arxiv: 2601.17262 · v2 · submitted 2026-01-24 · ❄️ cond-mat.mtrl-sci · eess.IV

Recognition: no theorem link

Unsupervised segmentation and clustering workflow for efficient processing of 4D-STEM and 5D-STEM data

Serin Lee , Stephanie M. Ribet , Arthur R. C. McCray , Andrew Barnum , Jennifer A. Dionne , Colin Ophus

Authors on Pith no claims yet

Pith reviewed 2026-05-16 11:47 UTC · model grok-4.3

classification ❄️ cond-mat.mtrl-sci eess.IV

keywords 4D-STEM5D-STEMclusteringsegmentationdiffraction patternsdata compressionin situ microscopynanoparticle growth

0 comments

The pith

Clustering by local diffraction pattern similarity segments 4D-STEM data into contiguous domains.

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

The paper introduces an unsupervised clustering workflow for 4D-STEM and 5D-STEM datasets that groups positions according to similarity in their local diffraction patterns. This similarity metric defines closed contours around spatially connected regions of consistent crystallographic character. Averaging the patterns inside each cluster raises signal quality while shrinking the overall data size by orders of magnitude. The resulting compact representation supports fast, accurate extraction of orientation, phase, and strain maps, as shown on in-situ liquid-cell observations of gold nanoparticle growth. The workflow is presented as a scalable, general tool for handling high-dimensional scanning diffraction data across multiple experimental modalities.

Core claim

A clustering framework identifies crystallographically distinct domains from 4D-STEM datasets by using local diffraction-pattern similarity as a metric. The method extracts closed contours delineating spatially contiguous regions. This produces cluster-averaged diffraction patterns that improve signal quality while reducing data volume by orders of magnitude, enabling rapid and accurate orientation, phase, and strain mapping. The approach is demonstrated on in situ liquid-cell 4D-STEM data of gold nanoparticle growth and offered as a general route for spatially coherent segmentation and data compression.

What carries the argument

Local diffraction-pattern similarity metric that compares patterns at neighboring scan positions to define cluster boundaries and extract closed contours around contiguous regions.

If this is right

Data volume is reduced by orders of magnitude, making storage and downstream analysis of large 5D-STEM sequences practical.
Cluster-averaged diffraction patterns yield higher signal quality for orientation, phase, and strain calculations.
Spatially coherent segmentation enables quantitative mapping across entire in-situ time series without manual region selection.
The same similarity-driven contour extraction applies to multiple 4D-STEM modalities beyond liquid-cell experiments.

Where Pith is reading between the lines

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

The method could be embedded in real-time acquisition software to flag emerging domains during scanning and adjust scan parameters on the fly.
Similar pattern-similarity clustering might transfer to other high-dimensional diffraction or spectroscopy datasets that lack explicit spatial labels.
Compressed cluster representations could serve as a standardized format for sharing large 4D-STEM archives while preserving quantitative diffraction information.

Load-bearing premise

Local diffraction-pattern similarity reliably identifies crystallographically distinct domains without being misled by noise, overlapping patterns, or gradual structural transitions.

What would settle it

Apply the workflow to a 4D-STEM dataset containing a known gradual structural transition and check whether the output clusters produce artificially sharp boundaries instead of reflecting the continuous change.

Figures

Figures reproduced from arXiv: 2601.17262 by Andrew Barnum, Arthur R. C. McCray, Colin Ophus, Jennifer A. Dionne, Serin Lee, Stephanie M. Ribet.

**Figure 2.** Figure 2: Applying clustering to the 4D-STEM dataset. a. Virtual dark field image of the Au nanoparticles formed by the electron-beam [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗

**Figure 3.** Figure 3: Comparing diffraction patterns with and without clustering. (a) Virtual dark-field image with colored markers indicating the six [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗

**Figure 4.** Figure 4: Bragg disk detection with clustering. Bragg disk detector (a) and orientation mapping by ACOM template matching (b) on [PITH_FULL_IMAGE:figures/full_fig_p005_4.png] view at source ↗

