A probabilistic framework for prostate core-needle biopsies in HDR brachytherapy : from clinical quality assurance to translational research

Prostate core-needle biopsies acquired during high-dose-rate (HDR) brachytherapy are clinically useful for diagnosis and can also support translational research when linked to sampled tissue, imaging-defined anatomy, and delivered dose. The main obstacle is that the effective biopsy location is uncertain by several millimetres because of registration error, contouring variability, needle placement uncertainty, and specimen-related effects. This thesis develops a probabilistic framework for mapping tissue, dose, and imaging data, including image information derived from multiparametric MRI (mpMRI), to reconstructed prostate biopsy cores under explicit localisation uncertainty. Biopsies were represented as voxelised core segments in the clinical imaging and treatment-planning reference frame. A shared Monte Carlo localisation model was used to propagate uncertainty into along-core tissue-class probabilities, voxel- and biopsy-level dose summaries, distributional dose-volume histogram (DVH) metrics, and cohort-level robustness measures. The framework was first applied to estimate the probability that each position along a biopsy sampled dominant intraprostatic lesion and other nearby organs at risk. These tissue-assignment probabilities can be used for targeting quality assurance because they quantify how plausibly a biopsy sampled the intended imaging-defined target. They may also serve as imaging-derived covariates for testing associations with pathology and malignancy-related measures. The framework was then extended to biopsy-scale HDR dosimetry, showing that nominal assignments can differ materially from uncertainty-propagated quantities and that biopsy dose is heterogeneous along the core. At cohort level, DVH-style pass probabilities and robustness classes were derived under the localisation model, yielding a probabilistic form of dosimetric quality assurance in which nominal distance from threshold emerged as the dominant determinant of high-confidence passing. Finally, Gaussian process regression was used to model along-core spatial correlation in dose, yielding more stable covariance-informed uncertainty summaries than independent voxel-wise Monte Carlo reporting alone. In this setting, biopsy-linked tissue and dosimetry analysis is better framed as probabilistic spatial inference than as deterministic assignment. By parameterising and propagating localisation uncertainty, and by explicitly modelling along-core spatial correlation, the framework yields interpretable biopsy-scale descriptors that can support quality assurance, help with interpreting tissue assays, and inform future dose-biology studies in prostate HDR brachytherapy.