Real‐time automatic image‐based slice tracking of gadolinium‐filled balloon wedge catheter during MR‐guided cardiac catheterization: A proof‐of‐concept study

Abstract Purpose MR‐guided cardiac catheterization procedures currently use passive tracking approaches to follow a gadolinium‐filled catheter balloon during catheter navigation. This requires frequent manual tracking and repositioning of the imaging slice during navigation. In this study, a novel framework for automatic real‐time catheter tracking during MR‐guided cardiac catheterization is presented. Methods The proposed framework includes two imaging modes (Calibration and Runtime). The sequence starts in Calibration mode, in which the 3D catheter coordinates are determined using a stack of 10–20 contiguous saturated slices combined with real‐time image processing. The sequence then automatically switches to Runtime mode, where three contiguous slices (acquired with partial saturation), initially centered on the catheter balloon using the Calibration feedback, are acquired continuously. The 3D catheter balloon coordinates are estimated in real time from each Runtime slice stack using image processing. Each Runtime stack is repositioned to maintain the catheter balloon in the central slice based on the prior Runtime feedback. The sequence switches back to Calibration mode if the catheter is not detected. This framework was evaluated in a heart phantom and 3 patients undergoing MR‐guided cardiac catheterization. Catheter detection accuracy and rate of catheter visibility were evaluated. Results The automatic detection accuracy for the catheter balloon during the Calibration/Runtime mode was 100%/95% in phantom and 100%/97 ± 3% in patients. During Runtime, the catheter was visible in 82% and 98 ± 2% of the real‐time measurements in the phantom and patients, respectively. Conclusion The proposed framework enabled real‐time continuous automatic tracking of a gadolinium‐filled catheter balloon during MR‐guided cardiac catheterization.


Conclusion:
The proposed framework enabled real-time continuous automatic tracking of a gadolinium-filled catheter balloon during MR-guided cardiac catheterization.

K E Y W O R D S
cardiac catheterization, image processing, MR guidance, partial saturation, passive tracking, real time

INTRODUCTION
2][3] During these procedures, catheters are navigated through the cardiovascular system and used for interventional purposes (such as balloon dilation) or for diagnostic purposes (pressure measurements).Guidewires can also be used to facilitate placement of the catheter.][11][12][14][15][16]22 These approaches commonly use balloon wedge catheters, which can be filled with CO 2 for negative contrast visualization (hypo-intense signal) or diluted Gadolinium (Gd) for positive contrast visualization (hyperintense signal) in the images.Gd-filled balloon wedge catheters tend to be more conspicuous and faster to navigate compared with air-filled balloons. 9Various magnetization preparation schemes have been used for improved visualization of the Gd-filled catheter balloon.A saturation prepulse that can be turned on to visualize the balloon and turned off to visualize the soft tissues and blood was proposed. 9Black blood preparation with flow-sensitive gradients was used for simultaneous visualization of the catheter balloon and soft tissues. 23A saturation prepulse with a reduced saturation angle to achieve partial saturation (pSAT) preparation was proposed for simultaneous high-contrast visualization of the catheter balloon, soft tissues, and blood. 15his technique offered excellent visualization capabilities during MR-guided cardiac catheterization in CHD patients. 16l passive tracking approaches that are used clinically require frequent manual slice tracking and manipulation of the imaging plane to follow the balloon during catheter navigation.This is usually achieved by the interventionist using foot pedals or via the scanner console.In a recent study, it was reported that the catheter fell out of plane in more than 30% of real-time measurement frames during navigation. 16This diverts the focus of the operator from the catheter navigation and prolongs the intervention, reducing the value of image guidance.
A T 1 overlay method in which brighter signals from T 1 -weighted images were overlaid to 3D multiplanar reconstruction views increased the catheter visualization time to 90%. 6 This study used a thick 20-mm slice, which could partly explain the reduced out-of-plane time of the catheter; however, this could also potentially reduce the value of the images in narrower and complex anatomies.Furthermore, physiological motion and any deformation induced by the interventional device would not be captured using a 3D visualization from a static volume, which could result in the misregistration of the real-time balloon signal relative to the underlying anatomical roadmap.
In this study, we present a novel framework that enables real-time image-based tracking of the catheter balloon and automatic repositioning of the imaging slice (i.e., slice tracking) for continuous high-contrast visualization of the balloon and anatomy during catheter navigation.We demonstrate the proposed technique in a phantom and subsequently present its feasibility in patients undergoing MR-guided cardiac catheterization.Part of this work has previously been presented as conference proceedings. 24,25

