Development and Validation of a Non-Invasive AI Diagnostic Model for Prostate Cancer Using Multiparametric MRI and PSMA PET/CT

Study Design:

This is a retrospective and prospective, multicenter, case-control study. The study aims to develop and validate a non-invasive artificial intelligence (AI) diagnostic model for prostate cancer by integrating multiparametric MRI (mpMRI) and PSMA PET/CT imaging.

Participants:

Patients who underwent mpMRI, PSMA PET/CT, and prostate biopsy or radical prostatectomy at participating centers will be retrospectively enrolled. Eligible participants include pathologically confirmed prostate cancer patients (cases) and benign prostatic hyperplasia (BPH) patients (controls). Inclusion criteria include age ≥18 years, ECOG performance status 0-2, life expectancy >6 months, and availability of complete clinical data (PSA, Gleason score, PI-RADS, SUVmax, prostate volume, etc.). Exclusion criteria include prior prostate cancer treatment (endocrine therapy or radiotherapy), previous prostate surgery (e.g., TURP), severe renal insufficiency, other malignancies. Informed consent is waived for retrospective patients, while signed informed consent is required for prospective patients.

Sample Size:

Approximately 1,000 to 1,500 participants will be enrolled from six hospitals in China: Xiangya Hospital of Central South University, Qilu Hospital of Shandong University, Chinese PLA General Hospital, The First Affiliated Hospital of Guangzhou Medical University, Beijing Hospital and Renji Hospital, Shanghai Jiao Tong University School of Medicine.

AI Model Development:

The AI model will be developed using deep learning and radiomics techniques. mpMRI sequences (including T2-weighted, DWI/ADC, and DCE) and PSMA PET/CT images will be preprocessed, coregistered, and fused. The model will be trained to distinguish clinically significant prostate cancer (csPCa) from non-csPCa or benign conditions. The reference standard is histopathology from prostate biopsy or radical prostatectomy.

Outcome Measures:

Primary outcome measures include the area under the receiver operating characteristic curve (AUC), sensitivity, and specificity of the AI model for csPCa detection. Secondary outcome measures include positive predictive value (PPV), negative predictive value (NPV), overall accuracy, decision curve analysis (DCA) net benefit, diagnostic performance in the PSA gray zone (4-20 ng/mL) subgroup, and the proportion of patients who could avoid biopsy at 100% specificity threshold.

Statistical Analysis:

Model performance will be evaluated using internal cross-validation and external validation on data from different centers. Calibration curves and decision curve analysis will be used to assess clinical utility. Subgroup analyses will be performed for PSA gray zone patients and different Gleason grade groups.