pith. sign in

arxiv: 2605.13455 · v2 · pith:JYQEBVZSnew · submitted 2026-05-13 · 💻 cs.CV

Bayesian In Vivo Tracking of Synapses using Joint Poisson Deconvolution and Diffeomorphic Registration

Shashwat Kumar , Dominic M. Padova , Binish Narang , Gabrielle I. Coste , Austin R. Graves , Richard L. Huganir , Adam S. Charles , Michael I. Miller
show 1 more author
Anuj Srivastava
This is my paper

Pith reviewed 2026-05-14 20:37 UTC · model grok-4.3

classification 💻 cs.CV
keywords Bayesian trackingsynapse imagingdiffeomorphic registrationPoisson deconvolutionin vivo microscopypoint source modelinglongitudinal imagingfluorescence microscopy
0
0 comments X

The pith

A single Bayesian posterior simultaneously builds a probabilistic synapse template, denoises and deconvolves images, infers intensities, registers tissue motion diffeomorphically, and reports uncertainty for in vivo tracking.

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

The paper develops a unified Bayesian framework that models synapses as varying-luminance point sources whose positions follow a nonlinear tissue deformation. It combines this model with a Gaussian point-spread function and a Poisson observation model for raw photon counts to derive a joint posterior over all unknowns. The approach solves detection, denoising, intensity estimation, and registration in one optimization rather than as separate steps. A sympathetic reader would care because it addresses the low signal-to-noise ratio, shot noise, and day-to-day tissue motion that normally make longitudinal synapse tracking unreliable in dense brain regions.

Core claim

By deriving a posterior that incorporates a diffeomorphic mapping for domain warping, a Gaussian point spread function for the imaging process, and a Poisson observation model for raw photon counts, the Bayesian solution simultaneously constructs a probabilistic template of synapse locations, denoises and deconvolves the image data, infers fluorescence intensities, performs diffeomorphic image registration to correct for tissue motion, and provides confidence regions for these parameter estimates.

What carries the argument

The joint posterior over synapse template locations, intensities, and diffeomorphic deformation parameters, under a Poisson likelihood with Gaussian point-spread function.

If this is right

  • The method produces uncertainty maps around each synapse location and intensity without separate post-processing.
  • It enables direct comparison of the same synapses across multiple imaging sessions in living animals.
  • Joint optimization reduces error propagation that occurs when registration and detection are performed sequentially.
  • The probabilistic template can be updated incrementally as new time points arrive.
  • The framework applies to both simulated 2D+t data and real 3D+t in vivo mouse datasets spanning two weeks.

Where Pith is reading between the lines

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

  • The same joint model could be tested on other sparse fluorescent structures such as dendritic spines or axonal boutons.
  • If the diffeomorphic assumption holds across longer time scales, the approach might support studies linking synaptic turnover to behavior.
  • Replacing the Gaussian PSF with a measured or learned kernel would be a direct extension that preserves the rest of the pipeline.
  • The confidence regions could serve as input for downstream statistical tests on synaptic plasticity across animals.

Load-bearing premise

Synapses can be treated as point sources whose motion is fully captured by one diffeomorphic tissue deformation, with the microscope blur well approximated by a Gaussian and photon counts following Poisson statistics.

What would settle it

A simulation or real dataset in which known synapse positions deviate systematically from any single diffeomorphic warp, or in which the recovered template fails to match ground-truth locations once non-Poisson noise or non-Gaussian blur is introduced.

Figures

Figures reproduced from arXiv: 2605.13455 by Adam S. Charles, Anuj Srivastava, Austin R. Graves, Binish Narang, Dominic M. Padova, Gabrielle I. Coste, Michael I. Miller, Richard L. Huganir, Shashwat Kumar.

Figure 1
Figure 1. Figure 1: Longitudinal in vivo synapse imaging is noisy and exhibits nonlinear motion. (a) Raw XY slices across days show low signal-to-noise ratio and substantial shot noise. (b) Output of a denoising autoencoder (XTC). Insets visualize synaptic motion by comparing their positions with first day. between observations. Additionally, longitudinal deep tissue imaging requires lower laser power to avoid photobleaching … view at source ↗
Figure 2
Figure 2. Figure 2: Failure modes of the denoising-segmentation-tracking pipeline. Multiple XY slices at [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Accurate recovery of signal and motion in simulated data. Observed photon counts [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Joint modeling produces temporally coherent intensity estimates. Observed photon counts [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Stabilized Tracks, Displacement Fields and motion under the Displacement Field. [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Posterior distributions concentrate around ground truth. Marginal posteriors over template [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Posterior distributions obtained from model. [PITH_FULL_IMAGE:figures/full_fig_p013_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Degenerate case in Simulated data. Fluorescence error: 697/2000. [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
read the original abstract

