4D MRI: Robust sorting of free breathing MRI slices for use in interventional settings.

PURPOSE: We aim to develop a robust 4D MRI method for large FOVs enabling the extraction of irregular respiratory motion that is readily usable with all MRI machines and thus applicable to support a wide range of interventional settings.
METHOD: We propose a 4D MRI reconstruction method to capture an arbitrary number of breathing states. It uses template updates in navigator slices and search regions for fast and robust vessel cross-section tracking. It captures FOVs of 255 mm x 320 mm x 228 mm at a spatial resolution of 1.82 mm x 1.82 mm x 4mm and temporal resolution of 200ms. A total of 37 4D MRIs of 13 healthy subjects were reconstructed to validate the method. A quantitative evaluation of the reconstruction rate and speed of both the new and baseline method was performed. Additionally, a study with ten radiologists was conducted to assess the subjective reconstruction quality of both methods.
RESULTS: Our results indicate improved mean reconstruction rates compared to the baseline method (79.4% vs. 45.5%) and improved mean reconstruction times (24s vs. 73s) per subject. Interventional radiologists perceive the reconstruction quality of our method as higher compared to the baseline (262.5 points vs. 217.5 points, p = 0.02).
CONCLUSIONS: Template updates are an effective and efficient way to increase 4D MRI reconstruction rates and to achieve better reconstruction quality. Search regions reduce reconstruction time. These improvements increase the applicability of 4D MRI as a base for seamless support of interventional image guidance in percutaneous interventions.

Introduction

During the last decade, 4D MRI has gained considerable interest in research, because it promises access to information on the respiratory motion of the thorax and abdomen free of radiation. Respiratory motion information is vital for many medical applications in diagnostics [1], treatment planning [2] and execution [3]. Our application scenarios are MRI guided percutaneous interventions on the liver like radio frequency-, microwave- and cryoablation, biopsies, or brachytherapy, where the challenge of a moving target exists. 4D MRI methods have been proposed, but none satisfy all the needs for our interventional application. These needs are first, physiological correctness of the 4D sequence, and second, robustness against the out-of-plane motion. In this study, we propose a new 4D MRI reconstruction method. It utilizes retrospective sorting of dynamic 2D TRUFI MRI slices and is capable of imaging the whole liver during free breathing and capturing organ deformations caused by respiration. It reconstructs a physiologically meaningful sequence of respiratory states by utilizing a dedicated navigator frame and copes with out-of-plane motion.

Related work

To our knowledge, there exist two approaches to acquiring 4D MRI, each with its unique advantages and disadvantages. The first is to acquire 3D MRI sequences in real-time, as done by Kim et al. [4] and Bled et al. [5]. The advantages of this approach are that it does not rely on gating and thus supports imaging events that do not occur repeatedly, i.e., events that are not periodic. The disadvantages of this approach are its low temporal and spatial resolution [6, 7] and its relatively small FOV, rendering it impossible to capture the respiratory motion of large organs like the liver.

The second approach is to reconstruct volumes for different organ states or breathing phases in retrospection by binning previously acquired data. Two main types of this approach exist. In the first type, the k-space data is sparsely sampled and binned before reconstructing a volume for a given organ state [8-10]. The strength of this type lies in capturing periodic organ state changes with a large FOV within a few minutes, depending on the length of the motion cycle. Its weaknesses are its assumption of strictly periodic organ motion. Thus, it can only reconstruct an average motion cycle of the target organ, which is not ensured to be physiologically meaningful. Furthermore, this type introduces image artifacts [11, 12] that could hinder motion estimation from the reconstructed 4D MRI.

The second type of the second approach reconstructs fast dynamic 2D sequences at all slice positions to cover the organ of interest. Then retrospective gating is applied to the resulting 2D images, binning them by different organ states, i.e., breathing states, and sorting them in their respective volumes. Its advantages are its applicability for non-periodic or quasi-periodic changes in the organ state and its high temporal and spatial resolution. Hence it is well-suited to capture motion variation, e.g., deep or shallow, abdominal or thoracic breaths within one session. It can work with a navigator or respiratory signal to ensure the physiological correctness of reconstructed motion. A further advantage of the binning strategy is its availability because it is readily usable with all MRI machines and all 2D sequences. Its disadvantages are that it is more time-intensive than the k-space binning and that much of the acquired data is redundant. The latter, however, can advantageously be used to increase the SNR of the reconstructed 4D images.

For both types, the surrogate can be intrinsic, relying on image information or k-space information, or extrinsic, relying on externally recorded signals, e.g., from using a breathing belt or form tracking markers that are placed on the abdomen of the subject. Siebenthal et al. [13, 14] utilize navigator slices as surrogate and vessel cross-section tracking as a matching criterion. Cai et al. [15] use the body area. Lee et al. [16] use sagittal diaphragm profiles and reconstruct one breathing cycle. Tong et al. [17] propose a graph-based sorting where the weights are based on image information and semi-automatic assigned respiratory phase although, they are only able to reconstruct one best breathing cycle and not a variety of breathing cycles. Romaguera et al. [18] propose a graph-based approach using pseudo-navigators. A drawback of the graph-based navigator-less approach is that physiological correctness cannot be ensured even if temporal coherence is ensured.

Materials and methods

We decided to follow the retrospective sorting approach because, as set out in the related work section, it is the only one suited for capturing physiologically meaningful, non-periodic organ motion with high temporal and spatial resolution and large field of views. Its only disadvantage is the long acquisition time, which can be overcome, as shown in this work. Specifically, we build upon the proposed method of von Siebenthal et al. [13, 14].

The Otto-von-Guericke-University Magdeburg ethics board approves our study “Studies with healthy subjects in 3 Tesla for methodological development of MRI experiments” (approval number 172/12), stating they concluded that there are no ethical concerns and that this approving assessment is made based on unchanged conditions. Oral and written consent was obtained during the study.

In the following three sections, we describe the general concept behind the baseline method and our method. In section Template updates and search region, we describe how we build upon the baseline to improve it and overcome the named drawbacks.

