Review Reports

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This manuscript presents a well-conducted and clinically relevant study investigating the use of Dual-Energy CT (DECT) for non-invasive characterization of axillary lymph nodes in breast cancer patients. The authors propose a predictive model that integrates morphological and spectral imaging parameters to differentiate metastatic from benign lymph nodes.

Strengths

1. Clinical Relevance and Innovation: The study addresses a critical aspect of breast cancer staging and proposes an innovative imaging-based approach that could reduce the need for invasive procedures such as sentinel lymph node biopsy.

2. Robust Methodology: The retrospective analysis includes a substantial cohort (117 patients, 375 lymph nodes), with rigorous statistical modeling and internal validation. The use of logistic regression and cross-validation enhances the reliability of the findings.

3. Clear Presentation of Results: The manuscript is well-organized, with comprehensive tables and figures that effectively illustrate the diagnostic performance of the proposed models. The inclusion of water concentration (WC) as a spectral parameter adds a novel dimension to lymph node assessment.

4. Balanced Discussion: The authors critically compare their findings with existing literature, acknowledge limitations, and suggest future directions, including multicenter validation and integration into standardized frameworks like Node-RADS.

Minor Limitations

• Modest Incremental Value of DECT Parameters: Although the addition of WC slightly improves the model’s AUC (from 0.871 to 0.883), the difference is not statistically significant. Nonetheless, the physiological rationale and potential clinical utility are well-articulated.

• Single-Center Retrospective Design: The study’s design may limit generalizability, but the authors appropriately propose multicenter prospective trials to address this.

• Manual ROI Placement: While interobserver agreement was high, future studies could benefit from automated segmentation to enhance reproducibility.

Conclusion and Recommendation

The manuscript is scientifically sound, methodologically rigorous, and contributes valuable insights to the field of oncologic imaging. The proposed integrated model combining DECT-derived water concentration with morphological criteria is promising and warrants further exploration.

Recommendation: Accept for publication with minor editorial revisions if needed.

Author Response

We thank the reviewer for his kind comments and for having understood the goal of the manuscript. We briefly responded and articulated the Minor limitations he addressed.

1. Modest Incremental Value of DECT Parameters: Although the addition of WC slightly improves the model’s AUC (from 0.871 to 0.883), the difference is not statistically significant. Nonetheless, the physiological rationale and potential clinical utility are well-articulated.

We agree with the reviewer that the incremental value of DECT-derived parameters over conventional morphological features is modest in our cohort. The small, non-significant increase in AUC observed after adding WC (from 0.871 to 0.883) is in line with the findings of Kong et al., who reported in their meta-analysis that the role of DECT in lymph node characterization appears to provide only limited additional benefit over standard imaging.
We believe that several technical factors may partly explain this modest incremental value, including the lack of standardized acquisition protocols, the availability of a single late arterial phase, and the still imperfect spectral separation of current systems for very small targets such as axillary lymph nodes. Our results therefore temper the most optimistic expectations regarding DECT as a stand-alone solution, but at the same time support its physiological rationale and potential as a complementary tool.

 

2. Single-Center Retrospective Design: The study’s design may limit generalizability, but the authors appropriately propose multicenter prospective trials to address this.

We thank the reviewer for this comment and fully agree that the single-center, retrospective design represents an important limitation of our study and restricts the generalizability of our findings. As correctly noted, our results are influenced by the specific institutional protocol, scanner technology, and patient population of our center.

 

3. Manual ROI Placement: While interobserver agreement was high, future studies could benefit from automated segmentation to enhance reproducibility

We thank the reviewer for this helpful suggestion. In the revised manuscript, we have added an interobserver reproducibility analysis for quantitative DECT parameters in a subset of 50 lymph nodes, showing good-to-excellent agreement. We fully agree that automated or semi-automated segmentation could further enhance reproducibility in future studies. However, at present, fully automatic segmentation of small axillary lymph nodes remains technically challenging, and available semi-automatic tools often have limited reliability in this setting because of the small size, variable shape, and close proximity to surrounding fat and vessels. We have briefly discussed this point in the limitations section and highlighted automated segmentation as an important direction for future research.

