Submit to this Journal Review for this Journal Propose a Special Issue

Article Menu

Share Help Cite Discuss in SciProfiles

Open AccessArticle

Peer-Review Record

Investigating the Feasibility of Predicting KRAS Status, Tumor Staging, and Extramural Venous Invasion in Colorectal Cancer Using Inter-Platform Magnetic Resonance Imaging Radiomic Features

Diagnostics 2023, 13(23), 3541; https://doi.org/10.3390/diagnostics13233541

by Mohammed S. Alshuhri^1,*

, Abdulaziz Alduhyyim^2,†, Haitham Al-Mubarak^3,†, Ahmad A. Alhulail¹

, Othman I. Alomair⁴

, Yahia Madkhali⁵

, Rakan A. Alghuraybi⁶, Abdullah M. Alotaibi⁷

and Abdullalh G. M. Alqahtani¹

Reviewer 1: Anonymous

Reviewer 2: Anonymous

Reviewer 3: Anonymous

Diagnostics 2023, 13(23), 3541; https://doi.org/10.3390/diagnostics13233541

Submission received: 18 September 2023 / Revised: 19 November 2023 / Accepted: 21 November 2023 / Published: 27 November 2023

(This article belongs to the Special Issue Application of AI in Diagnosis of Colorectal Cancer)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript topic is very interesting

However, a schematic digaram or graphical abstract is highly needed to summarize the work flow and the results obtained from the study

Minor Linguistic and Grammatical mistakes have to be corrected

The discussion section has to be enriched by more comparitive studies in other cancers other than colorectal cancer

Detailed or stratified clinical data shoudl be included for each patient beside the collective clinical data summarized in table 2

Comments on the Quality of English Language

Minor edits are needed

Author Response

Reviewer 1:

Comments and Suggestions for Authors

The manuscript topic is very interesting. However, a schematic diagram or graphical abstract is highly needed to summarize the work flow and the results obtained from the study

We appreciate the reviewer's feedback. The following graphical abstract was added in the manuscript.

Minor Linguistic and Grammatical mistakes have to be corrected

Reviewer comments have been considered and the most recent version of the paper has undergone comprehensive editing by the MDPI editing service.

The discussion section has to be enriched by more comparitive studies in other cancers other than colorectal cancer:

We appreciate the reviewer's suggestion and have incorporated a more detailed as the following:

“..Moreover, the present investigation demonstrates similar performance to studies on various cancer types. Specifically, the findings in melanoma [21], thyroid cancer [22], head and neck cancer [23], and adrenal gland carcinoma [24], all indicate a significant correlation between the biomarker and radiomics, with AUC values ranging from 0.62 to 0.78. However, it's worth noting that the current predictive performance is comparatively lower when compared to other studies. For instance, in the case of liver cancer (AUC = 0.94, cross-validation), breast cancer (AUC = 0.81–0.93) [25], [26], and cervical cancer (95%, 0.91). “

Detailed or stratified clinical data should be included for each patient beside the collective clinical data summarized in table 2

We have taken into account the feedback from the reviewers and have subsequently included clinical data for individual patients. This information will be presented as a supplementary table in the manuscript.

Patient Code Number

SEX

weight

age

KRAS (N/P)

Tumour staging (TNM staging system)

T stage

N stage

Metastasis

Effects of mesorectal fascia (MRF) (N/P)

Extramural vascular invasion (EMVI) (N/P)