**Figure 5.** Figure 5: Orientation and strain mapping with (a) and without clustering (b). a. (left) In-plane orientation map, (middle) out-of-plane [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗

read the original abstract

Four-dimensional scanning transmission electron microscopy (4D-STEM) enables mapping of diffraction information with nanometer-scale spatial resolution, offering detailed insight into local structure, orientation, and strain. However, as data dimensionality and sampling density increase, particularly for in situ scanning diffraction experiments (5D-STEM), robust segmentation of structurally consistent behavior across sequential measurements becomes essential for efficient and physically meaningful analysis. Here, we introduce a clustering framework that identifies crystallographically distinct domains from 4D-STEM datasets. By using local diffraction-pattern similarity as a metric, the method extracts closed contours delineating spatially contiguous regions. This approach produces cluster-averaged diffraction patterns that improve signal quality while reducing data volume by orders of magnitude, enabling rapid and accurate orientation, phase, and strain mapping. We demonstrate the applicability of this approach to in situ liquid-cell 4D-STEM data of gold nanoparticle growth. Our method provides a scalable and generalizable route for spatially coherent segmentation, data compression, and quantitative structure-strain mapping across diverse 4D-STEM modalities. The full analysis code and example workflows are publicly available to support reproducibility and reuse.

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.

Desk Editor's Note private letter to a colleague

A practical unsupervised clustering workflow for 4D-STEM segmentation with public code, though short on quantitative benchmarks.

read the letter

This paper gives a clustering workflow that groups 4D-STEM diffraction patterns by local similarity and pulls out closed spatial contours for distinct domains. They apply it to in-situ liquid-cell data on gold nanoparticle growth to support orientation, phase, and strain mapping while cutting data volume through cluster averaging. The code and workflows are released publicly, which is the most immediately useful part for anyone who needs to process these large datasets quickly. What is new is the specific combination of similarity-based clustering with contour extraction tailored to 4D-STEM and 5D-STEM in-situ experiments, where manual or supervised segmentation becomes impractical. The approach is straightforward and matches the practical bottleneck described in the abstract. It does well at laying out the motivation and showing a working demonstration on real nanoparticle growth data. The method produces averaged patterns that should improve signal quality, and the emphasis on scalability for high-dimensional scans is on target. The main soft spots are the lack of quantitative metrics, error bars, or direct comparisons against standard clustering or existing STEM segmentation tools. The orders-of-magnitude data reduction claim is stated but not backed by specific numbers or timing results here. The stress-test note about closed contours is worth checking in the full text: plain similarity clustering often produces disconnected labels, so the paper needs to show exactly how spatial connectivity is enforced, whether through regularization, post-processing, or another step. Without that detail or robustness tests against noise and overlapping patterns, it is hard to judge how reliably the contours hold up. This work is for materials researchers and microscopists who run 4D-STEM experiments and need faster ways to segment large in-situ volumes. A reader who processes these datasets would get value from the workflow and the open code even before full validation. It deserves peer review because the core idea is coherent, the demo is relevant, and the code release supports reuse, though the referees will likely ask for the missing benchmarks and connectivity details.

Referee Report

3 major / 1 minor

Summary. The manuscript introduces an unsupervised clustering workflow for segmenting 4D-STEM and 5D-STEM datasets. It employs local diffraction-pattern similarity as a metric to identify crystallographically distinct domains, extracts closed contours around spatially contiguous regions, and generates cluster-averaged diffraction patterns that improve signal quality while compressing data volume by orders of magnitude. This enables efficient orientation, phase, and strain mapping. The method is demonstrated on in situ liquid-cell 4D-STEM data of gold nanoparticle growth, with full analysis code and workflows made publicly available.

Significance. If the similarity-to-contour mapping and data-reduction claims hold under quantitative scrutiny, the workflow could substantially accelerate analysis of large in situ 4D-STEM datasets by providing a scalable, generalizable route to spatially coherent segmentation and improved signal-to-noise via averaging. The public release of code and example workflows is a clear strength that supports reproducibility and reuse in the materials-science community.

major comments (3)

