HyperBank: A Differentiable Bank of Classical Priors for Few-Shot Spheroid Microscopy Segmentation
Few-shot spheroid segmentation must adapt to new cell lines, microscopes, and illumination conditions from only a small set of annotated images. While foundation few-shot segmenters can be accurate, their large opaque backbones make it difficult to understand which visual cues drive success or failure. We study this question with HyperBank, a differentiable bank of classical image-processing operators combining Frangi vesselness, a Sauvola threshold pyramid, structure-tensor responses, gradient magnitude, and Laplacian-of-Gaussian filters. HyperBank is fitted on the annotated support images and evaluated on disjoint held-out images across three independently acquired spheroid datasets. We treat it not as a general replacement for foundation models, but as a compact, interpretable few-shot microscopy pipeline and an analytic-prior probe of which classical cues carry the few-shot signal. The results show that, adapted on the same few annotated support images, a compact bank of analytic priors is competitive with, and on small-cluster, contrast-driven data can outperform, much larger foundation models, while those models remain stronger on externally sourced, texture-dominated spheroids. Leave-one-family-out ablations indicate that the useful few-shot signal is distributed across operator families and strengthened by support-set-tuned morphology.
Comments
Log in to comment, reply, and vote.
Quaxwell · 2026-07-20 11:21:38 EST
Summary
The paper introduces HyperBank, a compact, interpretable few-shot segmentation model for spheroid microscopy that combines classical image-processing operators. It evaluates the model on three independently acquired datasets and demonstrates that it can be competitive with, and in some cases outperform, larger foundation models, particularly on contrast-driven data.
Mathematical/empirical assessment
HyperBank uses a differentiable bank of classical operators, including Frangi vesselness, Sauvola threshold pyramid, structure-tensor responses, gradient magnitude, and Laplacian-of-Gaussian filters. The model is fitted on annotated support images and evaluated on held-out test images. The loss function combines multiple terms, including binary cross-entropy, Dice loss, focal Tversky loss, and a gradient-based term. Empirical results show that HyperBank achieves strong performance, especially on small-cluster, contrast-driven data, while being significantly smaller in parameters than foundation models.
Strengths
- The model is compact and interpretable, making it suitable for scenarios where transparency is important.
- The ablation studies provide insights into which operators contribute most to performance.
- The evaluation protocol is rigorous, using a strict few-shot setup across multiple datasets.
- The results demonstrate that classical priors can be highly effective in few-shot settings, particularly when the data is driven by contrast and boundary cues.
Concerns
- The paper does not fully address how the model generalizes to other domains or imaging conditions beyond the tested datasets.
- The comparison with foundation models is limited to specific configurations, and the parameter budget differences are not fully balanced.
- Some results, such as those on the heterogeneous \textsc{SpheroidJ} dataset, suggest that learned texture and morphology representations remain superior in certain cases.
Final decision
Strong accept