Type of treatments

10172719

T4b, N2, M0/ yT4aN0Mx

T4b

yes

Positive

3 cycles of NAD XELOX

60661

T3N2M0/ after treatment T1 N1

Negative

Neoadjuvant chemoradiotherapy

340433

110

T3N2

Negative

Neoadjuvant chemoradiotherapy

449272

t4n2

Yes

Positive

CHEMO (XELOX) X 2

2322853

T4b, N2, M0/ NO FU

T4b

Yes

Positive

No data

2539669

T3 N2

Yes

Positive

Neoadjuvant chemoradiotherapy

10010705

T4AN2/ T3N0Mx after treatment

T4A

Yes

Positive

Neoadjuvant chemoradiotherapy

10087230

T2N2/ NO FU

Negative

CHEMO (XELOX) X 2

10089371

100

T4AN2/ NO FU

T4A

Yes

Negative

CHEMO (XELOX) X 3

10096833

T2N2 / AFTER TREAT T1-2 N2

Negative

Neoadjuvant chemoradiotherapy

10103691

100

T3N2 / AFTER TREATMENT T3N2

Yes

Positive

Neoadjuvant chemoradiotherapy

10134636

T2N1/ AFTER T2N1

Negative

Neoadjuvant chemoradiotherapy

10135968

T4N2M1

Yes

Positive

Neoadjuvant chemoradiotherapy

10138041

T4 N2 M0/ NO FU

Yes

10139959

T3 N0

Yes

Negative

cycles of NAD XELOX

10140593

T4B N2/ NO FU

T4B

Yes

Positive

No data

10143049

100

T3N2M0

Yes

Negative

CHEMO (XELOX) X 2

10179989

T3N0

Yes

Positive

CHEMO (XELOX) X 3

10184428

T3 N0 MX/ypT1 N0 Mx

Negative

CHEMO (XELOX) X 4

10188227

T3N2

Yes

Positive

Xeloda

317873

110

T2N2M0 / SURGERY

Yes

Negative

534327

T4N2/ AFTER TREAT ALMOST GONE

Yes

Positive

Neoadjuvant chemoradiotherapy

1135947

T3 N1 M0

Yes

Positive

1199271

T2N2M0/ NO FU

Positive

Neoadjuvant chemoradiotherapy

2132915

T3 N2 M1 / NO FU

Yes

Positive

Neoadjuvant chemoradiotherapy

10050980

T4B N2 M1/ yT4N0/1

T4B

Positive

chemo

10051270

T4N2M0/ NO FU

Positive

chemo

10205635

T3N2M0/ NO FU

Yes

Positive

CHEMO (XELOX) X 4

10206130

T3/ NO FU

Positive

Chemo

10207469

T4bN2M1/ No FU

T4B

Positive

Neoadjuvant chemoradiotherapy

10214476

T3N2M0

Yes

Positive

Neoadjuvant chemoradiotherapy

10216119

T3N2/ NO FU

Yes

Positive

chemo

10233788

100

T4N2M1/ NO FU

Yes

Positive

140777

T2N2M1

Yes

Negative

Neoadjuvant chemoradiotherapy

340243

T4AN2M0 / NO FU

T4A

Yes

Positive

Neoadjuvant chemoradiotherapy

368356

T4N2M0/ NO FU

Yes

Positive

Neoadjuvant chemoradiotherapy

502757

T3 N 2 M0 / yT3, N1, Mx

Positive

Neoadjuvant chemoradiotherapy

544515

T4AN2M0

T4A

Negative

chemo

2236146

T3N1 M0 / AFTER TREAT T4 N2 M0

Yes

Positive

Neoadjuvant chemoradiotherapy

2443955

T4N2M0

Yes

Positive

Neoadjuvant chemoradiotherapy

10015390

T3 N 2 M0

Positive

chemo

10019202

T2-T3 N2 M0/ AFTER TREAT T2N2M0

T2-T3

Positive

10026773

T4N1M0

Yes

Positive

Neoadjuvant chemoradiotherapy

10076837

T4N1M1/ NO FU

Yes

Positive

Neoadjuvant chemoradiotherapy

10150880

T4N2 M0

Yes

Positive

chemo

10155918

T3N1M1

Yes

Positive

Neoadjuvant chemoradiotherapy

432974

100

T3N2/ Interval reduction on size with no change on stage

Positive

Neoadjuvant chemoradiotherapy

450518

T2N0M0/ NO FU

Positive

Neoadjuvant chemoradiotherapy

1207020

T3 N1 / NO FU

Negative

Neoadjuvant chemoradiotherapy

10047266

T4 N1 M1/ POST SURGERY

Yes

Positive

Neoadjuvant chemoradiotherapy

10057014

T3N1M0 / NO FU

Positive

Neoadjuvant chemoradiotherapy

10096937

T3N2/ POST SURGERY

Yes

Positive

chemo

10134812

T2 N2/ NO FU

Negative

chemo

10147507

T4 N2 M0

Negative

Neoadjuvant chemoradiotherapy

10208084

T3N2/ AFTER TREATMENT T3 N2

Positive

Neoadjuvant chemoradiotherapy

10214476

T3 N2 M0

Yes

Positive

Neoadjuvant chemoradiotherapy

176761

T4AN2M0 / NO FU

T4A

Yes

Positive

chemo

517223

T4B N2 M1/ NO FU

T4B

Positive

chemo

10246553

T3N2

Negative

Neoadjuvant chemoradiotherapy

10254537

T3N2 M0

Yes

Negative

Neoadjuvant chemoradiotherapy

10256859

T4bN2M1/ Post surgery

T4B

CONCOMITTENT chemo rad

