MitraClip Device Automated Localization in 3D Transesophageal Echocardiography via Deep Learning Journal: ToBeDefined
Riccardo Munafò, Simone Saitta, Luca Vicentini, Davide Tondi, Veronica Ruozzi, Francesco Sturla, Giacomo Ingallina, Andrea Guidotti, Eustachio Agricola, Emiliano Votta · 2024
Read MitraClip Device Automated Localization in 3D Transesophageal Echocardiography via Deep Learning Journal: ToBeDefined by Riccardo Munafò, Simone Saitta, Luca Vicentini, Davide Tondi, Veronica Ruozzi, Francesco Sturla, Giacomo Ingallina, Andrea Guidotti, Eustachio Agricola, Emiliano Votta (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-15013.txt.
Introduction
Riccardo Munafò and co-authors published this 2024 paper on finding a heart device in ultrasound scans. The device, called a MitraClip, treats a leaking heart valve and is guided by live 3D scans of the heart. Noise and low contrast in those scans make the small clip hard to see.
The team built an automated pipeline that segments the clip and classifies its state, from fully closed to fully open. A computer model of the clip is then fitted to refine the outline and measure the device. Trained on 196 scans from a heart simulator, the system located the clip with sub-millimeter average error. Better views of the clip could help surgeons place it with more care.
Catalog record
| CREATOR | Riccardo Munafò, Simone Saitta, Luca Vicentini, Davide Tondi, Veronica Ruozzi, Francesco Sturla, Giacomo Ingallina, Andrea Guidotti, Eustachio Agricola, Emiliano Votta |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 11 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-15013
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
MitraClip Device Automated Localization in 3D Transesophageal Echocardiography via Deep Learning Journal: ToBeDefined Riccardo Munafò, Simone Saitta, Luca Vicentini, Davide Tondi, Veronica Ruozzi, Francesco Sturla, Giacomo Ingallina, Andrea Guidotti, Eustachio Agricola, Emiliano Votta Abstract The MitraClip is the most widely percutaneous treatment for mitral regurgitation, typically performed under the real-time guidance of 3D transesophagel echocardiography (TEE). However, artifacts and low image contrast in echocardiography hinder accurate clip visualization. This study presents an automated pipeline for clip detection from 3D TEE images. An Attention UNet was employed to segment the device, while a DenseNet classifier predicted its configuration among ten possible states, ranging from fully closed to fully open. Based on the predicted configuration, a template model derived from computer-aided design (CAD) was automatically registered to refine the segmentation and enable quantitative characterization of the device. The pipeline was trained and validated on 196 3D TEE images acquired using a heart simulator, with ground-truth annotations refined through CAD-based templates. The Attention UNet achieved an average surface distance of 0.76 mm and 95% Hausdorff distance of 2.44 mm for segmentation, while the DenseNet achieved an average weighted F1-score of 0.75 for classification. Post-refinement, segmentation accuracy improved, with average surface distance and 95% Hausdorff distance reduced to 0.75 mm and 2.05 mm, respectively. This pipeline enhanced clip visualization, providing fast and accurate detection with quantitative feedback, potentially improving procedural efficiency and reducing adverse outcomes. 1 Introduction Transcatheter edge-to-edge repair (TEER)is the most widespread percutaneous treatment for mitral regurgitation (MR) [ Vahanian_2021 ] . It offers a safe and effective alternative for patients with contraindications for surgery or those at high operative risk [ Feldman_2015 , Buzzati_2019 ] . The MitraClip (Abbott Laboratories, California, USA) is a catheter-based technology for mitral valve (MV) TEER, designed to treat MRby clipping together the mitral leaflets. The procedure involves accessing the left atrium (LA)through the interatrial septum with a steerable sheath. A delivery catheter is then advanced in the sheath and steered through the LAto reach the MVregion where lack of coaptation is observed. Upon reaching this target region, a clip, located on the tip of the delivery catheter, is actuated to grasp the MVleaflets and it is finally deployed. To achieve optimal grasping, the clip should be oriented perpendicularly to the MV, with its arms locally orthogonal to the coaptation line to be restored. These steps, from transseptal puncture through the steering of the delivery catheter toward MVand to clip positioning, are performed under the guidance of 2D and 3D transesophageal echocardiography (TEE) [ Sherif_2017 ] . 3D TEEprovides an en face view of the MVand of the approaching clip, allowing clinicians to determine when the clip is adequately positioned and oriented with respect to the target region of the MVleaflets [ Sherif_2017 ] . 