Assessment of AI Model Using Multiparametric Ultrasound for Prostate Cancer Localization

Overview

This study evaluated a multiparametric ultrasound (mpUS)-based artificial intelligence (AI) model for detecting clinically significant prostate cancer (csPCa) through simulated targeted biopsies. The AI model demonstrated high sensitivity but moderate specificity in both internal and external patient cohorts, supporting its potential clinical utility.

Background

Prostate cancer diagnosis traditionally relies on MRI followed by targeted biopsies, improving detection of clinically significant cases while reducing insignificant findings. However, MRI has limitations including variability, availability, and cost. Advances in ultrasound imaging, such as mpUS, offer promising alternatives but lack standardized interpretation. AI-powered computer-aided diagnosis systems can automate mpUS interpretation, potentially overcoming these challenges and guiding biopsy procedures more effectively.

Data Highlights

Key Findings

• The AI model achieved a sensitivity of 82% internally and 81% externally for detecting ISUP grade ≥2 prostate cancer.
• Specificity was moderate, approximately 42–43% across cohorts for ISUP ≥2 detection.
• For higher-grade tumors (ISUP ≥3), sensitivity increased to 90% internally and 96% externally.
• Positive predictive values ranged from 40% to 76%, with higher PPV internally for ISUP ≥2.
• Negative predictive values were higher for ISUP ≥3, reaching up to 96% externally.
• Simulated targeted biopsy based on AI heatmaps provides a controlled method to estimate clinical performance prior to prospective trials.

Clinical Implications

The mpUS AI model shows promise as a non-MRI-based tool to guide targeted prostate biopsies, particularly for detecting clinically significant cancers with high sensitivity. Its moderate specificity suggests it may generate some false positives, highlighting the need for further clinical validation. This approach could improve accessibility and reduce costs associated with prostate cancer diagnosis.

Conclusion

This study demonstrates that a multiparametric ultrasound-based AI model can effectively localize clinically significant prostate cancer with high sensitivity in simulated biopsy settings, supporting its potential role in clinical practice pending prospective evaluation.

References

1. van der Leest M et al. 2020 -- MRI-Targeted Biopsy vs Standard Biopsy
2. Rouvière O et al. 2019 -- Diagnostic Accuracy of MRI in Prostate Cancer
3. Kasivisvanathan V et al. 2018 -- MRI-Targeted vs Systematic Biopsy
4. Ahmed HU et al. 2017 -- Limitations of MRI in Prostate Cancer
5. Wildeboer RR et al. 2020 -- Advances in Multiparametric Ultrasound
6. Siddiqui MM et al. 2019 -- Micro-Ultrasound in Prostate Cancer
7. Cornud F et al. 2021 -- Ultrasound Imaging Developments
8. Wildeboer RR et al. 2021 -- Diagnostic Performance of mpUS
9. Wildeboer RR et al. 2022 -- AI Model for csPCa Detection
10. Current Study Authors 2024 -- Assessment of a Multiparametric Transrectal Ultrasound-Based AI Model