Arsanjani, Reza

Abstract Background Topological data analysis (TDA) is the state-of-the-art unsupervised machine learning framework that can provide insight into the dataset and visualize condensed information via the topological network graph. This robust approach was never used to assess the heterogeneous MitraClip population. Purpose We aim to develop a TDA model that will identify prognostically-distinct phenogroups in MitraClip patients without a priori knowledge of the population and their outcomes. Method Patients who underwent MitraClip (June 2014-September 2020) at Mayo Clinic sites were identified from the institutional database for baseline and follow-up data. Thirteen variables were used for TDA. The topological network graph was created using the Python Scikit-TDA Kepler-Mapper package (v. 2.0.1), and clustering was performed at the graph level with Louvain's modularity method. Kaplan-Meier survival analysis was used to assess the all-cause mortality endpoint of each cluster identified in an unsupervised manner. The dataset with cluster labels was also used to train a Light Gradient Boosted Machine model, and SHapley Additive exPlanations (SHAP) analysis was applied to determine the feature importance. Results A total of 389 consecutive patients were included in the final analysis and two major clusters consisting of 384 patients were identified. The mean age was 80.3±8.7 years; 256 (65.8%) were male. The mean Society of Thoracic Surgeons Mitral Valve Replacement risk score was 9.6±6.9%. Fifty-five (14.5%) patients died during the mean follow duration (185 days). Kaplan-Meier analysis showed significant survival differences among the two clusters (HR: 2.38; 95% CI: 1.39–4.06, p=0.001; Figure 1). Clusters 1 (n=195) was associated with > mild residual mitral regurgitation and worse survival performance and was characterized with worse tricuspid regurgitation severity, a higher proportion of patients with atrial fibrillation/flutter, anterior leaflet prolapse, and mitral annular/leaflet calcification, as summarized in Table 1. Conclusion TDA can identify distinct phenotype clusters with prognostic significance in MitraClip patients based on mitral valve morphology and clinical risk factors. This simple model can facilitate clinical risk stratification for MitraClip patients regarding procedural success and survival performance. Funding Acknowledgement Type of funding sources: None.

Aims Increased LV wall thickness is frequently encountered in transthoracic echocardiography (TTE). While accurate and early diagnosis is clinically important, given the differences in available therapeutic options and prognosis, an extensive workup is often required to establish the diagnosis. We propose the first echo-based, automated deep learning model with a fusion architecture to facilitate the evaluation and diagnosis of increased left ventricular (LV) wall thickness. Methods and Results Patients with an established diagnosis of increased LV wall thickness (hypertrophic cardiomyopathy (HCM), cardiac amyloidosis (CA), and hypertensive heart disease (HTN)/others) between 1/2015 to 11/2019 at Mayo Clinic Arizona were identified. The cohort was divided into 80%/10%/10% for training, validation, and testing sets, respectively. Six baseline TTE views were used to optimize a pre-trained InceptionResnetV2 model. Each model output was used to train a meta-learner under a fusion architecture. Model performance was assessed by multiclass area under the receiver operating characteristic curve (AUROC). A total of 586 patients were used for the final analysis (194 HCM, 201 CA, and 191 HTN/others). The mean age was 55.0 years, and 57.8% were male. Among the individual view-dependent models, the apical 4 chamber model had the best performance (AUROC: HCM: 0.94, CA: 0.73, and HTN/other: 0.87). The final fusion model outperformed all the view-dependent models (AUROC: CA: 0.90, HCM: 0.93, and HTN/other: 0.92). Conclusion The echo-based InceptionResnetV2 fusion model can accurately classify the main etiologies of increased LV wall thickness and can facilitate the process of diagnosis and workup.

In patients with medically refractory heart failure or cardiogenic shock, both temporary and durable mechanical circulatory support devices can be used to support cardiac circulation. Both transthoracic echocardiography (TTE) and transesophageal echocardiography (TEE) are widely available, relatively noninvasive, and avoid radiation exposure. Thus, echocardiography is an invaluable tool that provides vital information aiding in preprocedure evaluation, placement, management, and weaning of cardiac assist devices. The purpose of this article is to review the utility of both TTE and TEE in managing patients with cardiac assist devices.

