Detection of Low Blood Hemoglobin Levels on Pulmonary CT Angiography: A Feasibility Study Combining Dual-Energy CT and Machine Learning

Round 1

Reviewer 1 Report

 

The paper aims to evaluate the feasibility of using dual-energy CT pulmonary angiography (CTPA) and machine learning to detect anemia in patients. The study found that virtual noncontrast (VNC) imaging and machine learning showed good diagnostic performance for detecting anemia on DECT CTPA.

-        The author should clearly state the motivation for this study and the novelty.  

-        The limitations of this study were not explicitly stated in the paper and future studies.

-        How might the results be applied in clinical settings to improve patient outcomes?

-        The study was a proof-of-concept study conducted on a small number of patients from a single center, and therefore, it needs validation with larger, multi-center studies.

-        The study needs validation in other DECT scanners, which may have different cut-off values for diagnosing anaemia.

-        A larger training set may be required to improve the method's accuracy.

-        The clinical impact of recognizing anemia in CTPA needs to be evaluated in large outcome studies.

 

-        The machine learning algorithms have the capabilities to extract attenuation values in the aorta and the heart, which will be evaluated in future studies.

Author Response

We sincerely appreciate the time and effort you have dedicated to reviewing our manuscript. Your constructive feedback has been invaluable in refining our work. Below, please find our responses to the comments and concerns raised.

 

The author should clearly state the motivation for this study and the novelty.  

Response: Thank you for your constructive feedback. To address your comment, we have revised the last paragraph of the introduction to more explicitly outline the motivation behind our research and emphasize the innovative facets of our study.

 

The limitations of this study were not explicitly stated in the paper and future studies.

Response: Thank you for pointing out the concern regarding the study's limitations. We have indeed addressed the limitations in the Discussion section on page 14 (second to last paragraph). Within this section, we also delineate the subsequent steps required to further validate our approach. We appreciate any specific limitations or future directions the reviewers might suggest and are open to incorporating them for clarity and comprehensiveness.

 

How might the results be applied in clinical settings to improve patient outcomes?

Response: Thank you for highlighting this pertinent aspect. We've incorporated a paragraph in the Discussion section (page 14) detailing potential clinical applications of our findings. This elaboration offers insights into how our technique might be integrated into clinical workflows to enhance patient outcomes.

 

The study was a proof-of-concept study conducted on a small number of patients from a single center, and therefore, it needs validation with larger, multi-center studies.

Response: We appreciate your observation. Indeed, the single-center nature of our proof-of-concept study is a limitation. We hope to have clearly addressed this concern in the Discussion section (page 14) and recognize the need for validation through larger, multi-center studies.

 

The study needs validation in other DECT scanners, which may have different cut-off values for diagnosing anaemia.

Response: We value your insight on the potential variability among different DECT scanners. We've acknowledged this aspect in the Discussion section, emphasizing the importance of validating our findings across various DECT scanner models to ensure broader applicability.

 

A larger training set may be required to improve the method's accuracy.

Response: We appreciate your feedback on the potential benefits of a larger training set. Indeed, we recognize this aspect and have clearly highlighted it as a limitation in our Discussion section (page 14). Expanding the dataset in future studies could refine the accuracy of our method.

 

The clinical impact of recognizing anemia in CTPA needs to be evaluated in large outcome studies.

Response: Thank you for emphasizing the need for assessing the clinical impact. We concur that understanding the implications of recognizing anemia in CTPA is crucial. We've noted this aspect in our Discussion section (page 14) and anticipate that future large-scale outcome studies will provide a more comprehensive evaluation.

 

The machine learning algorithms have the capabilities to extract attenuation values in the aorta and the heart, which will be evaluated in future studies.

Response: Thank you for underscoring the potential of machine learning algorithms in extracting attenuation values from the aorta and heart. As mentioned in our discussion, we see significant potential in this direction and aim to explore it in future studies. If there are any other related limitations or considerations you'd like us to address, we're open to including them.

Reviewer 2 Report

Line 40-42, 44: "blood gas", presume the authors referring to point of care testing?  The standard testing is CBC.  If point-of-care testing is being referred here, turnaround is usually 1-2 min.  And by the time CT with contrast is ordered in such setting (ED), which may be using point of care diagnostics, other baseline studies may have been done already (blood work).  This subsection needs to be revised to reflect clinical situations more accurately.  There are clinical circumstances in which knowing hgb while doing non-contrast CT exists.  The authors should explain this better in lines 46-48.

Line 53-54: again is less convincing than the introduction because most of these patients getting CTPA would have gotten complete blood count before CT is done, so anemia status is already known (prob 100% of these patients). Need to revise to be better clinically relevant.