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The manuscript "Investigating the feasibility of predicting KRAS status, tumor 2 staging and EMVI in colorectal cancer using MRI Radiomics 3 features in inter-platform" is an interesting manuscript with practical scientific relevance. The study has been planned meticulously. There are some minor observations that are pointed below:

1. Title: OK. Please remove the Full stop at the end (Page 1)

2. Abstract: OK

3. Introduction: OK.

4. Materials and Methods: OK.

5. Results: Table 1: Please correct Platform A for both Platform

Please elaborate the findings with justification of the obtained predictability.

Comments on the Quality of English Language

English: Needs some minor corrections for grammar and syntax.

Author Response

Reviewer 2:

Comments and Suggestions for Authors

Title: OK. Please remove the Full stop at the end (Page 1)

Full stop was removed from the title.

Abstract: OK
Introduction: OK.
Materials and Methods: OK.
Results: Table 1: Please correct Platform A for both Platform

The platform annotation has been corrected now.

Parameters	Platform A			Platform B
	T2W	DWI		T2W		DWI
Image orientations	Axial		Axial		Axial		Axial
Field of view (mm)	480-500		400-370		480-500		420-380
Matrix	564 x 260		160 x 200		564 x 260		150 x 180
Slice thickness (mm)	5		10		5		10
ETL or FA	Min		90 FA		Min		90 FA
TR (ms)	4000		8100		4200		8300
TE (ms)	85		67		95		73
b values (s/mm²)	0-1000-1500				0-800-1200
Number of Averaging (NSA)	1		1		1		1

Please elaborate the findings with justification of the obtained predictability.

We appreciate the valuable feedback provided by the reviewer. In response to the comments, we have expanded the discussion section to provide a more in-depth exploration of our findings.

“The present study has made significant strides in the categorization of partici-pants based on their KRAS mutation status, specifically into KRAS+ (wild type) and KRAS- (mutation) groups. The achievement of an impressive Area Under the Curve (AUC) of 0.7 in this classification may be used to predict treatment responses and providing valuable insights into patients’ outcomes [18]–[20]. The obtained AUC of 0.7 indicates a reasonable level of discrimination between the KRAS+ and KRAS- groups. In the context of predictive modeling, an AUC of 0.7 suggests that the model has a moderate ability to distinguish between the two categories. A value of 0.5 represents a random chance, so an AUC of 0.7 indicates a better-than-random predictive perfor-mance, though there is room for improvement. These findings align with and reinforce earlier research, underscoring the significance of mutation statuses in identifying the patients who are likely to respond positively to EGFR therapy. A study by Guo et al. [5], which utilized T1-weighted enhanced images and KRAS mutation status, achieved a slightly lower AUC value of 0.6. Similarly, Zhang et al. [6], who developed a radi-omics model using T2-weighted MRI, attained a noteworthy AUC of 0.7 (95% CI 0.6-0.8). These comparable AUC values across different studies utilizing varied imaging platforms and methodologies validate the robustness of this study's findings.

Furthermore, the current investigation's performance is not limited to a specific cancer type, as it demonstrates similarity to other studies on melanoma [21], thyroid cancer [22], head and neck cancer [23], and adrenal gland carcinoma [24]. Across these diverse cancer types, a significant correlation between the biomarker and radiomics is observed, with AUC values ranging from 0.62 to 0.78. However, it is worth noting that the current predictive performance is comparatively worse when compared to those of other studies. For instance, those on liver cancer (AUC = 0.94, cross-validation), breast cancer (AUC = 0.70–0.81) [25], [26], and cervical cancer (95%, 0.70–0.91).”

