Artificial Neural Network-Derived Cerebral Metabolic Rate of Oxygen for Differentiating Glioblastoma and Brain Metastasis in MRI: A Feasibility Study

Round 1

Reviewer 1 Report

This topic is interesting, but need some revision, look at these points:

1. Lines 20-21:" Glioblastoma often appears similar to cerebral metastasis in conventional MRI". This is highly debatable, because in most cases they look different. Please revise.
2. Lines 104-105: "15 patients with previously untreated GBM or cMET were prospectively included " What do authors want to say? Why do they not treat these patients? Maybe do authors want to say with "primary GBM or cMET never surgically treated"?
3. Lines 306-317. The role of the oxygen extraction fraction is closely associated with blood glucose. Please authors must consider at this point these 2 very important refs:  -- Clinical Risk and Overall Survival in Patients with Diabetes Mellitus, Hyperglycemia and Glioblastoma Multiforme. A Review of the Current Literature. Int J Environ Res Public Health. 2020 Nov 17;17(22):8501. doi: 10.3390/ijerph17228501  ---  Intra-operative characterization of gliomas by near-infrared spectroscopy. Acta Neurochir (Wien). 2003 Jun;145(6):453-59; discussion 459-60. doi: 10.1007/s00701-003-0035
4. Lines 342-343: "The same applies to GBM patients that were not stratified according to IDH or MGMT promoter methylation status" This should be move to Material section.
5. Lines 61-63: "Median survival for this highly malignant, infiltratively growing tumor is between 12 and 15 months with optimal treatment [7–9]" What about recurrent GBM? Look at these ref. Surgical outcome and molecular pattern characterization of recurrent glioblastoma multiforme: A single-center retrospective series. Clin Neurol Neurosurg. 2021 Aug;207:106735. doi: 10.1016/j.clineuro.2021.106735. 
6.  Lines 297-300: "This may be explained by the tumor microenvironment, in particular angiogenesis, that is similar in the contrast-enhancing part of GBM and many hematogenous cMETs". Can the authors expose better the GBM's and cMETs' microenvironment composition? Look at these refs:  -- Decipher the Glioblastoma Microenvironment: The First Milestone for New Groundbreaking Therapeutic Strategies. Genes (Basel). 2021 Mar 20;12(3):445. doi: 10.3390/genes12030445. ---- Blood-Brain Barrier, Cell Junctions, and Tumor Microenvironment in Brain Metastases. Front Cell Dev Biol. 2021 Aug 24;9:722917. doi: 10.3389/fcell.2021.5

Overall a good manuscript.

Author Response

Please find attached.

Kind regards

Hakim Baazaoui

Author Response File: Author Response.pdf

Reviewer 2 Report

In this study, the authors investigated the application of QSM+qBOLD model to estimate oxygen extraction fraction (OEF), cerebral blood flow (CBF) and cerebral metabolic rate of oxygen (CMRO2) for the differentiation of glioblastoma and cerebral metastasis. The research objectives are well defined, and the design of the study is appropriate, although the small sample size is a major limitation, even for a feasibility study like this one. The study methodology is appropriate, although some important details are missing. The results are clear and well presented. The discussion is informative, with appropriate references to the scientific literature and with the potential clinical implications of their results. The limitations of the study are also presented by the authors. The conclusions drawn by the authors are in general supported by the results, although some sentences require clarifications.

Please find below my specific comments.

1. “QSM+qBOLD allows for robust differentiation of glioblastoma and cerebral metastasis while yielding insights into tumor oxygenation and infiltration before disruption of the blood-brain barrier … for early detection of recurrence or infiltration before disruption of the blood-brain-barrier”. These sentences in the abstract and conclusions are not supported by any data in this manuscript. This is a cross-sectional study, and therefore it is not possible to demonstrate that the information of tumor oxygenation and infiltration precedes disruption of the blood-brain barrier.
2. It seems that the ROIs were drawn only in one (CET and necrosis) or few (NET2) slices. If so, how was the slice/s selected? Instead, if the ROIs include the entire feature (CET, necrosis, or NET2) across all the slices, please specify this in the manuscript.
3. The CBF was calculated in mL/100g/min, but there is no mention of an M0 image acquired for this purpose. Please clarify.
4. The authors claim that the CBF values were all within the normal limits, but in supplementary table 1 there are a few cases with abnormally low CBF values (e.g., patient n.8 of the cMET group). Please clarify.
5. The authors mention they used SPM to register the images, but there is no mention about the algorithm used. Additionally, there is no mention about verification of the correct registration, which can be particularly difficult for images like pCASL. Further details need to be provided.
6. “When comparing oxygenation and perfusion parameters between GBM and cMET, OEF was found to be significantly (p = 0.03) lower in GBM than in cMET.” It is unclear whether “OEF” here was measured in a specific ROI or not.
7. “Meanwhile, for the GBM group, neither the difference in OEF (p = 0.11), nor CBF (p = 0.15), nor CMRO2 (p = 0.08) was significant.” Please specify with which element the comparison has been made.
8. In figure 4, it would be great to report all the datapoints.
9. Several outliers are visible in Figure 4. Have these outliers been excluded from the statistical analysis or not? Please specify this in the manuscript.

