pith. machine review for the scientific record. sign in

arxiv: 2604.24769 · v1 · submitted 2026-04-15 · ⚛️ physics.app-ph · cond-mat.mtrl-sci· physics.ins-det

Recognition: unknown

SPARSE -- Efficient High-Resolution SEM Imaging of Rare Microstructural Features Across Large Areas by Selective Rescanning

Tom Reclik , Jan Gerlach , Maximilian A. Wollenweber , Yannis P. Korkolis , Sandra Korte-Kerzel , Ulrich Kerzel
Authors on Pith no claims yet

Pith reviewed 2026-05-10 12:11 UTC · model grok-4.3

classification ⚛️ physics.app-ph cond-mat.mtrl-sciphysics.ins-det
keywords SEM imagingmicrostructural featuresselective rescanningrare featurestwo-stage scanningdamage detectionacquisition timeopen-source framework
0
0 comments X

The pith

A two-stage SEM framework identifies rare microstructural features with a fast scan then rescans only those regions at high resolution.

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

The paper introduces an open-source Python framework that solves the problem of excessively long acquisition times when imaging large areas at high resolution to capture rare features in materials. It works by running a quick low-resolution scan to locate candidate regions, then selectively rescanning only those regions with parameters suited for quantitative analysis. Separate processes handle scanning, detection, and rescanning through queues so that computation does not add to total time. Validation on damage detection in dual-phase steel demonstrates 99 percent feature detection at roughly 58 percent of conventional acquisition time and 95 percent detection at 19 percent. The approach makes detailed characterization of infrequent microstructures feasible over much larger sample areas.

Core claim

The framework defines a generic microscope interface and a modular detection interface to enable a two-stage workflow: a fast scan identifies regions of interest that are then selectively rescanned at high resolution. Parallel execution via separate processes and queue-based communication ensures detection overhead does not increase acquisition time. On a Tescan Clara SEM examining damage in DP800 steel, representative settings achieve 99 percent detection at approximately 58 percent of standard time and 95 percent detection at 19 percent, with time savings expressed as lower bounds from the ratio of scanned pixels.

What carries the argument

The two-stage selective rescanning workflow that uses a fast overview scan to locate regions of interest for targeted high-resolution follow-up imaging.

If this is right

  • Large-area high-resolution characterization of rare features becomes practical because total acquisition time scales with the number of detected regions rather than the full area.
  • The modular interfaces allow the same framework to work with different SEM platforms and alternative detection algorithms without rewriting core scanning logic.
  • Parallel processing of detection keeps computational cost from extending the physical scan duration.
  • Time savings scale directly with how sparse the features are, since fewer pixels require high-resolution acquisition.
  • The lower-bound estimates based on pixel ratios indicate that actual savings could be even larger once minor overheads are reduced.

Where Pith is reading between the lines

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

  • The same selective-rescan logic could be applied to other imaging modalities such as optical microscopy or atomic force microscopy when rare surface features must be examined over large areas.
  • Reduced acquisition time would enable statistical sampling over sample sizes that are currently impractical, improving the reliability of microstructure statistics.
  • Integration with real-time analysis could allow the detection step to adapt scan parameters on the fly for even higher efficiency.
  • The open-source release would let researchers test the framework on their own instruments and feature types to quantify time savings for specific use cases.

Load-bearing premise

The initial fast scan together with the chosen detection method will locate essentially all rare microstructural features of interest without missing any that would need high-resolution imaging.

What would settle it

A controlled test that reveals one or more known rare features present in the sample but undetected by the fast scan and detection step, resulting in incomplete high-resolution coverage.

Figures

Figures reproduced from arXiv: 2604.24769 by Jan Gerlach, Maximilian A. Wollenweber, Sandra Korte-Kerzel, Tom Reclik, Ulrich Kerzel, Yannis P. Korkolis.