Reviewer 2 Report

Comments and Suggestions for Authors

Using Ultrasound to evaluated the axilla is very reliable with the aded advantage to be able to do Fine Needle Aspiration. What is the real benefit of using a more time consuming technique that will provide similar results?  

I can see some use of this technique in assessing for residual positive LN after SLNB. Any experience?

Haw can you be sure that the node retrieve with surgery is the same one that was evaluated with the CT scan. Normally we clip the node that is suspicious to be sure it has been removed. Please comment.

Will the Medical Oncologist be satisfied with assessing the LN with this technique? Will they feel confident make decisions just base on this results. The same goes for Radiation Oncology

Author Response

We thank the reviewer for his comments that highlight a in person clinical experience with the management of this issue. Following we listed responses to the comments provided.

 

Using Ultrasound to evaluate the axilla is very reliable with the added advantage to be able to do Fine Needle Aspiration. What is the real benefit of using a more time-consuming technique that will provide similar results?  

We fully agree that ultrasound (US) is a reliable and widely available technique for assessing axillary lymph node status, with the important additional advantage of allowing US-guided fine-needle aspiration (FNA) when indicated. Our intent was not to question the value of axillary US, but rather to critically evaluate the clinical utility of DECT in a setting where previous studies have suggested a potential advantage over US in terms of diagnostic performance [1,2].
In our consecutive surgical cohort, we found that DECT provides good, but not outstanding, accuracy, and that its performance is broadly comparable to the accuracy reported for axillary US in the literature. We consider this an important result, as it indicates that some of the previously published DECT results may have been overly optimistic.
Regarding the clinical benefit, we see DECT as a complementary rather than alternative tool. US remains the first-line modality, but it has well-known limitations, including operator dependence and reduced sensitivity for deeper or higher axillary nodes, especially in patients with unfavorable acoustic windows or post-surgical changes. DECT, on the other hand, offers a standardized, operator-independent, three-dimensional evaluation of the entire axilla and thorax, and in many institutions contrast-enhanced CT is already performed for preoperative staging. In this context, DECT-derived spectral parameters may provide additional non-invasive information at no extra scan time, particularly in cases in which US findings are equivocal or when US-guided sampling is not feasible or not planned.

1. X. Zhang et al., “Axillary sentinel lymph nodes in breast cancer: Quantitative evaluation at dual-energy CT,” Radiology, vol. 289, no. 2, pp. 337–346, Nov. 2018, doi: 10.1148/radiol.2018180544.
2. Terada et al., “Predicting axillary lymph node metastasis in breast cancer using the similarity of quantitative dual-energy CT parameters between the primary lesion and axillary lymph node.” Japanese journal of radiology, vol. 40, no. 12, pp. 1272–1281. Dec 2022, doi:10.1007/s11604-022-01316-8.

I can see some use of this technique in assessing for residual positive LN after SLNB. Any experience?

We thank the reviewer for this insightful comment. We agree that DECT could be of potential value in assessing residual axillary disease after SLNB, and this has been suggested as an emerging application in the literature. In our institution, DECT has occasionally been used in this setting and we have observed a few anecdotal cases in which it aided in identifying residual suspicious nodes; however, these experiences are neither systematic nor sufficient to draw firm conclusions. In the present study, DECT was performed exclusively in the preoperative staging setting, and our main objective was to evaluate its reliability in characterizing axillary lymph nodes before surgery.

 

How can you be sure that the node retrieve with surgery is the same one that was evaluated with the CT scan. Normally we clip the node that is suspicious to be sure it has been removed. Please comment.

