Pith. sign in

REVIEW 3 major objections 6 minor 48 references

Smart membrane: high content in situ monitoring barrier on chip with artificial neural network

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Continuous impedance spectra from a nanoporous chip, read by lightweight neural networks, can identify endothelial barrier formation stages and drug-induced barrier changes in situ, without microscopy.

desk verdict Real engineering win; overclaimed biology — send to review with major-revision expectations. read the letter →

arxiv 2608.01239 v2 pith:UJJT2X42 submitted 2026-08-02 eess.SP q-bio.CB

Bo Tang , Victor Krajka , Mengxi Liu , Wei Zhao , Gazal Goekkus , Paul Lukowicz , Lili Zhu , Pu Chen
show 2 more authors
Stephan Reichl Andreas Dietzel
This is my paper
classification eess.SPq-bio.CB
keywords organ-on-chipECISbioelectricalimpedancespectroscopynanoporousmembranetight-junctionsensingendothelialbarrierformationConv1dKolmogorov-Arnoldnetwork
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper sets out to make ECIS impedance spectroscopy a complete substitute for microscope-based monitoring of endothelial barriers on chips. Its central claim is that Nyquist diagrams recorded continuously from a 700 nm nanoporous membrane chip contain enough information for a lightweight neural network to identify the four phases of endothelial monolayer formation—adherence, outspreading, confluence, and mature tight-junction barrier—with 95% confidence, and to distinguish reversible from irreversible drug-induced barrier weakening. The authors demonstrate this with HUVECs cultured on wafer-fabricated chips, training Conv1d and Kolmogorov-Arnold networks on a small number of impedance spectra labeled by bright-field and fluorescence microscopy. If the claim holds, barrier-on-chip experiments can be monitored automatically and non-invasively, and a permeation test can be started at the right moment without removing the chip for microscopy or endpoint staining.

What carries the argument

The load-bearing object is the nanoporous 'smart membrane' itself: a 700 nm ultra-low-stress SixNy membrane with 500 nm pores, monolithically integrated in a silicon chip, carrying coplanar comb electrodes that measure impedance from 1 kHz to 200 kHz at 5 mV excitation. The argument then rests on the Nyquist diagram, the complex-plane plot of the measured impedance, as a compact fingerprint of cell state; a color-coded Nyquist chronogram and two fitted equivalent-circuit parameters (Cdl and Rtight) make the phase progression visible, and t-SNE clustering shows that the spectra separate into the four phase clusters without time information. The classification itself is carried by two small ne

What would settle it

At each predicted phase, measure the paracellular flux of a fluorescent tracer of known molecular weight across the nanoporous membrane in the same chip. If the network reports phase IV while tracer flux remains high, the claim that impedance alone detects barrier maturity is falsified.

Watch

Extended reading notes

Core claim

The authors claim that ECIS impedance spectra alone—displayed as Nyquist diagrams and classified by two small neural networks—carry enough information to identify all four phases of endothelial barrier formation and to detect the effect of barrier modulators. The phases are defined optically as adherence (≤30% membrane coverage), outspreading (30–99%), confluence (≥99%), and a dense mature monolayer (≥99% coverage with higher cell density and increasing ZO-1/PECAM-1 junction staining). In their measurements the Nyquist trace evolves from an almost straight line to a Randles-like semicircle, the fitted tight-junction resistance Rtight rises monotonically to an average near 510 Ω in phase IV,

Load-bearing premise

The pipeline assumes that the phase labels set by bright-field coverage and ZO-1/PECAM-1 staining—not by a functional permeability measurement—correctly represent barrier maturity, and that impedance spectra from glass-bottom chips resemble spectra on nanoporous membrane chips closely enough for the trained classifier to transfer.

