Clinical Scorecard: Assessment of a multiparametric transrectal ultrasound-based artificial intelligence model for the localization of clinically significant prostate cancer through targeted biopsy simulations

At a Glance

Key Highlights

• MRI is the current standard for csPCa detection but has limitations including diagnostic accuracy, inter-observer variability, availability, and cost.
• The mpUS AI model demonstrated promising diagnostic performance with AUROC of 0.87 at voxel level and was validated at patient level via simulated targeted biopsies.
• Simulated biopsy results showed high sensitivity (~0.81–0.90) but moderate specificity (~0.39–0.43) for detecting csPCa (ISUP ≥ 2 and ≥ 3) in internal and external cohorts.

Guideline-Based Recommendations

Diagnosis

• Use mpUS combined with AI-based heatmap analysis to identify suspicious lesions ≥ 0.07 cc for targeted biopsy simulation.
• Consider AI model outputs to guide biopsy targeting, potentially improving detection of csPCa while reducing unnecessary biopsies.

Management

• Perform targeted biopsies on AI-identified lesions to improve csPCa localization prior to radical prostatectomy or further management.
• Integrate AI mpUS analysis as a complementary tool to MRI, especially where MRI availability or interpretation is limited.

Monitoring & Follow-up

• Evaluate biopsy outcomes in conjunction with AI predictions to refine diagnostic accuracy and guide patient follow-up.
• Monitor AI model performance in clinical practice to ensure consistent sensitivity and specificity across populations.

Risks

• Potential false positives due to moderate specificity may lead to unnecessary biopsies.
• Sampling errors and spatial discrepancies between AI predictions and actual tumor location may affect biopsy accuracy.

Patient & Prescribing Data

Men undergoing evaluation for prostate cancer with suspicion of csPCa or negative prior imaging/biopsy

AI mpUS model can enhance targeted biopsy accuracy, potentially improving csPCa detection rates and reducing detection of insignificant PCa.

Clinical Best Practices

• Use AI mpUS heatmaps to identify and prioritize biopsy targets, focusing on the two largest predicted lesions ≥ 0.07 cc.
• Simulate biopsy needle paths in planning to maximize overlap with AI-predicted lesions and histologically confirmed csPCa.
• Combine AI mpUS findings with clinical and MRI data to optimize patient selection and biopsy strategy.

References

• MRI improves patient selection prior to biopsy
• Limitations of MRI in prostate cancer diagnosis
• Advances in multiparametric ultrasound imaging
• AI model for mpUS in csPCa detection