Figure 1
Figure 1. Figure 1: Overview of the automated scanning framework SPARSE. From left to right: the panoramic imaging strategy defines the tile grid [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Flowchart of the scanning procedure. Blue boxes indi [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Parallelized workflow for a single tile divided into four partitions. Blue indicates SEM operations (fast scanning or rescanning); orange [PITH_FULL_IMAGE:figures/full_fig_p005_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Schematic representation of the scanned area on the left [PITH_FULL_IMAGE:figures/full_fig_p008_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Detection performance for τ = 55, ε = 1, nmin = 5. (a) Pareto front showing the trade-off between detection rate and efficiency. The dashed line indicates the theoretical maximum efficiency assuming perfect detection of all ground truth clusters. (b) Detection rate as a function of cluster size for three operating points corresponding to 99%, 95%, and 90% overall detection rate. Lines connecting data point… view at source ↗
Figure 6
Figure 6. Figure 6: Comparison of rescan regions for two operating points on the Pareto front. Outlines of proposed rescan regions are shown in yellow, [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
read the original abstract

Characterisation of rare microstructural features in scanning electron microscopy (SEM) requires imaging large areas at high resolution. This leads to prohibitively long acquisition times. We present an open-source Python framework that addresses this bottleneck through a two-stage approach: a fast scan identifies regions of interest, which are then selectively rescanned with imaging parameters suitable for quantitative analysis. The framework defines a generic microscope interface and a modular detection interface, allowing adaptation to different microscope platforms and detection methods. Scanning, detection, and rescanning are parallelized using separate processes, ensuring that computation time does not extend acquisition time. The two processes communicate exclusively through queues, avoiding shared mutable state and eliminating the need for explicit synchronization. We validate the framework on damage detection in dual-phase DP800 steel using a Tescan Clara SEM. For a representative configuration a detection rate of 99 % is achieved at approximately 58 % of the conventional acquisition time. At 95 % detection rate, acquisition time drops to 19 %. These time savings estimates represent lower bounds based on the ratio of scanned pixels. The complete implementation will be made available upon publication and upon request during peer-review.

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

1 major / 2 minor

Summary. The manuscript presents SPARSE, an open-source Python framework for efficient high-resolution SEM imaging of rare microstructural features over large areas. It implements a two-stage workflow: a fast low-resolution scan identifies candidate regions of interest using a modular detection interface, followed by targeted high-resolution rescanning of those regions. The system uses separate processes for scanning, detection, and rescanning that communicate exclusively via queues to enable parallelization without shared mutable state or added acquisition time. Validation is reported on damage detection in dual-phase DP800 steel using a Tescan Clara SEM, with a representative configuration achieving 99% detection at ~58% of conventional acquisition time and 95% detection at 19% time; these savings are characterized as lower bounds based on the ratio of scanned pixels.

Significance. If the fast initial scan plus chosen detector reliably locates essentially all instances of the target rare features, the framework could meaningfully reduce acquisition times for large-area quantitative SEM characterization in materials science. Notable strengths include the generic microscope interface for platform adaptation, the modular detection interface, the queue-based parallelization that keeps computation off the critical path, and the commitment to open-source release. The concrete empirical results on a commercial instrument provide practical evidence of utility for one specific use case.

major comments (1)
  1. Validation section: the reported 99% detection rate at 58% acquisition time (and 95% at 19%) for DP800 steel damage is presented as the central performance result, yet the manuscript supplies no description of the detection algorithm, its parameters/thresholds, false-negative rates, or direct comparison against exhaustive high-resolution ground truth across the full scanned area. This assumption—that the fast scan locates every relevant rare feature—is load-bearing for the efficiency claims; without such characterization the time savings cannot be guaranteed to correspond to complete population coverage.
minor comments (2)
  1. Abstract: the statement that 'the complete implementation will be made available upon publication' would be strengthened by including a repository URL or DOI even if under embargo during review.
  2. The manuscript would benefit from a concise flowchart or pseudocode in the methods section illustrating the queue-based inter-process communication to clarify how parallelism is achieved without synchronization primitives.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for their constructive feedback on our manuscript. The single major comment raises an important point about the characterization of detection performance in the validation experiments. We address it directly below and will revise the manuscript accordingly.