Editorial extensions

If this is right

  • A mature, confluent monolayer can be recognized from impedance alone, so a barrier-on-chip system can automatically signal when a permeation experiment should be started.
  • The trained network generalizes from cheaper glass-bottom training chips to the nanoporous membrane chips, so routine monitoring does not require every training chip to be the costly membrane version.
  • Rtight and Cdl, fitted from the spectra, track the same maturation signal as ZO-1 fluorescence, giving a continuous electrical correlate of tight-junction formation.
  • Reversible and irreversible barrier weakening produce different classifier trajectories (phase II with recovery versus phase I without recovery), enabling automated distinction of modulator effects.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper, the same spectral-to-phase mapping could serve as an automated quality-control trigger in routine organ-on-chip screening: whenever the network reports phase IV, the system could decide by itself that a permeability assay is ready to start.
  • A natural next test, not reported here, is whether the phase labels track a functional permeability measurement: if phase IV predicted from impedance corresponds to low paracellular flux of a fluorescent tracer, the 'barrier maturity' label would be grounded in function rather than only morphology.
  • The reversible-versus-irreversible discrimination demonstrated with PN159 and BAC suggests the classifier could also be used to follow pathological or inflammatory barrier opening in real time, since it detects departure from the mature-phase cluster rather than requiring a pre-defined drug.
  • The KAN result—similar generalization from roughly one-third as much training data—suggests that even smaller labeled datasets may suffice for other cell types, making the approach easier to port to epithelial or co-culture barriers.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper reports an ECIS-based 'smart membrane' barrier-on-chip device: coplanar electrodes are integrated on an ultrathin nanoporous SixNy membrane, sealed in a microfluidic chip, and connected to custom readout electronics. HUVECs are cultured on the membrane/glass-bottom chips, and impedance spectra are recorded over several days while bright-field and immunofluorescence images are used to assign four phases of monolayer formation (I adherence, II outspreading, III confluence, IV barrier maturity). The authors then train two lightweight classifiers, a Conv1d network and a Kolmogorov-Arnold network, on spectra from glass-bottom chips and validate them on held-out spectra from glass-bottom chips and on an independent set of nanoporous-membrane chips. They also apply the trained model to recordings during treatment with PN159 and BAC, reporting that the model can distinguish reversible (PN159) from irreversible (BAC) barrier weakening. The central claim is that the impedance spectra alone, interpreted by the neural networks, can automatically identify barrier-formation phases and modulator-induced changes, eliminating the need for microscopy and endpoint staining.

Significance. If the main claim holds, the paper has a useful contribution: a wafer-level fabricated, nanoporous-membrane ECIS chip with integrated electrodes and a proof-of-concept automated phase classifier. The cross-system validation on nanoporous membrane chips—where the classifier was trained only on glass-bottom chips—is a genuine strength and goes beyond simple same-chip testing. The comparison of Conv1d with a KAN and the use of t-SNE for unsupervised confirmation of phase structure are also useful. However, the central 'barrier function' claim is not yet fully established. The ground-truth phases are based on bright-field area coverage and ZO-1/PECAM-1 staining, not on a functional permeability assay, and Rtight—used as supportive evidence—is fitted from the same impedance spectra. The modulator experiments show small or no changes in junction-marker intensity while the classifier reports phase transitions, so independent functional validation is missing. The internal evaluation also has unaddressed leakage and class-imbalance risks. The significance is therefore moderate: the device and dataset are valuable, but the end-to-end 'barrier dynamics without microscopy' claim needs stronger