We thank the reviewer for this important methodological comment. We fully agree that establishing a robust correspondence between the lymph node evaluated on imaging and the node retrieved at surgery is crucial, and that preoperative clipping of suspicious nodes is an excellent strategy when routinely implemented.
In our institution, preoperative clipping of axillary nodes is not systematically performed in patients undergoing upfront surgery. However, several measures were adopted to maximize a reliable imaging–pathology correspondence. As detailed in the revised Methods section, the operating breast surgeon systematically participated in the preoperative review of CT images together with the radiologist to identify suspicious nodes and to document their location with respect to fixed anatomical landmarks (axillary level, relationship to vessels and muscles, and depth from the skin surface). During surgery, sentinel lymph nodes were localized using blue dye and/or radioactive tracer and indocyanine green, and all removed sentinel and non-sentinel nodes were mapped according to axillary level and anatomical site before histopathological examination. Importantly, for the present analysis we only included patients for which a consistent match among CT findings, intraoperative mapping (level I axilla in almost all cases), and histopathological reports could be established. On this basis, we are confident that node-level correspondence was achieved in the large majority of evaluated nodes.

Will the Medical Oncologist be satisfied with assessing the LN with this technique? Will they feel confident make decisions just base on these results. The same goes for Radiation Oncology

We thank the reviewer for raising this important clinical point. We fully agree that axillary management decisions in both medical and radiation oncology should not rely on a single imaging technique. In current practice, lymph node status is assessed by integrating clinical examination, ultrasound (with the option of image-guided sampling), cross-sectional imaging, pathological findings, and multidisciplinary discussion.

The aim of our study was not to propose DECT as a stand-alone decision tool, but to explore its potential contribution. Our results indicate that DECT, when combined with conventional morphological features, can provide additional quantitative information on nodal status; however, we do not suggest that medical or radiation oncologists should base treatment decisions solely on DECT findings.

Reviewer 3 Report

Comments and Suggestions for Authors

Dear authors, 

The manuscript is well-written. 

But please: 

The manuscript would benefit from additional references to support several key statements, particularly those related to the clinical relevance of DECT parameters. 

Strengthening the literature context will help justify the study rationale and place your findings more clearly within the existing studies.

Author Response

The manuscript would benefit from additional references to support several key statements, particularly those related to the clinical relevance of DECT parameters. 
Strengthening the literature context will help justify the study rationale and place your findings more clearly within the existing studies.

We thank the reviewer for the comment; we added additional references to support key statements related to the clinical relevance of DECT parameters.

Reviewer 4 Report

Comments and Suggestions for Authors

This manuscript presents a clinically meaningful retrospective analysis evaluating whether Dual-Energy CT (DECT)–derived spectral parameters, combined with traditional morphological features, can distinguish benign from metastatic axillary lymph nodes in patients with breast cancer.

Strengths:

Clinical relevance and novelty: The study addresses a clear clinical gap: the need for accurate, non-invasive assessment of axillary nodal status. The integration of DECT parameters with classical morphological criteria is innovative and potentially impactful for future workflows.

Major Comments:

1. Patient selection and risk of bias: Out of 638 eligible patients, only 117 were included. The manuscript should better detail the exclusion process.
2. Limited representation of nodal levels: Nearly all lymph nodes evaluated were at level I, which limits applicability to higher axillary levels. The authors should comment on whether DECT performance may differ across nodal levels.
3. Use of a single arterial phase for DECT acquisition: Perfusion-dependent parameters, especially iodine concentration, are highly sensitive to timing. Relying solely on the late arterial phase may attenuate the discriminatory potential of IC and could partly explain the prominence of WC in the multivariate model. This limitation deserves more emphasis.
4. ROI placement and reproducibility concerns: While morphological interobserver agreement is reported, the reproducibility of quantitative DECT measurements is not. Given the sensitivity of spectral parameters to ROI positioning, ICC or Bland–Altman analysis would significantly strengthen confidence in the findings.
5. Non-independence of lymph node data: Each patient contributed multiple nodes (mean 3.2), yet the analysis treats nodes as independent observations. A clustered or mixed-effects model would be more appropriate. If not feasible, the authors should clearly acknowledge the statistical implications.
6. External validity and risk of overfitting: Even with internal cross-validation, the modest sample size and number of predictors raise the possibility of overfitting. The authors should underscore the need for external, multicenter validation before broad clinical implementation.
7. Interpretation of clinical impact: To improve the quality of their manuscript, authors should include important recent references PMID: 39335140 regarding the consequences of axillary surgery and its implication of clinical and oncological prognosis.
8. Comparison with Node-RADS: The study emphasizes morphological features that overlap with Node-RADS. A direct comparison, or at minimum a discussion of how the proposed models align with Node-RADS criteria, would enhance interpretability.

