List of Papers Browse by Subject Areas Author List
Abstract
Diagnosis of hematological malignancies depends on accurate identification of white blood cells in peripheral blood smears. Deep learning techniques are emerging as a viable solution to scale and optimize this process by automatic cell classification. However, these techniques face several challenges such as limited generalizability, sensitivity to domain shifts, and lack of explainability. Here, we introduce a novel approach for white blood cell classification based on neural cellular automata (NCA). We test our approach on three datasets of white blood cell images and show that we achieve competitive performance compared to conventional methods. Our NCA-based method is significantly smaller in terms of parameters and exhibits robustness to domain shifts. Furthermore, the architecture is inherently explainable, providing insights into the decision process for each classification, which helps to understand and validate model predictions. Our results demonstrate that NCA can be used for image classification, and that they address key challenges of conventional methods, indicating a high potential for applicability in clinical practice.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2024/paper/3422_paper.pdf
SharedIt Link: https://rdcu.be/dV1Xf
SpringerLink (DOI): https://doi.org/10.1007/978-3-031-72384-1_65
Supplementary Material: N/A
Link to the Code Repository
https://github.com/marrlab/WBC-NCA
Link to the Dataset(s)
N/A
BibTex
@InProceedings{Deu_Neural_MICCAI2024,
author = { Deutges, Michael and Sadafi, Ario and Navab, Nassir and Marr, Carsten},
title = { { Neural Cellular Automata for Lightweight, Robust and Explainable Classification of White Blood Cell Images } },
booktitle = {proceedings of Medical Image Computing and Computer Assisted Intervention -- MICCAI 2024},
year = {2024},
publisher = {Springer Nature Switzerland},
volume = {LNCS 15003},
month = {October},
page = {693 -- 702}
}
Reviews
Review #1
- Please describe the contribution of the paper
Neural cellular automata (NCA) are a light-weight, trainable class of models that have been demonstrated to be useful in image generation, image denoising and most recently in image segmentation. This work presents a model based on NCA for performing image classification, with a focus on white blood cell classification. The proposed method is novel, light-weight, and shows interesting properties (transfer to new domans, and explainability).
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
-
NCA for image classification has not been addressed in the literature and the authors propose a method using NCA in this work.
-
The method is simple, and has fewer trainable parameters than the baselines reported.
-
Experimental evaluation on three datasets, and the cross-dataset performance show the proposed method shows better performance compared to the ResNext model in some cases.
-
The inherent explainability achieved using the proposed method can be useful in medical image analysis applications.
-
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
-
One of the main concerns with the work is unclear description of the method. While the Fig. 1, presents an overview in three stages (A,B,C) there is no clear correspondence to these operations in the text or equations. For instance, in Step B, there is a “conv” and “identity” operation mentioned. Where are these described in the text, and what is the use of these operations? Also, does this mean the method not only uses dense layers, it also uses convolution operations? This dilutes the contribution of the proposed method.
-
It is unclear as to how the 80k parameters are shared between NCA, conv, dense layers. I would suggest the authors report these numbers in order to emphasize the role of NCAs. It might even help strengthen their case for using NCAs. On a related note, the memory consumption or training time of the proposed model could also shed light on the additional overhead compared to using standard deep learning models.
-
Why are the results for ResNext not available for Matek-19 dataset in Table 1., even though the method was proposed in Matek-19?
-
Some key references might be missing in the work. [1,2]
[1] Florindo, J. B., & Metze, K. (2021). A cellular automata approach to local patterns for texture recognition. Expert Systems with Applications, 179, 115027.
[2] Tesfaldet, M., Nowrouzezahrai, D., & Pal, C. (2022). Attention-based neural cellular automata. Advances in Neural Information Processing Systems, 35, 8174-8186.
-
- Please rate the clarity and organization of this paper
Poor
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
See comments above.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Reject — could be rejected, dependent on rebuttal (3)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
This is an interesting methodological contribution but currently lacks sufficient clarity in the method description. My rating is primarily based on this deficit in the paper.
- Reviewer confidence
Confident but not absolutely certain (3)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Weak Accept — could be accepted, dependent on rebuttal (4)
- [Post rebuttal] Please justify your decision
The authors have clarified most of my concerns, and have committed to adjusting some of the description in the paper to improve clarity. I generally think this is a promising line of work, and this method has potential. This being said, the specific contribution of conv layers in this setting remains unclear.
I will raise my score from weak reject to weak accept.
