A geometric feature tracking approach for noninvasive patient specific estimation of leaflet strain from 3D images of heart valves

TL;DR

This feature-tracking framework provides a generalizable method for noninvasive quantification of atrioventricular valve leaflet strain from clinical 3DE images and demonstrated greater accuracy in quantifying anatomical alignment and leaflet strain than conventional point-based approaches.

Abstract

Valvular heart disease is prevalent and a major contributor to heart failure. Valve leaflet strain is a promising metric for evaluating the mechanics underlying the initiation and progression of valvular pathology. However, generalizable methods for noninvasively quantifying valvular strain from clinically acquired patient images remain limited. To address this limitation, we developed a geometric feature-tracking framework to quantify in vivo leaflet strain from 3DE images. The method integrates a cohort-derived geometric reference atlas to establish geometric correspondence and introduces a novel distance-weighted coherent point drift algorithm for non-rigid registration. We evaluated performance against a finite element benchmark model and compared the approach with conventional point-based tracking methods. The framework was applied to pediatric and adult patient datasets (N = 31) across variable valve morphologies. The proposed method demonstrated greater accuracy in quantifying anatomical alignment and leaflet strain than conventional point-based approaches. Validation against the finite element benchmark confirmed improved strain estimation. The framework achieved reliable inter-phase tracking of valve deformation across diverse morphologies in pediatric and adult patients. Analysis identified a consistent distribution pattern of the 1st principal strain associated with leaflet billow (prolapse). This feature-tracking framework provides a generalizable method for noninvasive quantification of atrioventricular valve leaflet strain from clinical 3DE images. Characterization of biomechanical strain patterns may improve prognostic assessment and support longitudinal evaluation of valvular heart disease. Further investigation of the biomechanical signatures of heart valve disease has the potential to enhance prognostic assessment and longitudinal evaluation of valvular heart disease.

Figures (6)

• Figure 1: Feature tracking-informed strain quantification framework. (A) The atrial surfaces of the valves were extracted from segmentation models and triangulated with 1000 points uniformly distributed across each leaflet. (B) Within each patient cohort, one diastolic valve was randomly selected as the reference. Leaflets of the same anatomical category were registered to the corresponding reference leaflet to establish point-to-point correspondence. The registered leaflets were then aligned, and the mean shape (reference atlas) was estimated using generalized procrustes analysis. The mean shape was registered to both the diastolic and systolic frames of that leaflet and leaflet deformation was derived to compute strain on individual leaflets. (C) For patient cohort analysis, mean valve shapes at diastole and systole were first estimated using generalized procrustes analysis. Principal component analysis was applied to determine shape variation within each cohort. Finally, valves were registered between the open and closed phases to estimate systolic valve strain.
• Figure 2: Feature tracking verification results. (A) Ground truth areal and 1$^\text{st}$ principal strain of an image-derived MV finite element model were used to compare the performance of the proposed and established feature-tracking methods. (B) Qualitative comparisons of the deformed geometry and strain distributions are provided. Both CPD and DW-CPD successfully captured the large valve deformation and the strain patterns. (C) DW-CPD yielded more accurate estimates of mean symmetric distance, area strain MAE, and 1$^{\text{st}}$ principal strain MAE compared to existing CPD approaches. (D) The discrepancy in strain estimation using DW-CPD is most noticeable in the commissural folds. (E) Scatter plots show strong correlations between areal and1$^{\text{st}}$ principal strain generated from Dw-CPD and FEA.
• Figure 3: Feature tracking performance. (A) This example illustrates that CPD resulted in self-penetration at the leaflet edge, while the proposed DW-CPD eliminated this defect with improved feature tracking fidelity. (B) DW-CPD was applied to 3 groups of atrioventricular valves and verified against ground truth segmentation. Strong shape agreement was observed, with MSD values below 0.4 mm and HD$_{95}$ values below 1 mm across all leaflets.
• Figure 4: Pediatric tricuspid valve analysis (N=11). (A) The estimated systolic atrial surfaces were superimposed onto 3D TEE images along with the ground truth segmentation models. Excellent feature tracking accuracy was obtained, as demonstrated in the visualization of the three cut planes shown. (B) The study cohorts included patients spanning a wide age range, from 5 days to 17 years. For each patient, the surgical stage at the time of image acquisition, as well as the pathological characteristics of the TVs, were reported. (C) The distributions of 1$^{\text{st}}$ principal strain, as well as their median and IQRs, were compared across the three leaflets in all 11 patients. Billowing leaflets were typically characterized by a high median and strain IQR, with a relatively uniform strain distribution profile. (D) Average areal strain and 1$^{\text{st}}$ principal strain were compared between the trivial to mild and mild or greater TR cohorts. The trivial to mild cohort exhibited lower areal and 1$^{\text{st}}$ principal strain in the anterior leaflet but higher values in the posterior leaflet relative to the mild or greater cohort. Both cohorts showed near zero areal strain in the septal leaflet, while the mild or greater cohort demonstrated higher 1$^\text{st}$ principal strain in the septal leaflet. (E) Population-based strain analysis was performed. 1$^\text{st}$ principal strain and IQRs increase as the valve morphology deviates from the mean shape.
• Figure 5: Pediatric mitral valve analysis (N=10). (A) Visualization of the ground truth segmentation models and estimated atrial surfaces were superimposed onto 3D TEE images on the three cut planes shown. (B) The study cohorts included patients from 2 to 16 years old at the time of image acquisition. The MV pathological characteristics for each patient were provided. (C) The median, IQR, and distribution of 1$^{\text{st}}$ principal strain were compared across the anterior and posterior leaflets in all 10 patients. Patient 10 was the only case with leaflet strain exhibiting a median and IQR greater than 0.5, which also corresponded to billowing morphology. (D) Average areal strain and 1$^{\text{st}}$ principal strain were compared between the normal and diseased cohorts. Both cohorts showed negligible changes in leaflet surface area and demonstrated higher average 1$^{\text{st}}$ principal strain in the posterior leaflet than in the anterior leaflet. (E) Population-based strain analysis showed that higher average 1$^{\text{st}}$ principal strains at both ends of the shape spectrum compared with the mean shape. In addition, strain SD increased as valve morphology deviated from the mean.