pith. sign in

arxiv: 2511.19953 · v2 · pith:3FAR577Ynew · submitted 2025-11-25 · 💻 cs.CV

Supervise Less, See More: Training-free Nuclear Instance Segmentation with Prototype-Guided Prompting

Wen Zhang , Qin Ren , Wenjing Liu , Haibin Ling , Chenyu You This is my paper

Pith reviewed 2026-05-21 18:46 UTC · model grok-4.3

classification 💻 cs.CV
keywords nuclear instance segmentationtraining-free segmentationprototype-guided promptingSegment Anything Modelhistopathologypartial optimal transportzero-shot biomedical segmentation
0
0 comments X

The pith

SPROUT segments cell nuclei in pathology slides without training or annotations by prompting SAM with slide-specific prototypes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents SPROUT, a fully training-free framework for nuclear instance segmentation in histopathology images. It builds reference prototypes from the current slide using typical cell appearance and staining patterns, then aligns image features to these prototypes with partial optimal transport. This alignment produces positive and negative point prompts that guide the Segment Anything Model to draw accurate nuclear boundaries. A sympathetic reader would care because the method removes the need for labeled training data and model retraining, which currently limits large-scale analysis of pathology archives.

Core claim

By constructing slide-specific reference prototypes from histology-informed priors and progressively aligning foreground and background features through a partial optimal transport scheme, the resulting features can be converted into effective positive and negative point prompts that enable the Segment Anything Model to produce precise nuclear delineations without any parameter updates or supervision.

What carries the argument

Slide-specific reference prototypes built from histology-informed priors, aligned to image features via partial optimal transport to generate positive and negative point prompts for the Segment Anything Model.

If this is right

  • Pathology labs can apply nuclear segmentation to new staining protocols or tissue types without collecting annotations or running retraining.
  • Large archives of whole-slide images become feasible to process at scale because no per-slide supervision is required.
  • The same prompting pipeline can support downstream tasks such as nuclear counting or morphology measurement directly from the SAM masks.
  • Zero-shot performance becomes competitive with supervised methods on standard benchmarks without domain-specific fine-tuning.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The prototype-plus-transport idea could be tested on other instance segmentation problems where object shape priors are known, such as cell segmentation in microscopy outside pathology.
  • Replacing SAM with a different foundation model might change prompt effectiveness and reveal how much the gains depend on the particular model.
  • If the priors prove robust, the method could reduce annotation costs in clinical AI pipelines by orders of magnitude.

Load-bearing premise

Histology-informed priors can be used to build slide-specific prototypes whose partial optimal transport alignment reliably yields prompts that let SAM delineate nuclei accurately on diverse pathology images.

What would settle it

Running SPROUT on a held-out histopathology dataset and observing that the prompted SAM outputs yield substantially lower Dice scores or more frequent nuclear mergers than fully supervised baselines would falsify the central claim.

read the original abstract

Accurate nuclear instance segmentation is a pivotal task in computational pathology, supporting data-driven clinical insights and facilitating downstream translational applications. While large vision foundation models have shown promise for zero-shot biomedical segmentation, most existing approaches still depend on dense supervision and computationally expensive fine-tuning. Consequently, training-free methods present a compelling research direction, yet remain largely unexplored. In this work, we introduce SPROUT, a fully training- and annotation-free prompting framework for nuclear instance segmentation. SPROUT leverages histology-informed priors to construct slide-specific reference prototypes that mitigate domain gaps. These prototypes progressively guide feature alignment through a partial optimal transport scheme. The resulting foreground and background features are transformed into positive and negative point prompts, enabling the Segment Anything Model (SAM) to produce precise nuclear delineations without any parameter updates. Extensive experiments across multiple histopathology benchmarks demonstrate that SPROUT achieves competitive performance without supervision or retraining, establishing a novel paradigm for scalable, training-free nuclear instance segmentation in pathology.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The paper introduces SPROUT, a fully training- and annotation-free prompting framework for nuclear instance segmentation in histopathology. It constructs slide-specific reference prototypes from histology-informed priors, aligns foreground/background features via partial optimal transport, converts the aligned features into positive and negative point prompts, and feeds these to an off-the-shelf SAM to produce instance masks without any parameter updates or supervision.

Significance. If the central claim holds, the work would establish a scalable, annotation-free paradigm for nuclear segmentation that bypasses the usual costs of dense supervision and fine-tuning, with potential impact on large-scale computational pathology pipelines.

major comments (3)
  1. [Abstract and Experiments] The abstract and method overview assert competitive performance across multiple histopathology benchmarks, yet supply no quantitative results, baselines, error bars, dataset statistics, or statistical tests. Without these in the experiments section, the central claim that SPROUT matches or exceeds supervised methods cannot be evaluated.
  2. [§3.2] §3.2 (Prototype construction and partial optimal transport): the transport objective, cost function, mass-preservation parameter, and regularization are not formulated explicitly. Because the method relies on this alignment step to produce reliable positive/negative point prompts for SAM, the absence of the mathematical definition and any ablation on these choices leaves the robustness to stain/organ/scanner shifts unverified.
  3. [§3.3 and §4] The claim that the generated prompts achieve the localization precision required for instance-level (rather than semantic) segmentation is load-bearing, yet no quantitative evaluation of prompt accuracy (e.g., distance to true nuclear centroids, false-positive rate on background) is reported. This gap directly affects whether the downstream SAM outputs can be trusted for overlapping or small nuclei.
minor comments (2)
  1. [§3] Notation for the prototype features and the partial OT plan should be introduced once and used consistently; current description mixes “reference prototypes,” “foreground features,” and “aligned embeddings” without clear mapping.
  2. [Figure 2] Figure 2 (pipeline diagram) would benefit from explicit arrows showing how the OT plan is converted into point coordinates for SAM.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the insightful comments. We address each major comment below and will make revisions to improve the clarity and completeness of the manuscript.

read point-by-point responses

  1. Referee: [Abstract and Experiments] The abstract and method overview assert competitive performance across multiple histopathology benchmarks, yet supply no quantitative results, baselines, error bars, dataset statistics, or statistical tests. Without these in the experiments section, the central claim that SPROUT matches or exceeds supervised methods cannot be evaluated.

