Retrospective Motion Artifact Reduction by Spatial Scaling of Liver Diffusion-Weighted Images

Round 1

Reviewer 1 Report (Previous Reviewer 1)

Dear Authors,

Interesting article.

However, I would like to draw attention to an important aspect that is implicitly noted. However, the potential consequences are not fully shown.

The issue of systematic errors related to the spatial non-uniformity of magnetic field gradients. Which will be potentially important for off-center measurements.

1. It is primarily a physical effect known experimentally. The first specific description is Bammer's work from 2003, where the author associates this fact with the heterogeneity of gradient coils, described more precisely by the function of spherical harmonics.

2. Later experimental and theoretical studies showed that the effect is more complex (BSD-DTI), but that it is repeatable and may be a source of large systematic errors. It is also quite difficult to measure (phantoms - anisotropic and isotropic, additional time).

3. This effect, the spatial inhomogeneity of magnetic field gradients, does not depend on the imaged object. This is solely a feature of the MR scanner, the MR sequence used and its parameters.

4. The effect of the object itself (inhomogeneities in magnetic susceptibility or movement effects) may be dominant (e.g. cores of some rocks - shales, are practically not measurable by MRI). But certainly taking into account the actual spatial distribution of magnetic field gradients, the measurement will be more accurate than assuming a constant gradient in space, as is always done in the standard approach (constant matrix-b for a given DWI).

5. And we need to remember this when creating subsequent approaches and algorithms. It would be best to verify each time the level of systematic errors for DWI/DTI (spatial distribution of magnetic field gradients) related to the hardware itself, specific to the scanner and MR sequence. Knowing their impact, you can better assess the effectiveness of new methods.

minor

Author Response

Dear reviewer,

thank you for reviewing our manuscript. Please find attached our response letter.

With best regards,

Johannes Raspe

Author Response File: Author Response.pdf

Reviewer 2 Report (Previous Reviewer 3)

In the manuscript, a data-driven algorithm has been proposed to mitigate the signal loss in the DWI MR liver images. Outlier rejection with spatial scaling has been utilized to mitigate the signal loss in the DWI images. However, the article lacks numerous critical pieces of information which must be addressed to be considered for publication. Below are my comments and suggestions.

1.       Please clarify the rationale behind choosing the threshold value to be five. How this value has been determined?

2.       The rejection process comprises of both repetition rejection and voxel rejection. How the risk of over-rejection or under-rejection of data has been mitigated?

3.       How does the algorithm handle other sources of motion artifacts, such as residual respiratory motion?

4.       Please clarify how the scores have been defined for the criterion such as image quality, liver homogeneity etc.

5.       In this study, only one radiologist was involved in the evaluation of images. Will the results vary if multiple radiologists were involved? Was there any assessment of inter-rater reliability?

6.       How does the method compare with other existing methods in terms of efficiency, accuracy, and clinical applicability?

Author Response

Dear reviewer,

thank you for reviewing our manuscript. Please find attached our point-by-point response.

With best regards,

Johannes Raspe

Author Response File: Author Response.pdf

Reviewer 3 Report (New Reviewer)

The author proposed a data-driven algorithm to handle severely corrupted images and improve the quality of DWI. One notable outcome is the restoration of signal in the left liver lobe and the overall homogeneity of liver signal. This improvement is crucial in enhancing the diagnostic value of DWI in the left liver lobe. Additionally, the algorithm effectively reduces the overestimation of the apparent ADC in the left liver lobe.  Overall, this work demonstrates its potential to contribute significantly to the field by addressing image corruption issues, improving image quality, and reducing bias in DWI of the liver. Here  are some minors:

(1) The article exhibits certain issues in its formatting, specifically regarding inconsistent paragraph indentation. Some paragraphs are properly indented at the beginning, while others lack indentation.

(2) There are some concerns regarding the images utilized in the article. Firstly, the descriptions accompanying the subfigures lack clarity and precision, impeding a comprehensive understanding of their content. Additionally, the textual elements within the images are excessively small, rendering them difficult to read effectively.

(3)The description of the study subjects is succinct and lacks essential demographic information such as age, gender, and disease status. It is imperative to supplement the participant profiles with comprehensive sociodemographic details for a more thorough understanding of the study population.

Author Response

Dear reviewer,

thank you for reviewing our manuscript. Please find attached our point-by-point response.

With best regards,

Johannes Raspe

Author Response File: Author Response.pdf

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

Quite well written, interesting study the use of the motion artifact correction procedure in DWI of the liver. However, several significant changes and extensions are necessary.

