Pith. sign in

REVIEW 4 major objections 5 minor 34 references

A hybrid YOLOv5–MobileNetV2–logistic-regression pipeline detects and classifies white blood cells into four types at high reported accuracy on the BCCD dataset.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-11 18:45 UTC pith:EG4ILX5E

load-bearing objection Standard YOLOv5 + MobileNetV2 + LR pipeline on BCCD with high reported numbers, but the headline RBC F1 claim is arithmetically impossible from the paper's own precision/recall table. the 4 major comments →

arxiv 2607.04486 v1 pith:EG4ILX5E submitted 2026-07-05 cs.LG cs.CV

LeukocyteCount: Automatic Identification and Counting for leukocytes using Deep Learning

Ahmed M. Sayed (1) , Sondos A. Refaat (1) , Abdallah M. Mostafa (1) , Mariam S. El-Rahmany (1) , Ensaf Hussein Mohamed (1 , 2) ((1) Faculty of Computers , Artificial Intelligence , Helwan University
show 7 more authors
Cairo Egypt (2) School of Information Technology Computer Science (ITCS) Nile University Giza Egypt)
This is my paper
classification cs.LG cs.CV
keywords white blood cellsleukocyte countingYOLOv5MobileNetV2logistic regressionBCCD datasetblood cell detectiondeep learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Manual white-blood-cell counting and typing is slow and error-prone, yet the numbers matter for diagnosing infection and immune disorders. This paper builds an automated three-stage pipeline that first uses YOLOv5 to locate red cells, white cells and platelets in smear images, then crops each white cell and feeds it to a MobileNetV2 feature extractor whose output is classified by logistic regression into eosinophil, lymphocyte, monocyte or neutrophil. Trained and tested on the public BCCD collection, the system reports 98–100 percent detection accuracy for white cells, 99.04 percent (or 98.14 percent) four-class accuracy, and an F1 of 99.73 percent for red-cell detection that the authors say exceeds a prior baseline by more than thirteen points. The practical claim is that the same pipeline can replace or assist the laborious microscopic differential count, delivering both the total count and the subtype percentages that clinicians use for diagnosis.

Core claim

A hybrid cascade of YOLOv5 detection followed by MobileNetV2 feature extraction and logistic-regression classification can locate the three major blood-cell classes and then assign every white cell to one of four subtypes with reported accuracies of 98 percent detection and 99.04 percent classification on the BCCD data, while the red-cell detector alone reaches an F1 of 99.73 percent.

What carries the argument

The staged hybrid model: YOLOv5 produces bounding boxes for red cells, white cells and platelets; each white-cell crop is resized to 224 imes224 and passed through unfrozen MobileNetV2 (ImageNet weights) to obtain a feature vector that a logistic-regression classifier maps to one of four white-cell types; a simple counting loop then tallies each type.

Load-bearing premise

That the reported high F1 and accuracy numbers are internally consistent and were measured on a test set large enough and representative enough of real clinical smears to support the claimed performance.

What would settle it

Re-running the identical YOLOv5 configuration on the same 36-image BCCD test split and recomputing red-cell precision, recall and F1; if the arithmetic identity F1 = 2PR/(P+R) does not recover the claimed 99.73 percent, or if performance collapses on an independent multi-center smear set, the central performance claim fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a hybrid pipeline for automated detection, counting, and four-class classification of white blood cells (plus RBC/platelet detection) on the BCCD dataset. YOLOv5 is used for multi-class blood-cell detection and cropping of WBCs; cropped cells are then featurized by a fully unfrozen MobileNetV2 and classified by logistic regression. The authors claim ~98–100% WBC detection accuracy, 98.14–99.04% four-class classification accuracy, and an RBC detection F1 of 99.73% that substantially outperforms a prior RetinaNet baseline of 86.49% on a 36-image held-out set.