“…..The logistic regression model's performance in identifying EMVI status is noteworthy. The high accuracy and AUC values indicate that radiomic features can be indicative of EMVI, which can guide the clinical decisions. However, the inability to definitively ascertain the status for a portion of participants highlights the complexity of the problem. Nevertheless, the model provides valuable insights into EMVI status in the majority of cases, which can aid in more personalized treatment approaches.”

Reviewer 3 Report

Comments and Suggestions for Authors

In this study, the authors investigated the feasibility of predicting KRAS status, tumor staging and EMVI in CRC using MRI radiomics feature. And the results showed that it can distinguish KRAS genes, tumor grading, and EMVI. However, the reviewer has the following concerns.

(1) There are some spelling or tense errors, please check the whole paper.

(2) Should the patient information from line 113-115 be moved to result part?

(3） What is the time interval between MRI scans and pathological testing, which is essential to minimize the gap to maintain the accuracy of the results.

(4) In line 236, it is said that "In the majority of cases (n=42, 72.4%), the tumor had spread into the lymph nodes at stage 2". Is it should be stage 3 if there is lymph nodes metastasis.

(5) In figure 1, it looks a bit odd that there are both sets are 1.5T MRI and ADC+T2W images.

(6) Why was T2W chosen as the morphological analysis rather than T1W enhanced providing clearer morphological details of the lesions?

(7) In table 1, why are there two "platform A" columns?

(8) In table 2, there are statistical significant between KRAS-wild and -mut for Age and Weigh. Is it true? Please explain it.

(9) In table 3, there is a little bit different between the training and testing groups. Is there a statistical significant between the groups?

(10) The study design is "one scanner for training and another scanner for testing". Why not randomly mix and match patients from both scanners and then assign them to groups randomly?

Comments on the Quality of English Language

The English language of the paper is good. The paper has a nice design and organization, although there are some minor spelling or tense errors.

Author Response

Reviewer 3:

Comments and Suggestions for Authors

There are some spelling or tense errors, please check the whole paper.

Reviewer comments have been considered and the most recent version of the paper has undergone comprehensive editing by the MDPI editing service.

Should the patient information from line 113-115 be moved to result part?

We agree with the reviewer's suggestion to relocate the patient information from lines 113-115 to the results section. The following text has now been relocated to the results section.

“ …Specifically, 32 patients were allocated to the training and 30 to the test datasets. Pathological analysis of surgical specimens in the training cohort confirmed the pres-ence of KRAS mutations in 19 patients, while 13 were confirmed negative. Similarly, in the test cohort, 9 patients tested positive for KRAS mutations, while 19 were negative. Subsequently, the study established a predictive model for KRAS mutations using the training dataset, which was subsequently assessed using the internal test dataset.”

What is the time interval between MRI scans and pathological testing, which is essential to minimize the gap to maintain the accuracy of the results.

We appreciate the reviewer's valuable comments. To maintain result accuracy, it's crucial to minimize the gap between MRI scans and pathological testing. In our current study, we've taken steps to address this concern by conducting our analysis based on MRI images taken before treatment and subsequent confirmation of rectal staging and other pathological tests. This approach ensures a close alignment between the imaging data and pathological findings, enhancing the relevance and accuracy of our results.

In line 236, it is said that "In the majority of cases (n=42, 72.4%), the tumor had spread into the lymph nodes at stage 2". Is it should be stage 3 if there is lymph nodes metastasis.

We appreciate the insightful comments from the reviewer. It's important to clarify the distinction between T stage and N stage in our study. T3 stage indicates evidence of extra mural tumor extension beyond the wall to the surrounding mesorectal fat, while T4 signifies that the tumor has reached the peritoneal reflection or adjacent organs (T4a/T4b). On the other hand, T2 stage implies that the tumor is confined within the mucosal layer.

Regarding the statement in line 236, where we mentioned that "In the majority of cases (n=42, 72.4%), the tumor had spread into the lymph nodes at stage 2," we want to emphasize that this refers to the N stage, which denotes metastasis to the loco-regional lymph nodes. The identification of normal and abnormal lymph nodes, as well as the determination of the number of affected nodes, is crucial for the radiologist to classify whether it is N1 or N2.

