Pith. sign in

REVIEW 4 major objections 7 minor 45 references

MicroAnalyzer: A Python Tool for Automated Bacterial Analysis with Fluorescence Microscopy

T0 review · 4 major / 7 minor · reviewed 2026-08-27 · deepseek-v4-flash

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

Pith's one-line read MicroAnalyzer claims to fully automate bacterial fluorescence image analysis from raw microscope files to a database, using deep-learning segmentation and a new validity metric.

desk verdict Honest engineering paper with a genuinely reusable evaluation metric, but the end-to-end claim rests on validation from just two images while the database fields go unchecked. read the letter →

arxiv 2009.12684 v1 pith:KP2KDCUJ submitted 2020-09-26 cs.CV cs.LGeess.IV

Jonathan Reiner , Guy Azran , Gal Hyams This is my paper · ORCID
classification cs.CVcs.LGeess.IV
keywords fluorescencemicroscopybacterialimageanalysiscellsegmentationclusterdeeplearningMaskR-CNNopen-sourcesoftwareevaluationmetric
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

MicroAnalyzer is presented as an open-source Python tool that turns raw fluorescence-microscopy images of rod-shaped bacteria into a complete analysis database without manual segmentation. The paper's central claim is that this end-to-end pipeline—Mask R-CNN for cell detection, FPN for fluorescence-cluster segmentation, and a modified Colicoords module for measurements—produces valid segmentations under the stated experimental assumptions, namely that crowded or out-of-focus cells are discarded and false-positive cell detections are strongly penalized. To support that claim, the paper introduces an "experimental lex-error" metric that judges a prediction against two independent researcher segmentations and calls it valid only if it is no farther from them than the researchers are from each other. Under that criterion, Mask R-CNN yields valid cell segmentation whereas thresholding and U-Net do not, while FPN and U-Net both give valid cluster segmentation. A sympathetic reader would care because this replaces the tedious, eye-straining manual segmentation bottleneck with a trainable, adaptable pipeline.

What carries the argument

The load-bearing object is the paper's validity criterion built from "experimental lex-error." For a prediction $p_d$ and two ground-truth segmentations $G_1, G_2$, the criterion counts false-positive and false-negative connected components relative to an IoU threshold ($T_{cell}=0.84$, $T_{fluo}=0.6$), weights them with $\beta$ ($0.7$ for cells to penalize false positives, $0.15$ for clusters to emphasize recall), and divides by the number of objects in either ground truth; $p_d$ is "valid" when its average lex-error against the two researchers' segmentations is no larger than their experimental distance $d_{ex}(G_1,G_2)$, i.e. the prediction could pass as a third researcher. This criterion is what selects Mask R-CNN and FPN, and the paper argues that classic pixel-level metrics and average-precision-style scores fail to respect the experiment's assumption that false positives are worse than false negatives.

What would settle it

Take the trained Mask R-CNN and FPN models, run them on the designated 5-cell-image and 4-fluorescence-image test set, and compare each prediction with two independent researcher segmentations using the paper's lex-error criterion; if most test predictions fail the validity inequality $(\ell_{ex}(p_d,G_1)+\ell_{ex}(p_d,G_2))/2 \le d_{ex}(G_1,G_2)$, then the central claim of valid fully automated analysis is refuted.

Watch

Extended reading notes

Core claim

On its own terms, the paper discovers that a specific combination of existing components is enough to automate bacterial fluorescence image analysis "from microscope to database." Given raw ND2 files, the pipeline separates the bright-field bacteria channel from fluorescence channels, runs Mask R-CNN on the bacteria image and FPN on an RGB image whose R channel is the bacteria image and G/B channels are the fluorescence channel, filters the resulting masks with deterministic size and proximity rules, then hands the accepted masks to Colicoords to fit per-cell coordinate systems and compute database fields including cluster positions and polar/nonpolar labels. The paper's own evaluation, using two validation images annotated by two researchers, reports that only Mask R-CNN among the tested cell-segmentation approaches meets the new validity criterion, and that FPN and U-Net are interchangeable for cluster segmentation; both yield valid predictions. In other words, the central discovery is not a new architecture but a validated workflow plus an evaluation criterion that treats multiple human ground truths seriously.

Load-bearing premise

The evaluation rests on two validation images that two researchers annotated; if those two images are not representative of the lab's actual imaging conditions, the conclusion that MicroAnalyzer produces valid segmentations in general does not follow.

