A probabilistic tissue classification metric for MR-US guided prostate core-needle biopsies with explicit modeling of localization uncertainty

Abstract Objective. Develop a spatially resolved probabilistic framework that explicitly models localization uncertainty to map along-core tissue-class sampling probabilities P i ( z ) for MR-informed, US-guided transperineal prostate biopsies, yielding millimetre-scale dominant intraprostatic lesion (DIL)-sampling descriptors for planning, quality assurance, and biology-related research. We also outline an exploratory linkage to core-level pathology; formal clinical validation remains future work. Approach. Using retrospectively analyzed data from 15 high-dose-rate-brachytherapy patients enrolled on a prospective trial, we linked 51 transrectal ultrasound biopsy tracks to multiparametric magnetic resonance imaging (mpMRI) DICOM structure sets with 26 DILs contoured. Procedural localization uncertainty was modeled as independent rigid translations for each structure type, sampled from zero-mean Gaussians (SDs 1.25– 2.2 mm ) and propagated via a 10 000-trial Monte Carlo method to obtain P i ( z ) and nominal labels B i ( z ) . Core-level DIL sampling metrics ( ⟨ P D ⟩ , max ( P D ) ) were reported per core and at cohort level. Main results. Continuous along-core probability maps that propagate sampling-location and delineation uncertainties go beyond a nominal along-core hit/miss trace, capturing lesion-enriched sub-segments predicted by the mpMRI derived structure set, transition-band width, and benign prostatic stretches. Across cores, median DIL-sampling descriptors were ⟨ P D ⟩ = 0.24 and max ( P D ) = 0.46 ; urethral and rectal sampling probabilities were near zero, consistent with safe practice. Significance. The framework converts measured localization uncertainty into interpretable, millimetre-scale tissue sampling metrics. These descriptors can inform pre-procedure plan checks and biopsy pre-planning and, where localization is available, intra-procedural estimates of expected DIL sampling. At the clinic level they offer QA summaries by tracking DIL-sampling metrics such as ⟨ P D ⟩ and