Significance. If the reported numbers were internally consistent and obtained on adequately sized, multi-source clinical data, the work would be a useful incremental engineering contribution: a practical YOLOv5 + MobileNetV2 + LR cascade that jointly counts and subtypes leukocytes. The pipeline is straightforward, re-uses public BCCD data, and includes a limitations section. However, the headline comparative claim (RBC F1 99.73% vs 86.49%) is arithmetically impossible from the authors’ own precision/recall figures, and accuracy figures for both stages are stated inconsistently across abstract, contributions, tables, body text and conclusion. These load-bearing inconsistencies currently prevent the results from being usable or citable.

major comments (4)
  1. [Abstract; §1 contribution 4; Table 4; Table 5] Table 4 reports RBC precision = 0.769 and recall = 0.841, whose harmonic mean is F1 ≈ 0.803 (exactly the F1 also listed in Table 4). Yet the abstract, contribution 4, and Table 5 all claim an RBC F1 of 99.73% (and an outperformance of Dralus et al.’s 86.49% by 13.24 points) on the identical 36-image test split. This is an arithmetic contradiction that nullifies the paper’s strongest quantitative novelty claim for the detection stage. The authors must either recompute and correct every occurrence of the 99.73% figure or supply the raw confusion-matrix counts that could justify a different F1.
  2. [Abstract; §4.2.1–4.2.2; Tables 5–6, 9; §5 Conclusion] Detection and classification accuracies are stated inconsistently: abstract claims 98% detection / 99.04% classification; body text and Table 6 claim 100% WBC detection; Table 5 and conclusion claim 99.14% segmentation accuracy; Table 9 and conclusion claim 98.14% classification accuracy while the abstract and §4.2.2 again claim 99.04%. These cannot all be true for the same experiments. A single, carefully defined set of metrics (with the exact test-set sizes and class definitions) must be used throughout.
  3. [§3.2.1; Tables 5–7; §4.3 Limitations] All detection claims rest on a 36-image held-out split (Tables 5–7). With only 36 images the reported point estimates (especially the 100% WBC detection and the disputed RBC F1) have extremely wide confidence intervals and cannot support statements of clinical-grade or state-of-the-art performance. The authors should either enlarge the test set, report bootstrap/CI intervals, or substantially temper the claims.
  4. [§4.2.2; Table 3; §3.1] §4.2.2 reports per-class AP values that include a Basophil class (AP 0.964), yet the paper repeatedly states that only four WBC types are considered (Eosinophil, Lymphocyte, Monocyte, Neutrophil) and the BCCD classification set used contains only those four. The appearance of a fifth class is unexplained and further undermines confidence in the reported classification metrics.
minor comments (5)
  1. [References] Several references appear incomplete or placeholder-like (e.g., [5] “Author, A., Author, B., Author, C.”; [1–2] are generic SOM citations that add little). Clean the bibliography.
  2. [§3.2] Figure 1 is described but the actual pipeline diagram is not visible in the supplied text; ensure all figures are present and legible.
  3. [Abstract; §1; §4.2.2] The abstract and introduction still speak of “four distinct types” while later text and the mAP list introduce Basophils; align the class inventory everywhere.
  4. [Throughout] Typographical inconsistencies: “Yolov5” vs “YOLOv5”, “MobileNetv2” vs “MobileNetV2”, and occasional missing spaces or duplicated phrases (“Xia Xia [15]”).
  5. [Table 5] Table 5 lists Precision 76.9 / Recall 84.1 yet F1 99.73; even after correcting the F1, the precision/recall should be reported to three decimal places consistently with Table 4.

Circularity Check

1 steps flagged

Empirical ML pipeline with only mild non-load-bearing self-citation for classifier choice; no derivation reduces to inputs by construction.

specific steps
  1. self citation load bearing [§3.2.4 Classification Module; also §4.2.2 and Table 9]
    "Based on Ensaf et al. [30], Logistic Regression has been chosen at this stage. ... We built this module using the research conducted by Ensaf et al. [8] as a foundation. ... Table 9 ... Ensaf et al. [30] (MobileNetv1 + Logistic Regression) 97.03% Our model (MobileNetv2 + Logistic Regression) 98.14%"