MRI acquisition

Our MR data were acquired on a MAGNETOM Skyra MRI scanner (Siemens Medical Solutions, Erlangen, Germany). All images were acquired with a TRUFI sequence (TR = 39.96 ms, echo spacing = 3.33 ms, TE = 1.49 ms, flip angle = 30 degree, readout bandwidth = 676 Hz/px, base resolution = 176 k, phase resolution = 80% yielding a matrix size of 140 x 176, in-plane resolution 1.82mm x 1.82mm, out of plane resolution 4 mm, FOV: 255 mm x 320 mm). For faster measurement, a partial Fourier was used sampling 5/8 of the k-space asymmetrically in phase-encoding direction, i.e., roughly 60% of the k lines, resulting in 88 actually acquired k lines. Using this setup, we achieve acquisition times of 200 ms per slice. The acquisition setup was chosen to mimic an interventional setup as closely as possible. This specifically means high acquisition speed and just good enough contrast to detect the respiratory motion. No body array coil (surface array coil comprised of multiple elements) was used. Only the bore fixed receiver coil was used, which makes this 4D MRI method compatible with a wide range of external surrogates, including those that need a free line of sight to the abdomen of the subject. This includes, but is not limited to, surrogates based on a scan of the abdomen’s surface or marker tracking on the abdomen. This is important to make the gathered motion information available for a wide range of interventional scenarios where different surrogates may be used to track breathing. A total of 19 data sets of 13 healthy subjects were acquired. One subject was imaged three times, four subjects were imaged twice, and eight subjects were imaged once. If a subject was imaged multiple times, then each data set acquisition was performed on different days to include variations that occur in between imaging sessions. Each data set consists of two reference sequences and several interleaved sequences. Both will be described in the following.

A reference sequence is a dynamic 2D MRI sequence of so-called navigator frames. A schematic depiction can be found in Fig 1. The navigator frames picture an image plane, in which the respiratory motion is visible. In our case, we used a slice in the sagittal orientation that intersects the target organ—the liver—and shows vessel cross-sections, because their spatial distribution describes the breathing state well. This sequence is the reference for the 4D reconstruction. The reference contains a natural succession of different breathing patterns, like shallow or deep, thoracic or abdominal breathing, and is thus physiologically and profoundly meaningful. One reference sequence was acquired at the beginning and one at the end of each session. A reference sequence comprises 513 images (time points) covering a time of 102 seconds (about 20 breathing cycles).

Fig 1

Schematic depiction of a reference sequence.

A reference sequence shows a physiologically meaningful breathing curve and consists only of navigator frames that were imaged at the same slice position.

Each interleaved sequence consists of equal parts of data frames and navigator frames (between 150 and 200 each), see Fig 2. The former are sorted into the 4D MRI sequences based on information extracted from the latter. Data slices and navigator slices were imaged alternatingly, facilitating the interleaved character of the sequence. The navigator slices are positioned exactly as in the reference sequence, rendering temporal reconstruction possible. The data slice sweeps over the target organ in 4 mm gaps during acquisition (see Fig 3), rendering spatial reconstruction possible. For each slice position of the reconstructed volume one interleaved sequences is acquired. The total number of interleaved sequences per subject ranges between 38 and 57 (mean = 46.68), depending on the size of the subjects’ target organ to capture its whole volume. Thus, the total acquisition time for a subject ranged between 40 min and 80 min, excluding time for imaging localizers, determining navigator position and setting up the interleaved sequences. The total acquisition time is the time it took to capture all MRI images necessary for 4D MRI reconstruction, i.e., reference sequences and interleaved sequences. In the use case this acquisition would be made during planning before the actual intervention. The imaging of localizers, determining the navigator position and setting up the interleaved sequences took roughly 15 min per subject.

Fig 2

Schematic depiction of an interleaved sequence.

An interleaved sequence consists of navigator frames and data frames that were imaged alternatingly. It shows a different breathing curve than the navigator sequence but contains similar breathing patterns.

Fig 3

Schematic depiction of slice positions capturing the target volume.

Slices are in sagittal orientation. The position of the navigator slice is the same for all sequences per subject. The slice positions for the data frames are distinct and correspond to different interleaved sequences from the 1’st to the N’th. Interleaved sequences are acquired from right to left.

The acquisition time can be halved when using matching criteria that do not depend on a navigator slice. The total acquisition time can be further reduced by optimizing the acquisition scheme, allowing in-time breathing instructions for the subject for more efficient use of the acquisition time. During the intervention itself, only a surrogate, e.g., a navigator frame, has to be acquired in real-time as a query to the reconstructed 4D MRI or to a breathing model that was derived from the 4D MRI. All acquired MRI sequences used for 4D reconstruction, and a detailed acquisition protocol are publicly available [19].

4D MRI reconstruction

Our method and the baseline method use the reference sequence as grounds for the temporal reconstruction of a 4D MRI sequence showing a physiologically meaningful course of breathing states. The general scheme of the reconstruction process is depicted in Fig 4. For each time point in the reference sequence, i.e., for each frame, a volume is reconstructed. First, the breathing state of the frame is determined. Second, in each interleaved sequence, all data frames are found that match the breathing state, using a matching criterion, see Fig 5. Third, the found frames are averaged (binned) to one slice to improve the SNR (signal-to-noise ratio). Fourth, the averaged slice is inserted (sorted) into the volume at its designated position, which is known and unique for each interleaved sequence. Doing this for all reference frames results in a continuous 4D MRI sequence. The reconstructed FOV’s range from 255 mm x 320 mm x 152 mm to 228 mm (140 x 176 x 38 to 57 voxels) depending on the size of the target organ. In the next section, the matching criterion is described in detail.

Fig 4

Scheme of 4D MRI reconstruction.

For each time point in the reference sequence, a volume is reconstructed. For that in each interleaved sequence, the data slices are found that match the breathing state. They are then averaged and sorted into a volume.

Fig 5

Scheme of finding data slices that match specific breathing state.

On the left hand, the reference sequence is depicted. The red bracket represents the third breathing state. It is found in the interleaved sequence, depicted on the right, by comparing the enclosing navigator slices.

Matching criterion

A matching criterion is used to find all data slices showing the reference breathing state within an interleaved sequence. The respiratory state of a frame is determined by its enclosing navigator frames. Hence, the matching criterion acts on pairs of navigator frames that encase another frame (navigator or data frame); see brackets in Fig 5. It is based on the displacement of tracked vessels within the navigator frames. Assume a navigator frame at time point t in the reference sequence that shows a reference breathing state BS. We want to find a data frame with the same breathing state as . To this end, the enclosing navigator frames of both and are used. The enclosing navigator frames of are and and the enclosing frames of are and . The vessel displacements from to and from to are calculated. When the sum of all vessel displacements for two pairs of navigator frames is under a certain threshold, then the two enclosed frames are assumed to be a match, i.e., to show the same breathing state. The threshold is the only parameter of the method. It determines the maximally allowed displacements for two frames to be counted as a match.