The analysis and interpretation of cardiac magnetic resonance (CMR) images are often time-consuming. The automated segmentation of cardiac structures can reduce the time required for image analysis. Spatial similarities between different CMR image types were leveraged to jointly segment multiple sequences using a segmentation model termed a multi-image type UNet (MI-UNet). This model was developed from 72 exams (46% female, mean age 63 ± 11 years) performed on patients with hypertrophic cardiomyopathy. The MI-UNet for steady-state free precession (SSFP) images achieved a superior Dice similarity coefficient (DSC) of 0.92 ± 0.06 compared to 0.87 ± 0.08 for a single-image type UNet (p < 0.001). The MI-UNet for late gadolinium enhancement (LGE) images also had a superior DSC of 0.86 ± 0.11 compared to 0.78 ± 0.11 for a single-image type UNet (p = 0.001). The difference across image types was most evident for the left ventricular myocardium in SSFP images and for both the left ventricular cavity and the left ventricular myocardium in LGE images. For the right ventricle, there were no differences in DCS when comparing the MI-UNet with single-image type UNets. The joint segmentation of multiple image types increases segmentation accuracy for CMR images of the left ventricle compared to single-image models. In clinical practice, the MI-UNet model may expedite the analysis and interpretation of CMR images of multiple types.

Introduction: Multiple supervised machine learning (ML) models have been used to assess the outcomes in patients who underwent transcatheter aortic valve replacement (TAVR) procedure, however, supervised ML requires significant expert time for annotation and labeling. Unsupervised ML can provide insight into the dataset without this process. Hypothesis: We hypothesized that unsupervised k-means ML can identify distinct phenogroups in TAVR patients that can predict prognosis without a priori knowledge. Methods: A retrospective study of patients who underwent TAVR (Jan 2012-Dec 2017) at Mayo Clinic sites was performed. Baseline demographics, EKG, cardiac computed tomography, and echocardiographic data were abstracted. Variables with ≥ 20% missing data were excluded. Prior to the k-means clustering algorithm, principal component analysis was used for dimensionality reduction. Input variables were analyzed according to k-means determined clusters. Kaplan-Meier survival analysis was used to assess the survival of each cluster. Results: A total of 1071 consecutive patients were included in the final analysis. A total of 141 unique variables were used. The mean age was 80.7±8.6 years, 58.2% were male. Patients were divided into 2 distinct phenogroups. In the two phenogroups, cluster 0 (n=639) has significantly better survival than cluster 1 (n= 432) (log-rank p=0.0002) (Figure 1). Cluster 1 has features of low flow-low gradient (LFLG) hemodynamic profile, including lower systolic blood pressure, mean aortic valve gradient, peak aortic valve velocity, and worse LV ejection fraction. Conclusions: Unsupervised ML can identify distinct phenotype features with prognostic significance in TAVR patients. Our results suggest that patients with LFLG hemodynamic features are intrinsically different from other TAVR patients and require more attention to avoid adverse outcomes.

Introduction: Unsupervised machine learning (ML) can provide insight into a dataset without significant expert time for annotation and labeling, but was never used to assess the heterogeneous MitraClip population. Hypothesis: We hypothesized that unsupervised k-means ML could identify prognostically-distinct phenogroups in MitraClip patients without a priori knowledge of the dataset. Methods: Patients who underwent MitraClip (June 2014-September 2020) at Mayo Clinic sites were identified from the Mayo institutional NCDR database for baseline and follow-up data. Variables with ≥ 50% missing data were excluded. A k-means algorithm was used for clustering analysis. Input variables were grouped according to 4 distinct k-means determined clusters. Kaplan-Meier survival analysis was used to assess the survival of each cluster. Results: A total of 389 consecutive patients were included in the final analysis, and 95 unique variables were used. The mean age was 80.3±8.7 years; 256 (65.8%) were male. The mean STS MV Replace score was 9.1±5.7. Fifty-five (14.5%) patients died during the mean follow duration (185 days). Kaplan-Meier analysis showed significant survival differences among clusters (Figure 1). The cluster (cluster 2, n=120) with best survival performance (all-cause mortality 3.3%) had features of non-calcified mitral leaflets, nonischemic cardiomyopathy, and less than moderate tricuspid regurgitation. Cluster 3 (n=66) had the worst survival performance (all-cause mortality 27.3%) and opposite features than cluster 2. Conclusions: Unsupervised ML can identify distinct phenotype clusters with prognostic significance in MitraClip patients. Patients with non-calcified mitral leaflets, nonischemic cardiomyopathy, and less than moderate tricuspid regurgitation are intrinsically different from other MitraClip patients and should be considered the potential optimal candidates.