    The architectural decision to pair a MobileNet feature extractor with Logistic Regression is justified solely by the authors' own prior paper rather than by an independent ablation or external theorem. The comparison baseline is likewise self-cited. The circularity is mild: the reported accuracy numbers themselves are fresh measurements on a new backbone, not algebraic restatements of the cited result.

full rationale

This is a standard empirical deep-learning paper reporting measured detection/classification accuracies of a YOLOv5 + MobileNetV2 + Logistic Regression pipeline on BCCD variants. There is no first-principles derivation, uniqueness theorem, fitted constant renamed as prediction, or ansatz smuggled via citation. The sole self-citation (to the corresponding author's prior MobileNet+LR hybrid [30]) merely motivates the choice of Logistic Regression as the final classifier and supplies a baseline accuracy number that the present experiments then re-measure independently with MobileNetV2; the new F1/accuracy figures are not algebraically forced by that citation. Internal numerical contradictions (e.g., Table 4 RBC F1 = 0.803 vs. claimed 99.73 %) are correctness defects, not circularity. Hence only a minor self-citation flag is warranted.

Axiom & Free-Parameter Ledger

4 free parameters · 3 axioms · 0 invented entities

The paper is an empirical deep-learning application; its claims rest on standard ML tooling assumptions, public dataset labels, and a handful of training choices rather than novel physical or mathematical postulates. No new entities are invented. Free parameters are the usual training hyper-parameters and the decision to leave all MobileNetV2 layers unfrozen.

free parameters (4)
  • YOLOv5 training epochs = 100
    Fixed at 100; convergence claimed after ~70 epochs. Directly affects reported detection metrics.
  • MobileNetV2 fine-tuning (all layers unfrozen) = all layers trainable
    Authors unfreeze every layer rather than freeze the backbone; this choice is not ablated and influences the 98–99% classification numbers.
  • Logistic Regression defaults (C=1.0, L2, L-BFGS) = C=1.0, L2, tol=1e-4
    Default sklearn settings taken from prior work; no hyper-parameter search reported.
  • Roboflow augmentation set and 416×416 resize = flips, 90° rotations, saturation/exposure
    Augmentations expand the detection set from 364 to 874 images; exact parameters affect generalization claims.
axioms (3)
  • domain assumption BCCD ground-truth bounding boxes and four-class labels are correct and complete.
    All accuracy, precision, recall and F1 numbers are computed against these labels; no independent re-annotation is performed.
  • ad hoc to paper A 36-image detection test split and the single-source BCCD classification split are sufficient to claim clinical-grade performance.
    Stated in §4 and Limitations; the paper itself later notes that external clinical data are needed.
  • domain assumption Transfer learning from ImageNet via MobileNetV2 yields features adequate for stained blood-cell discrimination once all layers are fine-tuned.
    Standard transfer-learning premise used in §3.2.3; not re-validated against medical-domain pre-training.

pith-pipeline@v1.1.0-grok45 · 17082 in / 3059 out tokens · 35670 ms · 2026-07-11T18:45:55.254545+00:00 · methodology

0 comments
read the original abstract

Diagnosing and monitoring diseases frequently involves the analysis of human biological samples, with blood analysis being pivotal. Specifically, leukocytes, or white blood cells (WBCs), are essential markers for evaluating the body's defense mechanisms against infections. Traditional methods for WBC counting and classification are labor-intensive and prone to inaccuracies, primarily due to human error. The conventional processes for blood cell analysis, especially those concerning WBCs, are beset with difficulties. These include the laborious nature of manual counting and the susceptibility to errors, which can significantly impact the accuracy and reliability of disease diagnosis and monitoring. This study proposes an automated, machine learning-based solution aimed at mitigating the identified challenges. By employing a hybrid model that integrates Yolov5 for the detection of WBCs, coupled with a finely tuned, pre-trained MobileNetV2 model and a Logistic Regression classifier, the study innovates in the accurate identification, counting, and classification of WBCs into four distinct types. The methodology leverages the BCCD dataset for training and validation purposes. The application of the proposed hybrid machine learning model has yielded remarkable results, demonstrating a detection accuracy rate of 98\% through the Yolov5 stage, and an unparalleled classification accuracy of 99.04\% in subsequent stages utilizing MobileNetV2 and Logistic Regression. Additionally, Our proposed YOLOv5-based RBC detection module achieves an F1 score of 99.73\%, which outperforms the baseline. These findings underscore the model's potential in transforming traditional laboratory practices for WBC analysis, offering a path towards more accurate, efficient, and reliable disease diagnostics and monitoring.

