Review Reports - Utilization of AI to Diagnose Aortic Stenosis in Patients Undergoing Hemodialysis

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript addresses an important clinical challenge — the early detection of aortic stenosis (AS) in patients undergoing hemodialysis (HD) — by evaluating an AI-enabled phonocardiography device (“Super Stethoscope”). The topic is timely, given the rising interest in AI-assisted diagnostics and the high cardiovascular risk profile of HD patients. The study is well-organized, with clear sections and a concise narrative. The reported sensitivity and specificity for moderate/severe AS detection are promising.

However, I have some suggestions, that may help improve the paper.

1) Methods: The inclusion/exclusion criteria are not fully detailed. It is unclear whether patients were consecutively enrolled or if any selection occurred that might bias results toward more symptomatic individuals.

2) Echocardiography within the “past 18 months” is used as the comparator. Given the potential for disease progression in HD patients, this time gap could underestimate or overestimate agreement.

3) While the discussion acknowledges hemodialysis-related hemodynamic changes and coexisting valve diseases as sources of false positives, the study does not stratify results by presence of mitral regurgitation, valve calcification, or other structural abnormalities.

4) I would integrate the discussion section dealing with the importance of renal resistive index (RI) in your findings. For example, you should cite and discuss the study by Provenzano et al Renal resistive index in chronic kidney disease patients: Possible determinants and risk profile. PLoS One. 2020 Apr 1;15(4):e0230020. In fact, in this study the authors identified that lower eGFR, prior cardiovascular disease, diabetes, smoking, and higher serum phosphorus independently predict elevated renal resistive index in CKD patients. These findings underscore the shared pathophysiological mechanisms—such as vascular calcification and remodeling—between renal microvascular damage and valvular heart disease. In this context, the high sensitivity of the AI-powered ‘Super Stethoscope’ in detecting moderate to severe AS adds to a growing body of research supporting non-invasive screening tools for cardiovascular-renal complications in high-risk renal populations.

Author Response

September 6th, 2025

Manuscript ID: kidneydial-3791363

Type of manuscript: Article

The response to the reviewer’s comments

Thank you very much for your kind advice and very instructive suggestions. As we agreed to the reviewers’ instruction, we provided a point-by-point response to the reviewer’s comments.

Reply to the Reviewer 1

However, I have some suggestions, that may help improve the paper.

Methods: The inclusion/exclusion criteria are not fully detailed. It is unclear whether patients were consecutively enrolled or if any selection occurred that might bias results toward more symptomatic individuals.

Reply

Thank you for your kind advice. We do agree to this comment. We changed the expression to clarify the inclusion criteria of enrolled patients in the present study in the Methods section

Before: Consecutive patients receiving HD who underwent super stethoscope were enrolled in this study.

After: Consecutive patients receiving HD who underwent super stethoscope at the same time of routine ECG examination were enrolled in this study. In our institute, ECG examination was routinely performed on admission or every six months for patients receiving dialysis on an outpatient basis.

Echocardiography within the “past 18 months” is used as the comparator. Given the potential for disease progression in HD patients, this time gap could underestimate or overestimate agreement.

Reply

We completely agree to the comments provided by reviewer 1, because it is reasonable to think that the 18 months can be long enough for the patients on HD to encounter the progression of heart valve disease or deterioration of cardiac functions. Therefore, we added the following sentences as fourth reason why the rate of false positive is increased in the patients receiving HD in the section of Discussion.

Before: Third, confounding factors such as the medications, ages, blood pressure, and serum cholesterol levels cannot be neglected when interpreting the obtained data [19, 22-24].

After: Third, confounding factors such as the medications, ages, blood pressure, and serum cholesterol levels cannot be neglected when interpreting the obtained data [19, 23-25]. Fourth, the findings of echocardiography within the past 18 months were utilized to compare to those obtained from super stethoscope in the present analysis. The progression of valvular heart disease or deterioration of cardiac functions might occur during the 18 months among the patients on HD who frequently suffered from various complications.

While the discussion acknowledges hemodialysis-related hemodynamic changes and coexisting valve diseases as sources of false positives, the study does not stratify results by presence of mitral regurgitation, valve calcification, or other structural abnormalities.

