Artificial Intelligence-Enhanced Echocardiographic Assessment of the Aortic Valve Stenosis Continuum

Summary Background Transthoracic echocardiography (TTE) is the primary modality for diagnosing aortic stenosis (AS), yet it requires skilled operators and can be resource-intensive. We developed and validated an artificial intelligence (AI)-based system for evaluating AS that is effective in both resource-limited and advanced settings. Methods We created a dual-pathway AI system for AS evaluation using a nationwide echocardiographic dataset (developmental dataset, n=8,427): 1) a deep learning (DL)-based AS continuum assessment algorithm using limited 2D TTE videos, and 2) automating conventional AS evaluation. We performed internal (internal test dataset ITDS, n=841) and external validation (distinct hospital dataset DHDS, n=1,696; temporally distinct dataset TDDS, n=772) for diagnostic value across various stages of AS and prognostic value for composite endpoints (cardiovascular death, heart failure, and aortic valve replacement) Findings The DL index for the AS continuum (DLi-ASc, range 0-100) increases with worsening AS severity and demonstrated excellent discrimination for any AS (AUC 0.91– 0.99), significant AS (0.95–0.98), and severe AS (0.97–0.99). DLi-ASc was independent predictor for composite endpoint (adjusted hazard ratios 2.19, 1.64, and 1.61 per 10-point increase in ITDS, DHDS, and TDDS, respectively). Automatic measurement of conventional AS parameters demonstrated excellent correlation with manual measurement, resulting in high accuracy for AS staging (98.2% for ITDS, 82.1% for DHDS, and 96.8% for TDDS) and comparable prognostic value to manually-derived parameters. Interpretation The AI-based system provides accurate and prognostically valuable AS assessment, suitable for various clinical settings. Further validation studies are planned to confirm its effectiveness across diverse environments. Research in Context Evidence before this study We screened all English-based research articles in PubMed up to December 2023 using the keywords “artificial intelligence," "echocardiography," “aortic stenosis,” and "aortic valve stenosis." While some studies have used artificial intelligence (AI) to evaluate aortic stenosis (AS) in echocardiography, these efforts were typically focused on either predicting significant AS or automating conventional measurements, not both. For instance, Wesler BS et al. trained a deep learning model on 338 patients and validated it with 119 patients, achieving an area under the receiver operating characteristic curve (AUC) of 0.86 for distinguishing significant AS from non-significant AS. In a larger-scale study, Holste G. et al. trained a deep learning model on 5,257 studies and validated it using two external datasets (4,226 and 3,072 studies), achieving high accuracy in detecting severe AS (AUC: 0.942–0.952). However, both models were limited to the parasternal long-axis view and did not provide conventional quantitative analysis. In contrast, Krishna H. et al. automated conventional AS evaluation, demonstrating that AI could accurately measure AS parameters like aortic valve maximal velocity, mean pressure gradient, and aortic valve area in 256 patients, comparable to human measurements, but did not perform qualitative assessment of AS. Additionally, while Strange G et al. identified AI-based AS phenotypes linked to mortality risk using data from echocardiographic reports, this approach was based on tabular data rather than direct image analysis, thus lacking the capability to assess AS severity from imaging data. Added value of this study In this study, we developed a comprehensive AI-based system to evaluate AS through a dual pathway: 1) assessing AS presence and severity by deriving a DL index for the AS continuum (DLi-ASc) from parasternal long and/or short axis videos only, and 2) automatically measuring AS parameters and providing conventional quantitative AS evaluation if additional images are available. The system was validated internally and in two independent external datasets, where DLi-ASc increased with AS severity and demonstrated excellent discrimination for any AS (AUC 0.91–0.99), significant AS (0.95–0.98), and severe AS (0.97–0.99). Additionally, DLi-ASc independently predicted adverse cardiovascular events. The automatic measurement of conventional AS parameters showed a strong correlation with manual measurement, resulting in high accuracy for AS staging (98.2% for internal test set, 81.0%, and 96.8% for external test sets) and offered prognostic value comparable to manually-derived parameters. Implications of all the available evidence AI-enhanced echocardiographic evaluation of AS allows for accurate diagnosis of significant AS and prediction of severity using only parasternal long or short axis views, typically obtained in the first step of echocardiographic evaluation. This capability can enhance AS assessment in resource-limited settings and provide novices with guidance on when quantitative analysis is necessary. If additional views are appropriately acquired, the system automatically analyses them, potentially enabling conventional quantitative evaluation, thereby saving time and effort while ensuring accurate assessment. However, further comparative prospective studies are necessary to assess whether this AI-based approach ensure these efficiencies without inadvertently increasing diagnostic errors or adverse cardiac outcomes compared to conventional, manual AS evaluation.