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

34 extracted references · 12 canonical work pages · 1 internal anchor

  1. [1]

    Springer, ??? (2001)

    Kohonen, T.: Self-Organizing Maps, 3rd edn. Springer, ??? (2001)

    2001

  2. [2]

    IEEE Transactions on Medical Imaging20(12), 1234–1243 (2001)

    Zhang, Y.,et al.: Medical image segmentation using self-organizing maps and neural networks. IEEE Transactions on Medical Imaging20(12), 1234–1243 (2001)

    2001

  3. [3]

    Medical Image Analysis42, 60–88 (2017)

    Litjens, G.,et al.: A survey on deep learning in medical image analysis. Medical Image Analysis42, 60–88 (2017)

    2017

  4. [4]

    Journal of Big Data8(1), 1–74 (2021)

    Alzubaidi, L.,et al.: Review of deep learning: concepts, cnn architectures, challenges, applications, future directions. Journal of Big Data8(1), 1–74 (2021)

    2021

  5. [5]

    International Journal of Medical Informatics170, 104974 (2023)

    Author, A., Author, B., Author, C.: A fast and efficient CNN model for B-ALL diagnosis and its subtypes classification using peripheral blood smear images. International Journal of Medical Informatics170, 104974 (2023). https://doi.org/10.1016/j.ijmedinf.2022.104974

    work page doi:10.1016/j.ijmedinf.2022.104974 2023

  6. [6]

    International Journal of Medical, Health, Biomedical and Pharmaceutical Engineering7(1), 15–33 (2013)

    Putzu, L., Di Ruberto, C.: White blood cells identification and counting from microscopic blood images. International Journal of Medical, Health, Biomedical and Pharmaceutical Engineering7(1), 15–33 (2013)

    2013

  7. [7]

    University of Texas at Arlington, Pearson Education (2015)

    Frederick, L.A.: Chapter 19: Blood. University of Texas at Arlington, Pearson Education (2015)

    2015

  8. [8]

    arXiv e-prints (2019) https://arxiv.org/abs/1807.05511 Springer Nature 2021 LATEX template 18LeukocyteCount: Automatic Identification & Counting for leukocytes using DL

    Zhao, Z.-Q., Xu, S.-t., Wu, X.: Object detection with deep learning: A review. arXiv e-prints (2019) https://arxiv.org/abs/1807.05511 Springer Nature 2021 LATEX template 18LeukocyteCount: Automatic Identification & Counting for leukocytes using DL

    Pith/arXiv arXiv 2019

  9. [9]

    Hassaballah, M., Abdelmgeid, A.A., Alshazly, H.A.: Image features detection, description and matching, pp. 11–45. Springer, ??? (2016)

    2016

  10. [10]

    The Scientific World Journal2014(2014)

    Su, M.C., Cheng, C.Y., Wang, P.C.: A neural-network based approach to white blood cell classification. The Scientific World Journal2014(2014). https://doi.org/10.1155/2014/796371

    work page doi:10.1155/2014/796371 2014

  11. [11]

    Radiological Physics and Technology10(3), 257–273 (2017)

    Suzuki, K.: Overview of deep learning in medical imaging. Radiological Physics and Technology10(3), 257–273 (2017). https://doi.org/10.1007/ s12194-017-0406-5

    2017

  12. [12]

    Healthcare Technology Letters 6(4), 103–108 (2019)

    Alam, M.M., Islam, M.T.: A machine learning approach of automatic identification and counting of blood cells. Healthcare Technology Letters 6(4), 103–108 (2019)

    2019

  13. [13]

    Uni- versity of Washington (2020)

    Redmon, J., Farhadi, A.: YOLOv3: An incremental improvement. Uni- versity of Washington (2020)