Author Response

We deeply thank the reviewer for his insightful comments. We strongly believe that our paper can be ameliorated thanks to suggestions provided.
Here you can find your points and our response listed.

 

1. Patient selection and risk of bias: Out of 638 eligible patients, only 117 were included. The manuscript should better detail the exclusion process.
   We thank the reviewer for this important comment. We agree that detailed description of the patient selection process is essential to assess potential selection bias.
   The list of causes for patient exclusion is reported in Figure 1. However, in the revised version we have reported the exclusion criteria also in the text and expanded some criteria with a short explanation.
   We also clarify that the cohort represents a series of consecutive patients treated at our institution during the study period, which we believe minimizes selection bias.
   
   
2. Limited representation of nodal levels: Nearly all lymph nodes evaluated were at level I, which limits applicability to higher axillary levels. The authors should comment on whether DECT performance may differ across nodal levels
   We fully agree that the predominance of level I nodes represents an important limitation of our study. As reported in the Results, 366/375 nodes were located at level I and only 9/375 at level II, with no level III nodes in our dataset. In the revised Discussion, we emphasize that our findings therefore primarily apply to level I nodes, which are those most frequently targeted in clinical practice for early-stage breast cancer. We also note that this distribution likely reflects our focus on patients undergoing upfront surgery; including patients treated with neoadjuvant therapy would have introduced a substantially different axillary disease profile and treatment induced bias, making the interpretation of DECT findings less straightforward and beyond the scope of the present analysis
3. Use of a single arterial phase for DECT acquisition: Perfusion-dependent parameters, especially iodine concentration, are highly sensitive to timing. Relying solely on the late arterial phase may attenuate the discriminatory potential of IC and could partly explain the prominence of WC in the multivariate model. This limitation deserves more emphasis.
   We thank the reviewer for highlighting this important methodological issue. We agree that the use of a single late arterial phase potentially affects perfusion-sensitive parameters such as iodine concentration and spectral slope, and may partly contribute to the limited independent role of iodine metrics in the multivariate model.
   However the technology available allow a single DECT acquisition cause repeating the acquisition with the DE mode would lead to tube overheating.
   We strengthened the Limitations paragraph in the Discussion to explicitly state that the single-phase, late arterial acquisition may have attenuated the discriminative performance of iodine concentration and other perfusion-related parameters.
4. ROI placement and reproducibility concerns: While morphological interobserver agreement is reported, the reproducibility of quantitative DECT measurements is not. Given the sensitivity of spectral parameters to ROI positioning, ICC or Bland–Altman analysis would significantly strengthen confidence in the findings.
   We appreciate this valuable suggestion and fully agree that the reproducibility of quantitative DECT measurements is crucial.
   First, we have expanded the “Quantitative Analysis” subsection to provide a more detailed description of ROI placement. We now report the distribution of ROI size (mean, interquartile range, minimum and maximum values) and explicitly state that the fatty hilum, calcifications, necrotic areas, and adjacent vessels were systematically excluded. In addition, we describe the use of an attenuation threshold to exclude voxels with negative HU values. This thresholding step, which is a specific feature of our protocol, allowed us to sample most of the cortical component of the lymph nodes while minimizing the inclusion of surrounding adipose tissue.
   Second, in response to the reviewer’s comment, we performed an interobserver reproducibility analysis for the main quantitative DECT parameters on a subset of 50 lymph nodes independently evaluated by both readers. Intraclass correlation coefficients (ICCs) showed good-to-excellent agreement for the key spectral parameters, supporting the robustness of the quantitative measurements.
5. Non-independence of lymph node data: Each patient contributed multiple nodes (mean 3.2), yet the analysis treats nodes as independent observations. A clustered or mixed-effects model would be more appropriate. If not feasible, the authors should clearly acknowledge the statistical implications.
   We thank the reviewer for raising this important statistical point. We agree that treating all nodes as independent may underestimate variability because multiple lymph nodes are clustered within the same patient.
   In the Statistical Analysis section, we now explicitly acknowledge that the primary logistic regression models were built at the lymph-node level and therefore assume independence of observations. We know that multiple nodes per patient could induce within-patient correlation and we explicitly discuss the potential impact on standard error estimation in the discussion.
   In the Discussion, we now clearly state that this non-independence represents a limitation of our study and underline that, although the number of events per variable was adequate, residual clustering may lead to over-optimistic confidence intervals and p-values. Furthermore, in exploratory analyses using patient-level summaries (e.g. considering the most suspicious node per patient), the direction and magnitude of the associations did not materially change, suggesting that the main conclusions are robust.
6. External validity and risk of overfitting: Even with internal cross-validation, the modest sample size and number of predictors raise the possibility of overfitting. The authors should underscore the need for external, multicenter validation before broad clinical implementation.
   We fully agree with the reviewer that, despite internal validation, the risk of overfitting cannot be completely excluded.
   In the Discussion and Conclusions, we have now strengthened the message that our models, although internally cross-validated, remain exploratory tools developed in a single-center retrospective cohort with a limited and relatively homogeneous sample. We explicitly state that these models should not yet be used for routine clinical decision-making and that external, ideally multicenter prospective validation with harmonized DECT protocols is mandatory to confirm performance, refine thresholds, and test generalizability across different scanners and patient populations.