major comments (3)
  1. [§2.2, §2.6, Fig. 6, Note 2] The ground truth is not tied to a functional permeability readout. Phases are defined by bright-field area-coverage thresholds (I ≤30%, II 30–99%, III ≥99%, IV ≥99% with higher cell density) and by ZO-1/PECAM-1 fluorescence intensity; no tracer-flux or equivalent permeability assay is used. Rtight—cited as evidence for barrier maturity—is fitted from the same ECIS spectra used as classifier input, so it is not independent support. More importantly, in the modulator experiments (Fig. 6 and Note 5/Fig. S5), ZO-1 and PECAM-1 border-to-soma ratios are essentially unchanged after PN159 or BAC, yet the classifier reports IV→II or IV→I transitions and subsequent recovery. The classifier may therefore be recognizing a morphology/coverage proxy rather than tight-junction function. The authors should add a permeability assay (e.g., FITC-dextran flux) or explicitly narrow the claims to classificati
  2. [§2.5, §4.4] The evaluation protocol does not rule out leakage and does not account for dependence between spectra. The five-fold cross-validation on the 303 glass-bottom curves is described only in terms of random five-fold splitting; because multiple spectra are acquired from the same chip at successive time points, a random split can place temporally adjacent spectra from the same chip in both training and test, inflating the reported ~95% precision. The class imbalance is admitted but not compensated. The nanoporous-membrane test set (123 curves, six chips) is a genuine cross-system check, but metrics are pooled over curves rather than reported per chip, and no chip-level variability is given. Please perform grouped cross-validation (group by chip, or at least by non-overlapping time windows), report per-chip accuracy, and show the class distribution per chip.
  3. [Abstract, §2.5] The abstract states that phases 'could be recognized with 95% confidence.' In §2.5, the ~95% precision/recall/macro-F1 values are for the glass-bottom self-evaluation; the cross-system nanoporous-membrane results are lower (~89% precision, ~87% recall/macro-F1 for Conv1d, with somewhat better KAN results). No confidence intervals are given for the classification metrics. The abstract should quote the cross-system numbers and avoid '95% confidence' phrasing unless actual confidence bounds are reported.
minor comments (6)
  1. [§2.2 vs §2.3] The number of glass-bottom chips is inconsistent: §2.2 says '18 glass-bottom chips and six nanopore chips,' while the Fig. 4 caption says '24 less costly glass bottom chips.' Please reconcile.
  2. [§4.4 vs §2.5] The training/validation split is unclear: §4.4 describes five-fold cross-validation (80/20), while §2.5 says KAN was 'trained only with 30% of glass bottom chips and Conv1d with 80%.' Please clarify the actual split and ensure the comparison is fair.
  3. [Fig. 6 caption] 'PN159 10 mM' appears to be a typo; the text uses 10 µM for the high concentration. Please correct.
  4. [Table S1 vs §2.2] Table S1 lists Rtight = 446 Ω in phase IV, while §2.2 says 'highest average value of 510 Ω in phase IV.' If these come from different datasets or normalizations, state so explicitly.
  5. [§4.4] The text definition of precision is incorrect: 'Precision was defined as the ratio of correctly predicted positive observations (i.e., TP + TN) divided by the total predicted positive observations.' Precision is TP/(TP+FP). Equation 3 is correct; the prose should be fixed.
  6. [References and data availability] Reference [48] is listed only as '2014' with no authors/title. Also, the data availability statement says 'upon reasonable request' but no code or trained model weights are provided; for an ML reproducibility claim, please make the code and, where possible, the dataset available.

Circularity Check

0 steps flagged · score 1.0 of 10

No significant circularity: the NN classification is supervised on microscopy-derived phase labels, validated on held-out and cross-system impedance data; self-citations are prior device work and not load-bearing.

full rationale

The central derivation is a supervised learning pipeline, not a self-referential one. Section 2.2 explicitly states that 'Once the phases of barrier formation had been determined by microscopic images of cell morphology and fluorescence staining, the ECIS data could be used to train artificial neural network models.' Thus the network labels are external to the impedance input, and the reported confidence is measured against held-out manual labels, with additional cross-system validation on nanoporous membrane chips (Section 2.5). The equivalent-circuit parameter Rtight is a fitted quantity derived from the same Nyquist spectra, but it is used for interpretative correlation with ZO-1 immunofluorescence (Section 2.2, Note 2), not renamed as a prediction; no fitted parameter is fed back as the network's target. The paper's self-citations (refs 8-11, 15, 23) concern prior barrier-chip and electrode-simulation work and are not used to justify the classification claim. The reviewer's concern that phase IV is a coverage/morphology proxy rather than a functional permeability measure is a construct-validity limitation, not a circularity: the absence of a permeability assay weakens the biological interpretation but does not make the impedance-to-phase mapping definitionally equivalent to its inputs. No uniqueness theorem or ansatz is imported from the authors' own prior work. Under the hard rules requiring a quotable equation-level reduction, no circular step is present.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The ledger is dominated by supervised-learning choices: the phase labels come from manual microscopy thresholds, the equivalent-circuit values are fit to the spectra, and the classifier is a collection of hyperparameters and learned weights. There are no invented physical entities. The largest unfunded assumption is cross-system transfer from glass-bottom chips to nanoporous membrane chips.