[Abstract] Abstract: the central claim that local diffraction-pattern similarity alone 'extracts closed contours delineating spatially contiguous regions' is load-bearing for the entire workflow, yet the description supplies no indication of spatial regularization, graph-cut, watershed, or connected-component post-processing; standard similarity clustering (Euclidean or cosine distance on flattened patterns) typically produces disconnected or noisy label fields, so the mapping from similarity to closed contours remains unverified.
[Abstract] Abstract: the assertion of 'reducing data volume by orders of magnitude' lacks any supporting quantitative metrics, raw-vs-compressed size ratios, timing benchmarks, or comparison against baselines (e.g., raw 4D-STEM storage or alternative compression schemes), rendering the efficiency claim impossible to evaluate.
[Results/Demonstration] Demonstration on liquid-cell gold-nanoparticle data: no error analysis, segmentation accuracy metrics (e.g., overlap with manual labels), or robustness tests against beam-induced motion and overlapping patterns are reported, leaving the weakest assumption—that similarity reliably delineates domains without fragmentation or leakage—unquantified.

minor comments (1)

[Abstract] Abstract: the distinction between 4D-STEM and 5D-STEM is introduced without a concise definition of the additional dimension; a single clarifying sentence would improve accessibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed review of our manuscript. We address each major comment point by point below, providing clarifications from the full text and indicating where revisions have been made to improve clarity, quantification, and validation.

read point-by-point responses

Referee: [Abstract] Abstract: the central claim that local diffraction-pattern similarity alone 'extracts closed contours delineating spatially contiguous regions' is load-bearing for the entire workflow, yet the description supplies no indication of spatial regularization, graph-cut, watershed, or connected-component post-processing; standard similarity clustering (Euclidean or cosine distance on flattened patterns) typically produces disconnected or noisy label fields, so the mapping from similarity to closed contours remains unverified.

Authors: We agree that the abstract, as a concise summary, does not detail the post-processing steps. The full manuscript (Methods section) specifies that similarity-based clustering is followed by connected-component labeling on the resulting label field to identify spatially contiguous regions and extract closed contours. This step enforces spatial coherence and prevents disconnected labels. To address the concern, we have revised the abstract to include a brief clause referencing the connected-component post-processing that maps similarity clusters to closed contours. revision: yes
Referee: [Abstract] Abstract: the assertion of 'reducing data volume by orders of magnitude' lacks any supporting quantitative metrics, raw-vs-compressed size ratios, timing benchmarks, or comparison against baselines (e.g., raw 4D-STEM storage or alternative compression schemes), rendering the efficiency claim impossible to evaluate.

Authors: We acknowledge that the efficiency claim requires quantitative support. In the revised manuscript, we have added explicit metrics in the Results section: raw 4D-STEM dataset size versus the compressed representation (cluster-averaged patterns plus metadata), the achieved reduction factor, and wall-clock timing for the full workflow on the demonstration dataset. We also include a brief comparison to storing the uncompressed raw data. revision: yes
Referee: [Results/Demonstration] Demonstration on liquid-cell gold-nanoparticle data: no error analysis, segmentation accuracy metrics (e.g., overlap with manual labels), or robustness tests against beam-induced motion and overlapping patterns are reported, leaving the weakest assumption—that similarity reliably delineates domains without fragmentation or leakage—unquantified.

Authors: The referee correctly identifies the lack of quantitative segmentation metrics. For this in situ liquid-cell experiment, pixel-level ground-truth labels are unavailable due to the dynamic growth process and overlapping diffraction signals. We have added qualitative validation via expert visual comparison and new robustness tests (simulated beam motion and pattern overlap) in the supplementary information. These additions quantify consistency with physical expectations of nanoparticle domains. A full synthetic benchmark with overlap metrics is beyond the current scope but is noted as future work. revision: partial

Circularity Check

0 steps flagged

No circularity: direct application of similarity clustering to 4D-STEM data

full rationale

The paper presents an algorithmic workflow that applies standard similarity metrics (e.g., local diffraction-pattern comparison) followed by clustering to identify domains and produce averaged patterns. No equations, parameters, or derivations are defined in terms of their own outputs. The extraction of closed contours is described as a direct consequence of the similarity-based segmentation step without any self-referential fitting loop, self-citation load-bearing the central claim, or renaming of known results. The method is self-contained as a data-processing pipeline whose outputs (cluster-averaged patterns, reduced data volume) follow logically from the input data and chosen similarity metric without reduction to fitted inputs by construction.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review; no explicit free parameters, axioms, or invented entities are stated. The method appears to rest on standard clustering algorithms applied to diffraction data without new postulates.