Reply

Coexisting valvular diseases are relatively common in the patients receiving hemodialysis and we agree that coexisting valvular heart disease should decrease the specificity to detect aortic stenosis (AS) by visualized phonocardiogram; however, we did not determine how these disorders influence the rate of false positive. We, therefore, just described the coexisting valvular heart diseases and their incidences in the text of “The reasons why the rate of false positive is increased in the patients receiving HD” in Discussion.

Before: Fourth, patients on HD sometimes have other valvular heart disease comorbidities than AS.

After: Fourth, patients on HD sometimes have other valvular heart disease comorbidities than AS [Mitral valve regurgitation (MR), mild 67, moderate 16, severe 0;

Aortic valve regurgitation (AR), mild 40, moderate 18, severe 1; Tricuspid regurgitation (TR), mild 57, moderate 14, severe 1; Mitral valve stenosis (MS), mild 0, moderate 1, severe 0].

Reply

We really appreciated introducing the concept of renal resistive index (RI). We did not measure RI in the routine echocardiography at all. In addition, the usefulness of RI was identified in the patients with non-dialysis CKD, but not determined in the patients on HD. Therefore, instead of demonstrating the importance of RI, we speculated the relevance of RI and added the following sentences in the section of Discussion.

Recently, Provenzano M, et al. proposed that having a risk profile of chronic kidney disease (CKD) patients with abnormal renal resistance index (RI) may be relevant for the clinicians [22]. They identified decreased eGFR, cardiovascular disease, diabetes, smoking, and high serum phosphorus as independent predictors of elevated RI in non-dialysis CKD. These findings emphasize a common pathophysiological mechanism linking renal microvascular injury and valvular heart disease—such as vascular calcification and remodeling. Therefore, the AI-equipped “Super Stethoscope” is expected to contribute as a non-invasive screening tool for cardiovascular disease in patients with CKD as well as those on HD.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

This study evaluates an AI-powered phonocardiogram device, the "super stethoscope," as a screening tool for aortic stenosis (AS) in hemodialysis patients by comparing its diagnostic performance against TTE. the authors found the device to have high sensitivity (0.9) for detecting moderate to severe aortic stenosis in this high-risk population and moderate specificity (0.7). the research provides a valuable look into how new technology might aid in early cardiac disease detection in a clinical dialysis setting. however, there are some relatively minor areas that could benefit from revision/clarification.

the interpretation of the correlation is overly optimistic. the paper reports a "significant correlation" of r = 0.578 between the ai-based murmur grade and peak aortic jet velocity. while statistically significant (p < 0.05), a Pearson coefficient of 0.578 indicates only a moderate, not a strong, relationship. the scatter plot (labeled figure 1) visually confirms this, showing substantial overlap between grades and a wide dispersion of data points. this suggests that while a trend exists, the AI grade alone is a somewhat imprecise predictor of the actual jet peak velocity, a weakness that should be more candidly addressed.
the clinical implication of the modest specificity is underdeveloped. a specificity of 0.70 for detecting moderate or severe AS means that 30% of patients without significant AS will be flagged for further investigation. in a busy dialysis unit, this false positive rate could lead to a significant number of unnecessary TTEs, creating patient anxiety, increasing healthcare costs, and adding to the burden on diagnostic services. please add a discussion of the cost-benefit implications and the practical feasibility of managing this referral rate to help readers assess this tools’ real-world utility.
lack of direct comparison to existing standard of care: human auscultation. the introduction appropriately critiques the unreliability of traditional auscultation, yet the study does not compare the "super stethoscope's" performance to that of a physician (nephrologist or cardiologist) examining the same patients. without this direct comparison, it is difficult to determine the true incremental value of the device. the ai could be better, worse, or the same as a trained clinician, but this crucial piece of evidence is absent.
the ai algorithm's methodology lacks transparency. the manuscript describes the ai-based classification as grades a, b, c, or d based on a prediction algorithm from a previous study. however, it provides no detail on what acoustic features (e.g., murmur intensity, timing, frequency profile, shape) the algorithm uses to determine these grades. this "black box" approach prevents readers from critically evaluating the technology itself and makes it difficult for other researchers to replicate or build upon the findings.
timeframe between phonograms and TTEs = long. using TTEs from the "past 18 months" is a potential flaw in a population where cardiovascular disease can progress materially in that time. a patient's cardiac status could have changed significantly between the echo and the phonocardiogram, confounding the results. a prospective study with a much narrower and more consistent timeframe between the phonocardiogram and the chosen gold standard (TTE) would be required to validate these preliminary findings. would help to acknowledge this in the limitations section.