Editorial extensions

If this is right

  • Labs that adopt the same four assumptions can feed overnight batches of ND2 files into MicroAnalyzer and obtain per-cell and per-cluster measurements without manual intervention.
  • The validity criterion gives a concrete way to compare segmentation models whenever two human annotators disagree on what counts as a cell or cluster.
  • Because the pipeline is open source and the networks are retrainable, a new experiment's restrictions (cell size, spacing, focus rules) can be encoded by retraining rather than by hand-tuning thresholds.
  • The reported runtimes (about one hour to train Mask R-CNN and under 30 minutes for FPN on the described hardware) make the workflow practical for a single lab.
  • MicroAnalyzer extends Colicoords by adding fluorescence-cluster calculations and database construction, so measurements such as cluster count, cluster center, and polar localization are produced in the same automated pass.

Reading between the lines

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

  • I infer that the same validity criterion could be reused as a quality gate in routine annotation pipelines: if a newly trained model's lex-error against two annotators exceeds their mutual distance, that flags either a bad model or an ambiguous image worth re-examining.
  • A natural testable extension is to run the held-out test sets (5 cell images, 4 fluorescence images) through the trained models and report lex-error there; the paper currently evaluates only the two-image validation set.
  • The connected-component equivalence argument behind the criterion holds for any IoU threshold above 0.5, so the metric could be adapted to other object types, including 3D volumetric segmentations, without changing the mathematics.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Request a human review

A listed scientist reviews the paper for a fee and the review publishes here regardless of verdict. See the reviewers or get listed.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 7 minor

Summary. MicroAnalyzer is an open-source Python pipeline for bacterial fluorescence microscopy analysis. It takes Nikon ND2 images, segments cells with Mask R-CNN and fluorescence clusters with FPN, post-processes these masks in a module called CellAnalyzer (a modified Colicoords), and outputs a database with cell and cluster measurements. The paper introduces a new evaluation metric, the experimental lex-error, which counts object-level false positives and false negatives with a false-positive weight beta, and defines a prediction as valid when its average lex-error relative to two independent human ground truths is no larger than the inter-rater distance dex. On a validation set of two images, the authors report that Mask R-CNN yields valid cell segmentations while U-Net, thresholding, and Cellpose do not, and that both U-Net and FPN yield valid fluorescence-cluster segmentations. The abstract claims that MicroAnalyzer automates the entire process 'from microscope to database' with no further researcher input after initial model training.

Significance. If the end-to-end claim were fully established, MicroAnalyzer would be a useful open-source contribution: it combines existing segmentation networks with Colicoords-based analysis, adds fluorescence cluster detection, and proposes an evaluation criterion that explicitly accounts for inter-rater disagreement. The manuscript ships code, notebooks, data links, and a sample database, which is a practical strength. The derivation of the simplified dex expression is correct, and the idea of anchoring validity to inter-rater distance is a reasonable operationalization. However, the current evidence base is too narrow: the validity of the segmentation is established on only two images, the designated test sets are never scored, and the database fields that constitute the 'database' part of the pipeline are never checked against independent measurements. These gaps are load-bearing for the abstract's central claim.