free parameters (4)
  • Equivalent-circuit component values (Rs, Ws, Ccell, Rcell, Cdl, Rtight) = Rtight 51 to 446 Ohm across phases (510 Ohm quoted in text); see Table S1
    Obtained by fitting the Nyquist spectra with ZView; used to support the phase interpretation and drug-effect narrative. These are fitted values, not independent measurements.
  • Phase coverage thresholds used as labels = Phase I <=30%, II 30-99%, III >=99%, IV >=99% with higher density
    Manually chosen thresholds define the supervised classification targets; changing them changes reported accuracy.
  • Neural network hyperparameters (Conv1d and KAN) = Conv1d: two layers, 120 filters, kernel 3; KAN: one layer, order-3 polynomial; lr 2e-4 and 1; 100 epochs
    Selected by the authors without a reported systematic search; classifier metrics depend on them.
  • Trained neural network weights = not released
    The classifier's predictions are determined by these fitted weights; without them the exact model cannot be reproduced.
assumptions (5)
  • domain assumption The 500 nm pore SixNy membrane is permeable to medium and drugs but prevents cell migration, as established in prior work [9].
    The device's dual barrier and permeation function is inherited from earlier work by the same group, not re-measured here.
  • domain assumption Bright-field coverage and ZO-1/PECAM-1 fluorescence are valid ground truth for the four barrier phases in HUVECs.
    Phases are labeled from morphology and staining; no functional permeability assay confirms barrier maturity.
  • domain assumption The equivalent circuit in Figure S2 adequately represents the cell-covered electrode impedance.
    Used to extract Rtight; the introduction itself notes EIS data can be fit by many circuit models.
  • domain assumption Glass-bottom chip impedance data transfer to nanoporous membrane chips.
    The training set is almost entirely glass-bottom data (18 chips versus 6 nanopore chips); the paper validates transfer empirically but does not derive it.
  • domain assumption The 5 mV sinusoidal excitation does not alter cell behavior.
    Assumed to justify the non-invasive claim; this is not directly tested in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Smart membrane: high content in situ monitoring barrier on chip with artificial neural network." pith.science (2026). https://pith.science/paper/UJJT2X42

@misc{pith2026260801239,
  author       = {Pith},
  title        = {Pith review of: Smart membrane: high content in situ monitoring barrier on chip with artificial neural network},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UJJT2X42}},
  note         = {Machine review of arXiv:2608.01239}
}
read the original abstract

Conventional transepithelial electrical resistance (TEER) technique provides only a low-content analysis of cell-layer conditions, necessitating repeated microscopic assessments of morphology and cell-cell contacts outside the incubator for barrier-on-chip systems. This work presents a novel high-content TEER device in the form of a novel nanoporous membrane that facilitates continuous electrical measurement of cell-substrate impedance sensing (ECIS). The ultrathin (700 nm) membrane, composed of ultra-low-stress SixNy, is monolithically integrated into wafer-level fabricated chips sealed with glass lids. Coplanar ECIS electrodes were connected to custom electronics to record impedance under sinusoidal excitation. Human umbilical vein endothelial cells (HUVECs) were seeded and continuously recorded impedance spectra were compared with bright-field and fluorescence microscopy, revealing distinct phases of monolayer formation. With one-dimensional convolutional neural network (Conv1d) and Kolmogorov-Arnold Network (KAN) trained with a small amount of Nyquist-diagrams, phases of (I) adherence, (II) outspreading, (III) confluence and (IV) barrier maturity with tight junction formation could be recognized with 95% confidence. As further proof of concept, reversible and irreversible barrier weakening using modulators PN159 and BAC was identified in this way. Our studies have demonstrated that an immediate and automatable non-invasive detection of in-vitro barrier dynamics within barrier-on-chip systems, eliminating the need for microscopy and endpoint staining. We expect this ECIS technique will find broad applications in organ-on-chip systems for in situ monitoring physiological or pathological states of tissue barrier.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

48 extracted references · 48 canonical work pages

  1. [1]

    S. B. Campbell, Q. Wu, J. Yazbeck, C. Liu, S. Okhovatian, M. Radisic, ACS Biomater. Sci. Eng. 2021, 7, 2880

    work page 2021

  2. [2]

    Taiedinejad, P

    E. Taiedinejad, P. Erfle, C. Bausch, A. Dietzel, SPhERe Proceedings: 4th International Symposium on Pharmaceutical Engineering Research 2021, DOI 10.24355/DBBS.084- 202110261752-0

    work page doi:10.24355/dbbs.084- 2021

  3. [3]

    C. M. Leung, P. De Haan, K. Ronaldson-Bouchard, G.-A. Kim, J. Ko, H. S. Rho, Z. Chen, P. Habibovic, N. L. Jeon, S. Takayama, M. L. Shuler, G. Vunjak-Novakovic, O. Frey, E. Verpoorte, Y.-C. Toh, Nat Rev Methods Primers 2022, 2, 33

    work page 2022

  4. [4]