pith-pipeline@v0.9.0 · 5528 in / 1047 out tokens · 54734 ms · 2026-05-16T11:47:16.655049+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

15 extracted references · 15 canonical work pages

[1]

Four-dimensional scanning transmission electron microscopy (4d-stem): From scanning nanodiffraction to ptychography and beyond.Microscopy and Microanalysis, 25 (3):563–582, 2019

Colin Ophus. Four-dimensional scanning transmission electron microscopy (4d-stem): From scanning nanodiffraction to ptychography and beyond.Microscopy and Microanalysis, 25 (3):563–582, 2019. Edgar F Rauch, Joaquin Portillo, Stavros Nicolopoulos, Daniel

work page 2019
[2]

Bultreys, Sergei Rouvimov, and Peter Moeck. Automated nanocrystal orientation and phase mapping in the transmission electron microscope on the basis of precession electron diffraction.Zeitschrift f¨ ur Kristallographie, 225(2-3):103–109, 2010. Colin Ophus, Steven E Zeltmann, Alexandra Bruefach, Alexander

work page 2010
[3]

Automated crystal orientation mapping in py4dstem using sparse correlation matching.Microscopy and microanalysis, 28(2):390–403, 2022

Rakowski, Benjamin H Savitzky, Andrew M Minor, and Mary C Scott. Automated crystal orientation mapping in py4dstem using sparse correlation matching.Microscopy and microanalysis, 28(2):390–403, 2022. Niels Cautaerts, Phillip Crout, H˚ akon W ˚Anes, Eric Prestat, Jiwon Jeong, Gerhard Dehm, and Christian H Liebscher. Free, flexible and fast: Orientation map...

work page 2022
[4]

Accurate measurement of strain at interfaces in 4d-stem: A comparison of various methods.Ultramicroscopy, 221:113196, 2021

Krause, Marco Schowalter, and Andreas Rosenauer. Accurate measurement of strain at interfaces in 4d-stem: A comparison of various methods.Ultramicroscopy, 221:113196, 2021. Ambarneil Saha, Alexander J Pattison, Karen C Bustillo, David W

work page 2021
[5]

Reuniting crystallography with real space: Ab initio structure elucidation with 4d-stem.Proceedings of the National Academy of Sciences, 122(42):e2508185122, 2025

Mittan-Moreau, Aaron S Brewster, Jian Zhang, and Peter Ercius. Reuniting crystallography with real space: Ab initio structure elucidation with 4d-stem.Proceedings of the National Academy of Sciences, 122(42):e2508185122, 2025. Chang Liu, Oliver Lin, Saran Pidaparthy, Haoyang Ni, Zhiheng

work page 2025
[6]

4d-stem mapping of nanocrystal reaction dynamics and heterogeneity in a graphene liquid cell.Nano letters, 24(13):3890–3897, 2024

Lyu, Jian-Min Zuo, and Qian Chen. 4d-stem mapping of nanocrystal reaction dynamics and heterogeneity in a graphene liquid cell.Nano letters, 24(13):3890–3897, 2024. Sungin Kim, Valentin Briega-Martos, Shikai Liu, Kwanghwi Je, Chuqiao Shi, Katherine Marusak Stephens, Steven E Zeltmann, Zhijing Zhang, Rafael Guzman-Soriano, Wenqi Li, et al. Operando heating...

work page 2024
[7]

Non-negative matrix factorization for mining big data obtained using four-dimensional scanning transmission electron microscopy

Nagai, Kazutaka Mitsuishi, and Koji Kimoto. Non-negative matrix factorization for mining big data obtained using four-dimensional scanning transmission electron microscopy. Ultramicroscopy, 221:113168, 2021. Frances I Allen, Thomas C Pekin, Arun Persaud, Steven J

work page 2021
[8]

Fast grain mapping with sub-nanometer resolution using 4d-stem with grain classification by principal component analysis and non-negative matrix factorization

Rozeveld, Gregory F Meyers, Jim Ciston, Colin Ophus, and Andrew M Minor. Fast grain mapping with sub-nanometer resolution using 4d-stem with grain classification by principal component analysis and non-negative matrix factorization. Microscopy and microanalysis, 27(4):794–803, 2021. Koji Kimoto, Fumihiko Uesugi, Koji Harano, Jun Kikkawa, Ovidiu Cretu, Yuk...