Motion artifacts and systematic errors of a spatial nature (BSD-DWI/DTI) ?

1. First, the problem needs to be defined more precisely. Techniques based on DWI (such as measuring 3 orthogonal DWIs) are subject to the following rules. What is theoretically missing here?

In order to precisely determine the values of diffusion coefficients, diffusion coefficient, ADC, diffusion tensor, tensor kurtosis, etc. we need precise DWI measurements. To do this, we need a precise determination of the matrix b, which is a 3x3 matrix, symmetric, different for each diffusion gradient vector (DWI direction). Typically, there is a single matrix for each DWI provided by the manufacturer, potentially leading to large biases. And as recent works show, it actually has a spatial distribution.

Therefore, I am asking you to supplement this information and literature, I propose an analysis of the following issues:

- b-matrix,

- b-matrix spatial distribution,

- Generalized ST equation for non-uniform gradients,

- b-matrix spatial distribution in DWI/DTI

2. Another important issue is the discussion of the consequences of typical measurement approaches, i.e. not using the full b matrix or its actual spatial distribution.

As shown by theoretical works, numerical simulations, methodical experimental works on phantoms, single patients, in fact we have a spatial b-matrix distribution. Failure to take this into account leads to large systematic errors of a spatial nature. These errors cannot be removed by additional accumulations or by increasing the number of DWI directions.

This seems to be an important direction to improve techniques based on DWI. The topics for discussion here are:

- Systematic Errors in DWI/DTI

- Correction of Errors in DWI/DTI

- Validation of BSD-DTI

- Phantoms for BSD-DTI

- Anisotropic phantoms for BSD-DTI

In general, BSD-DTI seems to be the most comprehensive description of this problem at present.

This undoubtedly also has an impact on the any measured object, by ADC, DTI metrics or tractography (liver muscels).

It should be noted that the liver is outside the isocentre of the magnet, which greatly increases the systematic errors associated with non-uniform magnetic field gradients.

I give broader examples of the use of DWI, such as DTI. But DTI is actually at least 6 DWIs with different diffusion gradient vectors. All these descriptions, however, apply to a single DWI.

3. Which diffusion model do we use and what about perfusion. Below is a proposal for a more in-depth discussion. By manipulating the parameters (number of b, value of b, number of D values, D tensor or ADC) we are in different regimes (diffusion, perfusion, ...).

4.

Another quite important issue is the variety of measurement parameters, value of b, resolution, times TE, RT, field B, generally the parameters of a single DWI measurement. These are quantities that affect from a physical point of view how we "see" diffusion coefficients from a physical point of view. This is also an important issue and requires a certain consensus in the future as well as additional information from MRI scanner manufacturers, for example about diffusion times, small and large delta. This can be analyzed in the context of the issue:

- NMR Diffusometry and Cellular System

The article is an attempt to model non-Gaussian diffusion. But which approach best reflects reality, how many components of the diffusion coefficient or tensor should we take into account? There is no consensus here, there are trials. What we can do is provide the best measured image without systematic errors. Then we can build models.

These are a few important points whose discussion, substantive and literature supplementation will enrich this undoubtedly promising manuscript.

minor

Author Response

Dear reviewer,

thank you for your revision of our manuscript. Please find the point-by-point response letter attached here.

Author Response File: Author Response.pdf

Reviewer 2 Report

 The paper presents a data-driven algorithm aimed at mitigating motion-induced signal loss in respiratory-triggered diffusion-weighted magnetic resonance imaging (DWI) of the liver, with a focus on the left liver lobe. The algorithm involves exclusion of severely corrupted images and subsequent spatially dependent image scaling to improve the quality of multi-average diffusion-weighted images. The paper provides a comprehensive overview of the challenges faced in liver DWI and the existing techniques to address motion artifacts. The proposed algorithm is evaluated using in vivo DWI data from patients receiving an abdominal MRI examination. Overall, the study is relevant and valuable for the field of liver DWI. However, there are some areas that need improvement and clarification. Below are the specific points of feedback:

Clarity and Organization:

The abstract provides a concise overview of the study but lacks specific details about the proposed algorithm. Consider adding a sentence briefly describing the steps involved in the algorithm and its novelty.

The introduction is well-written and sets the context for the study. However, the objectives of the research and the main contributions of the proposed algorithm should be explicitly stated.