Could you provide a description and reason for how and why 50 cases and 50 controls were selected and why the remaining 341 patients were excluded?  Need to explain the inclusion and exclusion process here to evaluate for a possibility of biased selection.

LIne 108: Edit "two radiologists (_,_)"

Results section seems a bit disconnected.  Transitioning to highlighting desc aorta as the best location seems abrupt and unclear except the ROC analysis provided.  Provide more details analysis of performance using signals from other locations.

Question... if contrast effects are inversely proportional to the distance from the entry to the vasculature (IV site), the sequence of relative importance will be RV_VNC to LV_VNC to ascAo_VNC to descAo_VNC.  Any explanation as to why AscAo_VNC fell off?  (Fig 4)?

I would like to see the performance of the RV and LV as at least online data.

If scientific rigor is present in the conclusion, should there be some acceptable performance from the RV and LV data?

Again, overall, the clinical relevance or use of CT-PA to assess anemia needs to be better presented as the clinical context seems not clinically accurate.

Author Response

We sincerely appreciate the time and effort you have dedicated to reviewing our manuscript. Your constructive feedback has been invaluable in refining our work. Below, please find our responses to the comments and concerns raised.

 

Line 40-42, 44: "blood gas", presume the authors referring to point of care testing?  The standard testing is CBC.  If point-of-care testing is being referred here, turnaround is usually 1-2 min.  And by the time CT with contrast is ordered in such setting (ED), which may be using point of care diagnostics, other baseline studies may have been done already (blood work).  This subsection needs to be revised to reflect clinical situations more accurately.  There are clinical circumstances in which knowing hgb while doing non-contrast CT exists.  The authors should explain this better in lines 46-48.

Response: Thank you for the detailed feedback, highlighting the discrepancies between our descriptions and real-world clinical situations. We acknowledge that the primary standard for anemia detection is indeed the CBC. We've revised the mentioned lines to accurately represent the timeline and sequence of diagnostic procedures in emergency settings. Additionally, we have expanded our explanation in lines 53-54 to clarify the clinical circumstances where assessing hemoglobin during non-contrast CT is pertinent, ensuring our study aligns more closely with clinical realities.

 

Line 53-54: again is less convincing than the introduction because most of these patients getting CTPA would have gotten complete blood count before CT is done, so anemia status is already known (prob 100% of these patients). Need to revise to be better clinically relevant.

Response: Thank you for highlighting the clinical context, emphasizing that most patients undergoing CTPA likely have prior CBC results. In light of this, we've revised lines 53-54 to more accurately reflect scenarios where the supplementary information about Hb from imaging could provide added value, ensuring a more relevant clinical narrative.

 

Could you provide a description and reason for how and why 50 cases and 50 controls were selected and why the remaining 341 patients were excluded?  Need to explain the inclusion and exclusion process here to evaluate for a possibility of biased selection.

Response: Thank you for bringing attention to the clarity and rationale behind our patient selection process. Here's our methodology explained in a more logical sequence:

• Objective Setting: Our study aimed to test our hypothesis with adequate statistical power. Based on our preliminary sample size calculation, we determined that 50 cases and 50 control subjects would be required to achieve this power.
• Timeframe Selection: To achieve the requisite number of cases and controls, we looked at consecutive patients undergoing CTPA. We chose the first two months of 2017 for this purpose. We anticipated, based on past data and patient volume, that this timeframe would yield the number of cases we needed.
• Exclusion Explanation: Not all patients from these two months were included because our primary goal was to meet the specific number of 50 cases and 50 controls. We prioritized cases based on chronological order and excluded nondiagnostic studies by quality criteria. Any patient beyond this count or not meeting our eligibility criteria was not considered for analysis.

We hope this explanation provides a clearer picture of our patient selection process, and we assure you that the intention was a systematic approach rather than biased selection.

 

Line 108: Edit "two radiologists (_,_)"

Response: Thank you for pointing that out. The omission was to maintain the blinding during the review process. We have now rectified line 114 by including the initials of the two radiologists.

 

Results section seems a bit disconnected.  Transitioning to highlighting desc aorta as the best location seems abrupt and unclear except the ROC analysis provided.  Provide more details analysis of performance using signals from other locations.

Response: Thank you for highlighting the coherence concerns in the Results section. We recognize that the emphasis on the descending aorta as the prime location may have come across as abrupt. In our intent to be concise, we might have sacrificed clarity. The section on the top-ranked location for blood pool measurements, derived from the machine learning model, was densely packed in the statistical analysis subsection of the methods. For improved clarity, we've now isolated this description and presented it in a standalone paragraph in the Methods section (line 171). This should make the flow of information smoother and more comprehensible for the reader.

