Deep EM with Hierarchical Latent Label Modelling for Multi-Site Prostate Lesion Segmentation

2026-03-15

AI Summary

This study addresses the challenge of limited model generalizability in multi-center prostate lesion segmentation due to inconsistencies in annotation protocols across sites. To mitigate this, the authors propose a hierarchical Expectation-Maximization (HierEM) framework that treats each center’s annotations as noisy observations of a latent “clean” mask. The method alternately infers voxel-level soft targets via Bayesian posterior estimation and uses these to train a CNN. A hierarchical prior is introduced to jointly model the global mean and center- or case-specific biases, enabling simultaneous estimation of each center’s sensitivity and specificity. Evaluated in leave-one-center-out experiments across three cohorts, the approach achieves Dice similarity coefficients (DSC) of 27.91%–32.67%, significantly outperforming existing methods (p < 0.039), and yields interpretable center-level annotation quality metrics—sensitivity ranging from 31.5% to 47.3% and specificity approximately 99%.

Technology Category

• Machine Learning: Multi-instance/Multi-view Learning
• Computer Vision: Segmentation
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Security and Privacy: Data transparency and provenance
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs

Abstract

Label variability is a major challenge for prostate lesion segmentation. In multi-site datasets, annotations often reflect centre-specific contouring protocols, causing segmentation networks to overfit to local styles and generalise poorly to unseen sites in inference. We treat each observed annotation as a noisy observation of an underlying latent 'clean' lesion mask, and propose a hierarchical expectation-maximisation (HierEM) framework that alternates between: (1) inferring a voxel-wise posterior distribution over the latent mask, and (2) training a CNN using this posterior as a soft target and estimate site-specific sensitivity and specificity under a hierarchical prior. This hierarchical prior decomposes label-quality into a global mean with site- and case-level deviations, reducing site-specific bias by penalising the likelihood term contributed only by site deviations. Experiments on three cohorts demonstrate that the proposed hierarchical EM framework enhances cross-site generalisation compared to state-of-the-art methods. For pooled-dataset evaluation, the per-site mean DSC ranges from 29.50% to 39.69%; for leave-one-site-out generalisation, it ranges from 27.91% to 32.67%, yielding statistically significant improvements over comparison methods (p<0.039). The method also produces interpretable per-site latent label-quality estimates (sensitivity alpha ranges from 31.5% to 47.3% at specificity beta approximates 0.99), supporting post-hoc analyses of cross-site annotation variability. These results indicate that explicitly modelling site-dependent annotation can improve cross-site generalisation.

Problem

• label variability
• multi-site segmentation
• annotation bias
• cross-site generalisation
• prostate lesion segmentation

Innovation

• hierarchical EM
• latent label modelling
• multi-site generalisation
• annotation variability
• prostate lesion segmentation