 

7. Interpretation of clinical impact: To improve the quality of their manuscript, authors should include important recent references PMID: 39335140 regarding the consequences of axillary surgery and its implication of clinical and oncological prognosis.

We thank for the comment and add important recent reference.

8. Comparison with Node-RADS: The study emphasizes morphological features that overlap with Node-RADS. A direct comparison, or at minimum a discussion of how the proposed models align with Node-RADS criteria, would enhance interpretability.
We appreciate this insightful comment and agree that a more systematic comparison with Node-RADS improves interpretability. In our study, the “morphologic” model was built using four CT-based features: lymph node size (short-axis diameter), cortical morphology (including focal or diffuse cortical thickening), fatty hilum status (preserved vs. effaced), and extranodal extension. These variables correspond closely to the core Node-RADS descriptors “size”, “cortex/shape”, “hilum” and “extranodal features”.
In the revised Discussion, we now explicitly state this overlap and clarify that our morphologic model can be regarded as a simplified, Node-RADS–like approach that focuses on the main structural criteria, while other Node-RADS items (such as nodal grouping or some site-specific features) were not systematically collected in our retrospective dataset. We also emphasize that we did not prospectively assign Node-RADS categories, and therefore no formal head-to-head diagnostic comparison with Node-RADS scoring could be performed. Instead, we conceptually position our integrated model as an extension of a Node-RADS–like morphological assessment by adding a single quantitative spectral parameter (post-contrast WC). In our cohort, this resulted in only a marginal, non-significant increase in AUC, suggesting that morphology remains the major driver of diagnostic performance.

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

The manuscript can be accepted in the present form