Proposed framework
The proposed prototype sequence consists of two imaging modes: Calibration and Runtime (Figure 1).To achieve optimal contrast between the catheter balloon and surrounding anatomy, all image acquisitions are preceded by nonselective pSAT and chemical shift-selective fat-suppression pulses.The sequence begins with the Calibration mode in which a fixed stack of contiguous slices (n = 10-20, slice thickness = 10 mm, pSAT = 90 0 ) is acquired in under 3 s.Real-time image processing of this slice stack (described in Section 2.2) is performed to identify the initial 3D coordinates of the balloon without any prior knowledge of its location.The sequence then automatically switches to the Runtime mode, where three contiguous slices in the orientation of interest (slice thickness = 10 mm, pSAT = 30 • -50 • , adjustable via the scanner console) are acquired continuously.A pSAT of 30 • -50 • has previously been shown to provide a good compromise between visualization of the cardiovascular anatomy and balloon/blood contrast. 15Initially, based on the 3D coordinates obtained from the Calibration mode, the first three Runtime slices are automatically positioned to intersect the balloon in the central slice.During Runtime, the 3D balloon position is continuously estimated from real-time image processing of the three slices (detailed in Section 2.2).If the balloon is detected in either of the outer slices, the three slices are automatically repositioned to ensure that the catheter is in the central slice (i.e., the Runtime stack is shifted by one slice thickness toward the outer slice containing the catheter balloon).Furthermore, the sequence automatically switches back to the Calibration mode if the balloon is lost for more than 3 s (i.e., > 5 real-time measurements, adjustable via the scanner console), such as when the catheter leaps beyond the three-slice through-plane range in the Runtime mode.The different feedback scenarios are shown in Figure 1.
The proposed sequence and postprocessing were developed within the manufacturer's programming environments to enable online real-time acquisition and tracking.Images were visualized using a standard commercially available inline display window on the scanner console.This screen was mirrored onto an LCD monitor located inside the scanner room to provide the same view to the interventionist.

Real-time image processing
Each stack of images acquired during the Calibration and Runtime modes is processed using the following steps, unless specified otherwise (the corresponding pseudo code is provided in Script S1 in Data S1).
1. First, the image stack is binarized using a signal intensity threshold to identify hyperintense pixels and generate a binary image stack.Two different thresholds are used for the Calibration and Runtime modes (90% and 40% of the maximum signal intensity of the stack, respectively, adjustable via the scanner console).This processing is restricted to the volume of interest corresponding to the prescribed shim box encompassing the entire cardiovascular system navigated during the intervention to eliminate any spurious signals from the edge of the FOV. 2. Morphological closing, which involves the processes of dilation followed by erosion (using a 3 × 3 square structuring element), is performed on this binary image stack to fill/merge any small holes into the background.3.All detected pixels are clustered into multiple contiguous regions.4. Any region with a large area (> disc area with an 8-mm radius) or a maximum distance to its center of mass > 8 mm is then discarded.Note that 8 mm was selected as an upper bound of the balloon radius and is adjustable via the scanner console. 5. Additionally, the catheter displacement is expected to show some temporal consistency between consecutives frames.To account for this, in the Runtime mode, regions that are considered too distant (using a maximum distance threshold empirically set to 14 mm, adjustable via the scanner console) from the previous catheter position are discarded.6.For each slice, if multiple regions remain, the region closest to the previous catheter location is selected in that slice.7. If multiple regions are identified across all slices, the region with the brightest signal is selected as the balloon and its 3D coordinates are computed.
The user-defined parameters used in this real-time image processing pipeline were optimized in a phantom experiment presented in Tables S1-S4 in Data S1.

Experimental evaluation
All imaging experiments were performed on a 1.5T MRI scanner (MAGNETOM Aera; Siemens Healthcare, Erlangen, Germany).The balloon of the wedge catheter (Arrow; Teleflex, Wayne, PA, USA) was filled with 1% Gd (Dotarem; Guerbet, Villepint, France) for positive contrast visualization in all the experiments.This study was approved by the local institutional review board (REC reference: 21/LO/0650 IRAS project ID: 304329).

Phantom experiment
The proposed framework was tested in a 3D-printed heart phantom, which was printed from the segmented anatomy of a healthy adult subject obtained using a high-resolution MRI scan.Both the Calibration and Runtime modes used a 2D single-shot acquisition with the same balanced SSFP readout and the following parameters: TR/TE = 2.44/0.99ms, flip angle (FA) = 50 • , FOV = 450 × 450 mm 2 , reconstructed resolution = 2.8 × 2.8 mm 2 , slice thickness = 10 mm, temporal resolution = 209.4ms, number of real-time measurements = 99, bandwidth = 1010 Hz/px, GRAPPA factor = 2, partial Fourier = 5/8.The Calibration stack was prescribed in the coronal orientation and remained fixed during the procedure.The Runtime slices were prescribed along the coronal orientation.The catheter was manipulated through the phantom during the entire acquisition.In two instances, the catheter was deliberately moved rapidly in the through-plane direction to force the loss of the balloon from the Runtime slices during navigation.