 

Question... if contrast effects are inversely proportional to the distance from the entry to the vasculature (IV site), the sequence of relative importance will be RV_VNC to LV_VNC to ascAo_VNC to descAo_VNC.  Any explanation as to why AscAo_VNC fell off?  (Fig 4)?

Response: Thank you for raising this insightful observation. Indeed, the sequence you highlighted, based on the distance from the IV site, is theoretically sound. Regarding the drop in importance of AscAo_VNC, while we can't pinpoint a definitive reason, there are a couple of factors that may contribute. It's possible that the dense concentration of the contrast, especially when it arrives via the SVC, or artifacts related to cardiac pulsations, might have affected the signal from the AscAo_VNC, leading to its decreased variable importance. We appreciate this question as it offers an avenue for deeper exploration in future studies.

 

I would like to see the performance of the RV and LV as at least online data.

If scientific rigor is present in the conclusion, should there be some acceptable performance from the RV and LV data?

Response: Thank you for emphasizing the importance of examining the performance metrics for RV and LV. We've included the performance data for both RV and LV as an attachment, ensuring that the reviewer can access these additional data for a more comprehensive understanding. Our preliminary examination indicates that the correlation of HU values from both RV and LV align closely with the observations from the descending aorta, reaffirming the validity and consistency of our main findings. While this added data does bolster the scientific rigor of our conclusions, in adherence to our original research plan and to maintain the conciseness of the manuscript, we have opted to not integrate these graphs directly into the primary manuscript.

 

Again, overall, the clinical relevance or use of CT-PA to assess anemia needs to be better presented as the clinical context seems not clinically accurate.

Response: Thank you for your feedback on the clinical relevance of CT-PA in assessing anemia. We respect your perspective and understand where you're coming from. However, based on our research and findings, we believe that the use of CT-PA does have merit in certain clinical scenarios. Our intent was to highlight a potential additional utility for a commonly used diagnostic tool, especially in situations where timely CBC results might be unavailable or delayed. We have attempted to clarify these scenarios more explicitly in our revised manuscript. Nevertheless, we value your insight and will continue to consider such feedback in our future research endeavors.

Author Response File: Author Response.pdf

Reviewer 3 Report

The presented manuscript provides a feasibility / proof of conept-study regarding detection of anemia in DECT CTPA with the aid of machine-learning algorithms. As indicated in the title, the study cohort is rather small (n=100), which suits the study aim, though. The work is well written and easily comprehensible.

I only have minor remarks:

It would be helpful if the authors stated mean Hb levels of the cases and controls instead of merely above or below 12 g/dL.

I would consider it beneficial to the manuscript if the authors actually defined „anemia“. Moreover, the WHO defines Anemia differently for men an women. Was this taken into account?

I consider the use of two different DECT systems a little problematic and possibly unnecessary considering the overall sample size. The authors do state that VNC and TNC images differ, however different DECT systems may differ in direct comparison, too. Please discuss.

Author Response

We sincerely appreciate the time and effort you have dedicated to reviewing our manuscript. Your constructive feedback has been invaluable in refining our work. Below, please find our responses to the comments and concerns raised.

 

It would be helpful if the authors stated mean Hb levels of the cases and controls instead of merely above or below 12 g/dL.

Response: Thank you for pointing that out. We recognize the importance of providing a detailed account of the Hb levels, as it provides a clearer picture of the patient profiles in our study. The mean Hb levels for both the case and control groups have been presented in Table 1 for clarity.

 

I would consider it beneficial to the manuscript if the authors actually defined „anemia“. Moreover, the WHO defines Anemia differently for men an women. Was this taken into account?

Response: Thank you for highlighting this important aspect. Indeed, the WHO has different definitions of anemia for men and women. In our study, we utilized a single threshold to distinguish cases from controls. We understand that this may not strictly adhere to the WHO definition, which is why we have opted to use the term "low Hb level" rather than "anemia" throughout the manuscript. This decision was made to ensure clarity and to avoid misrepresentation. We have emphasized this distinction more clearly in our revised manuscript to provide clarity to our readers (lines 355– 359).

 

I consider the use of two different DECT systems a little problematic and possibly unnecessary considering the overall sample size. The authors do state that VNC and TNC images differ, however different DECT systems may differ in direct comparison, too. Please discuss.

Response: Thank you for pointing out the potential issue with using two different DECT systems in our study. We acknowledge the variability that might arise from using different machines, even when comparing VNC and TNC images. The intention behind using two DECT systems was to provide a broader perspective and to test the feasibility of our approach across different platforms. Given this is a preliminary study, our aim was to lay groundwork for future investigations. We have now expanded the discussion section to address the potential differences between DECT systems and the implications of these differences on our findings (line 352 – 355).