    2020

  14. [14]

    OSA Continuum4(2), 649–658 (2021)

    Jiang, Z., Liu, X., Yan, Z., Gu, W., Jiang, J.: Improved detection per- formance in blood cell count by an attention-guided deep learning. OSA Continuum4(2), 649–658 (2021). https://doi.org/10.1364/OSAC.423771

    work page doi:10.1364/osac.423771 2021

  15. [15]

    Materials Science and Engineering646, 1–6 (2018)

    Xia, T., Jiang, R., Fu, Y., Jin, N.: Automated blood cell detection and counting via deep learning for microfluidic point-of-care medical devices. Materials Science and Engineering646, 1–6 (2018)

    2018

  16. [16]

    Entropy23, 1522 (2020)

    Dralus, G., Mazur, D., Czmil, A.: Automatic detection and counting of blood cells in smear images using retinanet. Entropy23, 1522 (2020). https://doi.org/10.3390/e23111522

    work page doi:10.3390/e23111522 2020

  17. [17]

    In: Proceedings of the 1st IEEE Conference on Indus- trial Electronics and Applications, pp

    Yampri, P., Pintavirooj, C., Daochai, S., Teartulakarn, S.: White blood cell classification based on the combination of eigen cell and parametric feature detection. In: Proceedings of the 1st IEEE Conference on Indus- trial Electronics and Applications, pp. 1–4 (2006). https://doi.org/10. 1109/ICIEA.2006.257341

    arXiv 2006

  18. [18]

    In: Cuba-Flanders Workshop on Machine Learning and Knowl- edge Discovery (2010)

    Falcon, A., Taboada-Crispi, A., Monteagudo, M.O., Perez, M.A., Sahli, H.: Classification of white blood cells using morphometric features of nucleus. In: Cuba-Flanders Workshop on Machine Learning and Knowl- edge Discovery (2010)

    2010

  19. [19]

    Journal of Theoretical and Applied Computer Science 7(1), 20–35 (2013)

    Habibzadeh, M., Krzy˙ zak, A., Fevens, T.: Comparative study of shape, intensity, and texture features and support vector machine for white blood cell classification. Journal of Theoretical and Applied Computer Science 7(1), 20–35 (2013)

    2013

  20. [20]

    In: 2014 International Conference on Advances in Computing, Communication and Informatics (ICACCI), pp

    Gautam, A., Bhadauria, H.: Classification of white blood cells based on Springer Nature 2021 LATEX template LeukocyteCount: Automatic Identification & Counting for leukocytes using DL19 morphological features. In: 2014 International Conference on Advances in Computing, Communication and Informatics (ICACCI), pp. 2363–2368 (2014). https://doi.org/10.1109/I...

    work page doi:10.1109/icacci.2014.6968362 2021

  21. [21]

    Biomedical Engineering Online14, 63 (2015)

    Prinyakupt, J., Pluempitiwiriyawej, C.: Segmentation of white blood cells and comparison of cell morphology by linear and na¨ ıve bayes classifiers. Biomedical Engineering Online14, 63 (2015). https://doi.org/10.1186/ s12938-015-0037-1

    2015

  22. [22]

    Applied Mathematical Sciences, 2227–2237 (2014)

    Ravikumar, S., Shanmugam, A.: Wbc image segmentation and classi- fication using rvm. Applied Mathematical Sciences, 2227–2237 (2014). https://doi.org/10.12988/ams.2014.43191

    work page doi:10.12988/ams.2014.43191 2014

  23. [23]

    Branching of Automorphic Fundamental Solutions

    Ravikumar, S.: Image segmentation and classification of white blood cells with the extreme learning machine and the fast relevance vector machine. Artificial Cells, Nanomedicine, and Biotechnology44(3), 985–989 (2016). https://doi.org/10.3109/21691401.2015.1008506

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.3109/21691401.2015.1008506 2016

  24. [24]

    Journal of Medical Signals and Sensors7(4), 220–227 (2017)