The vessel tracking is realized via template matching using OpenCV [20] and its similarity measure TM_CCOEFF_NORMED (see Eq 1). where

Here T′ is the template T minus its mean pixel intensity, and I′ is an image patch with the same size as the template. Its pixel values are also shifted by minus the patches mean pixel intensity. w and h are the width and height of the template and the patch.

R is the resulting image of the template matching. Each entry R(x, y) contains the similarity value of the template to the source image at position (x, y).

The templates are manually defined for each tracked vessel cross-section in the reference sequence. To this end, a user identifies trackable vessels in one slice of the reference sequence prior to the 4D reconstruction, which takes only a few seconds. In our case, trackable means that the vessel cross-section or cluster of cross-sections will be visible in most navigator frames throughout the whole navigator sequence and that the cross-section has a high contrast to the surrounding tissue as well as a high signal to noise ratio. This is mostly not the case for small cross-sections but true for larger ones.

Template updates and search region

One of the challenges for the template matching is the out-of-plane motion of the vessel cross-sections in the navigator frames. In these cases, the searched-for regions are changing their appearance throughout breathing; hence, they are difficult to find using fixed templates.

To increase robustness against the out-of-plane motion, we propose to apply template updates within the reference sequence. In Fig 6, one can see how the appearance of a vessel cross-section can change during a breathing cycle. The method starts with the templates that were defined manually on reference frame . Then, for each following navigator frame that was captured at time point t, the templates get automatically updated, as follows: The positions of all tracked vessels in are found with subpixel precision using the templates from time point t. Then a new set of templates is cut from based on the position of the matched templates. The template position is updated with floating-point precision. The updates ensure that changes in the appearance of the tracked vessel are represented in the updated templates. The subpixel precision in the updates is needed to avoid drift during the update.

Fig 6

Out-of-plane motion and template updates.

The figure shows a series of navigator slices. The green rectangle denotes a typical ROI that was manually determined as a template. In the red rectangles, one can see how the vessel cross-section changes its appearance during the breathing cycle. For viewing purposes only, the images gradation curve was altered globally to enhance contrast.

Another concern of the reconstruction approach is speed. In its original form, the method matches each template against each navigator frame, resulting in a substantial computational burden. We propose to speed up the vessel tracking by exploiting spatial coherence between temporally adjacent navigator frames. The underlying assumption is that the next searched-for match is in a small spatial neighborhood around the previously found match, which is the case due to fast and continuous acquisition. Therefore, we only use a small neighborhood around the last matched template position as a search area.

Moreover, we automatically detect breathing states that cannot be reconstructed entirely and use that knowledge to inform where (temporally and spatially) the 4D sequence is incomplete. This information is essential for the later application, because of the visual feedback that can be provided to the physician in real-time when the motion information is insufficient to fuse the planning data to the interventional data.

Evaluation

We compare our method with the baseline method of Siebenthal et al. through reconstruction rate and image quality. We define the reconstruction rate as the percentage of the number of slices in the volume that could be reconstructed by the method. Note that this does not account for false positives or false negatives because the ground truth is not available to us. We also investigate how the acquisition order of the reference sequence and interleaved sequence influences the method’s ability to find matching data frames. We evaluate the point of false positives indirectly using a qualitative assessment of both approaches. The image quality is assessed in a double-blind study with interventional radiologists.

Reconstruction rate

We compare the reconstruction rate of both methods for different parameterizations. This is possible because the baseline method uses the same parameters in its matching criterion. When a subject was imaged multiple times, the reconstruction rates of its respective data sets were averaged for the statistical analysis to avoid possible biases. We tested the parameters shown in Table 1. We tested the threshold, for the values 0.5, 1, and 2. Evaluating different thresholds from a quantitative point-of-view allows us to judge which method will be more suitable for different applications that differ in the kind of trade-off between precision and coverage that is preferable in the application. With lower (stricter) thresholds, the coverage goes down and the precision increases. With higher thresholds, the coverage increases and the precision decreases. We tested two similarity measures from OpenCV, namely TM_CCOEFF_NORMED (see Eq 1) and TM_CCORR_NORMED (see Eq 3), and we tested the influence of the chosen reference sequence, ref. 1 and ref. 2, where ref. 1 is acquired before and ref. 2 is acquired after the interleaved sequences. where T is the template, I is the image and R is the resulting image with the highest intensity in position (x, y), where the similarity was the highest.

Table 1

Tested parameter values.

Parameter	Value
Threshold	0.5; 1; 2
Similarity measure	TM_CCORR_NORMED; TM_CCOEFF_NORMED
Reference Sequence	ref. 1; ref. 2

A four-factorial analysis of variance (ANOVA) was conducted to test for the effects of the aforementioned factors on the reconstruction rate.

Reconstruction quality

We conducted a double-blind study with ten interventional radiologists to compare the reconstruction quality of both methods and to evaluate whether our method’s reconstruction quality improves over the baseline. Participants were recruited from a General Radiology clinic. Their professional experience ranged from 4 months to 20 years (median: 16 months, mean: 62 months).

The interviews were in no way invasive, and no data that would allow for participant identification was included in the analysis. Thus, IRB approval was not requested for the interviews. In all cases oral participation consent was obtained and recorded.

Each radiologist was shown a set of 48 slice image pairs. The images of a pair were reconstructed from the same subject and breathing state showing the same anatomical structure and having the same slice position and orientation. One slice in a pair was sampled from a reconstruction of the baseline method. The other was sampled from a reconstruction of our method. Slices of a reconstructed volume are depicted in Fig 7. The radiologists had to decide which of the images in a pair shows the anatomy of the target organ more faithfully, i.e., with fewer image artifacts. Participants did not see the two slices from each pair simultaneously but could switch back and forth between them as often as they wanted before picking one. Participants were asked to select the slice they considered better. A neutral option was provided. For the evaluation of reconstruction quality, the parameter set was chosen to be 1 px threshold and TM_CCOEFF_NORMED as a similarity measure for both methods. Only those volumes were considered for comparison, for which both methods had a reconstruction rate of at least 80%. For each radiologist, 48 volume pairs were chosen randomly.

Fig 7

Axial, coronal and sagittal slices of a reconstructed volume.

The images gradation curve was altered globally to enhance contrast for better viewing only. In the axial and coronal orientation, one can see that our method is capable of reconstructing smooth and continuous volumes from sagittal slices.

