Spatiotemporal Segmentation and Modeling of the Mitral Valve in Real-Time 3D Echocardiographic Images

Abstract

Transesophageal echocardiography is the primary imaging modality for preoperative assessment of mitral valves with ischemic mitral regurgitation (IMR). While there are well known echocardiographic insights into the 3D morphology of mitral valves with IMR, such as annular dilation and leaflet tethering, less is understood about how quantification of valve dynamics can inform surgical treatment of IMR or predict short-term recurrence of the disease. As a step towards filling this knowledge gap, we present a novel framework for 4D segmentation and geometric modeling of the mitral valve in real-time 3D echocardiography (rt-3DE). The framework integrates multi-atlas label fusion and template-based medial modeling to generate quantitatively descriptive models of valve dynamics. The novelty of this work is that temporal consistency in the rt-3DE segmentations is enforced during both the segmentation and modeling stages with the use of groupwise label fusion and Kalman filtering. The algorithm is evaluated on rt-3DE data series from 10 patients: five with normal mitral valve morphology and five with severe IMR. In these 10 data series that total 207 individual 3DE images, each 3DE segmentation is validated against manual tracing and temporal consistency between segmentations is demonstrated. The ultimate goal is to generate accurate and consistent representations of valve dynamics that can both visually and quantitatively provide insight into normal and pathological valve function.

A Deep Learning-Based Fully Automated Pipeline for Regurgitant Mitral Valve Anatomy Analysis From 3D Echocardiography

“…The average surface distance to the GT data for the full segmentation mask (0.43 ± 0.14 mm) was of the order of the spatial resolution of the TEE volumes in the dataset (0.30-0.70 mm/voxel). Furthermore, the reported distance was almost equal to the inter-user variability typical of manual segmentation, previously reported as 0.6 ± 0.17 mm [43], and compared favorably with previous semiautomatic (0.59 ± 0.49 mm [18] and 0.60 ± 0.20 mm [19]) and fully automated methods (1.54 ± 1.17 mm [20], 0.59 ± 0.23 mm [23] 0.925 ± 0.392 [24]). The MSD and 95% HD were observed to be slightly lower when computed for the complete mask as compared to each label separately.…”

“…Precise MV segmentation and identification of MV annulus and leaflets from 3DTEE would facilitate accurate and quantitative measurements of the regurgitant MV anatomy to support MR diagnosis and MV surgical or transcatheter repair [16]. In recent years, several works have introduced semi-[17]- [19] or fully automatic [20]- [24] methods to detect and segment MV structures. Many early methods proposed in the last decade are based on level set [17] or graph cut method [18], which require human-in-the-loop-interactions to work properly and reconstruct mitral leaflets as an inner surface lacking the preservation of leaflet thickness details.…”

“…In [20], a 3D geometrical model of MV was constructed by fitting key points identified by leveraging machine learning algorithms and an extensive collection of labelled images. In [19], a semiautomatic method to segment mitral leaflets was described using multi-atlas joint label fusion and a deformable model template. However, abnormal morphologies typical of pathological MV anatomy are often not accurately detected using a model fitting or atlas-based approach.…”

Multimodal image analysis and subvalvular dynamics in ischemic mitral regurgitation

“…In terms of clinical prospect, noninvasive multimodal image analysis using MRI for LV and subvalvular assessment and 3DE for the MV shows clinical promise. With the increased availability of fast automated algorithms, 26 , 27 the automatic fusion and analysis of both modalities may be possible. 28 …”

Three-Dimensional Transesophageal Echocardiography: Do We Have the Technology to Predict Outcomes in Ischemic Mitral Regurgitation?

“…6,7 Pouch et al have developed computational algorithms using segmentation analysis to automate 3D modeling of the mitral valve throughout the cardiac cycle. 8,9 Much of the current work being done in an effort to automate 3D echocardiographic image analysis involves the use of "deep learning" or "computer vision," which is a form of artificial intelligence. Technologic advances in this field combined with faster, more powerful, and more affordable processors, may make automated echocardiographic image analysis for realtime intraoperative decision-making a reality in the near future.…”