Review #2
- Please describe the contribution of the paper
The authors applied neural cellular automata (NCA) on white blood cells classification using RGB images. NCA modeling offers a lightweight backbone and a more developed explainability side compared to current deep-learning based methods. In addition, this work provided an in-house dataset that comprises approximately 42000 images from 18 different classes with a resolution of 288x288 pixels or 25x25 micrometers.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The main strength of this paper is the contribution of a rich in-house white blood cell dataset from 18 different classes. The authors also introduced a novel application of neural cellular automata(NCA) for white blood cells classifications. The explainability of this NCA method was explored by examining the features contribution, which is a nice addition compared to regular CNN-based deep learning architectures.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
The application of neural cellular automota (NCA) to white blood cells classification is novel. However, no significant improvement was made to the NCA baseline. Hence the novelty of the method is limited, and might not alighn with the scope of MICCAI.
- Please rate the clarity and organization of this paper
Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The authors claimed to release the source code and/or dataset upon acceptance of the submission.
- Do you have any additional comments regarding the paper’s reproducibility?
N/A
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
The paper was well-written with a detailed explanation of the methodology used. The analysis of the different NCA hyper-parameters on training and evaluation performance was insighteful. and valuable for future research in NCA development.
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Weak Reject — could be rejected, dependent on rebuttal (3)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
No significant improvement was made to the NCA baseline. Hence the novelty of the method is limited, and might not align with the scope of MICCAI. The in-house dataset that the authors used for evaluation and training could be very beneficial for the research community, if made public.
- Reviewer confidence
Very confident (4)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
N/A
- [Post rebuttal] Please justify your decision
N/A
Review #3
- Please describe the contribution of the paper
The authors introduce an explainable neural cellular automata-based modeling scheme for analysis of white blood cells.
- Please list the main strengths of the paper; you should write about a novel formulation, an original way to use data, demonstration of clinical feasibility, a novel application, a particularly strong evaluation, or anything else that is a strong aspect of this work. Please provide details, for instance, if a method is novel, explain what aspect is novel and why this is interesting.
The authors use a novel combination of white blood cell data and neural cellular automata (NCA) formulation to model it. The authors use a novel formulation for this problem, and formulate it well.
- Please list the main weaknesses of the paper. Please provide details, for instance, if you think a method is not novel, explain why and provide a reference to prior work.
Not much to criticize. Consider talking more about clinical practice.
- Please rate the clarity and organization of this paper
Very Good
- Please comment on the reproducibility of the paper. Please be aware that providing code and data is a plus, but not a requirement for acceptance.
The submission has provided an anonymized link to the source code, dataset, or any other dependencies.
- Do you have any additional comments regarding the paper’s reproducibility?
Will the dataset be released?
- Please provide detailed and constructive comments for the authors. Please also refer to our Reviewer’s guide on what makes a good review. Pay specific attention to the different assessment criteria for the different paper categories (MIC, CAI, Clinical Translation of Methodology, Health Equity): https://conferences.miccai.org/2024/en/REVIEWER-GUIDELINES.html
Good paper!
- Rate the paper on a scale of 1-6, 6 being the strongest (6-4: accept; 3-1: reject). Please use the entire range of the distribution. Spreading the score helps create a distribution for decision-making
Accept — should be accepted, independent of rebuttal (5)
- Please justify your recommendation. What were the major factors that led you to your overall score for this paper?
The unique formulation and the author’s use of multiple datasets including their own. The authors will release code.
- Reviewer confidence
Somewhat confident (2)
- [Post rebuttal] After reading the author’s rebuttal, state your overall opinion of the paper if it has been changed
Accept — should be accepted, independent of rebuttal (5)
- [Post rebuttal] Please justify your decision
This work seems unique and novel to me.