Furthermore, in both volumes, we automatically masked slices out (setting intensity values to black), where either of the methods did not find a matching data frame. We made both volumes identical in the amount and distribution of black slices. This was done because it is likely that a reduced reconstruction rate for a volume would be detrimental to its perceived reconstruction quality. Each slice pair was sampled at a random orientation and position chosen within a range, such that the sampled slice would show the target organ. Slices were sampled either in sagittal, coronal, or axial orientation. Due to a software error, the number of slices for different planes was slightly imbalanced: Overall, 100 slices were shown for the sagittal and axial orientation each, and 280 slices were shown for the coronal orientation. For each of the 480 image pairs shown to participants, we recorded which method was preferred, if either. For data analysis, the two methods were appointed one ‘point’ each for every time they had been preferred. For each neutral vote, both methods were appointed a half ‘point’. This led to a dichotomous variable that allows for a direct comparison of the two methods’ scores. A one-sided binomial test was conducted (H0: p ≤ 0.5, H1: p > 0.5).

Results

Table 2 shows the mean reconstruction rates for all parameter combinations. Our method has a consistently higher reconstruction rate than the baseline (about twice as high) for all parameter sets. Figs 8 and 9 show the respective distribution of reconstruction rates. All underlying reconstruction rates per reconstructed 4D MRI and all tested parameters are provided in S1 File.

Table 2

Mean reconstruction rates [%] of our method and baseline.

Reconstruction rates are given in percent reconstructed of a volume. Bold is the best rates for each parameter set.

		TM_CCORR_NORMED			TM_CCOEFF_NORMED
threshold		2px	1px	0.5px	2px	1px	0.5px
ref. 1	baseline	24.58	15.95	9.94	41.78	24.10	12.74
	our method	73.60	40.99	23.24	77.69	47.10	27
ref. 2	baseline	46.86	31.95	18.75	60.09	40.07	22.92
	our method	79.67	56.89	36.78	82.18	58.53	37.34
avrg.	baseline	35.72	23.95	14.34	50.93	32.08	17.83
	our method	76.63	48.94	30.01	79.93	52.82	32.17

Fig 8

Reconstruction rates for reference sequence one.

Fig 9

Reconstruction rates for reference sequence two.

The four-factorial ANOVA showed significant main effects for all four factors and one significant interaction effect for the reconstruction method and the threshold used (Table 3). This interaction effect describes that while our method performs better than the baseline method at all threshold levels, it achieves more significant improvements at higher thresholds (see also Figs 8 and 9).

Table 3

Main results of the ANOVA on the reconstruction rate.

Effect type	Factor	df	F	p
Main effects	Reconstruction method	1	134.99	<0.001
	Threshold	2	106.56	<0.001
	Similarity measure	1	8.33	0.004
	Reference sequence	1	37.40	<0.001
Interaction effect	Reconstruction method * Threshold	2	7.71	<0.001
	Rec method * Similarity measure	1	1.95	0.164
	Rec method * Reference sequence	1	1.41	0.236

On the tested data, it was also more robust against the chosen similarity measure used for the template matching and also more robust against whether the reference sequence was acquired in the beginning or at the end of the session. Though, these interaction effects could not be shown to be significant in the ANOVA.

A correlation between acquisition order of the slice positions relative to the reference sequence and the ability of the methods to reconstruct these slice positions can be seen in Fig 10. With the increasing temporal distance between the acquisition of an interleaved sequence and the reference sequence, both methods find fewer similar slices for the corresponding slice position. Reference sequence one (red graphs) is acquired before the interleaved sequences. Here both methods find more slices for the earlier slice positions. Reference sequence two (blue graphs) is acquired after all interleaved sequences. Here both methods find more slices for the later slice positions.

Fig 10

Correlation of slice position and number of slice matches.

Red graphs represent the average number of slice matches for the first reference sequence (averaged over all subjects). Blue graphs correspond likewise to the second reference sequence. Graphs with squares represent our method; graphs with crosses represent the baseline method. Error bars represent standard deviation and are scaled by 0.1 for better readability.

The mean reconstruction time of our method is 24.19 seconds, with a standard deviation of 6.82 seconds. The mean reconstruction time of the baseline is 73 seconds, with a standard deviation of 21.81 seconds.

In the double-blind study, overall, participants selected our method in 156 trials, the baseline method in 111 trials, and had no preference in 213 trials (see Fig 11). Following our analysis method, this yielded 262.5 ‘points’ for our method and 217.5 ‘points’ for the baseline method (p = 0.02). All acquired data of the study is provided in S2 File.

Fig 11

Participant choice.

The bars represent the number of times each option was chosen out of 480 trials.

The study shows that radiologists perceive the reconstruction quality of our method as significantly better than the baseline method, although the effect seems to be small.

Discussion and conclusion

The particular acquisition scheme shows difficulties with changes in breathing patterns that arise over a more extended period, like the typical flattening of the resting breath. Slice positions to the left are imaged only at the end of acquisition time, whereas slices to the right are only imaged at the beginning. As a consequence, if the reference sequence was captured in the beginning, it can show breathing states that do not occur later, when slice positions to the left are imaged. Deep breaths often can not be fully reconstructed since image data of the left slice positions was not acquired for deep breathing states. Generally speaking, the scheme has difficulties with breathing states that are less frequent. This problem can be solved in changing the acquisition scheme. Instead of first acquiring all slices in one position before moving on to the next slice position, it is beneficial to move the slice position after each acquisition while keeping the navigator position fixed. This rotating acquisition scheme could also be combined with intermediate reference sequences. This would directly counter the problem with flattening breath over time. Furthermore, with the new scheme, it is feasible to give a few commands, so the subject can take a few more deep breaths in the beginning before starting to relax more.

The rotating acquisition scheme was used by Siebenthal et al. on a 1.5T Philips Intera whole-body MRI system [14]. However, Siemens MRI machines do not allow this kind of scheme. A solution to the problem that is independent of the scanner used is to use external respiratory signals instead of navigator frames. Preiswerk et al. [21] had correlated 1D MR compatible ultrasound with 2D and multiplanar MRI. This allows for the continuous rotating acquisition of the data slices on any MRI machine. Celicanin et al. [22] propose a simultaneous multislice (SMS) imaging technique that allows for the simultaneous acquisition of navigator and data frames, increasing the temporal coherence of navigator and data frame. Barth et al. [23] give a current overview of parallel imaging and SMS imaging techniques. These would integrate well with the rotational acquisition scheme when using body array coils. No body array coil is used in our experiment to ensure a line of sight for external marker tracking. However, when external marker tracking is not needed, a body array coil can readily be used in conjunction with our method to have better image contrast and possible faster imaging with aforementioned SMS techniques applied. When flat, flexible array coils with an opening for operation become available, those benefits, i.e, higher SNR, faster acquisition and line of sight, could be combined.

