Acute Tumor Transition Angle on Computed Tomography Predicts Chromosomal Instability Status of Primary Gastric Cancer: Radiogenomics Analysis from TCGA and Independent Validation

Round 1

Reviewer 1 Report

In this study, the authors explore whether CT-derived morphological tumor features can be used to predict the molecular subtype of colon tumors. For this they train software on dataset of colon tumors for which most molecular features are known and determine the correlation between shape features of the tumors and the molecular subtype. Intriguingly, they find that the ‘tumor transition angle’ is a good indicator of the CIN subtype. These findings are relevant as a CT-scan is much less invasive than a tumor biopsy. However, these are some limitations to the study that probably need to be addressed before this paper is suitable for publication.

-      The key limitation of this study, also highlighted by the authors in the discussion, is the size of the training and validation cohorts. 10 out of 11 predictions were correct, which suggests that the tumor transition angle potentially is a powerful predictor, but numbers should be dramatically increased before this could become a widely-used feature to predict tumor subtype. As getting to significant numbers might be difficult for the current study, the authors could describe this more as a pilot study and in the discussion make more recommendations how to get to datasets sufficiently large to be able to conclude that the tumor transition angle is indeed a bona fide predictor of CIN colon cancer

-      As also mentioned in the discussion, the CIN tumors also correlate well with a less advanced Borrmann stage, which indicates that (probably because of the small cohort size) the CIN tumors are less advanced stage. Can the authors rule out that an acute angle only predicts early stage and that tumors grow more diffuse as they progress? This should be addressed somehow.

-      CIN comes in various grades: some tumors are more aneuploid than others: does the correlation improve when excluding tumors that are only mildly aneuploid: in other words, is it really the CIN that determines the morphologic features of the cancer or is it something else (e.g. the driver behind the CIN tumor). Possibly the authors could test this, since some of the CIN tumors in Fig. 1 show fewer copy number abnormalities than others.

-      There are quite big differences between the training and validation datasets (Table 3), presumably also the result of the limited dataset size. I am not a statistician, but it would be wise to have these data scrutinized by a statistician (if this has not been done yet). Furthermore, probably because of the dataset size, the tumor diameter size has a big impact on the quality of the predictor. 

-      Related: Table 2 displays quite a few parameters that were tested as possible predictors of a CIN phenotype. Also here with the disclaimer that I am not a statistics expert, but should the p-values not be corrected for multiple testing?

Author Response

Dear Reviewer,

We are grateful for the comments from Reviewers to improve this manuscript. Please find the point-by-point response as uploaded file. Thank you very much indeed.

Author Response File: Author Response.docx

Reviewer 2 Report

Chromosomal instability is the most important predictive and prognostic biomarker in the management of cancer patients. In this article, Ying-Chieh Lai et al. investigate the possible link between CT scan images and chromosomal instability in gastric cancer patients. They demonstrate that some morphological features may predict chromosomal instability in gastric cancers. It is an intriguing approach in the screening of chromosomal instability. Nevertheless, the small size of the cohorts limits the ability to make the definitive conclusion mentioned in the discussion.

Major remarks:

First, is the use of CT scan images sufficient for the detection of chromosomal instability in gastric cancers or is it a tool among other biomarkers. Certainly, CT scan images, especially the introduction of 3D reconstruction of the image, allow a preliminary and rapid approach to understanding the nature of the tumor. However, the complexity of chromosomal instability requires the use other biomarkers that can provide complementary information and eliminate all false negatives and false positives. This point can be presented in the introduction of this work.

Second, the validation of other simple biomarkers using peripheral blood, for example, may be of great value in detecting chromosomal instability in addition to CT scan images. This point could be introduced in the discussion. In addition, the use of an animal model, as well as a large cohort with various profiles, could advance your validation of CT scan images as a marker of chromosomal instability.

Third, the relationship between EBV status and chromosomal instability could be discussed with respect to the CT scan images in the used cohorts.

Minor remarks:

1-     The contrast of the images could be improved

2-     Give the appropriate definition of chromosomal instability with the reference in the sentence and not between parentheses.

3-     Define TNM

4-     Try to use short sentences to make the article clearer and easier to read.

Author Response

Dear Reviewer,

We are grateful for the comments from Reviewers to improve this manuscript. Please find the point-by-point response as uploaded file. Thank you very much indeed.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have satisfactory addressed my concerns. The paper now clearly states this is a pilot study and I hope that the group or the field wil use this study as a starting point to perform a full scale study to detect whether imaging can indeed be used to predict CIN in cancer.

Reviewer 2 Report

The article can be accepted.It is another tool for the detection of chromosomal instability and opportunity to build a database able to make the detection system more efficient artificial intelligence