read point-by-point responses

  1. Referee: Validation section: the reported 99% detection rate at 58% acquisition time (and 95% at 19%) for DP800 steel damage is presented as the central performance result, yet the manuscript supplies no description of the detection algorithm, its parameters/thresholds, false-negative rates, or direct comparison against exhaustive high-resolution ground truth across the full scanned area. This assumption—that the fast scan locates every relevant rare feature—is load-bearing for the efficiency claims; without such characterization the time savings cannot be guaranteed to correspond to complete population coverage.

    Authors: We agree that the validation section would benefit from greater detail on the detection component. In the revised manuscript we will add a dedicated subsection describing the specific detection algorithm employed for the DP800 damage features (including whether it is threshold-based, edge-detection, or a simple machine-learning classifier), all tunable parameters and thresholds used, and the procedure for estimating false-negative rates. These details were omitted from the original submission for brevity but are available from our experimental records. With respect to exhaustive high-resolution ground truth across the entire scanned area, we note that acquiring such a dataset would require a conventional full-area high-resolution scan—the very procedure whose time cost the framework is designed to reduce. Instead, detection performance was quantified by (i) manual expert review of all rescanned high-resolution images and (ii) comparison against a smaller set of independently acquired high-resolution reference images covering representative regions. We will make this evaluation protocol explicit, report the resulting false-negative statistics, and emphasize that the reported time savings are therefore lower bounds conditioned on the observed detection rate rather than an absolute guarantee of complete coverage. If the referee considers these additions insufficient, we are prepared to include additional supplementary figures showing example detections and missed features. revision: yes

Circularity Check

0 steps flagged

No circularity; empirical performance metrics from direct validation

full rationale

The paper introduces a two-stage SEM framework (fast scan + selective rescanning) and reports measured detection rates (99% at 58% time, 95% at 19%) from a single experimental configuration on DP800 steel. These are direct empirical outcomes, not predictions derived from equations, fitted parameters, or self-citations. No mathematical derivation chain exists, no parameters are fitted then renamed as predictions, and no load-bearing claims reduce to self-referential definitions or prior author work. The result is self-contained against external benchmarks via the reported pixel-ratio time savings and detection counts.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The framework rests on the domain assumption that a fast low-resolution scan plus detection algorithm can serve as a reliable proxy for identifying all regions needing high-resolution quantitative imaging; no free parameters or invented entities are introduced in the abstract.

axioms (1)
  • domain assumption A fast low-resolution scan combined with the detection method will locate essentially all rare features of interest.
    Stated implicitly in the two-stage approach and validation claims.

pith-pipeline@v0.9.0 · 5539 in / 1154 out tokens · 32762 ms · 2026-05-10T12:11:10.430011+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

22 extracted references · 18 canonical work pages

  1. [1]

    C. C. Tasan, M. Diehl, D. Yan, M. Bech- told, F. Roters, L. Schemmann, C. Zheng, N. Peranio, D. Ponge, M. Koyama, K. Tsuzaki, D. Raabe, An Overview of Dual-Phase Steels: Advances in Microstructure-Oriented Process- ing and Micromechanically Guided Design, An- nual Review of Materials Research 45 (V olume 45, 2015) (2015) 391–431.doi:10.1146/ annurev-matsc...

    2015

  2. [2]

    PLOS ONE9(12), 113490 (2014) https://doi.org/10.1371/journal

    C. Kusche, T. Reclik, M. Freund, T. Al- Samman, U. Kerzel, S. Korte-Kerzel, Large- area, high-resolution characterisation and clas- sification of damage mechanisms in dual-phase steel using deep learning, PLOS ONE 14 (5) (2019) e0216493.doi:10.1371/journal. pone.0216493

    work page doi:10.1371/journal 2019

  3. [3]