Regarding the acquisition time, the aforementioned changes to the acquisition scheme would half the acquisition time in our case to between 20 and 40 min.

Regarding the reconstruction rate, because of the lack of ground truth, it is not possible to account for false negatives and false positives in the evaluation. Based on this fact, the reconstruction rate of both methods will possibly be higher than measured in this study. This is because, in our test data, the number of reconstructable slice positions in each volume is lower than the number of slices in a volume, resulting from the acquisition scheme mentioned above.

An open issue arises when vessel cross-sections in the navigator frame are not continually visible. This frequently happens to depend on blood flow. To solve this, one could detect outliers in the template matching step and omit those for the calculation of the summed displacement.

We decided to use MRI data of healthy volunteers for the development and evaluation of the method. For a proof of concept of our method, this eliminates possible adverse effects of liver diseases on the respiration of the patient, making the evaluation environment more controlled. However, in future work, it has to be evaluated if typical diseases targeted by this method, like liver carcinoma, affect the method. This could be especially the case if the disease impairs the respiration of the patient. If the patient’s breathing shows no or few repetitions of patterns, this would be a challenge for the method because whilst allowing for irregular breathing, it assumes that patterns are recurring over time.

In its presented form, our method relies on a manual step in which the ROIs around the vessel cross-sections are defined. In a real clinical setting, this is intended to be done offline after the planning MRI session and before the date of the intervention on a suitable computer, not directly on the MRI machine. Even though this manual interaction is minimal and takes less than a minute to perform, it could and should be automated in future work. This could be solved as a classification problem in image space using the temporal information of the reference sequence as supporting information.

In our evaluation of the visual reconstruction quality, we only compare our method relative to the baseline. The provided neutral option does not differentiate between equally good and equally bad or unusable, and no absolute data was gathered. Hence, our analysis does not show whether the reconstructions are good enough for a given task or not. The analysis only indicates that our method’s reconstruction quality improves over the baseline.

In summary, our results clearly show that template updates are an effective and efficient means to increase reconstruction rates and image quality of the reconstruction result for template-based 4D MRI reconstruction methods. We reported that employing search regions significantly reduces reconstruction time. The results suggest that our method is preferable compared to the baseline. This is regardless of the application’s favorable trade-off between precision and coverage because, in all cases, reconstruction rates are higher than the baseline.

Reconstruction rates.

Reconstruction results of the experiments for all 4D MRI reconstructions and tested parameters.

(CSV)

Click here for additional data file.

Study results.

Participants choices in the image quality study.

(CSV)

Click here for additional data file.

(PDF)

Click here for additional data file.

11 Sep 2019

PONE-D-19-18726

4D MRI: Robust sorting of free breathing MRI slices for use in interventional settings

PLOS ONE

Dear Mr. Gulamhussene,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Additionally, please be aware of PLOS ONE policy for making the published research as reproducible as possible, in particular for what it concerns the availability of data. In order for your research to be published please elaborate on the reasons you are not allowed to share data (even anonymized) and/or the criteria upon which researchers can be granted access.

We would appreciate receiving your revised manuscript by Oct 26 2019 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Enrico Grisan

Academic Editor

PLOS ONE

Journal Requirements:

When submitting your revision, we need you to address these additional requirements.

1. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2.  We note that you have indicated that data from this study are available upon request. PLOS only allows data to be available upon request if there are legal or ethical restrictions on sharing data publicly. For more information on unacceptable data access restrictions, please see http://journals.plos.org/plosone/s/data-availability#loc-unacceptable-data-access-restrictions.

In your revised cover letter, please address the following prompts:

a) If there are ethical or legal restrictions on sharing a de-identified data set, please explain them in detail (e.g., data contain potentially sensitive information, data are owned by a third-party organization, etc.) and who has imposed them (e.g., an ethics committee). Please also provide contact information for a data access committee, ethics committee, or other institutional body to which data requests may be sent.

b) If there are no restrictions, please upload the minimal anonymized data set necessary to replicate your study findings as either Supporting Information files or to a stable, public repository and provide us with the relevant URLs, DOIs, or accession numbers. For a list of acceptable repositories, please see http://journals.plos.org/plosone/s/data-availability#loc-recommended-repositories.

We will update your Data Availability statement on your behalf to reflect the information you provide.

3.  Thank you for stating the following in the Financial Disclosure section:

'This study was supported by funding from the Investitionsbank Sachsen-Anhalt (https://www.ib-sachsen-anhalt.de/) to GG with the grant number 1704/00038. The funder had no role in the study design, data collection and analysis, decision to publish, or preparation of this manuscript.'

We note that you received funding from a commercial source: Investitionsbank Sachsen-Anhalt

Please provide an amended Competing Interests Statement that explicitly states this commercial funder, along with any other relevant declarations relating to employment, consultancy, patents, products in development, marketed products, etc.

Within this Competing Interests Statement, please confirm that this does not alter your adherence to all PLOS ONE policies on sharing data and materials by including the following statement: "This does not alter our adherence to PLOS ONE policies on sharing data and materials.” (as detailed online in our guide for authors http://journals.plos.org/plosone/s/competing-interests).  If there are restrictions on sharing of data and/or materials, please state these. Please note that we cannot proceed with consideration of your article until this information has been declared.

Please include your amended Competing Interests Statement within your cover letter. We will change the online submission form on your behalf.

Please know it is PLOS ONE policy for corresponding authors to declare, on behalf of all authors, all potential competing interests for the purposes of transparency. PLOS defines a competing interest as anything that interferes with, or could reasonably be perceived as interfering with, the full and objective presentation, peer review, editorial decision-making, or publication of research or non-research articles submitted to one of the journals. Competing interests can be financial or non-financial, professional, or personal. Competing interests can arise in relationship to an organization or another person. Please follow this link to our website for more details on competing interests: http://journals.plos.org/plosone/s/competing-interests

Additional Editor Comments (if provided):

Dear Authors,

the reviewers' have found scientific merit on your work and we are requesting a minor revision to address the point raised in their comments.

However, PLOS-ONE request the data to be made available, and it is not clear how anonymized MRI data can pose ethical challenges (just thinking about the many repositories targeting variuos diseases). In order to allow the publications, you should provide strong reasons why MRI data can not be shared and clarify the criteria for access.

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: No

Reviewer #2: No

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The paper presents a method for the reconstruction of 4D MRI sequences accounting for many states of respiratory motion. It finds that the proposed method, compared to a baseline, reconstructs sequences more often, more quickly, and to a higher perceived quality. Overall, while the method is a fairly incremental improvement over existing work, the paper is clearly written and the experiments well conceived.