major comments (4)
  1. [Section IV, Tables 1-2] The segmentation validity claim is based entirely on two validation images. The paper defines training and test splits (5 cell images, 4 fluorescence images) but then states 'All models are evaluated on the validation set' and reports no results on the held-out test images. With only two images, the selection of Mask R-CNN over U-Net and the validity verdict for FPN may not generalize to the lab's other images, which is exactly what the 'from microscope to database' claim requires. The authors should evaluate the held-out test sets and report per-image lex-error and validity for each model.
  2. [Appendix F, Section 3.2] None of the database fields (C-T: length, width, area, volume, intensity statistics, cluster size, center, polarity, etc.) is compared with an independent measurement or a manual gold standard. The lex-error criterion only validates the binary masks, not the coordinate-system fitting and intensity calculations produced by CellAnalyzer/Colicoords. Since the abstract promises a 'full analysis database,' the paper needs at least a comparison of key fields (e.g., cell length, area, fluorescence intensity) against manual or established-tool measurements on the same images.
  3. [Appendix B] The text states that for detections present in neither ground truth, 'this last evaluation has to be performed manually by the researchers.' This directly qualifies the automation claim in the Abstract, which says the tool 'does not require any further input from the researcher except for the initial deep-learning model training.' The manuscript should either integrate this manual review into the stated workflow or remove the phrase 'does not require any further input.'
  4. [Sections 3.1.1-3.1.4] The validity criterion uses hand-set thresholds T_Cell=0.84, T_fluo=0.6 and beta weights 0.7 and 0.15. No sensitivity analysis, confidence intervals, or alternative parameter values are reported. Because the choice of Mask R-CNN (Table 1) is made under this criterion, the conclusion could be sensitive to these arbitrary choices. A small parameter sweep (varying T and beta over reasonable ranges) would show whether the model ranking and validity verdicts are stable.
minor comments (7)
  1. [Section I heading] The Introduction heading is misspelled 'INTORDUCTION' and should be 'INTRODUCTION'.
  2. [Figure 4 caption] The caption contains a stray Hebrew character 'ש' that should be removed.
  3. [Figure 9 caption] The caption spells 'Maks-RCNN' instead of 'Mask-RCNN'.
  4. [Figure 1 caption] The caption says 'form microscope' instead of 'from microscope'.
  5. [Appendices] The appendices jump from C to F; Appendix D and E appear to be missing, which makes cross-references such as 'see appendix F' in Section 3.2 confusing and should be renumbered.
  6. [Section 2.2] The phrase 'givan such a file' should read 'given such a file'.
  7. [Tables 1-3] Tables 1-3 are referenced in the text and captioned, but the actual table contents are not included in the manuscript; please supply the full tables so readers can inspect the numerical results.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the validity criterion is anchored to external inter-rater agreement, and thresholds/weights are set a priori rather than fitted to model outputs.

full rationale

MicroAnalyzer's derivation chain does not reduce to its own inputs. The new lex-error criterion is defined independently of the trained models: it compares a prediction with two researcher-created ground-truth segmentations and normalizes by the inter-rater distance, so the benchmark is an external reference rather than a fitted quantity. The thresholds T_cell=0.84 and T_fluo=0.6 and the beta weights beta_cell=0.7, beta_fluo=0.15 are stated as experimental choices reflecting the lab's assumptions and Colicoords' input requirements, not as parameters fitted to maximize the validation score. Model selection on the validation set and the failure to evaluate the reserved test sets are validity and overfitting concerns, not circularity: the selected models' outputs are not constructed to equal the ground truths by definition. Appendix B's admission that the final judgment of uncertain 'rogue' detections must be performed manually is a limitation of the automation claim, not a self-referential derivation. No self-citations or imported uniqueness theorems carry load, and no prediction is equivalent to its inputs by construction.

Assumptions & free parameters 5 free parameters · 6 assumptions · 0 invented entities

The central evaluation and model selection depend on hand-set thresholds and beta weights, on the assumption that connected-component matching captures segmentation quality, and on the representativeness of a two-image validation set. The math itself is simple; the burden is in these choices.

free parameters (5)
  • T_Cell = 0.84
    IoU threshold for cell matching in lex-error; chosen because lower thresholds accepted cell masks that did not provide enough context to Colicoords, causing runtime errors (Section 3.1.1, note 4).
  • T_fluo = 0.6
    IoU threshold for fluorescence cluster matching in lex-error; set by hand for this experiment (Section 3.1.1).
  • beta_cell = 0.7
    Weight for false positives in cell lex-error; chosen to encode assumption (c) that false positive detections are worse than false negatives (Section 3.1.4).
  • beta_fluo = 0.15
    Weight for false positives in cluster lex-error; chosen to favor finding many clusters even at the cost of extra detections (Section 3.1.4).
  • CellAnalyzer size and proximity filters = not specified
    CellAnalyzer filters invalid cells and clusters using deterministic algorithms based on minimal object size and proximity; exact thresholds are not given (Section 2.2 and Figure 2).
assumptions (6)
  • domain assumption Segmented cells and fluorescence clusters can be represented as connected components in a binary mask.
    Section 3.1.1 assumes all cells and clusters appear as connected components in the images.
  • domain assumption The experimental assumptions (a)-(d) from the collaborating lab are the correct validity criteria for this application.
    Assumptions listed in Section 1.2 define which cells are valid and drive the evaluation metric.
  • ad hoc to paper A valid prediction is one whose average lex-error to two ground truths is no larger than the distance between the ground truths.
    Definition in Section 3.1.4; this is a new criterion introduced by the paper, not an external standard.
  • standard math IoU matching with threshold T > 0.5 yields an equivalence relation between connected components.
    Proved in Section 3.1.1 via the pigeonhole principle; standard set-theoretic argument.
  • domain assumption Deep learning models trained on 40 cell images and 27 fluorescence images generalize to the lab's images.
    Section IV; no statistical evidence beyond 2 validation images supports generalization.
  • domain assumption Fluorescence clusters have a roughly 3D Gaussian intensity profile.
    Section 4.2 states 'the algorithms should search for a three-dimensional Gaussian shape in the image' and uses this as motivation for model selection.