Background/Introduction: More than moderate tricuspid regurgitation (TR) at baseline is associated with worse outcome in MitraClip patients. The presence of iatrogenic ASD (iASD) may cause right-sided volume overload and lead to adverse events. However, the real effects of iASD are not well-investigated in these patients who are vulnerable to right-sided volume overload. Methods: Patients with more than moderate TR who underwent a successful MitraClip procedure in the period from June 2014 to September 2020 were identified from the Mayo institutional NCDR database. Follow-up transthoracic echocardiography (TTE) at about 1 year after the procedure was used to determine the presence of iASD, and speckle-tracking strain analysis was done. Patients were grouped according to the presence of iASD (iASD+ group vs. iASD- group) for statistical analysis. Results: A total of 141 consecutive patients were included for the final analysis (iASD+ group n=44 and iASD- group n=97). The mean age was 79.3 ± 8.8 years, and 72.3% were male. Follow-up echocardiography and speckle-tracking strain data did not show a significant difference in either left or right ventricular function in conventional echocardiographic parameters and strain. LA reservoir strain (13.7 ± 6.9 vs 11.5±7.5, P=0.0418) was the only parameter with significant difference among the two groups. Details are summarized in Table 1 . Conclusion: The presence of iASD can decompress the left atrium, but does not affect left or right ventricular strain function in MitraClip patients with more than moderate baseline TR.

Abstract Objectives: We aimed to compare the segmentation performance of the current prominent deep learning (DL) algorithms with ground-truth segmentations and to validate the reproducibility of the manually created 2D echocardiographic four cardiac chamber ground-truth annotation.Background: Recently emerged DL based fully-automated chamber segmentation and function assessment methods have shown great potential for future application in aiding image acquisition, quantification, and suggestion for diagnosis. However, the performance of current DL algorithms have not previously been compared with each other. In addition, the reproducibility of ground-truth annotations which are the basis of these algorithms have not yet been fully validated.Methods: We retrospectively enrolled 500 consecutive patients who underwent transthoracic echocardiogram (TTE) from December 2019 to December 2020. Simple U-net, Res-U-net, and Dense-U-net algorithms were compared for the segmentation performances and clinical indices such as left atrial volume (LAV), left ventricular end diastolic volume (LVEDV), LV end systolic volume (LVESV), LV mass, and ejection fraction (EF) were evaluated. The inter- and intra- observer variability analysis was performed by two expert sonographers for a randomly selected echocardiographic view in 100 patients (apical 2-chamber, apical 4-chamber, and parasternal short axis views).Results: The overall performance of all DL methods was excellent (average Dice similarity coefficient (DSC) 0.91 to 0.95 and average Intersection over union (IOU) 0.83 to 0.90), with the exception of LV wall area on PSAX view (average DSC of 0.83, IOU 0.72). In addition, there were no significant difference in clinical indices between ground truth and automated DL measurements. For inter- and intra observer variability analysis, the overall intra observer reproducibility was excellent: LAV (ICC = 0.995), LVEDV (ICC = 0.996), LVESV (ICC = 0.997), LV mass (ICC = 0.991) and EF (ICC = 0.984). The inter-observer reproducibility was slightly lower as compared to intraobserver agreement: LAV (ICC = 0.976), LVEDV (ICC = 0.982), LVESV (ICC =0.970), LV mass (ICC = 0.971), and EF (ICC = 0.899).Conclusions: The three current prominent DL-based fully automated methods are able to reliably perform four-chamber segmentation and quantification of clinical indices. Furthermore, we were able to validate the four cardiac chamber ground-truth annotation and demonstrate an overall excellent reproducibility, but still with some degree of inter-observer variability.