On the Viability of Semi-Supervised Segmentation Methods for Statistical Shape Modeling

TL;DR

The paper tackles the annotation bottleneck in Statistical Shape Modeling by evaluating semi-supervised segmentation as a substitute for manual labeling in SSM construction. It introduces a benchmark comparing eight semi-supervised methods across two datasets, FEMUR and NAMIC Left Atrium, under two data-label strategies, and uses ShapeWorks to build and compare $SSM$ representations. Results show that several methods (e.g., BCP, CAML, DeSCO, DTC, MCF) can approximate GT SSMs with substantial reductions in labeled data on FEMUR (60-80%), while NAMIC results are inconsistent, particularly for small structures like ventricles. The findings illustrate the potential and limits of weak supervision for population-level shape analysis and point to directions for broader weak/self-supervised approaches and future integration with foundational segmentation models.

Abstract

Statistical Shape Models (SSMs) excel at identifying population level anatomical variations, which is at the core of various clinical and biomedical applications, including morphology-based diagnostics and surgical planning. However, the effectiveness of SSM is often constrained by the necessity for expert-driven manual segmentation, a process that is both time-intensive and expensive, thereby restricting their broader application and utility. Recent deep learning approaches enable the direct estimation of Statistical Shape Models (SSMs) from unsegmented images. While these models can predict SSMs without segmentation during deployment, they do not address the challenge of acquiring the manual annotations needed for training, particularly in resource-limited settings. Semi-supervised models for anatomy segmentation can mitigate the annotation burden. Yet, despite the abundance of available approaches, there are no established guidelines to inform end-users on their effectiveness for the downstream task of constructing SSMs. In this study, we systematically evaluate the potential of semi-supervised methods as viable alternatives to manual segmentations for building SSMs. We establish a new performance benchmark by employing various semi-supervised methods for anatomy segmentation under low annotation settings, utilizing the predicted segmentations for the task of SSM. Our results indicate that some methods produce noisy segmentation, which is very unfavorable for SSM tasks, while others can capture the correct modes of variations in the population cohort with 60-80% reduction in required manual annotation

Figures (8)

• Figure 1: Can manual annotation in constructing shape models be replaced by semi-supervised segmentation methods? Figure shows the proposed new pipeline using Semi-supervised methods.
• Figure 2: Quantitative Results for strategy 1. Generalization and log of Grassmannian distance reported for both FEMUR and NAMIC datasets when using 20% and 40% of annotations for training different semi-supervised methods.
• Figure 3: Results when using 20% of Annotation for strategy 2. Compactness, Specificity, Generalization and log of Grassmannian distance were reported for both both FEMUR and NAMIC datasets when using 20% of annotations for training different semi-supervised methods.
• Figure 4: Qualitative Results for FEMUR and NAMIC dataset showing Second mode of variation for Strategy 1. Second Mode of Variation when SSM is created using segmentation predicted by semi-supervised models trained on 20% of the training data.
• Figure 5: First Mode Of variation Results for FEMUR and NAMIC using 20% of labelled data for strategy 1 . We show first mode of variations for both Femur and Left Atrium datasets showing mean shape($\mu$), first ($\pm \sigma$) and second order ($\pm 2\sigma$) variations from mean shape.