how reviews work

0 comments
Cite this review

Pith. "Pith review of MicroAnalyzer: A Python Tool for Automated Bacterial Analysis with Fluorescence Microscopy." pith.science (2026). https://pith.science/paper/KP2KDCUJ

@misc{pith2026200912684,
  author       = {Pith},
  title        = {Pith review of: MicroAnalyzer: A Python Tool for Automated Bacterial Analysis with Fluorescence Microscopy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/KP2KDCUJ}},
  note         = {Machine review of arXiv:2009.12684}
}
read the original abstract

Fluorescence microscopy is a widely used method among cell biologists for studying the localization and co-localization of fluorescent protein. For microbial cell biologists, these studies often include tedious and time-consuming manual segmentation of bacteria and of the fluorescence clusters or working with multiple programs. Here, we present MicroAnalyzer - a tool that automates these tasks by providing an end-to-end platform for microscope image analysis. While such tools do exist, they are costly, black-boxed programs. Microanalyzer offers an open-source alternative to these tools, allowing flexibility and expandability by advanced users. MicroAnalyzer provides accurate cell and fluorescence cluster segmentation based on state-of-the-art deep-learning segmentation models, combined with ad-hoc post-processing and Colicoords - an open-source cell image analysis tool for calculating general cell and fluorescence measurements. Using these methods, it performs better than generic approaches since the dynamic nature of neural networks allows for a quick adaptation to experiment restrictions and assumptions. Other existing tools do not consider experiment assumptions, nor do they provide fluorescence cluster detection without the need for any specialized equipment. The key goal of MicroAnalyzer is to automate the entire process of cell and fluorescence image analysis "from microscope to database", meaning it does not require any further input from the researcher except for the initial deep-learning model training. In this fashion, it allows the researchers to concentrate on the bigger picture instead of granular, eye-straining labor

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

45 extracted references · 45 canonical work pages

  1. [1]

    The compartmentalized vessel,

    O. Amster-Choder, “The compartmentalized vessel,” Cell. Logist., vol. 1, no. 2, pp. 77–81, Mar. 2011, doi: 10.4161/cl.1.2.16152

    work page doi:10.4161/cl.1.2.16152 2011

  2. [2]

    Spatiotemporal Organization of the E. coli Transcriptome: Translation Independence and Engagement in Regulation,

    S. Kannaiah, J. Livny, and O. Amster-Choder, “Spatiotemporal Organization of the E. coli Transcriptome: Translation Independence and Engagement in Regulation,” Mol. Cell, vol. 76, no. 4, Nov. 2019, doi: 10.1016/j.molcel.2019.08.013

    work page doi:10.1016/j.molcel.2019.08.013 2019

  3. [3]

    Streptococcus pneumoniae PBP2x mid-cell localization requires the C-terminal PASTA domains and is essential for cell shape maintenance,

    K. Peters et al., “Streptococcus pneumoniae PBP2x mid-cell localization requires the C-terminal PASTA domains and is essential for cell shape maintenance,” Mol. Microbiol., vol. 92, no. 4, 2014, doi: 10.1111/mmi.12588

    work page doi:10.1111/mmi.12588 2014

  4. [4]

    Three-dimensional localization precision of the double-helix point spread function versus astigmatism and biplane,

    M. Badieirostami, M. D. Lew, M. A. Thompson, and W. E. Moerner, “Three-dimensional localization precision of the double-helix point spread function versus astigmatism and biplane,” Appl. Phys. Lett., vol. 97, no. 16, Oct. 2010, doi: 10.1063/1.3499652

    work page doi:10.1063/1.3499652 2010

  5. [5]

