Integrating 3D Volumetric Segmentation and LLM-Based Classification for csPCa Detection on mpMRI: Multi-Institutional External Validation.
Combining two AI approaches for prostate cancer detection on MRI achieved near-expert accuracy across different hospital scanners.
By fusing a 3D segmentation model's volumetric metrics with a large language model's slice-level classification output, this multi-institutional study achieved AUROC 0.900 for clinically significant prostate cancer detection on mpMRI in an independent external cohort of 1,154 patients with heterogeneous scanner protocols. Both combined models significantly outperformed single-modality approaches on discrimination, reclassification, and clinical decision curve analysis.
What the study was
- Study design
- multicenter_retrospective_validation
- Category
- ai_ml_diagnostics
- Maturity
- Potentially Practice-Changing
- Journal
- Acad Radiol
Why it surfaced
Multi-institutional external validation of 5,050-patient cohort with rigorous DCA and NRI reporting; MedGemma LLM integration with 3D segmentation is technically novel; csPCa over-diagnosis/under-diagnosis on MRI is a high-stakes clinical problem; Academic Radiology appropriate for imaging AI validation.
A plain-language summary of published research — not medical advice. Talk to a clinician about your own care.