Background and Related Work:

The background section adequately introduces the concept of DWI and its applications in liver imaging. However, additional references or examples could be included to support the information provided. There are also some recent references missing, e.g.,

1)     Robust Feature Matching for Remote Sensing Image Registration via Guided Hyperplane Fitting, IEEE Transactions on Geoscience and Remote Sensing, 2022, 60, 1-14

2)     Deterministic Model Fitting by Local-Neighbor Preservation and Global-Residual Optimization, IEEE Transactions on Image Processing

The related work section mentions various existing techniques for motion compensation in liver DWI. It would be beneficial to include a brief summary or comparison table of these methods, highlighting their strengths and limitations.

Methodology:

The methodology section provides an overview of the proposed algorithm but lacks sufficient technical details. Consider providing step-by-step explanations of the data-driven algorithm, including the criteria for excluding severely corrupted images and the specific spatially dependent image scaling technique employed.

The proposed algorithm's mathematical formulation, if available, should be included to enhance the technical rigor of the paper.

The conclusion provides a concise summary of the study's findings. However, it could be strengthened by emphasizing the significance of the proposed algorithm and its potential impact on clinical liver DWI evaluation.

Suggest including a call-to-action for future research directions or practical applications of the algorithm in a clinical setting.

Language and Formatting:

The paper's language is generally clear, but some sentences are quite lengthy and complex. Consider breaking them into smaller, more digestible sentences for better readability.

Ensure consistent formatting and referencing throughout the paper.

Overall, the paper presents an interesting and valuable contribution to the field of liver DWI. Addressing the above points will enhance the clarity, technical rigor, and overall impact of the paper.

See above

Author Response

Dear reviewer,

Thank you for reviewing our manuscript. Please find attached our point-by-point responses.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this manuscript, a post-processing approach to reduce the motion-induced signal loss in the DWI of the liver has been proposed.  A two-step process involving the dismissal of slice and voxel outliers based on data thresholds followed by scaled averaging of multi-repetition data has been deployed to reduce signal loss and correct ADC estimation. However, the following concerns must be addressed, and necessary revisions need to be made in the article for it to be considered for publication.

1.       Can you please annotate Fig 1(a) to explicitly identify the region of the left liver lobe?

2.       In Section 2.2.1, it's not clear how the empirical value of 5 was chosen for λ(rep). Could you provide an explanation or any statistical evidence supporting this choice? Could this value vary under different imaging conditions?

3.       In Section 3.3, you're comparing ADC values across various liver segments using the standard averaging method. Have you implemented a correction for multiple comparisons to account for this multi-segment analysis? If not, could you elaborate on why it was not necessary?

4.       In the same Section 3.3, how does the variation in ADC values across different liver segments relate to the liver's health and structure? Could you elaborate on what these differences imply?

5.       In the study, in addition to the ADC bias in the left liver lobe, a significant change in ADC in the right liver lobe can also be noticed. Could you provide more explanation on what could be causing this change?

6.       Please discuss why the proposed method improved homogeneity in only 35.8% of evaluated cases. Why did it not work effectively in the remaining cases?

7.       Please discuss the implication of the proposed technique in clinical settings and how it can improve the diagnostic accuracy of liver DWI.

.

Author Response

Dear reviewer,

thank you for reviewing our manuscript. Attached, please find our point-by-point response.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

Dear authors,

It's hard to talk about motion artifacts without the influence of other effects.

I understand that the work is about the impact of motion artifacts. But in the introduction and discussion, the possible influence of other effects that affects the measurement should be indicated.

1. Indeed, BSD-DTI-type effects are best studied in the brain, but there are examples in other organs as well.

2. The spatial non-uniformity of the magnetic field gradients and, consequently, the spatial distribution of the b matrix for a single DWI are a physical fact. We know from brain research that the effect can be huge. In addition, we know from the literature that for the off-center positions of the magnet, it increases. Thus, the effect on the liver is to be expected to be greater than on the brain. Hence my suggestion.

3. The BSD-DWI(DTI) effect is only related to hardware, MR sequence and its parameters. It has nothing to do with the tested object. Yes, the heterogeneity of the object (differences in magnetic susceptibility, large motion artefacts) may dominate systematic errors resulting from the non-uniformity of the magnetic field gradients generated by the MR sequence and interactions with the environment (cross-terms, radiation damping, gradient coils, eddy currents..). However, these factors undoubtedly generate a systematic pattern, which may be the source of large systematic errors.

I suggest taking this fact into consideration.

ok

Reviewer 2 Report

After carefully reading the response report and paper, we donot think the authors have considered my comments. So please read the last version of my comments.

see above.