Author Response

Please find attached.

Kind regards

Hakim Baazaoui

Author Response File: Author Response.pdf

Reviewer 3 Report

Dear authors,

First of all, I’d like to give a great congratulation to them for nice and graceful study. I think that the article has been professionally written focused on the clinical meanings of the differential diagnosis between high grade glioma and brain metastasis. The topic is interesting enough to attract the reader’s interests. The analytic method is scientific and reasonable. However, I am afraid that the application of their study in clinical field can be limited. Because the differential diagnosis between high grade glioma and brain metastasis is not difficult. Despite the MR images are similar each other, the clinical characteristics are much different. Systemic work-up for primary cancer using PET-CT, tumor makers, clinical manifestation of B symptoms can make physician differentiate two brain tumors easily without functional MR analysis. In fact, unmet needs in clinical practice could be differential diagnosis between each grade of gliomas. Also, unique features of functional MRI according to the type of primary cancer can be more informative to readers. And many results have been already published focused on similar studies which suggest the metabolic difference in functional MRI of several brain tumors. Briefly, I am not sure what is the unique feature in their manuscript. It is essential for authors to high light the unique feature comparing with other studies. This review point had better be added in the section of Discussion.

I would like to appreciate reading an interesting article. Good Luck.

Author Response

Please find attached.

Kind regards

Hakim Baazaoui

Author Response File: Author Response.pdf

Reviewer 4 Report

In this study, the authors propose using an artificial neural network model for evaluating the cerebral metabolic rate of oxygen for differentiating glioblastoma and brain metastasis in MRI images. Furthermore, the aim is to classify these by using an SVM linear classifier.

In my opinion, the paper shows some issues that need to be fixed first to accept it for publication:

1) As stated (see lines 92-95 and lines 174-177), the reliability of the ANN approach has been already demonstrated and tested in [1]. , On this premise, the relevant novelty of the present study lies exclusively on the side of the classification of the MRI images; The paper’s organization and title should be taken into account of this aspect and should be re-written accordingly.

2) Regarding the approach used to classify the MRI images, the authors declared: “A support-vector machine was chosen since it emerged as the best classifier, delivering the highest accuracy in previous studies that compared it to different approaches such as Naïve Bayes, KNN and decision trees for binary classification of glioblastoma and cerebral metastasis” (see lines 284-288). However, other [2] and better approaches to binary classification exist [3]. Therefore, the authors should provide a fair comparison among them referred to the case study discussed in their paper to evaluate the suitability of the selected approach.

References

[1] Hubertus, S.; Thomas, S.; Cho, J.; Zhang, S.; Wang, Y.; Schad, L. R., Using an artificial neural network for fast mapping of the

oxygen extraction fraction with combined QSM and quantitative BOLD. Magn Reson Med 2019, 82 (6), 2199-2211

[2] https://medium.com/analytics-vidhya/comparison-between-linearsvc-svm-and-sgdclassifier-results-comparison-showcase-on-iris-dataset-8437657df276

[3] Jeatrakul, P., & Wong, K. W. (2009, October). Comparing the performance of different neural networks for binary classification problems. In 2009 Eighth International Symposium on Natural Language Processing (pp. 111-115). IEEE.

Author Response

Please find attached.

Kind regards

Hakim Baazaoui

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Authors solved all my criticisms.

Reviewer 3 Report

October 19, 2021

Reviewer’s opinion

Dear Editor,

I’d like to say my great thank for a kind invitation for review of revised version of manuscript (MS number applsci-1376359, whose title is “Artificial neural network-derived cerebral metabolic rate of oxygen for differentiating glioblastoma and brain metastasis in MRI: a feasibility study”. It’s a great honor for me to review a manuscript submitted to such a great journal of Applied Science.

Baazaoui et al. suggested that functional MR-derived cerebral metabolic rate of oxygen in contrast-enhancing and peritumoral non-enhancing regions, as calculated by an artificial neural network, should allow for robust differentiation of glioblastoma and brain metastasis. I think that the authors had given their best effort to revise the manuscript, as a result, it got to be refined after revision. Therefore, I would like to accept for publication on such a nice journal as Applied Science.

Best regards,

Young Zoon Kim, M.D., Ph.D.

Professor, Sungkyunkwan University School of Medicine,

Director, Division of Neuro-Oncology, Sungkyunkwan University Samsung Changwon Hospital.

Reviewer 4 Report

The paper has been sufficiently improved: it  can be accepted in present form.