ROC) analysis.
Dice score and linear correlation coefficient of CSPCa distribution on radical prostatectomy (RP) specimens and ROI generated by GL AI, DL AI and PIRADS	Up to 3 months	Distribution of CSPCa on the RP specimen will be annotated on the digitized slides and 3-dimensional reconstructed as ground truth. ROI segmentations according to PIRADS, GL AI prediction heatmap, and DL AI prediction heatmap also will be 3D reconstructed. The spatial concordance between GL-ROI, DL AI-ROI, PIRADS-ROI, and CSPCa distribution on RP specimen will be assessed. The accuracy of the spatial concordance and volume estimation will be analyzed by the Dice score and linear correlation analysis, respectively.

Trial Locations

Locations (1)

USC / Norris Comprehensive Cancer Center; 🇺🇸 Los Angeles, California, United States