work page 2021
[9]

End-to- end automated segmentation framework for four-dimensional scanning transmission electron microscopy data.Microscopy and Microanalysis, 31(5):ozaf094, 2025

Dravid, Wei Chen, and Daniel W Apley. End-to- end automated segmentation framework for four-dimensional scanning transmission electron microscopy data.Microscopy and Microanalysis, 31(5):ozaf094, 2025. Zhiquan Kho, Andy Bridger, Keith Butler, Ercin C Duran, Mohsen Danaie, and Alexander S Eggeman. On the use of clustering workflows for automated microstruc...

work page 2025
[10]

Assessment of active dopants and p–n junction abruptness using in situ biased 4d-stem.Nano Letters, 22(23): 9544–9550, 2022

Okuno, Jean-Luc Rouviere, David Cooper, and Martien Ilse Den Hertog. Assessment of active dopants and p–n junction abruptness using in situ biased 4d-stem.Nano Letters, 22(23): 9544–9550, 2022. 8 Lee et al. Robert Winkler, Alexander Zintler, Oscar Recalde-Benitez, Tianshu Jiang, D´ espina Nasiou, Esmaeil Adabifiroozjaei, Philipp Schreyer, Taewook Kim, Esz...

work page 2022
[11]

In situ 4d-stem imaging of the orientation of lamellar clusters in polymer crystallization.Macromolecular Rapid Communications, page e00450, 2025

Ciston, Brooks A Abel, Xi Jiang, Nitash P Balsara, and Andrew M Minor. In situ 4d-stem imaging of the orientation of lamellar clusters in polymer crystallization.Macromolecular Rapid Communications, page e00450, 2025. Yongwen Sun, Ying Han, Dan Zhou, Hugo Perez Garza, Alejandro Gomez Perez, Thanos Galanis, Starvos Nicolopoulos, and Yang Yang. In-situ 4d-s...

work page 2025
[12]

In situ grain growth experiments: Tem imaging & automated segmentation with correlative 4d-stem orientation mapping.Microscopy and Microanalysis, 31(Supplement 1):ozaf048–909, 2025

Ma, Jeffrey M Rickman, and Katayun Barmak. In situ grain growth experiments: Tem imaging & automated segmentation with correlative 4d-stem orientation mapping.Microscopy and Microanalysis, 31(Supplement 1):ozaf048–909, 2025. Delphic Chen and Jui-Chao Kuo. Bilateral filter based orientation smoothing of ebsd data.Ultramicroscopy, 110(10):1297–1305, 2010. S...

work page 2025
[13]

From STEM to 4D STEM: Ultrafast diffraction mapping with a hybrid-pixel detector.Microscopy Today, 31(2):10–14, 2023

Zambon, Darya Bachevskaya, Herv´ e Remigy, Clemens Schulze- Briese, and Luca Piazza. From STEM to 4D STEM: Ultrafast diffraction mapping with a hybrid-pixel detector.Microscopy Today, 31(2):10–14, 2023. Stefan Van der Walt, Johannes L Sch¨ onberger, Juan Nunez-

work page 2023
[14]

scikit-image: image processing in python.PeerJ, 2:e453, 2014

Iglesias, Fran¸ cois Boulogne, Joshua D Warner, Neil Yager, Emmanuelle Gouillart, and Tony Yu. scikit-image: image processing in python.PeerJ, 2:e453, 2014. Benjamin H Savitzky, Steven E Zeltmann, Lauren A Hughes, Hamish G Brown, Shiteng Zhao, Philipp M Pelz, Thomas C

work page 2014
[15]

py4dstem: A software package for four- dimensional scanning transmission electron microscopy data analysis.Microscopy and Microanalysis, 27(4):712–743, 2021

Pekin, Edward S Barnard, Jennifer Donohue, Luis Rangel DaCosta, et al. py4dstem: A software package for four- dimensional scanning transmission electron microscopy data analysis.Microscopy and Microanalysis, 27(4):712–743, 2021. Stephanie M Ribet, Rohan Dhall, Colin Ophus, and Karen C Bustillo. Multi-angle precession electron diffraction (maped): a versat...

work page 2021