    Authors: We acknowledge the referee's concern regarding the presentation of results. The experiments in Section 4 do include quantitative comparisons on several histopathology datasets, reporting metrics like Dice coefficient and Aggregated Jaccard Index against supervised baselines. To address the specific request, we will revise the section to include error bars (e.g., standard deviation over multiple images or cross-validation), detailed dataset statistics, and statistical tests (such as paired t-tests) to rigorously support the competitive performance claims. revision: yes

  2. Referee: [§3.2] §3.2 (Prototype construction and partial optimal transport): the transport objective, cost function, mass-preservation parameter, and regularization are not formulated explicitly. Because the method relies on this alignment step to produce reliable positive/negative point prompts for SAM, the absence of the mathematical definition and any ablation on these choices leaves the robustness to stain/organ/scanner shifts unverified.

    Authors: We agree that explicit mathematical formulations would enhance reproducibility and understanding. In the revised manuscript, we will provide the full formulation of the partial optimal transport problem in §3.2, including the objective function, the cost matrix based on feature similarities, the mass preservation constraint, and the entropic regularization term. Furthermore, we will conduct and report ablations on key parameters to verify robustness under variations in staining, organs, and scanners. revision: yes

  3. Referee: [§3.3 and §4] The claim that the generated prompts achieve the localization precision required for instance-level (rather than semantic) segmentation is load-bearing, yet no quantitative evaluation of prompt accuracy (e.g., distance to true nuclear centroids, false-positive rate on background) is reported. This gap directly affects whether the downstream SAM outputs can be trusted for overlapping or small nuclei.

    Authors: This observation is well-taken. While the end-to-end instance segmentation results are provided, direct assessment of prompt quality is indeed missing. We will add quantitative evaluations of the generated prompts in the revised version, including metrics such as mean distance from positive prompts to annotated nuclear centroids and the rate of negative prompts incorrectly placed on foreground regions. These additions will better substantiate the suitability for instance-level segmentation in complex scenarios. revision: yes

Circularity Check

0 steps flagged

No circularity: procedural framework relies on external priors and off-the-shelf SAM

full rationale

The paper describes SPROUT as a training-free procedural pipeline that builds slide-specific prototypes from histology-informed priors, aligns them via partial optimal transport, and converts the results into point prompts for an unmodified SAM model. No equations, fitted parameters, or self-referential definitions appear in the provided text; the method does not derive its outputs from quantities defined by its own results. Performance claims rest on external benchmark experiments rather than internal consistency loops, rendering the derivation self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The framework rests on the unstated assumption that histology priors transfer across slides and that partial optimal transport produces sufficiently discriminative foreground/background features for prompting; no free parameters or invented entities are named in the abstract.

axioms (2)
  • domain assumption Histology-informed priors can be used to construct slide-specific reference prototypes that mitigate domain gaps.
    Invoked in the abstract as the basis for prototype construction.
  • domain assumption Partial optimal transport can progressively guide feature alignment between prototypes and image features.
    Described as the mechanism that transforms features into prompts.

pith-pipeline@v0.9.0 · 5709 in / 1390 out tokens · 40123 ms · 2026-05-21T18:46:28.828349+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Hystar: Hypernetwork-driven Style-adaptive Retrieval via Dynamic SVD Modulation

    cs.CV 2026-05 unverdicted novelty 6.0

    Hystar adapts CLIP-like models to unseen query styles by generating per-input singular-value perturbations with a hypernetwork for attention layers and a new StyleNCE contrastive loss.

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages · cited by 1 Pith paper

  1. [1]

    @esa (Ref

    \@ifxundefined[1] #1\@undefined \@firstoftwo \@secondoftwo \@ifnum[1] #1 \@firstoftwo \@secondoftwo \@ifx[1] #1 \@firstoftwo \@secondoftwo [2] @ #1 \@temptokena #2 #1 @ \@temptokena \@ifclassloaded agu2001 natbib The agu2001 class already includes natbib coding, so you should not add it explicitly Type <Return> for now, but then later remove the command n...

    work page

  2. [2]

    \@lbibitem[] @bibitem@first@sw\@secondoftwo \@lbibitem[#1]#2 \@extra@b@citeb \@ifundefined br@#2\@extra@b@citeb \@namedef br@#2 \@nameuse br@#2\@extra@b@citeb \@ifundefined b@#2\@extra@b@citeb @num @parse #2 @tmp #1 NAT@b@open@#2 NAT@b@shut@#2 \@ifnum @merge>\@ne @bibitem@first@sw \@firstoftwo \@ifundefined NAT@b*@#2 \@firstoftwo @num @NAT@ctr \@secondoft...

    work page

  3. [3]

    (QGT+  o/߸ ;fQ Zt鐒gvZxG*J Y ȮY! dZs (HE E 2 n=#R

    @open @close @open @close and [1] URL: #1 \@ifundefined chapter * \@mkboth \@ifxundefined @sectionbib * \@mkboth * \@mkboth\@gobbletwo \@ifclassloaded amsart * \@ifclassloaded amsbook * \@ifxundefined @heading @heading NAT@ctr thebibliography [1] @ \@biblabel @NAT@ctr \@bibsetup #1 @NAT@ctr @ @openbib .11em \@plus.33em \@minus.07em 4000 4000 `\.\@m @bibit...

    work page arXiv