3D TEEimages can be non-trivial to interpret, since they are displayed on 2D monitors, with a loss of depth perception. Thus, clinicians often rely on 2D views for clip positioning, which requires adjusting the TEEtransducer to align the clip in 2D TEEviews or extracting specific slices from the 3D volume for better visualization on the fly [ Nyman_2018 ] . However, both actions are time-consuming and operator-dependent, and require specific expertise. Moreover, the presence of the catheter and of the clip generates artifacts that hamper the assessment of clip configuration, in particular when it is close to native tissues. To address these challenges, an automated, real-time clip detection system for 3D TEEcould provide significant benefits. Such a system would enable the automatic extraction of relevant 2D views encompassing the device, and, when integrated with automated MVanalysis [ Munafo_2024a ] , facilitate precise quantification of the spatial relationship between the clip and its target region. This would enhance procedural guidance by providing standardized, operator-independent visualization. 1.1 Related works Despite the clinical importance of intraoperative guidance, no prior study has tackled the automated segmentation of TEEfor characterizing the position and configuration of clip devices for TEER. Published studies have primarily focused on catheter localization in TEEimages, leveraging detection or segmentation techniques [ Yang_2023 ] . Detection methods aim to identify the general location or shape of the catheter by using representations such as bounding box (a rectangular region enclosing the catheter), axis (a central line running along the catheter’s length), or skeleton (a simplified line-based representation capturing the catheter’s structure). These methods typically rely on carefully designed filters or instrument templates to extract catheter-related information from the image [ Yang_2023 ] . On the other hand, segmentation involves classifying each voxel associated with the catheter to reconstruct the latter within the image. Convolutional neural networks (CNNs)are the state-of-the-art tools for automatic segmentation of 3D echocardiography images, and recent studies have explored their application to cardiac catheter localization. Yang et al. applied a 2D CNNarchitecture for catheter segmentation in 3D TEEby stacking together adjacent 2D axial slices [ Yang_2019 ] . To mitigate the possible loss of 3D semantic information after slicing, they subsequently introduced image patch extraction and multi-planar slicing [ Yang_2019b ] . Although the patch-based approach benefits from speed and low GPU memory usage, it may limit the network’s ability to capture the full contextual information of the image. To address this, Yang et al. adopted a hybrid loss function that integrates voxel-wise loss with contextual loss for 3D UNet training, encouraging the network to learn a better contextual representation [ Yang_2019c , Yang_2021 ] . Alternatively, full 3D image information has been leveraged by combining 3D encoder and projection layers for features dimension reduction along axial and lateral dimensions, achieving efficient catheter segmentation in 3D TEE [ Yang_2020 ] . Despite these advancements, automatic catheter segmentation still requires additional complex post-processing to extract meaningful clinical information, such as catheter position and orientation. Furthermore, CNNmethods for semantic segmentation rely on accurate ground truth (GT)annotations, which are difficult to obtain due to the noise and artifacts present in TEEimages when visualizing metallic catheters. This frequently leads to over-segmentation of the catheter, producing numerous false positives that negatively affect the localization [ Mastmeyer_2017 ] . Consequently, these approaches have been applied to detect intracardiac catheters with simple shapes, such as ablation catheters or guide wires that are typically characterized by a straight configuration during the procedure. 1.2 Main contribution In this work, we present the first automated approach for detecting, localizing, and characterizing the clip on the tip of the MitraClip delivery catheter. For clarity, we will refer to the MitraClip device, specifically the component involved in the grasping of MVleaflets during the procedure, simply as the clip. Unlike previous efforts that focused solely on catheter segmentation, our method is designed to accurately extract the clip’s spatial features, which are essential for intraoperative guidance during TEER. We introduced a novel deep learning-based framework for clip detection in volumetric echocardiography data. Beyond simple segmentation, our approach enabled a comprehensive characterization of clip orientation and positioning, providing key parameters to support real-time surgical decision-making. Our method reduces the reliance on manual slice selection and expert interpretation, allowing for standardized and efficient intraoperative visualization. By addressing these challenges, our approach has the potential to enhance procedural accuracy, reduce operative time, and improve overall patient outcomes in TEERinterventions. 2 Methods 2.1 Dataset Collection To implement and validate