    Sarrafzadeh, O., Dehnavi, A.M., Banaem, H.Y., Talebi, A., Gharibi, A.: The best texture features for leukocytes recognition. Journal of Medical Signals and Sensors7(4), 220–227 (2017). https://doi.org/10.4103/jmss. JMSS-7-17

    work page doi:10.4103/jmss 2017

  25. [25]

    In: Medical Image Computing and Computer-Assisted Intervention (MICCAI), pp

    Fu, H., Xu, Y., Lin, S., Wong, D.K., Liu, J.: Deepvessel: Retinal vessel segmentation via deep learning and conditional random field. In: Medical Image Computing and Computer-Assisted Intervention (MICCAI), pp. 132–139 (2016). https://doi.org/10.1007/978-3-319-46723-8 16

    doi:10.1007/978-3-319-46723-8 2016

  26. [26]

    Computer Methods and Programs in Biomedicine168, 69–80 (2019)

    Shahin, A., Guo, Y., Amin, K., Sharawi, A.A.: White blood cells identi- fication system based on convolutional deep neural networks. Computer Methods and Programs in Biomedicine168, 69–80 (2019). https://doi. org/10.1016/j.cmpb.2017.11.015

    work page doi:10.1016/j.cmpb.2017.11.015 2019

  27. [27]

    PeerJ6, 4568 (2018)

    Rajaraman, S., Antani, S.K., Poostchi, M., Silamut, K., Hossain, M.A., Maude, R.J., Jaeger, S., Thoma, G.R.: Pre-trained convolutional neural networks as feature extractors toward improved malaria parasite detec- tion in thin blood smear images. PeerJ6, 4568 (2018). https://doi.org/ 10.7717/peerj.4568

    work page doi:10.7717/peerj.4568 2018

  28. [28]

    In: Proceedings of SPIE 10696, Tenth International Conference on Machine Vision, p

    Habibzadeh, M., Jannesari, M., Rezaei, Z., Baharvand, H., Totonchi, M.: Automatic white blood cell classification using pre-trained deep learning models: Resnet and inception. In: Proceedings of SPIE 10696, Tenth International Conference on Machine Vision, p. 1069612 (2018). https://doi.org/10.1117/12.2311282

    work page doi:10.1117/12.2311282 2018

  29. [29]

    IEEE Access6, 1–12 (2018)

    Liang, G., Hong, H., Xie, W., Zheng, L.: Combining convolutional neural Springer Nature 2021 LATEX template 20LeukocyteCount: Automatic Identification & Counting for leukocytes using DL network with recursive neural network for blood cell image classifica- tion. IEEE Access6, 1–12 (2018). https://doi.org/10.1109/ACCESS. 2018.2846685

    doi:10.1109/access 2021

  30. [30]

    Journal of Communication Software and Systems16(1), 1–9 (2020)

    Mohamed, E.H., El-Behaidy, W.H., Khoriba, G., Li, J.: Improved white blood cells classification based on pre-trained deep learning models. Journal of Communication Software and Systems16(1), 1–9 (2020)

    2020

  31. [31]

    Journal of Healthcare Engineering2021(2021)

    Yao, J., Huang, X., Wei, M., Han, W., Xu, X., Wang, R., Sun, L.: High- efficiency classification of white blood cells based on object detection. Journal of Healthcare Engineering2021(2021). https://doi.org/10.1155/ 2021/1615192

    2021

  32. [32]

    Biomedical Signal Processing and Control100, 106981 (2025)

    Lu, S.-Y., Zhu, Z., Tang, Y., Zhang, X., Liu, X.: Ctbvit: A novel vit for tuberculosis classification with efficient block and randomized classifier. Biomedical Signal Processing and Control100, 106981 (2025). https:// doi.org/10.1016/j.bspc.2024.106981

    work page doi:10.1016/j.bspc.2024.106981 2025

  33. [33]

    https://www.kaggle.com/datasets/ paultimothymooney/blood-cells

    Blood Cell Images. https://www.kaggle.com/datasets/ paultimothymooney/blood-cells

  34. [34]

    https://github.com/Shenggan/BCCD Dataset

    BCCD Dataset. https://github.com/Shenggan/BCCD Dataset