    Localization of MreB in Rhodobacter sphaeroides under conditions causing changes in cell shape and membrane structure,

    P. M. Slovak, G. H. Wadhams, and J. P. Armitage, “Localization of MreB in Rhodobacter sphaeroides under conditions causing changes in cell shape and membrane structure,” in Journal of Bacteriology, Jan. 2005, vol. 187, no. 1, doi: 10.1128/JB.187.1.54- 64.2005

    work page doi:10.1128/jb.187.1.54- 2005

  6. [6]

    Phenotypic Heterogeneity in Sugar Utilization by E. coli Is Generated by Stochastic Dispersal of the General PTS Protein EI from Polar Clusters,

    S. Govindarajan, N. Albocher, T. Szoke, A. Nussbaum-Shochat, and O. Amster-Choder, “Phenotypic Heterogeneity in Sugar Utilization by E. coli Is Generated by Stochastic Dispersal of the General PTS Protein EI from Polar Clusters,” Front. Microbiol., vol. 8, no. JAN, Jan. 2018, doi: 10.3389/fmicb.2017.02695

    work page arXiv 2018

  7. [7]

    How do bacteria localize proteins to the cell pole?,

    G. Laloux and C. Jacobs-Wagner, “How do bacteria localize proteins to the cell pole?,” Journal of Cell Science, vol. 127, no. 1. Company of Biologists, Jan. 01, 2014, doi: 10.1242/jcs.138628

    work page doi:10.1242/jcs.138628 2014

  8. [8]

    Polar localization of a bacterial chemoreceptor,

    M. R. K. Alley, J. R. Maddock, and L. Shapiro, “Polar localization of a bacterial chemoreceptor,” 17 Genes Dev., vol. 6, no. 5, 1992, doi: 10.1101/gad.6.5.825

    work page doi:10.1101/gad.6.5.825 1992

Show all 45 references
  1. [9]

    Automated microscopy for high-content RNAi screening,

    C. Conrad and D. W. Gerlich, “Automated microscopy for high-content RNAi screening,” Journal of Cell Biology, vol. 188, no. 4. The Rockefeller University Press, Feb. 22, 2010, doi: 10.1083/jcb.200910105

    2010 doi

  2. [10]

    High-throughput fluorescence microscopy for systems biology,

    R. Pepperkok and J. Ellenberg, “High-throughput fluorescence microscopy for systems biology,” Nature Reviews Molecular Cell Biology, vol. 7, no

  3. [11]

    Automated quality assessment of autonomously acquired microscopic images of fluorescently stained bacteria,

    M. Zeder, E. Kohler, and J. Pernthaler, “Automated quality assessment of autonomously acquired microscopic images of fluorescently stained bacteria,” Cytom. Part A, vol. 77, no. 1, Jan. 2010, doi: 10.1002/cyto.a.20810

    2010 doi

  4. [12]

    NIH Image to ImageJ: 25 years of image analysis,

    C. A. Schneider, W. S. Rasband, and K. W. Eliceiri, “NIH Image to ImageJ: 25 years of image analysis,” Nature Methods, vol. 9, no. 7. NIH Public Access, Jul. 2012, doi: 10.1038/nmeth.2089

    2012 doi

  5. [13]

    19, 2006, doi: 10.1038/nrm1979

    Nat Rev Mol Cell Biol, Sep. 19, 2006, doi: 10.1038/nrm1979

    2006 doi

  6. [14]

    Super-resolution microscopy demystified,

    L. Schermelleh et al., “Super-resolution microscopy demystified,” Nature Cell Biology, vol. 21, no. 1. Nature Publishing Group, Jan. 01, 2019, doi: 10.1038/s41556-018-0251-8

    2019 doi

  7. [15]

    Ultra-high resolution imaging by fluorescence photoactivation localization microscopy,

    S. T. Hess, T. P. K. Girirajan, and M. D. Mason, “Ultra-high resolution imaging by fluorescence photoactivation localization microscopy,” Biophys. J., vol. 91, no. 11, pp. 4258–4272, Dec. 2006, doi: 10.1529/biophysj.106.091116

    2006 doi

  8. [16]

    NIS-Elements | Software | Products | Nikon Instruments Inc

    “NIS-Elements | Software | Products | Nikon Instruments Inc.” https://www.microscope.healthcare.nikon.com/prod ucts/software/nis-elements (accessed Sep. 06, 2020)

    2020

  9. [17]

    Cellpose: a generalist algorithm for cellular segmentation,

    C. Stringer, T. Wang, M. Michaelos, and M. Pachitariu, “Cellpose: a generalist algorithm for cellular segmentation,” bioRxiv, Feb. 2020, doi: 10.1101/2020.02.02.931238

    2020 doi

  10. [18]

    Analysis

    handle segmentation of fluorescence clusters which is required for calculating localization metrics of the observed material. After the information has been extracted from the images (cells and clusters segmentation) , the researchers must perform calculations on that data. Th...