Evaluation in patients
The in vivo feasibility of the proposed approach was investigated in 3 CHD patients (10 ± 3 years, all male, weight = 31 ± 10 kg) undergoing MR-guided cardiac catheterization.The 2D single-shot balanced SSFP acquisition parameters were the same as those for the phantom experiment except for the following: the number of real-time measurements was 19, 29, and 59 for Patients 1, 2, and 3, respectively.The sequence was run during catheter manipulation.The Calibration stack was prescribed in the coronal orientation for fast screening of the cardiovascular system because the anterior/posterior usually represents the smaller body dimension and remained fixed during the procedure.The Runtime slices were prescribed in the sagittal direction in all 3 patients.

Analysis
The accuracy of automatic identification of the catheter balloon was assessed in the phantom and patients.We verified whether the computed catheter coordinates matched the actual location of the catheter balloon in the magnitude images.A correct catheter detection was defined as the center of mass of the estimated balloon region falling within the true region of the balloon.Accuracy was determined for both the Calibration and Runtime modes and expressed as a percentage of the total number of acquired real-time measurements.For the Runtime mode, we also examined the percentage of true positive (balloon present and correctly detected), true negative (balloon absent and not detected), false positive (balloon absent and detected), and false negative (balloon present and not detected) outcomes for catheter detection of the real-time image processing detailed earlier.The percentage of time the sequence was in the Calibration and Runtime modes was calculated.Furthermore, we computed the percentage of real-time measurements the catheter balloon was visible in the magnitude images in the Runtime mode during catheter navigation.negative outcomes were 77%, 18%, 0%, and 5%, respectively.During the experiment, the sequence was in the Runtime mode for 97% of the time, with the remaining time in the Calibration mode.The balloon was visible in the magnitude images in 82% of all real-time measurements during the Runtime mode.The remaining 18% of real-time measurements correspond to the part of the experiment when the balloon was briefly lost due to deliberate fast displacement of the catheter beyond the three-slice through-plane range.In all 3 patients, the balloon was automatically identified with 100% accuracy when the sequence was in the Calibration mode.During Runtime, the detection accuracy was 97 ± 3%.During the in vivo procedures, the sequence was in the Runtime mode for 95%, 97%, and 98% of the time for Patients 1-3, respectively, with the remaining corresponding times in the Calibration mode.The balloon was clearly visible in the magnitude images during Runtime in 100%, 96%, and 98% of the real-time measurements in Patients 1-3, respectively.Additionally, the SNR values for the balloon and blood, and contrast-to-noise ratio (CNR) for balloon/blood for the 3 patients, were SNR balloon = 81 ± 2, SNR blood = 28 ± 2, and CNR balloon/blood = 52 ± 3. The results for the phantom and patients are summarized in Table 1.

Feasibility in patients
The computation time of the proposed real-time postprocessing was 20 ms for a stack of three Runtime slices, and the latency, defined as the reconstruction time of the three Runtime slices (∼200 ms/slice), postprocessing, and feedback sent and received by the acquisition process, was 650 ms.

F I G U R E 3
Representative Runtime images demonstrating the proposed approach in Patient 1. Automatic slice tracking and repositioning (between real-time measurements 6 and 7 as well as 17 and 18) were performed when the balloon was detected in one of the outer slices.The white arrows indicate the location of the automatically identified balloon.

F I G U R E 4
Runtime slices illustrating the proposed approach in Patient 2. The three-slice stack was automatically adjusted to ensure balloon visibility and follow the catheter in the central slice.The white arrows show the detected location of the catheter balloon.

T A B L E 1
The accuracy of automatic detection of the catheter balloon and rate of catheter visibility in the phantom and patients, along with the SNR and contrast-to-noise ratio (CNR) values of the balloon/blood in vivo.a In the magnitude images (Runtime).b In two instances, the catheter was deliberately moved rapidly in the through-plane direction to force the loss of the balloon from the Runtime slices during navigation.

