Deep Transfer Learning in Diagnosing Leukemia in Blood Cells

Round 1

Reviewer 1 Report

This paper uses a transfer learning method to diagnose leukemia. The authors use AlexNet and achieve 100% accuracy. Technical novelty is low but this application is important for the social good. Thus, I agree to publish if you respond to my comments.

Majors:
- Please evaluate the second model by 10-fold cross-validation.
- Please explain why you use AlexNet. Why didn't you use ResNet?
- If there is no problem, please upload the dataset.
- Please add the explanation of the transfer learning in Section 2. Here are suggested references:
[1]Sawada et.al, "Transfer learning method using multi-prediction deep boltzmann machines for a small scale dataset"
[2]Sawada et.al, "All-Transfer Learning for Deep Neural Networks and its Application to Sepsis Classification"
[3]Sawada et.al, "Improvement in Classification Performance Based on Target Vector Modification for All-Transfer Deep Learning"
[4]Gu et.al, "Progressive Transfer Learning and Adversarial Domain Adaptation for Cross-Domain Skin Disease Classification"
[5]Ganin and Lempitsky, "Unsupervised Domain Adaptation by Backpropagation"
[6]Tzeng et.al, "Adversarial Discriminative Domain Adaptation"
[7]Zamir et.al, "Taskonomy: Disentangling Task Transfer Learning"

[5]-[7] does not the medical field papers, but are important as the transfer learning method. Also, here is not the transfer learning paper, but very important.
[8]Esteva et.al, "Dermatologist-level classification of skin cancer with deep neural networks"

Minors:
- All figures are low-resolution. Please increase the resolution.
- l.38: history.[6] -> history[6].
- l.38: "examination" was text garbled.
- l.184: Line break mistake

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

Deep learning /machine learning already outperforms humans as a diagnostic especially in radiology, cardiology (reading of ECGs) and now in pathology. This work is both timely and topical and has the potential too translate to clinical use. The early detection of leukemia can help effectively in its treatment. This study proposed two 287 classification models distinguishing between leukemia-free and leukemia-affected blood 288 microscopic images. Both models employ transfer learning. In the first model, a pre-trained CNN 289 known as AlexNet is employed to extract the discriminant features and other well-known classifiers, 290 such as DT, LD, SVM, and K-NN, are employed for classification. Experiments demonstrated the 291 superiority of the SVM classifier. The second model employs AlexNet for both feature extraction and 292 classification. Experiments for this model demonstrated its superiority to the first model with 293 respect to various performance metrics. 

Author Response

Thank you for reviewing the manuscript

Round 2

Reviewer 1 Report

Majors:

- Please write the relationship between your work and related works, in Sec.2.
In the current version, readers cannot understand the technical flow of research from related works to your work.

- Please write the reason you choose AlexNet in your manuscript because the choosing network architecture is important to apply DNN to medical fields.

Minor:
l.172, 181: We -> They? I think 'we' is odd.

Author Response

Please see the attachment

Author Response File: Author Response.pdf