Echocardiographic Detection of Regional Wall Motion Abnormalities Using Artificial Intelligence Compared to Human Readers.

BACKGROUND: Although regional wall motion abnormality (RWMA) detection is foundational to transthoracic echocardiography, current methods are prone to interobserver variability. We aimed to develop a deep learning (DL) model for RWMA assessment and compare it to expert and novice readers. METHODS: We used 15,746 transthoracic echocardiography studies-including 25,529 apical videos-which were split into training, validation, and test datasets. A convolutional neural network was trained and validated using apical 2-, 3-, and 4-chamber videos to predict the presence of RWMA in 7 regions defined by coronary perfusion territories, using the ground truth derived from clinical transthoracic echocardiography reports. Within the test cohort, DL model accuracy was compared to 6 expert and 3 novice readers using F1 score evaluation, with the ground truth of RWMA defined by expert readers. Significance between the DL model and novices was assessed using the permutation test. RESULTS: Within the test cohort, the DL model accurately identified any RWMA with an area under the curve of 0.96 (0.92-0.98). The mean F1 scores of the experts and the DL model were numerically similar for 6 of 7 regions: anterior (86 vs 84), anterolateral (80 vs 74), inferolateral (83 vs 87), inferoseptal (86 vs 86), apical (88 vs 87), inferior (79 vs 81), and any RWMA (90 vs 94), respectively, while in the anteroseptal region, the F1 score of the DL model was lower than the experts (75 vs 89). Using F1 scores, the DL model outperformed both novices 1 (P = .002) and 2 (P = .02) for the detection of any RWMA. CONCLUSIONS: Deep learning provides accurate detection of RWMA, which was comparable to experts and outperformed a majority of novices. Deep learning may improve the efficiency of RWMA assessment and serve as a teaching tool for novices.

CFD- and Bernoulli-based pressure drop estimates: A comparison using patient anatomies from heart and aortic valve segmentation of CT images.

PURPOSE: An aortic valve stenosis is an abnormal narrowing of the aortic valve (AV). It impedes blood flow and is often quantified by the geometric orifice area of the AV (AVA) and the pressure drop (PD). Using the Bernoulli equation, a relation between the PD and the effective orifice area (EOA) represented by the area of the vena contracta (VC) downstream of the AV can be derived. We investigate the relation between the AVA and the EOA using patient anatomies derived from cardiac computed tomography (CT) angiography images and computational fluid dynamic (CFD) simulations. METHODS: We developed a shape-constrained deformable model for segmenting the AV, the ascending aorta (AA), and the left ventricle (LV) in cardiac CT images. In particular, we designed a structured AV mesh model, trained the model on CT scans, and integrated it with an available model for heart segmentation. The planimetric AVA was determined from the cross-sectional slice with minimum AV opening area. In addition, the AVA was determined as the nonobstructed area along the AV axis by projecting the AV leaflet rims on a plane perpendicular to the AV axis. The flow rate was derived from the LV volume change. Steady-state CFD simulations were performed on the patient anatomies resulting from segmentation. RESULTS: Heart and valve segmentation was used to retrospectively analyze 22 cardiac CT angiography image sequences of patients with noncalcified and (partially) severely calcified tricuspid AVs. Resulting AVAs were in the range of 1-4.5 cm2 and ejection fractions (EFs) between 20 and 75%. AVA values computed by projection were smaller than those computed by planimetry, and both were strongly correlated (R2 = 0.995). EOA values computed via the Bernoulli equation from CFD-based PD results were strongly correlated with both AVA values (R2 = 0.97). EOA values were ∼10% smaller than planimetric AVA values. For EOA values < 2.0 cm2 , the EOA was up to ∼15% larger than the projected AVA. CONCLUSIONS: The presented segmentation algorithm allowed to construct detailed AV models for 22 patient cases. Because of the crown-like 3D structure of the AV, the planimetric AVA is larger than the projected AVA formed by the free edges of the AV leaflets. The AVA formed by the free edges of the AV leaflets was smaller than the EOA for EOA values <2.0cm2. This contradiction with respect to previous studies that reported the EOA to be always smaller or equal to the geometric AVA is explained by the more detailed AV models used within this study.

Objective quantification of aortic valvular structures by cardiac computed tomography angiography in patients considered for transcatheter aortic valve implantation.

PURPOSE: To test the ability of a model-based segmentation of the aortic root for consistent assessment of aortic valve structures in patients considered for transcatheter aortic valve implantation (TAVI) who underwent 256-slice cardiac computed tomography (CT). METHODS: Consecutive patients (n = 49) with symptomatic severe aortic stenosis considered for TAVI and patients without aortic stenosis (n = 17) underwent cardiac CT. Images were evaluated by two independent observers who measured the diameter of the aortic annulus and its distance to both coronary ostia (1) manually and (2) software-assisted. All acquired measures were compared with each other and to (3) fully automatic quantification. RESULTS: High correlations were observed for 3D measures of the aortic annulus conducted on multiple oblique planes (r = 0.87 and 0.84 between observers and model-based measures, and r = 0.81 between observers). Reproducibility was further improved by software-assisted versus manual assessment for all the acquired variables (r = 0.98 versus 0.81 for annulus diameter, r = 0.94 versus 0.85 for distance to the left coronary ostium, P < 0.01 for both). Thus, using software-assisted measurements very low limits of agreement were observed for the annulus diameter (95%CI of -1.2 to 0.6 mm) and within very low time-spent (0.6 ± 0.1 min for software-assisted versus 1.6 ± 0.3 min per patient for manual assessment, P < 0.001). Assessment of the aortic annulus using the 3D model-based instead of manual 2D-coronal measurements would have modified the implantation strategy in 12 of 49 patients (25%) with aortic stenosis. Four of 12 patients with potentially modified implantation strategy yielded postprocedural moderate paravalvular regurgitation, which may have been avoided by implantation of a larger prosthesis, as suggested by automatic 3D measures. CONCLUSION: Our study highlights the usefulness of software-assisted preprocedural assessment of the aortic annulus in patients considered for TAVI.