Next Article in Journal
Optimal Fractional PID Controller for Buck Converter Using Cohort Intelligent Algorithm
Previous Article in Journal
Development of an Optical System for Non-Contact Type Measurement of Heart Rate and Heart Rate Variability
 
 
Article
Peer-Review Record

Classification of Alzheimer’s Disease Patients Using Texture Analysis and Machine Learning

Appl. Syst. Innov. 2021, 4(3), 49; https://doi.org/10.3390/asi4030049
by Sumit Salunkhe 1, Mrinal Bachute 1,*, Shilpa Gite 2, Nishad Vyas 1, Saanil Khanna 1, Keta Modi 1, Chinmay Katpatal 1 and Ketan Kotecha 2
Reviewer 1: Anonymous
Reviewer 2: Anonymous
Appl. Syst. Innov. 2021, 4(3), 49; https://doi.org/10.3390/asi4030049
Submission received: 6 June 2021 / Revised: 27 July 2021 / Accepted: 28 July 2021 / Published: 4 August 2021
(This article belongs to the Section Medical Informatics and Healthcare Engineering)

Round 1

Reviewer 1 Report

Please see the attached PDF file.

Comments for author File: Comments.pdf

Author Response

Please see the attachment for details responses to the review comments

Author Response File: Author Response.docx

Reviewer 2 Report

General comments:

The goals of this work are to detect minor changes in the hippocampus by means of 3D texture analysis and to classify human subjects into Alzheimer’s disease (AD) and control normal groups.

The authors describe their methods in detail. The open Alzheimer’s Disease Neuro Imaging (ADNI) database was used to provide MRI images with T1 contrast. In total, datasets from 234 subjects were available. Automated hippocampus segmentation was performed with the online platform volBrain and the software package MRIcron. The 3D Gray Level Co-occurrence Matrix (GLCM) was generated and different texture attributes (features) were derived from the GLCM. The GLCM features were then fed into three different machine learning algorithms (support vector machine, decision tree ands and random forest) for classification purposes.

The contribution of this work lies in the testing of many different combinations of GLCM features and machine learning algorithms. It was found that eight GCLM features (out of 20) contribute most to the classification. In addition, merging texture features with volume and age data increases the accuracy. The best combination can achieve up to 90.2 percent accuracy for the ADNI data set.

I have only a few comments:

  1. Although the working principle of the different machine learning algorithms are well described (including supporting Figures), more implementation details would be very helpful for the reader (e.g. programming language, architecture and hyperparameters)
  2. In recent years, there have been many publications on the classification of Alzheimer ’s disease based on MRI images and machine learning. It is difficult to evaluate the results of the present study in comparison with other methods, e.g. voxel-based morphometry or direct classification using (deep) convolutional neural networks. It would be helpful to point out advantages and disadvantages of the texture analysis compared to other approaches.
  3. Introduction: Further works on machine learning based texture analysis for AD classification are presented in the Introduction. However, it does not become clear how the present study relates to prior work. For example, what is the contribution of this work with respect to previous studies, e.g. compared to a similar work from Zhang J, et al. Brain Imaging and Behavior; 2012?
  4. The discussion is relatively short. For example, it would be helpful to discuss what needs to be done to bring this approach into clinical routine? Is the accuracy already sufficient?
  5. The reference list is relatively short and Reference 1 is not complete

Author Response

Please see the attachment

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

Lines 161-165 answered my first and most critical concern about whether or not the correct procedure was performed for training and validation. 

Lines 302-303 remove the following words which are incorrect: "hence hyper parameters were disabled due to which confidence intervals are not provided"

Line 247 Figure 7 is still incorrect. Thank you for trying to indicate the support vectors for hard-margin SVM, but the diagram is still incorrect because now the margin has been omitted. Please refer to figure 5 in: Burges CJ. A tutorial on support vector machines for pattern recognition. Data mining and knowledge discovery. 1998 Jun;2(2):121-67.

Line 321 Please remove (or fix) the extra text that is not part of a sentence: machine learning algorithm linear boundaries [30]. 

Line 123: R2 still does not have a superscript for the number 2

Author Response

"Please see the attachment."

Author Response File: Author Response.docx

Reviewer 2 Report

General comments:

The authors have addressed all of my comments. More details on the machine learning process and algorithms are provided and a more detailed discussion of similar previous studies is conducted.

I have only minor comments:

  1. Introduction (line 78): Although the abbreviation GLCM is explained in the abstract it may be also explained in the main manuscript body.
  2. Introduction (line 122): The sentence about the dispersion of features in the z-layer may be revised („to Complex linear relationship see if“ does not sound correct).
  3. Line 177: The abbreviation SPGR (Spoiled gradient echo) may be defined here.
  4. Line 192: The abbreviation NPM may be defined here.
  5. Table 3: Please include definitions of the variables in the caption, e.g. for N_g, M_ij, mu, sigma, …
  6. Line 306: “It is noticeable from Figure 10; for any Machine learning algorithm if the training and testing data is linearly separable, performs better than other nonlinear datasets.”. The meaning of this sentence does not entirely become clear. Please revise this sentence.
  7. Line 372 and Figure 11: Please define the abbreviation ROC.
  8. Figure 9: The meaning “Build K models ion K samples” is not clear (might be “Build K models on K samples”?)
  9. The reference list contains many conference abstracts. Please replace these references by peer-reviewed journals, if possible.
  10. Sometimes, details are missing in the reference list, e.g. in reference 33 the volume is missing, also in reference 31.

Author Response

"Please see the attachment."

Author Response File: Author Response.docx

Back to TopTop