    Q. Wu, J. Liu, X. Wang, L. Feng, J. Wu, X. Zhu, W. Wen, X. Gong, BioMed Eng OnLine 2020, 19, 9

    work page 2020

  5. [5]

    Zhang, A

    B. Zhang, A. Korolj, B. F. L. Lai, M. Radisic, Nat Rev Mater 2018, 3, 257

    work page 2018

  6. [6]

    A. Oddo, B. Peng, Z. Tong, Y. Wei, W. Y. Tong, H. Thissen, N. H. Voelcker, Trends in Biotechnology 2019, 37, 1295

    work page 2019

  7. [7]

    D. Vera, M. García-Díaz, N. Torras, M. Álvarez, R. Villa, E. Martinez, ACS Appl. Mater. Interfaces 2021, 13, 13920

    work page 2021

  8. [8]

    Mattern, N

    K. Mattern, N. Beißner, S. Reichl, A. Dietzel, European Journal of Pharmaceutics and Biopharmaceutics 2018, 126, 159

    work page 2018

Show all 48 references
  1. [9]

    B. Tang, S. Bendas, V. Krajka, T. May, A. Moritz, I. Constantinou, S. Reichl, A. Dietzel, Front. Sens. 2022, 3, 974895. 31

    2022

  2. [10]

    E. V. Koch, V. Ledwig, S. Bendas, S. Reichl, A. Dietzel, Pharmaceutics 2022, 14, 1451

    2022

  3. [11]

    E. V. Koch, S. Bendas, K. Nehlsen, T. May, S. Reichl, A. Dietzel, Pharmaceutics 2023, 15, 2439

    2023

  4. [12]

    Srinivasan, A

    B. Srinivasan, A. R. Kolli, M. B. Esch, H. E. Abaci, M. L. Shuler, J. J. Hickman, SLAS Technology 2015, 20, 107

    2015

  5. [13]

    Felix, S

    K. Felix, S. Tobias, H. Jan, S. Nicolas, M. Michael, Histochem Cell Biol 2021, DOI 10.1007/s00418-021-02026-4

    2021 doi

  6. [14]

    E. M. Shaughnessey, S. H. Kann, H. Azizgolshani, L. D. Black, J. L. Charest, E. M. Vedula, Sci Rep 2022, 12, 13182

    2022

  7. [15]

    Beiβner, K

    N. Beiβner, K. Mattern, A. Dietzel, S. Reichl, European Journal of Pharmaceutics and Biopharmaceutics 2018, 126, 166

    2018

  8. [16]

    F. Wu, B. Zhou, J. Wang, M. Zhong, A. Das, M. Watkinson, K. Hing, D.-W. Zhang, S. Krause, Anal. Chem. 2019, 91, 5896

    2019

  9. [17]

    Yoshinobu, M

    T. Yoshinobu, M. J. Schöning, Current Opinion in Electrochemistry 2021, 28, 100727

    2021

  10. [18]

    Y. Shi, G. Feng, X. Li, X. Yang, A. H. Ghanim, P. Ruchhoeft, D. Jackson, S. Mubeen, X. Shan, Anal. Chem. 2021, 93, 12320

    2021

  11. [19]

    Polcari, P

    D. Polcari, P. Dauphin-Ducharme, J. Mauzeroll, Chem. Rev. 2016, 116, 13234

    2016

  12. [20]

    T.-E. Lin, S. Rapino, H. H. Girault, A. Lesch, Chem. Sci. 2018, 9, 4546

    2018

  13. [21]

    S. E. De León, A. Pupovac, S. L. McArthur, Biotech & Bioengineering 2020, 117, 1230

    2020

  14. [22]

    Bounik, F

    R. Bounik, F. Cardes, H. Ulusan, M. M. Modena, A. Hierlemann, BME Front 2022, 2022, 9857485

    2022

  15. [23]

    B. Tang, M. Liu, A. Dietzel, Biosensors 2023, 13, 281

    2023

  16. [24]

    Y. Xu, X. Xie, Y. Duan, L. Wang, Z. Cheng, J. Cheng, Biosensors and Bioelectronics 2016, 77, 824

    2016

  17. [25]

    S. Wang, J. Zhang, O. Gharbi, V. Vivier, M. Gao, M. E. Orazem, Nat Rev Methods Primers 2021, 1, 41

    2021

  18. [26]

    D. M. Lewis, N. Mavrogiannis, Z. Gagnon, S. Gerecht, Biomicrofluidics 2018, 12, 042202

    2018

  19. [27]