In summary, our study distinguishes between T stage and N stage, and the statement accurately reflects the presence of lymph node metastasis at stage 2 based on our classification criteria. We hope this clarification addresses the reviewer's concern.

In figure 1, it looks a bit odd that there are both sets are 1.5T MRI and ADC+T2W images.

We agree with reviewer’s comment, The figure was updated with assigning the MRI scanner to platform A and B. also we removed the redundancy on T2/ADC images (see figure below).

Why was T2W chosen as the morphological analysis rather than T1W enhanced providing clearer morphological details of the lesions?

The reviewer's observation is accurate; T2W was selected for morphological analysis instead of T1W enhanced imaging, which often provides clearer morphological details of lesions. In our study, regrettably, not all patients underwent T1 post-contrast scans. Therefore, we made the decision to utilize T2W imaging, especially because the current hospital protocol involves acquiring high-resolution T2W images with specific slice thicknesses. This choice was made to ensure consistency and to work with the data that was available for the majority of the study participants.

In table 1, why are there two "platform A" columns?

This was a mistake. The platform annotation has been corrected now.

Parameters	Platform A			Platform B
	T2W	DWI		T2W		DWI
Image orientations	Axial		Axial		Axial		Axial
Field of view (mm)	480-500		400-370		480-500		420-380
Matrix	564 x 260		160 x 200		564 x 260		150 x 180
Slice thickness (mm)	5		10		5		10
ETL or FA	Min		90 FA		Min		90 FA
TR (ms)	4000		8100		4200		8300
TE (ms)	85		67		95		73
b values (s/mm²)	0-1000-1500				0-800-1200
Number of Averaging (NSA)	1		1		1		1

In table 2, there are statistical significant between KRAS-wild and -mut for Age and Weigh. Is it true? Please explain it.

In Table 2, initially, there appeared to be statistically significant differences between KRAS-wild and KRAS-mut groups for Age and Weight. However, upon re-evaluating the data, we discovered that there is no significant variance in Age (60.2 ± 19.3 for KRAS-wild and 56.5 ± 16.7 for KRAS-mut) and Weight (71.5 ± 17.7 for KRAS-wild and 79.8 ± 16.5 for KRAS-mut) between the groups (Table 2). The adjusted p-values of 0.14 and 0.09 suggest that these differences lack statistical significance. We attribute these initially observed disparities in means to our relatively small sample size. With a larger sample, we anticipate that these differences may diminish. We appreciate your attention to this matter.

In table 3, there is a little bit different between the training and testing groups. Is there a statistical significant between the groups?

“We acknowledge the reviewer's observation regarding the slight difference between the training and testing groups in Table 3. The dataset was divided into two groups: group 1 for training (n=32) from scanner 1 and group 2 for testing (n=28) from scanner 2. To assess the significance of this difference, we conducted a statistical analysis. The results revealed that *p < 0.05, indicating statistically significant differences between training and testing accuracies in the KRAS+/- class. This significance underscores distinctions in the machine learning model's performance between the training and testing phases for this specific category.”

The study design is "one scanner for training and another scanner for testing".

Why not randomly mix and match patients from both scanners and then assign them to groups randomly?

The study design "one scanner for training and another scanner for testing" was deliberately chosen to address specific research objectives. Our primary aim in this study is to identify radiomics features that are independent of scanner variations and focus on features that are robust and not influenced by the specific scanner used. By maintaining separate training and testing groups based on scanners, we can better isolate and evaluate the impact of scanner-related variables on the radiomics features.

This design helps us improve the reliability of our findings and ensures that the predictive models are not overly influenced by scanner-specific artifacts. It allows us to assess the generalizability of radiomics features across different imaging platforms, which is crucial for the applicability of the research in various clinical settings.

While randomly mixing and matching patients from both scanners is a valid approach for some research questions, our study's specific focus on scanner-independent radiomics features necessitates the separation of training and testing groups based on scanners.

Author Response File: Author Response.pdf

Article Menu

Investigating the Feasibility of Predicting KRAS Status, Tumor Staging, and Extramural Venous Invasion in Colorectal Cancer Using Inter-Platform Magnetic Resonance Imaging Radiomic Features

Further Information

Guidelines

MDPI Initiatives

Follow MDPI