I think there are a few minor issues with the paper which should be corrected in revisions.

- The two similarity measures used are referenced by their names in OpenCV - TM_CCORR_NORMED and TM_CCOEFF_NORMED.These measures should be described (mathematically) in the text. They are a key part of the method and discussed at length, but a reader who is unfamiliar with OpenCV software would have no idea what they are. Their exact implementations in OpenCV may even change in the future.

- There are a number of grammatical / spelling errors in the text which should be corrected. For example,

-- the sentence beginning on line 5 should read "Our application scenario[s] are ... where the challenge of a moving target exist[s] ... but none satisfy all [the] needs ..." (corrections in [])

-- the sentence beginning on line 108 seems to have had some words cut out of it

-- the caption for Figure 6 misspells 'contrast'

- Regarding the data availability, it is reasonable to withhold the actual MRI image data - the journal suggests though that the numerical scores used to generate the summary statistics in, for example, tables 2 and 3 could be provided.

Reviewer #2: The authors describe a method for reconstructing four-dimensional volumes (three spatial + one temporal) from multiple 2D acquisitions. Navigator images are interleaved with imaging frames to provide a method for assigning 2D slices to the 4D volume. The work is built upon a method from Siebenthal et al and describes a minor improvement by constraining the search space for the navigator tracking. The data are not publicly available but the authors do note it is available upon reasonable request.

Please find minor comments below:

1). Please contextualize the imaging time (40 to 80 minutes) for interventional imaging. For non-interventionists, this seems like a very long acquisition time.

2). Please comment on why only healthy volunteers were scanned and speculate on what limitations the method may face in the presence of irregular breathing that is more likely in a patient population.

3). Please briefly describe TM CCOEFF NORMED and TM CCORR NORMED

4). Please comment on the limitations of manual steps such as ROI selection in the scanner environment. Could such steps be automated?

5). Please provide reconstruction times in the body of the paper.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

30 Oct 2019

We thank PLOS ONE’s academic Editor Enrico Grisan and the anonymous reviewers for their

comments. We have attempted to address all remarks in the revised version of the manuscript.

Below we give detailed responses to all reviewers comments and describe the corresponding

changes in our manuscript.

Enrico Grisan Academic Editor:

01. Please be aware of PLOS ONE policy for making the published research as reproducible as

possible, in particular for what it concerns the availability of data. In order for your research

to be published, please elaborate on the reasons you are not allowed to share data (even

anonymized) and/or the criteria upon which researchers can be granted access.:

We submit the anonymized data set underlying the computed scores as supporting

information in a csv file with the revised submission. We are furthermore providing

the anonymized MRI image data through the open repository of the university library

of the Otto-von-Guericke-University Magdeburg under the creative commons license

CC-BY-SA (DOI: 10.24352/UB.OVGU-2019-093; https://doi.org/10.24352/UB.OVGU-2019-

093)

Journal Requirements:

02. Please ensure that your manuscript meets PLOS ONE's style requirements, including those

for file naming.:

We made sure to follow PLOS ONE's style requirements. In particular, we made sure

that figure references follow the style guides when referring to more than one figure

at once and made changes to the manuscript accordingly. We made changes so that

figures are cited in order of appearance. We checked that all images adhere to the file

naming rules of PLOS ONE.

03. We note that you have indicated that data from this study are available upon request.

PLOS only allows data to be available upon request if there are legal or ethical restrictions

on sharing data publicly.:

Please see our answer to point 01.

04. We note that you received funding from a commercial source: Investitionsbank Sachsen-

Anhalt:

We believe there is a misunderstanding. The Investitionsbank Sachsen-Anhalt is an

institute of the State of Saxony-Anhalt that has only administrative functions to

distribute public funding. It is not a commercial funder.

Reviewer 1

05. The two similarity measures used are referenced by their names in OpenCV -

TM_CCORR_NORMED and TM_CCOEFF_NORMED.These measures should be described

(mathematically) in the text. They are a key part of the method and discussed at length,

but a reader who is unfamiliar with OpenCV software would have no idea what they are.

Their exact implementations in OpenCV may even change in the future.:

Thank you for pointing that out. We added mathematical formulas and explanations

in the manuscript (cf. page 5 lines 147-153 and page 7 lines 210-212) to make the

used similarity measures more transparent to readers.

06. There are a number of grammatical / spelling errors in the text which should be corrected.

For example:

a. the sentence beginning on line 5 should read "Our application scenario[s] are ...

where the challenge of a moving target exist[s] ... but none satisfy all [the] needs

..." (corrections in []):

Thank you for pointing that out. We corrected the sentence.

b. the sentence beginning on line 108 seems to have had some words cut out of it

Thank you for pointing that out.: We corrected the sentence.

c. the caption for Figure 6 misspells 'contrast'

Thank you for pointing that out.: We corrected the sentence.

We also have had the revised manuscript proofread by a native speaker.

Regarding the data availability, it is reasonable to withhold the actual MRI image data - the

journal suggests though that the numerical scores used to generate the summary statistics

in, for example, tables 2 and 3 could be provided.:

Thank you for that point. Please see our answer to point 01.

Reviewer 2

07. Please contextualize the imaging time (40 to 80 minutes) for interventional imaging. For

non-interventionists, this seems like a very long acquisition time.:

Thank you for pointing that out. We made changes to the manuscript (cf. page 4 lines

105 - 118) to better point out how the total acquisition time is to understand. See

lines 107 to 122 in the revised manuscript with marked-up changes.

08. Please comment on why only healthy volunteers were scanned and speculate on what

limitations the method may face in the presence of irregular breathing that is more likely in

a patient population.:

Thank you for pointing that out. We changed the manuscript (cf. page 10 lines 322 -

330) to address this question. See lines 335 to 343 in the revised manuscript with

marked-up changes.

09. Please briefly describe TM CCOEFF NORMED and TM CCORR NORMED:

Thank you for pointing that out. Please see our answer to point 05.

10. Please comment on the limitations of manual steps such as ROI selection in the scanner

environment. Could such steps be automated?:

Thank you for pointing that out. We changed the manuscript (cf. page 10 lines 331 –

338) to address the limitations of manual steps in our method. See lines 344 to 351 in

the revised manuscript with marked-up changes.

11. Please provide reconstruction times in the body of the paper.:

Thank you for pointing that out. We changed the manuscript (cf. page 9 lines 272 -

274) to provide mean reconstruction times of both methods as well as the respective

standard deviations. See lines 284 to 286 in the revised manuscript with marked-up

changes.

