Artificial Intelligence-Enabled Imaging for Predicting Preoperative Extraprostatic Extension in Prostate Cancer:...

Area of Science:

• Radiology
• Artificial Intelligence
• Oncology

Background:

• Artificial intelligence (AI) techniques, particularly machine learning and deep learning, show promise in analyzing multimodal imaging data for enhanced preoperative prediction of extraprostatic extension (EPE) in prostate cancer.
• Accurate preoperative prediction of EPE is crucial for treatment planning and patient management in prostate cancer.

Purpose of the Study:

• To compare the diagnostic performance of AI-enabled imaging techniques against radiologists for predicting preoperative EPE in prostate cancer.
• To evaluate the effectiveness of multiparametric magnetic resonance imaging (mpMRI) and prostate-specific membrane antigen positron emission tomography (PSMA PET) based AI models.

Main Methods:

• A systematic literature search was conducted in PubMed, Embase, and Web of Science up to September 2025, adhering to PRISMA-DTA guidelines.
• Studies utilizing AI for EPE prediction with mpMRI and PSMA PET imaging were included.
• Pooled sensitivity, specificity, and area under the curve (AUC) were calculated using a bivariate random effects model, with study quality assessed by the QUADAS-2 tool.

Main Results:

• AI-based mpMRI demonstrated superior diagnostic performance compared to radiologists, with a higher AUC (0.81 vs. 0.77 in internal validation).
• No significant difference was found between mpMRI-based AI and radiologists in terms of sensitivity and specificity.
• PSMA PET-based AI showed moderate performance but no significant advantage over mpMRI-based AI or radiologists, particularly in external validation.

Conclusions:

• AI utilizing mpMRI significantly improves the preoperative prediction of EPE in prostate cancer compared to conventional radiological assessment.
• PSMA PET-based AI currently does not offer a significant advantage over mpMRI-based AI or radiologists.
• Further research with larger, diverse cohorts is needed to confirm findings and optimize AI integration into clinical practice, acknowledging limitations of retrospective designs and heterogeneity.