Author Response

September 6th, 2025

Manuscript ID: kidneydial-3791363

Type of manuscript: Article

The response to the reviewer’s comments

Thank you very much for your kind advice and very instructive suggestions. As we agreed to the reviewers’ instruction, we provided a point-by-point response to the reviewer’s comments.

Reply to the Reviewer 2

the interpretation of the correlation is overly optimistic. the paper reports a "significant correlation" of r = 0.578 between the ai-based murmur grade and peak aortic jet velocity. while statistically significant (p < 0.05), a Pearson coefficient of 0.578 indicates only a moderate, not a strong, relationship. the scatter plot (labeled figure 1) visually confirms this, showing substantial overlap between grades and a wide dispersion of data points. this suggests that while a trend exists, the AI grade alone is a somewhat imprecise predictor of the actual jet peak velocity, a weakness that should be more candidly addressed.

Reply

We do agree to this comment. We added the following sentences in section of Discussion.

We observed a significant correlation between the AI-based murmur grade and peak aortic jet velocity, but a Pearson coefficient of 0.578 indicated only a moderate, not a strong, relationship. This association might be improved if echocardiography were examined at the same day of super stethoscope instead of within the past 18 months; however, echocardiography involves high screening costs and requires physicians or trained technicians and it is sometimes difficult to promptly make a reservation of echocardiography in a dialysis unit.

the clinical implication of the modest specificity is underdeveloped. a specificity of 0.70 for detecting moderate or severe AS means that 30% of patients without significant AS will be flagged for further investigation. in a busy dialysis unit, this false positive rate could lead to a significant number of unnecessary TTEs, creating patient anxiety, increasing healthcare costs, and adding to the burden on diagnostic services. please add a discussion of the cost-benefit implications and the practical feasibility of managing this referral rate to help readers assess this tools’ real-world utility.

Reply

Thank you indeed for your kind comment of real world utility of the Super Stethoscope. Even though the specificity in detecting AS is not satisfactory, high sensitivity of super stethoscope to detect AS is very useful to recommend echocardiography to confirm the diagnosis of AS. The examination of echocardiography can also determine the presence of other valvular heart diseases, which sometimes require prompt treatment. Firstly, super stethoscope is easy to perform without any special equipment, highly portable, and the output of the results can be available within 1 minute. Secondly, echocardiography involves high screening costs and requires physicians or trained technicians. It is not easy to adjust the reservation date of echocardiogram as soon as possible in a dialysis unit. So we added the following sentences to explain the clinical utility of this device.

The characteristic of high sensitivity and low specificity to detect AS provided by this modality might be beneficial in clinical setting. In the outpatient hemodialysis clinics where the specialists in echocardiograhy are not a full-time position. On the other hand, screening to detect AS by super stethoscope, a low-cost medical device that aims to be easy for general practitioners to use thanks to its advanced auscultation technology and AI diagnosis, can also cover other valvular heart diseases, which sometimes require prompt treatment.

lack of direct comparison to existing standard of care: human auscultation. the introduction appropriately critiques the unreliability of traditional auscultation, yet the study does not compare the "super stethoscope's" performance to that of a physician (nephrologist or cardiologist) examining the same patients. without this direct comparison, it is difficult to determine the true incremental value of the device. the ai could be better, worse, or the same as a trained clinician, but this crucial piece of evidence is absent.

Reply

We do agree to the comments. Actual auscultation was performed by two physicians (one cardiologist and one nephrologist) using archived audio data. And then the specificity to detect AS between AI diagnosis and human auscultation were compared between AI diagnosis and human auscultation. The results were demonstrated in Table.

Table Comparison of Specificity to Detect AS Between AI Diagnosis and Human Auscultation

	Specificity
AS diagnosed by echocardiography	AI-based diagnosis B		Human auscultation by a cardiologist (YM)	Human auscultation by a nephrologist (MO)
	N=194	N=177	N=177	N=177
Mild and higher	0.72	0.71	0.39	0.38
Moderate and severe	0.70	0.69	0.37	0.37

Based on the above results, we added the following sentences in the section of Results.