DISCUSSION
In this study, we developed a novel cardiac MRI sequence that enables automatic real-time tracking and visualization of a Gd-filled balloon during catheter navigation.The proposed acquisition and image processing framework incorporates (i) partial saturation with high spatial coverage, (ii) automatic image-based estimation of the catheter balloon position, and (iii) real-time slice following and repositioning.This framework was successfully demonstrated in a 3D-printed heart phantom, and its in vivo feasibility was shown in 3 patients.The sequence achieved automatic continuous visualization of the balloon with high detection accuracy, and improved visibility of the catheter was also achieved compared with existing approaches.Furthermore, the SNR and CNR values for the balloon/blood in vivo were found to be comparable with the values reported in a previous pSAT study in patients. 16he image processing of the slice stacks in both imaging modes was based on several user-defined parameters.Catheter detection was restricted to the volume of interest covered by the prescribed shim box, which was implemented to reduce the identification of false/spurious structures (especially fat signal) at the edges of the FOV.The prescription of a region of interest encompassing the entire cardiovascular system and excluding bright structures (such as fat signal from the chest and back) may be challenging in certain patients using a rectangular box.Nevertheless, fat signal areas from the chest and back have a different spatial pattern than the catheter balloon and may be robustly discarded using these spatial pattern constraints.The need for a more robust rejection algorithm, such as using a nonlinear region of interest around the targeted anatomy, requires further investigation.Furthermore, spatiotemporal constraints were added to restrain possible solutions based on the expected catheter movement within the imaging interval.These parameters were suitable for the described experiments but may require adjustments to ensure the robustness of the framework in a larger patient cohort.
The proposed sequence can facilitate the fast detection of out-of-plane catheters, which are time-consuming to locate when manually tracked and can prolong the procedure.When the catheter suddenly moved beyond the three-slice through-plane range in the Runtime mode, a controllable lost limit ensured an automatic switch back to Calibration mode for re-estimation of the balloon coordinates within the prescribed Calibration stack.However, it may be desirable to introduce the possibility for the operator to manually switch to the Calibration mode, such as by using a pedal.This could be useful in cases in which the operator judges that the catheter may come back in-plane shortly or if the sequence identifies a different structure as the catheter.Although the latter case was not observed in our study, this could in theory occasionally happen when navigating the catheter in the vicinity of vessels depicting bright in-flow artifacts or small fat areas that could be mistaken for the catheter.
The Runtime mode is based on the real-time sequential acquisition of three contiguous slices, resulting in reduced temporal resolution compared with single-slice imaging.This can be avoided by using a single slice only during Runtime at the cost of a more frequent switch back to the Calibration mode (as demonstrated in the Supporting Information and corresponding Videos S4 and S5) or using accelerated imaging schemes such as simultaneous multislice imaging. 26Achieving a higher framerate may also facilitate the acquisition of orthogonal orientations during the Runtime mode, which may provide useful additional anatomical context to the user. 10,16ditionally, the availability of simultaneous multiple views may potentially improve the performance of the catheter balloon detection and tracking.Furthermore, all three slices have the same slice thickness.The use of larger slice thicknesses combined with higher pSAT angles for the two outer slices could be investigated to further reduce the likelihood of losing the catheter during Runtime.
The proposed approach was evaluated at 1.5 T using a Gd-filled catheter balloon.Despite reduced conspicuity, the use of a gas-filled catheter balloon may have benefits for MRI-guided cardiac catheterizations due to improved buoyancy. 27The current postprocessing pipeline would need adjustment for tracking of hypo-intense signal in the images, which will be the focus of future work.Furthermore, the use of low-field MRI scanners may be attractive for this application due to larger bore sizes facilitating patient access and catheter manipulation particularly in small patients.It may also offer the possibility to use off-the-shelf devices with better mechanical properties than MR-conditional ones. 28Evaluation of the proposed framework at low field will also be investigated in the future.
This study has some limitations.First, the accuracy of the computed 3D balloon coordinates depends on the selection of several user-defined parameters in the postprocessing.Although most of these parameters were kept constant in our experiments, a parameter-free approach could further improve the robustness of the technique and facilitate its application in a clinical setting.The use of deep learning approaches may prove useful in this context, in which a neural network could be trained for image-based detection of the catheter balloon, which could improve the robustness of the proposed technique and reduce its dependency on user-defined parameters. 29Second, in this proof-of-concept study, the proposed framework was evaluated in only 3 patients.Further clinical evaluation in a larger patient cohort during an entire catheterization procedure and comparison with a standard approach is required to establish its advantages, especially for potentially reducing the out-of-plane time of the catheter and overall procedure time.

CONCLUSIONS
A novel framework was developed for real-time automatic catheter tracking during MR-guided cardiac catheterization.This technique enabled robust automatic tracking of the catheter during navigation while simultaneously providing high-contrast visualization of the catheter and cardiovascular system.

Figure 2
Figure 2 shows still frames of the phantom experiment.A movie of the experiment is shown in Video S1.The balloon was initially detected in the first Calibration stack of the acquisition (Slice 8).The 3D balloon coordinates obtained via feedback from the Calibration

Figures 3 and 4
Figures 3 and 4 along with Figures S1 and S2 in Data S1 show the application of the proposed sequence in Patients 1 and 2 undergoing MR-guided cardiac catheterization.The catheter balloon was detected in the Calibration stack (Slice 7 in Patient 1 and Slice 6 in Patient 2), followed by automatic centering of the Runtime slices on the balloon.Catheter tracking with slice repositioning during Runtime was then successfully achieved while the catheter was manipulated.Video S2 shows a movie of the in vivo tracking in Patient 1.When the balloon was identified in one of the outer slices, the Runtime slices were