  11. [19]

    Stochastic optical reconstruction microscopy (STORM),

    J. Xu, H. Ma, and Y. Liu, “Stochastic optical reconstruction microscopy (STORM),” Curr. Protoc. Cytom., vol. 2017, Jul. 2017, doi: 10.1002/cpcy.23

    2017 doi

  12. [20]

    DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs,

    L. C. Chen, G. Papandreou, I. Kokkinos, K. Murphy, and A. L. Yuille, “DeepLab: Semantic Image Segmentation with Deep Convolutional Nets, Atrous Convolution, and Fully Connected CRFs,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 40, no. 4, Apr. 2018, doi: 10.1109/TPAMI.2017.2699184

    2018

  13. [21]

    PartSeg, a Tool for Quantitative Feature Extraction From 3D Microscopy Images for Dummies,

    G. Bokota et al., “PartSeg, a Tool for Quantitative Feature Extraction From 3D Microscopy Images for Dummies,” bioRxiv, Jul. 2020, doi: 10.1101/2020.07.16.206789

    2020 doi

  14. [22]

    ColiCoords: A Python package for the analysis of bacterial fluorescence microscopy data,

    J. H. Smit, Y. Li, E. M. Warszawik, A. Herrmann, and T. Cordes, “ColiCoords: A Python package for the analysis of bacterial fluorescence microscopy data,” PLoS One, vol. 14, no. 6, Jun. 2019, doi: 10.1371/journal.pone.0217524

    2019 doi

  15. [23]

    U-net: Convolutional networks for biomedical image segmentation,

    O. Ronneberger, P. Fischer, and T. Brox, “U-net: Convolutional networks for biomedical image segmentation,” in Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), May 2015, vol. 9351, doi: 10.100...

    2015 doi

  16. [24]

    A deep learning-based algorithm for 2-D cell segmentation in microscopy images,

    Y. Al-Kofahi, A. Zaltsman, R. Graves, W. Marshall, and M. Rusu, “A deep learning-based algorithm for 2-D cell segmentation in microscopy images,” BMC Bioinformatics, vol. 19, no. 1, Oct. 2018, doi: 10.1186/s12859-018-2375-z

    2018 doi

  17. [25]

    LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation,

    A. Chaurasia and E. Culurciello, “LinkNet: Exploiting Encoder Representations for Efficient Semantic Segmentation,” 2017 IEEE Vis. Commun. Image Process. VCIP 2017, vol. 2018-Janua, Jun. 2017, doi: 10.1109/VCIP.2017.8305148

    2017

  18. [26]

    H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “PSPNet,” Proc. - 30th IEEE Conf. Comput. Vis. Pattern Recognition, CVPR 2017, vol. 2017-Janua, Dec. 2017, doi: 10.1109/CVPR.2017.660

    2017 doi

  19. [27]

    UNet++: A Nested U-Net Architecture for Medical Image Segmentation,

    Z. Zhou, M. M. R. Siddiquee, N. Tajbakhsh, and J. Liang, “UNet++: A Nested U-Net Architecture for Medical Image Segmentation,” Jul. 2018, Accessed: 18 Aug. 26, 2020. [Online]. Available: http://arxiv.org/abs/1807.10165

    2018 arXiv

  20. [28]

    Faster R- CNN: Towards Real-Time Object Detection with Region Proposal Networks,

    S. Ren, K. He, R. Girshick, and J. Sun, “Faster R- CNN: Towards Real-Time Object Detection with Region Proposal Networks,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 39, no. 6, Jun. 2017, doi: 10.1109/TPAMI.2016.2577031

    2017

  21. [29]

    Rich feature hierarchies for accurate object detection and semantic segmentation,