Submitted filename: Response to Reviewers.pdf

Click here for additional data file.

6 Apr 2020

PONE-D-19-18726R1

4D MRI: Robust sorting of free breathing MRI slices for use in interventional settings

PLOS ONE

Dear Mr. Gulamhussene,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

Both reviewers are enthusiastic for the manuscript. However, there are few minor comments to be addressed. Please respond all carefully.

We would appreciate receiving your revised manuscript by May 21 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Haydar Celik, PhD

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

Reviewer #3: Partly

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

Reviewer #3: No

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: (No Response)

Reviewer #3: The manuscript titled “4D MRI: Robust sorting of free breathing MRI slices for use in interventional settings” presents a modification of a previously published technique for respiratory motion extraction and 4D image volume generation as pre-procedural imaging for MR-guided interventions. The manuscript is well written, and the results presented in a clear manner. Nevertheless, the manuscript could benefit from direct and succinct clarification as to what are the differences between “their method” and the “reference method” since their method builds upon the reference one. The Methods section clarifies the method, but it is not clear where the original ends and where the new one begins. This should be crystal clear for any reader without having to read the entirety of the work.

General Comments

1) Treatments of individual subjects: The authors acquire multiple datasets from some subjects (“One subject was imaged three times, four subjects were imaged twice…”). These repeated acquisitions are treated as independent measurements and it appears there is no distinction when calculating metrics and averages or in any of the statistical analysis. Yet, these samples are not statistically independent measures. It is clear that in MRI some volunteers naturally provide excellent images and others are simply less photogenic (usually move more in the scanner). Hence, acquiring images from one subject 3 times and treating those values as independent can bias the measurement. Any data from a single subject should be averaged/combined into a single point before statistical consideration (e.g. ANOVA, binomial test). Similarly, when constructing the cases for review by the 10 radiologists, the number of slices included from a given volunteer should be equal. Ultimately, in most of these analyses there are only 13 subjects, not 19.

2) Matched pairs and image quality: Please clarify in the text if the “slice pairs” shown to radiologists for evaluation where of the same slice on the same subject. This is implied in the text but not made explicit. If it is the case, then the analysis made sense. If it isn’t then the comparison analysis does not make sense at all and is probably the incorrect approach. Also, please clarify why there is no pathway for the radiologist to grade images in terms of quality? T

The current analysis allows an observer to pick one or the other image (or both), but it does not assess how many times the reconstruction produced images that were inadequate for interventional guidance either due to incoherent data or residual image artifact. That result is simply hidden by the observer choosing the alternative (especially since there is no “neither” option). In other words, though proposed technique may receive more “points” that the reference technique, if it fails to reconstruct an adequate image within the 4D volume, then it may not be worth pursuing. To claim superiority, the authors need to demonstrate that the images that are reconstructed are not junk any more than those reconstructed in the original method.

3) Scan time: The acquisition time for the proposed method is clearly an issue. I recommend the authors discuss potential techniques for speeding up acquisition. I would suggest that the authors should include the use of phased array coils which would allow for the use of parallel imaging for in plane acceleration, or even more aptly suited, the use of SMS (simultaneous multislice imaging) which is ideal for this approach. I have yet to see a single interventional MRI procedure in which phased array coils were not used during pre-procedural imaging and I believe that in an effort to increase breadth of applicability by relying only on the body coil for signal, the authors are hurting their actual applicability due prohibitively long scan times.

Specific Comments

Page 3/13, Methods, first paragraph: “TR = 40.0 ms” is manufacturer specific term since no modern TRUFI sequence can run with this TR successfully (especially at 3T). The parameters that should be reported in this section but aren’t include: (1) real repetition time (TR) which in Siemens-speak is referred to as echo spacing (likely 2*TE), (2) the actual number of ky-lines acquired within the matrix of 140 phase encodes and descriptions of whether partial Fourier sampling in ky (or assymetric echo in kx) is applied, (3) the readout bandwidth, (4) the min,max and average number of slices acquired per subject (assume 38-57 but average should be reported). In the current description, if 140 lines of k-space were acquired in 200 ms (ignoring start/stop subsequences), then each line would take 1.43 ms, not a feasible number given the matrix size. Hence, the authors need to specify the acquisition more correctly to ensure reproducibility for the reader to understand the current acquisition strategy since it has great impact on temporal resolution.

Page 3/13, Methods, first paragraph: “… body coil…” Clarifying terminology about body coils as there may be some confusion here. The term “body coil” is used to describe the large-bore fixed birdcage coil used for homogeneous RF transmission, rarely used for imaging in diagnostic MRI due to extremely low SNR. A body array (a coil composed of multiple elements) represents the standard surface coils arrays with multiple receive channels that are standard in diagnostic MRI. It is not clear which was used for imaging here, though by the description the authors used no body arrays and used only the body coil for imaging. If this is correct, please clarify the terminology.

Page 4/13: “localisators” is not MR terminology or standard English. The typically used term is "localizer(s)" or "localiser(s)".

Page 6/13: The phrase "well trackable" is awkward in English. Maybe "easily trackable" or "robustly trackable"? Also, since the term is inherently subjective, it may be helpful to display a figure/panel with several patterns that do meet the criteria and maybe even some patterns that do not meet the desired criteria. This is central to the paper as in the end, the authors are using cross correlation on a vessel pattern to estimate respiratory state.

Figure 10: There should be error bars for each point? If my understanding is correct, each point represents averages across all subjects?

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

Reviewer #3: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

20 May 2020

Reviewer 3 02.: The Methods section clarifies the method, but it is not clear where the original ends and where the new one begins. This should be crystal clear for any reader without having to read the entirety of the work.

> You are right, we were not absolutely clear regarding where the baseline method ends and our method starts. We made changes to the manuscript to clarify which sections describe the general concept and common aspects that our method shares with the baseline method. We also made clear which sections describe the parts that differentiate our method from the baseline. (cf. page 3 lines 71 – 74 in the revised manuscript with track changes)

Reviewer 3 03.: Any data from a single subject should be averaged/combined into a single point before statistical consideration (e.g. ANOVA, binomial test). Similarly, when constructing the cases for review by the 10 radiologists, the number of slices included from a given volunteer should be equal.

> Thank you for pointing out the possibility of a bias in our analysis. We made changes to the statistical analysis. We now average data points of the same subject and made changes to the manuscript to reflect that point. After the change, all reported main and interaction effects are present. (cf. page 7 lines 224 – 225 and page 11 table and page 12 table, in the revised manuscript with track changes) Please note that due to the annotation the table does not fit on the page. Please also refer to the manuscript without tracked changes.