Because the specificity of super stethoscope to detect moderate and severe AS was not high enough, auscultation was then performed by two physicians (one cardiologist and one nephrologist) using archived audio data to compare the specificity between AI diagnosis and human auscultation. Of the total 194 cases, 17 could not be evaluated due to environmental noise interference or lack of saved audio data. In contrast, AI diagnosis was able to make a determination even in cases where auscultation was impossible due to environmental factors. As results 177 cases could be evaluated where direct comparison between the Super Stethoscope and standard human auscultation was possible. The specificity of actual auscultation to detect moderate and severe AS was 0.37 for both a cardiologist and a nephrologist, which was much lower than the specificity of AI diagnosis (0.70).

the ai algorithm's methodology lacks transparency. the manuscript describes the ai-based classification as grades a, b, c, or d based on a prediction algorithm from a previous study. however, it provides no detail on what acoustic features (e.g., murmur intensity, timing, frequency profile, shape) the algorithm uses to determine these grades. this "black box" approach prevents readers from critically evaluating the technology itself and makes it difficult for other researchers to replicate or build upon the findings.

Reply

Thank you very much for your thoughtful comment. We have revised the manuscript (in blue letters）accordingly by adding a description of the AI framework for AS estimation, as well as details regarding the input data, in the Methods section.

Background

Therefore, we determined the presence of AS using the estimation of “Super Stethoscope” along with electrocardiography (ECG) and echocardiography among patients undergoing HD

Methods

For the estimation of AS using phonocardiograms, the entire raw waveform of 8-second heart sound recordings was utilized, and the results were ultimately classified into four categories of AS severity: A, B, C, or D Cutting-edge digital technologies, including AI and machine learning, have been utilized as analytical tools for systolic murmurs. AS severity was classified as A, B, C, or D according to the prediction algorithm proposed by Nomura et al. [16]. This AI framework was developed using a convolutional neural network (CNN), with the input consisting of phonocardiographic signals and the supervisory labels derived from AS severity determined by transthoracic echocardiography. The model was trained on approximately 2,000 paired datasets. During training, data augmentation techniques such as the superimposition of Gaussian noise were applied to enhance the model’s robustness to signal variations. The 1D-CNN was constructed with a ResNet backbone incorporating a self-attention pooling mechanism. Importantly, this AI framework is considered to learn not only the typical features of ejection systolic murmurs that we usually observe, but also additional information such as the width of the first heart sound and the presence of extra heart sounds, including the fourth sound. In this study, the AS estimation was performed in a stepwise manner: first, classifying cases as moderate AS or not; for those not classified as moderate, a second step classified them as mild AS or not. Finally, patients were categorized into four classes: A = none, B = mild, C = moderate, and D = severe.

The results of the AI-based estimation of AS were compared with the findings obtained using transthoracic echocardiography conducted within the past 18 months using the phonocardiograms.

timeframe between phonograms and TTEs = long. using TTEs from the "past 18 months" is a potential flaw in a population where cardiovascular disease can progress materially in that time. a patient's cardiac status could have changed significantly between the echo and the phonocardiogram, confounding the results. a prospective study with a much narrower and more consistent timeframe between the phonocardiogram and the chosen gold standard (TTE) would be required to validate these preliminary findings. would help to acknowledge this in the limitations section.

Reply

We completely agree to the comments provided by reviewer 2, because it is reasonable to think that the 18 months can be long enough for the patients on HD to encounter the progression of heart valve disease or deterioration of cardiac functions. Therefore, we added the following sentences as fourth reason why the rate of false positive is increased in the patients receiving HD in the limitation section of Discussion.

Before: Third, confounding factors such as the medications, ages, blood pressure, and serum cholesterol levels cannot be neglected when interpreting the obtained data [19, 22-24].

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Comments and Suggestions for Authors

The manuscript is now ready for publication.

Author Response

Thank you for your feedback.

Reviewer 2 Report

Comments and Suggestions for Authors

The Authors have addressed all of my critiques. Thank you.

Author Response

Thank you for your feedback.