    Matthiesen, D

    I. Matthiesen, D. Voulgaris, P. Nikolakopoulou, T. E. Winkler, A. Herland, Small 2021, 17, 2101785

    2021

  20. [28]

    Chebotarev, A

    O. Chebotarev, A. Ugodnikov, C. A. Simmons, ACS Appl. Bio Mater. 2024, acsabm.4c00114

    2024

  21. [29]

    I. O. K’Owino, O. A. Sadik, Electroanalysis 2005, 17, 2101

    2005

  22. [30]

    Gharbi, M

    O. Gharbi, M. T. T. Tran, M. E. Orazem, B. Tribollet, M. Turmine, V. Vivier, ChemPhysChem 2021, 22, 1371

    2021

  23. [31]

    Sarró, M

    E. Sarró, M. Lecina, A. Fontova, C. Solà, F. Gòdia, J. J. Cairó, R. Bragós, Biosensors and Bioelectronics 2012, 31, 257

    2012

  24. [32]

    H. P. Schwan, in Advances in Biological and Medical Physics, Elsevier, 1957, pp. 147– 209

    1957

  25. [33]

    Krukiewicz, Electrochemistry Communications 2020, 116, 106742

    K. Krukiewicz, Electrochemistry Communications 2020, 116, 106742

    2020

  26. [34]

    Skalová, V

    Š. Skalová, V. Vyskočil, J. Barek, T. Navrátil, Electroanalysis 2018, 30, 207

    2018

  27. [35]

    Claesson-Welsh, E

    L. Claesson-Welsh, E. Dejana, D. M. McDonald, Trends in Molecular Medicine 2021, 27, 314

    2021

  28. [36]

    A. Ch. Lazanas, M. I. Prodromidis, ACS Meas. Sci. Au 2023, 3, 162

    2023

  29. [37]

    W. Chen, Z. Li, J. Hu, Z. Zhang, L. Chen, Y. Chen, Z. Liu, PLoS ONE 2011, 6, e26103

    2011

  30. [38]

    Gravel, C

    J. Gravel, C. Paradis-Bleau, A. R. Schmitzer, Med. Chem. Commun. 2017, 8, 1408

    2017

  31. [39]

    Bocsik, I

    A. Bocsik, I. Gróf, L. Kiss, F. Ötvös, O. Zsíros, L. Daruka, L. Fülöp, M. Vastag, Á. Kittel, N. Imre, T. Martinek, C. Pál, P. Szabó-Révész, M. A. Deli, Pharmaceutics 2019, 11, 73

    2019

  32. [40]

    Tornavaca, M

    O. Tornavaca, M. Chia, N. Dufton, L. O. Almagro, D. E. Conway, A. M. Randi, M. A. Schwartz, K. Matter, M. S. Balda, Journal of Cell Biology 2015, 208, 821. 32

    2015

  33. [41]

    Huang, S.-M

    Q.-A. Huang, S.-M. Park, J. Phys. Chem. C 2012, 116, 16939

    2012

  34. [42]

    Ko, S.-M

    Y. Ko, S.-M. Park, J. Phys. Chem. C 2012, 116, 7260

    2012

  35. [43]

    Yukutake, R

    Y. Yukutake, R. Honda, M. Ukawa, K. Kurita, Earth Planets Space 2022, 74, 144

    2022

  36. [44]

    X. Hu, W. Yang, Sensor Review 2010, 30, 24

    2010

  37. [45]

    Pachitariu, C

    M. Pachitariu, C. Stringer, Nat Methods 2022, 19, 1634

    2022

  38. [46]

    pyclesperanto: GPU-accelerated image processing in python using OpenCL,

    S. Rigaud, “pyclesperanto: GPU-accelerated image processing in python using OpenCL,” can be found under https://pypi.org/project/pyclesperanto/, 2024

    2024

  39. [47]

    Z. Liu, Y. Wang, S. Vaidya, F. Ruehle, J. Halverson, M. Soljačić, T. Y. Hou, M. Tegmark, 2024

    2024

  40. [49]

    Sokolova, G

    M. Sokolova, G. Lapalme, Information Processing & Management 2009, 45, 427. 33 Supporting Information Note 1: Cell analysis procedure Figure S1. Cell analysis procedure using cellpose 2.0. (A-B) First, cells were segmented based on ZO-1 immunofluorescence staining images (size...

    2009

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.