    H. V . Atkinson, G. Shi, Characterization of inclu- sions in clean steels: a review including the statis- tics of extremes methods, Progress in Materials Science 48 (5) (2003) 457–520.doi:10.1016/ S0079-6425(02)00014-2

    2003

  4. [4]

    M. A. Wollenweber, C. F. Kusche, T. Al-Samman, S. Korte-Kerzel, On the automated characteri- sation of inclusion-induced damage in 16Mn- CrS5 case-hardening steel, Advances in Industrial and Manufacturing Engineering 7 (2023) 100123. doi:10.1016/j.aime.2023.100123

    work page doi:10.1016/j.aime.2023.100123 2023

  5. [5]

    Dahmen, M

    T. Dahmen, M. Engstler, C. Pauly, P. Trampert, N. de Jonge, F. Mücklich, P. Slusallek, Feature Adaptive Sampling for Scanning Electron Mi- croscopy, Scientific Reports 6 (1) (2016) 25350. doi:10.1038/srep25350

    work page doi:10.1038/srep25350 2016

  6. [6]

    Ultramicroscopy273, 114138 (2025) https://doi.org/10.1016/j.ultramic

    S. Clusiau, N. Piché, N. Brodusch, M. Strauss, R. Gauvin, Workflow automation of SEM acqui- sitions and feature tracking, Ultramicroscopy 269 (2025) 114093.doi:10.1016/j.ultramic. 2024.114093

    work page doi:10.1016/j.ultramic 2025

  7. [7]

    A. E. Tekkaya, N. Ben Khalifa, O. Hering, R. Meya, S. Myslicki, F. Walther, Forming- induced damage and its effects on product prop- erties, CIRP Annals 66 (1) (2017) 281–284.doi: 10.1016/j.cirp.2017.04.113

    work page doi:10.1016/j.cirp.2017.04.113 2017

  8. [8]

    Hering, A

    O. Hering, A. E. Tekkaya, Damage-induced per- formance variations of cold forged parts, Journal of Materials Processing Technology 279 (2020) 116556.doi:10.1016/j.jmatprotec.2019. 116556

    work page doi:10.1016/j.jmatprotec.2019 2020

  9. [9]

    Medghalchi, C

    S. Medghalchi, C. F. Kusche, E. Karimi, U. Kerzel, S. Korte-Kerzel, Damage Analysis in Dual-Phase Steel Using Deep Learning: Transfer from Uniaxial to Biaxial Straining Conditions by Image Data Augmentation, JOM 72 (12) (2020) 4420–4430, tLDR: Overall, the network perfor- mance could be greatly improved and an anal- ysis of changes in damage behavior, her...

    2020

  10. [10]

    Hannard, T

    F. Hannard, T. Pardoen, E. Maire, C. Le Bour- lot, R. Mokso, A. Simar, Characterization and mi- cromechanical modelling of microstructural het- erogeneity effects on ductile fracture of 6xxx alu- minium alloys, Acta Materialia 103 (2016) 558– 572.doi:10.1016/j.actamat.2015.10.008

    work page doi:10.1016/j.actamat.2015.10.008 2016

  11. [11]

    Avramovic-Cingara, Y

    G. Avramovic-Cingara, Y . Ososkov, M. K. Jain, D. S. Wilkinson, Effect of martensite distribution on damage behaviour in DP600 dual phase steels, Materials Science and Engineering: A 516 (1) (2009) 7–16.doi:10.1016/j.msea.2009.03. 055

    work page doi:10.1016/j.msea.2009.03 2009

  12. [12]

    E. E. A¸ sık, E. S. Perdahcıo ˘glu, A. H. van den Boogaard, Microscopic investigation of damage mechanisms and anisotropic evolution of damage in DP600, Materials Science and Engineering: A 739 (2019) 348–356.doi:10.1016/j.msea. 2018.10.018

    work page doi:10.1016/j.msea 2019

  13. [13]