Reviewer 3 04.: Please clarify in the text if the “slice pairs” shown to radiologists for evaluation where of the same slice on the same subject.

> Thank you for pointing that out. We made changes to the manuscript to clarify that slices of a pair were of the same subject, position and orientation. (cf. page 8 lines 250 – 254 in the revised manuscript with track changes)

Reviewer 3 05.: Also, please clarify why there is no pathway for the radiologist to grade images in terms of quality? The current analysis allows an observer to pick one or the other image (or both), but it does not assess how many times the reconstruction produced images that were inadequate for interventional guidance either due to incoherent data or residual image artifact. That result is simply hidden by the observer choosing the alternative (especially since there is no “neither” option). In other words, though proposed technique may receive more “points” that the reference technique, if it fails to reconstruct an adequate image within the 4D volume, then it may not be worth pursuing. To claim superiority, the authors need to demonstrate that the images that are reconstructed are not junk any more than those reconstructed in the original method.

> We understand that an absolute scale contains more information and is more potent than a mere comparison, which is a valid point. We made changes to the manuscript to make clear that we aim to evaluate whether our methods additions improve the reconstruction quality over the baseline showing that this is a valid path. We also made it clear that from our analysis we cannot and do not derive a conclusion whether the reconstruction quality is good enough for any given use case. (cf. page 8 lines 242 – 243 and page 11 lines 385 – 390 in the revised manuscript with track changes)

Reviewer 3 06.: Scan time: The acquisition time for the proposed method is clearly an issue. I recommend the authors discuss potential techniques for speeding up acquisition. I would suggest that the authors should include the use of phased array coils which would allow for the use of parallel imaging for in plane acceleration, or even more aptly suited, the use of SMS (simultaneous multislice imaging) which is ideal for this approach. I have yet to see a single interventional MRI procedure in which phased array coils were not used during pre-procedural imaging and I believe that in an effort to increase breadth of applicability by relying only on the body coil for signal, the authors are hurting their actual applicability due prohibitively long scan times.

> You noticed that phased array coils allow for SMS with faster imaging and better SNR. We agree that this in an improvement and have made changes to the manuscript to discuss the possibilities and how our method would integrate with that. (cf. page 10 lines 337 – 352 in the revised manuscript with track changes)

Reviewer 3 07.: Page 3/13, Methods, first paragraph: “TR = 40.0 ms” is manufacturer specific term since no modern TRUFI sequence can run with this TR successfully (especially at 3T). The parameters that should be reported in this section but aren’t include: (1) real repetition time (TR) which in Siemens-speak is referred to as echo spacing (likely 2*TE), (2) the actual number of ky-lines acquired within the matrix of 140 phase encodes and descriptions of whether partial Fourier sampling in ky (or assymetric echo in kx) is applied, (3) the readout bandwidth, (4) the min,max and average number of slices acquired per subject (assume 38-57 but average should be reported). In the current description, if 140 lines of k-space were acquired in 200 ms (ignoring start/stop subsequences), then each line would take 1.43 ms, not a feasible number given the matrix size. Hence, the authors need to specify the acquisition more correctly to ensure reproducibility for the reader to understand the current acquisition strategy since it has great impact on temporal resolution.

> Thank you for pointing that out. We made changes to the manuscript to add this information. (cf. page 3 lines 77 – 83, page 4 line 120 in the revised manuscript with track changes)

Reviewer 3 08.: Page 3/13, Methods, first paragraph: “… body coil…” Clarifying terminology about body coils as there may be some confusion here. The term “body coil” is used to describe the large-bore fixed birdcage coil used for homogeneous RF transmission, rarely used for imaging in diagnostic MRI due to extremely low SNR. A body array (a coil composed of multiple elements) represents the standard surface coils arrays with multiple receive channels that are standard in diagnostic MRI. It is not clear which was used for imaging here, though by the description the authors used no body arrays and used only the body coil for imaging. If this is correct, please clarify the terminology.

> Thank you for pointing that out. We made changes to the manuscript to clarify the terminology. (cf. page 3 lines 87 – 88 in the revised manuscript with track changes)

Reviewer 3 09.: Page 4/13: “localisators” is not MR terminology or standard English. The typically used term is "localizer(s)" or "localiser(s)".

> Thank you for pointing that out. We corrected this in the manuscript (cf. page 4 lines 122 and 126 in the revised manuscript with track changes)

Reviewer 3 10.: Page 6/13: The phrase "well trackable" is awkward in English. Maybe "easily trackable" or "robustly trackable"? Also, since the term is inherently subjective, it may be helpful to display a figure/panel with several patterns that do meet the criteria and maybe even some patterns that do not meet the desired criteria. This is central to the paper as in the end, the authors are using cross correlation on a vessel pattern to estimate respiratory state.

> Thank you for pointing that out. We made changes to the manuscript to clarify that point. (cf. page 6 lines 175 – 181 in the revised manuscript with track changes)

Reviewer 3 11.: Figure 10: There should be error bars for each point? If my understanding is correct, each point represents averages across all subjects?

> Thank you for this point. We added error bars to represent the standard deviation and made changes to the manuscript to reflect that point. (cf. page 10 Fig. 10 in the revised manuscript)

10 Jun 2020

4D MRI: Robust sorting of free breathing MRI slices for use in interventional settings

PONE-D-19-18726R2

Dear Dr. Gulamhussene,

We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.

Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication.

An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at http://www.editorialmanager.com/pone/, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to-date. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

Kind regards,

Haydar Celik, PhD

Academic Editor

PLOS ONE

Additional Editor Comments (optional):

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #2: All comments have been addressed

Reviewer #3: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #2: Yes

Reviewer #3: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #2: Yes

Reviewer #3: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #2: Yes

Reviewer #3: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #2: Yes

Reviewer #3: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #2: I believe the authors have addressed all of the concerns of the reviewers in their revised manuscript.

Reviewer #3: The majority of my comments have been addressed appropriately.

The majority of my comments have been addressed appropriately.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #2: No

Reviewer #3: No

12 Jun 2020

PONE-D-19-18726R2

4D MRI: Robust sorting of free breathing MRI slices for use in interventional settings

Dear Dr. Gulamhussene:

I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

If we can help with anything else, please email us at plosone@plos.org.

Thank you for submitting your work to PLOS ONE and supporting open access.

Kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Dr. Haydar Celik

Academic Editor

PLOS ONE