    R. Girshick, J. Donahue, T. Darrell, and J. Malik, “Rich feature hierarchies for accurate object detection and semantic segmentation,” in Proceedings of the IEEE Computer Society Conference on Computer Vision and Pattern Recognition, Sep. 2014, pp. 580–587, doi: 10.1109/CVPR.2014.81

    2014 doi

  22. [30]

    Feature Pyramid Networks for Object Detection,

    T.-Y. Lin, P. Dollár, R. Girshick, K. He, B. Hariharan, and S. Belongie, “Feature Pyramid Networks for Object Detection,” Dec. 2016, Accessed: Aug. 26, 2020. [Online]. Available: http://arxiv.org/abs/1612.03144

    2016 arXiv

  23. [31]

    Mask R-CNN,

    K. He, G. Gkioxari, P. Dollár, and R. Girshick, “Mask R-CNN,” IEEE Trans. Pattern Anal. Mach. Intell., vol. 42, no. 2, Feb. 2020, doi: 10.1109/TPAMI.2018.2844175

    2020

  24. [32]

    Validation

    thresholding. Using such algorithms, it is possible to find cell -shaped objects in the input images. Some outputs do require further processing, but all methods basically find the same set of cells in the images. Nonetheless, these methods with their default settings find all...

  25. [33]

    rbnvrw/nd2reader

    R. Verweij, “rbnvrw/nd2reader.” Jun. 2020, Accessed: Aug. 22, 2020. [Online]. Available: https://github.com/rbnvrw/nd2reader

    2020

  26. [34]

    Minimum error thresholding,

    J. Kittler and J. Illingworth, “Minimum error thresholding,” Pattern Recognit., vol. 19, no. 1, Jan. 1986, doi: 10.1016/0031-3203(86)90030-0

    1986 doi

  27. [35]

    A New Criterion for Automatic Multilevel Thresholding,

    J. C. Yen, F. J. Chang, and S. Chang, “A New Criterion for Automatic Multilevel Thresholding,” IEEE Trans. Image Process., vol. 4, no. 3, 1995, doi: 10.1109/83.366472

    1995 doi

  28. [36]

    Adam: A method for stochastic optimization,

    D. P. Kingma and J. L. Ba, “Adam: A method for stochastic optimization,” Dec. 2015, Accessed: Aug. 27, 2020. [Online]. Available: https://arxiv.org/abs/1412.6980v9. 19 APPENDICES Appendix A: Relevant Links

    2015 arXiv

  29. [37]

    MicroAnalyzer Repository - https://github.com/JG-codies/MicroAnalyzer

  30. [38]

    Neural Network Model Architectures - https://github.com/JG- codies/MicroAnalyzer/blob/master/Notebooks/m odel_summaries.ipynb

  31. [39]

    Dataset - https://drive.google.com/drive/folders/1byIX3Dt aSTsBLF8a91012ljtrNkqVGa8?usp=sharing

  32. [40]

    Neural Network Models Training8 - Cells – https://github.com/JG- codies/MicroAnalyzer/blob/master/Notebooks/ce ll_training.ipynb - Clusters – https://github.com/JG- codies/MicroAnalyzer/blob/master/Notebooks/fl uo_training.ipynb

  33. [41]

    Validation - Image 1 – https://github.com/JG- codies/MicroAnalyzer/blob/master/Notebooks/V alidation%20Image%201.ipynb - Image 2 – https://github.com/JG- codies/MicroAnalyzer/blob/master/Notebooks/V alidation%20Image%202.ipynb

  34. [42]

    Usage Demo - https://github.com/JG- codies/MicroAnalyzer/blob/master/Notebooks/us age_demo.ipynb

  35. [43]

    Google Colab

    Sample DB - https://github.com/JG- codies/MicroAnalyzer/blob/master/samp le_output/validation_images/1/database.c sv Appendix B: “Soft” Segmentation Evaluation Remember that since the experiment is evaluated using multiple ground truth segmentations, the definition the evaluat...

  36. [44]

    Convert the image to 8bit RGB image (all channels are the same)

  37. [45]

    <Fluorescence-name> cell mean intensity

    Pad the images such that they’re dimensions a re divisible by 25 . This allows us to down -sample the image evenly at least five times (required by the neural networks). Each fluorescence image is pre -processed in the same way as the bacteria image, except that the R channel ...

Pith tools

Reviewed August 27, 2026 · model on record in the stance chip above.