    L. Zhou, H. Wen, I. C. Kuschnerus, S. L. Y . Chang, Efficientand Robust Automated Segmentation of Nanoparticles and Aggregates from Transmission Electron Microscopy Images with Highly Com- plex Backgrounds, Nanomaterials 14 (14) (2024) 1169.doi:10.3390/nano14141169

    work page doi:10.3390/nano14141169 2024

  14. [14]

    A. Shah, J. A. Schiller, I. Ramos, J. Serrano, D. K. Adams, S. Tawfick, E. Ertekin, Auto- mated image segmentation of scanning electron microscopy images of graphene using U-Net Neu- ral Network, Materials Today Communications 35 (2023) 106127.doi:10.1016/j.mtcomm.2023. 106127

    work page doi:10.1016/j.mtcomm.2023 2023

  15. [15]

    Rocklin, Dask: Parallel computation with blocked algorithms and task scheduling, SciPy 2015 (2015), doi:10.25080/Majora-7b98e3ed-013

    M. Rocklin, Dask: Parallel Computa- tion with Blocked algorithms and Task Scheduling, SciPy 2015 (Jun. 2015). doi:10.25080/Majora-7b98e3ed-013. 15

    work page doi:10.25080/majora-7b98e3ed-013 2015

  16. [16]

    Kolpak, H

    F. Kolpak, H. Traphöner, O. Hering, A. E. Tekkaya, Large strain flow curves of sheet met- als by sheet extrusion, CIRP Annals 70 (1) (2021) 247–250.doi:10.1016/j.cirp.2021.03.023

    work page doi:10.1016/j.cirp.2021.03.023 2021

  17. [17]

    Ester, H.-P

    M. Ester, H.-P. Kriegel, J. Sander, X. Xu, A density-based algorithm for discovering clusters in large spatial databases with noise, in: Proceedings of the Second International Conference on Knowl- edge Discovery and Data Mining, KDD’96, AAAI Press, Portland, Oregon, 1996, pp. 226–231

    1996

  18. [18]

    C. E. Duchon, Lanczos Filtering in One and Two Dimensions, Journal of Applied Meteorology and Climatology 18 (8) (1979) 1016–1022. doi:10.1175/1520-0450(1979)018<1016: LFIOAT>2.0.CO;2

    work page doi:10.1175/1520-0450(1979)018 1979

  19. [19]

    L. R. Dice, Measures of the Amount of Ecologic Association Between Species, Ecology 26 (3) (1945) 297–302, _eprint: https://esajournals.onlinelibrary.wiley.com/doi/pdf/10.2307/1932409. doi:10.2307/1932409

    work page doi:10.2307/1932409 1945

  20. [20]

    Reclik, S

    T. Reclik, S. Medghalchi, P. Schumacher, M. A. Wollenweber, T. Al-Samman, S. Korte-Kerzel, U. Kerzel, Resolution enhancement of scanning electron micrographs using artificial intelligence, Materials & Design 253 (2025) 113955.doi: 10.1016/j.matdes.2025.113955

    work page doi:10.1016/j.matdes.2025.113955 2025

  21. [21]

    Nicolás, The bar derived category of a curved dg algebra, Journal of Pure and Applied Algebra 212 (2008) 2633–2659

    S. Medghalchi, E. Karimi, S.-H. Lee, B. Berkels, U. Kerzel, S. Korte-Kerzel, Three-dimensional characterisation of deformation-induced damage in dual phase steel using deep learning, Materials & Design 232 (2023) 112108.doi:10.1016/j. matdes.2023.112108

    work page doi:10.1016/j 2023

  22. [22]

    Medghalchi, J

    S. Medghalchi, J. Kortmann, S.-H. Lee, E. Karimi, U. Kerzel, S. Korte-Kerzel, Automated segmen- tation of large image datasets using artificial intelligence for microstructure characterisation and damage analysis, Materials & Design 243 (2024) 113031.doi:10.1016/j.matdes.2024. 113031. 16 Appendix A. 0.6 0.8 1.0 0.4 0.5 0.6 0.7 0.8 0.9 1.0Detection Rate = ...

    work page doi:10.1016/j.matdes.2024 2024