Author Feedback
Dear Area Chair, Dear Reviewers, We would like to extend our sincere thanks to the three reviewers for their valuable feedback and for highlighting the strengths of our work. We appreciate their time, effort and the constructive comments which will help us improve our manuscript. We are glad that all reviewers highlighted several strengths including the novelty of utilizing Neural Cellular Automata (NCA) for image classification (R1, R3, R4), its advantages in terms of the inherent explainability (R3, R4), and the simplicity and lightweight nature of our approach (R3). The recognition of our validation efforts across three datasets including our own (R1, R3, R4) and the analysis of parameters (R4) are encouraging, and we are pleased to receive positive comments regarding the clarity and organization of our paper (R1, R4), in particular their emphasis on the detailed explanation of our methodology (R4). We also appreciate the feedback from Reviewer 3 and are fully committed to addressing the suggestions to enhance clarity and transparency in our manuscript. We believe that with adjustments to the text and graphics, such as directly linking the NCA architecture description with the corresponding figure 1, we can significantly improve the clarity of our methods. Furthermore, we will reference part B of figure 1 directly in the section after equations 2 to 5 and include symbols used in the equations within the graphic, such as N_c and the perception vector f_p(N_c). Additionally, we will clarify in figure 1’s caption that section B corresponds to equations 2 to 5. Moreover, in the same figure, we will easily address how parameters are distributed between the classifier and the NCA, as well as within the NCA components (R3). Regarding the question about the missing values in table 1 (R3), we would like to clarify that both columns of “Matek et al” and “ResNeXt” are referring to the same architecture and same setup. In the column titled “Matek et al.”, we included the evaluation results reported by Matek et al., while in the column titled “ResNeXt” we trained the same approach on the remaining two datasets to provide a complete report of the performance of the mentioned model on all three datasets. Following the Reviewer’s remark, we will merge the two columns in the revised version of our manuscript, denote it “ResNeXt”, and describe the evaluation results accordingly in the caption, to avoid confusion for the reader. We are grateful for the suggestions regarding additional literature (R3) and will reference them while specifying the distinctions to our method in the Introduction and Discussion sections of our revised manuscript. With that, we are confident to appropriately address Reviewer 3’s concern about the “clarity in the method description”. Lastly, we wish to address the comment about the perceived limited novelty of our method due to the lack of advancements to the NCA backbone (R4). Our focus has been on demonstrating the potential of NCA for image classification, which has not been explored yet and is thus methodologically novel, while being “useful for medical image analysis” (R3). With our work, we were able to showcase the advantages of NCA, such as extreme lightweight storage for high accessibility, and generalizability to unseen data, and regard this as an innovative starting point for further exploration and refinement of NCA-based image classification within the research community. We thus believe that we are well within the scope of MICCAI 2024, which explicitly lists “New methods in medical image computing”, “Accessible medical imaging solutions”, and “Generalizable machine learning in medical imaging” as topics of interest. Thank you once again for your thorough review and insightful feedback.
Meta-Review
Meta-review #1
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
Although some reviewers admit that they are not familiar with the area, the author rebuttal seems to address most of the concerns.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
Although some reviewers admit that they are not familiar with the area, the author rebuttal seems to address most of the concerns.
Meta-review #2
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Reject
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
R1’s reviews are superficial. As highlighted by R3, some key references are missing where NCA is used for image classification [additional refs]. The paper emphasizes the use of NCA for image classification as its main strength. Given the missing references, the impact is less than claimed. If the paper is accepted, I suggest adjusting the tone of the claim and crediting previous studies. The relevance score of features is a nice addition. However, explainability does not directly come with feature visualization. Background segmentation, a standard attention mechanism, could have been done prior to analyzing this visualization further. [ref] Randazzo E, Mordvintsev A, Niklasson E, Levin M, Greydanus S. Self-classifying mnist digits. Distill. 2020 Aug 27;5(8):e00027-002. [ref] Yeşil Ç, Korkmaz EE. A novel cellular automata-based approach for generating convolutional filters. Machine Vision and Applications. 2023 May;34(3):38.
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
R1’s reviews are superficial. As highlighted by R3, some key references are missing where NCA is used for image classification [additional refs]. The paper emphasizes the use of NCA for image classification as its main strength. Given the missing references, the impact is less than claimed. If the paper is accepted, I suggest adjusting the tone of the claim and crediting previous studies. The relevance score of features is a nice addition. However, explainability does not directly come with feature visualization. Background segmentation, a standard attention mechanism, could have been done prior to analyzing this visualization further. [ref] Randazzo E, Mordvintsev A, Niklasson E, Levin M, Greydanus S. Self-classifying mnist digits. Distill. 2020 Aug 27;5(8):e00027-002. [ref] Yeşil Ç, Korkmaz EE. A novel cellular automata-based approach for generating convolutional filters. Machine Vision and Applications. 2023 May;34(3):38.
Meta-review #3
- After you have reviewed the rebuttal and updated reviews, please provide your recommendation based on all reviews and the authors’ rebuttal.
Accept
- Please justify your recommendation. You may optionally write justifications for ‘accepts’, but are expected to write a justification for ‘rejects’
N/A
- What is the rank of this paper among all your rebuttal papers? Use a number between 1/n (best paper in your stack) and n/n (worst paper in your stack of n papers). If this paper is among the bottom 30% of your stack, feel free to use NR (not ranked).
N/A