Synapses are densely packed submicron structures that dynamically reorganize during learning and memory formation. Longitudinal \textit{in vivo} imaging of fluorescently tagged synaptic receptors offers a promising opportunity to study large-scale synaptic dynamics and how these processes are disrupted in neurological disease. However, in vivo imaging with 2-photon microscopy uses low laser power and therefore suffers from low signal-to-noise ratio (SNR) and high shot noise, nonlinear tissue motion between days, nonstationary fluctuations in synaptic fluorescence, and significant blur induced by the microscope point spread function (PSF). Together, these factors make it challenging to detect and track synapses, especially in regions with high synaptic density. This paper presents a novel template-based framework for modeling synapses as varying luminance point sources that move under a nonlinear tissue deformation. Taking a unified Bayesian approach, we apply this model to microscopy data by deriving a posterior that incorporates a diffeomorphic mapping for domain warping, a Gaussian point spread function for the imaging process, and a Poisson observation model for raw photon counts. The Bayesian solution simultaneously: (1) Constructs a probabilistic template of synapse locations, (2) denoises and deconvolves the image data, (3) infers fluorescence intensities, (4) performs diffeomorphic image registration to correct for tissue motion, and (5) provides confidence regions for these parameter estimates. We demonstrate the framework on both a 2D+t simulated dataset and a 3D+t longitudinal \textit{in vivo} microscopy dataset of fluorescent synapses imaged in a mouse over two weeks.

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

0 major / 3 minor

Summary. The manuscript presents a Bayesian framework for longitudinal in vivo synapse tracking in 2-photon microscopy. Synapses are modeled as varying-luminance point sources whose positions are governed by a probabilistic template; tissue motion is captured by a diffeomorphic deformation field; imaging uses a Gaussian PSF and Poisson photon-count likelihood. The joint posterior is derived to simultaneously construct the template, perform deconvolution and denoising, infer intensities, register images, and supply credible regions. The approach is demonstrated on a simulated 2D+t dataset and a real 3D+t longitudinal dataset acquired over two weeks in mouse cortex.

Significance. If the numerical results hold, the work would provide a coherent probabilistic treatment of several coupled tasks that are currently handled separately in synaptic imaging pipelines. The use of standard components (Poisson likelihood, Gaussian PSF, diffeomorphic group action) together with a point-source template prior yields a well-defined posterior without hidden circularity or identifiability collapse. The reported existence of feasible inference on both simulated and real 3D+t data is a useful existence proof; the availability of confidence regions would be directly useful for downstream biological analysis of synaptic turnover.

minor comments (3)
  1. [Abstract] Abstract: the claim that the method 'simultaneously' achieves all five listed tasks is strong; the abstract should briefly indicate which quantitative metrics (e.g., localization error, registration Dice, intensity correlation) support each claim on the real data.
  2. [Methods] Methods section (inferred from derivation): the specific form of the diffeomorphic velocity field parameterization and the numerical scheme used to sample or optimize the joint posterior should be stated explicitly, including any hyper-parameter settings that were held fixed across experiments.
  3. [Results] Results: while demonstrations on simulated and real data are reported, the manuscript should include at least one ablation (e.g., with vs. without the diffeomorphic term) and direct numerical comparison to a standard sequential pipeline (template estimation followed by separate registration) to quantify the benefit of the joint formulation.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive summary and recommendation of minor revision. We are pleased that the unified Bayesian treatment of joint deconvolution, registration, and template inference is viewed as providing a coherent probabilistic framework without identifiability issues, and that the availability of credible regions is recognized as useful for downstream biological analysis.

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained from stated model

full rationale

The paper constructs a joint posterior over a probabilistic synapse template, per-synapse intensities, and diffeomorphic deformation field by direct application of the imaging model (Gaussian PSF plus Poisson likelihood) and the deformation group action to the point-source template prior. All components are defined independently of the target dataset; no parameter is fitted to a subset of the data and then renamed as a prediction, no self-citation supplies a load-bearing uniqueness theorem or ansatz, and the central claims follow from the standard Bayesian construction without reduction to input quantities by definition. Demonstrations on simulated and real data serve only as existence proofs of numerical feasibility, not as circular validation.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 0 invented entities

The central claim rests on several domain assumptions about synapse appearance and imaging physics; no new physical entities are postulated and no free parameters are explicitly named in the abstract.

axioms (3)
  • domain assumption Synapses modeled as varying luminance point sources
    Stated in the template-based framework description.
  • domain assumption Tissue motion captured by diffeomorphic deformation
    Used to correct nonlinear tissue motion between days.
  • domain assumption Imaging process uses Gaussian PSF and Poisson observation model
    Incorporated into the derived posterior.

pith-pipeline@v0.9.0 · 5616 in / 1369 out tokens · 47627 ms · 2026-05-14T20:37:23.529433+00:00 · methodology

discussion (0)

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