CMR-AI and Outcomes in AS

Artificial Intelligence (AI) and Machine Learning are reshaping our daily clinical practice, which has the potential to be more efficient, precise, and personalized. Adopting these technologies in cardiac imaging does not only affect decision making by improved accuracy and risk stratification but also significantly reduces scan times and post-imaging workup.

Current guidelines acknowledge cardiac magnetic resonance (CMR) imaging as gold standard for assessment of myocardial structure and function. Despite the fundamental importance in various cardiac diseases, measurements of size, mass, and ejection fraction (EF) are flawed by the inherent variability and subjectivity of human analysis. Recent developments in deep learning using convolutional neural networks (CNNs) allow for automated segmentation of the ventricular blood pool and myocardium using pre-existing CMR datasets. Importantly, these tools are integrated into CMR scanners generating real-time measurements without the need of time-consuming image post-processing. AI-based models have previously shown similar to superior precision in ventricular contouring, volumetry, and maximum wall thickness (MWT) measurements, outperforming clinical experts.

In patients with aortic stenosis (AS), changes in EF more often occur late in the disease process, whereas longitudinal shortening represents an earlier and more sensitive marker of left ventricular (LV) dysfunction. However, these CMR measurements are subjective, time-consuming, and therefore not routinely performed due to the lack of automated analysis. Recently, AI-measured global longitudinal shortening (GLS) and mitral annular plane systolic excursion (MAPSE) have been demonstrated to provide more reproducible and accurate results compared to human experts. We hypothesize that AI-based GLS and MAPSE could convey important prognostic information beyond LVEF in severe AS and represent early markers of adverse cardiac remodeling and outcome following aortic valve replacement (AVR). Furthermore, in our own working group, we could demonstrate that right ventricular (RV) dysfunction on CMR, rather than conventional parameters assessed by echocardiography, was independently associated with outcome in individuals with AS undergoing transcatheter aortic valve implantation. We aim to extend on our findings and investigate whether AI-based RV GLS and tricuspid annular plane systolic excursion (TAPSE) represent early markers of RV dysfunction indicating adverse prognosis.

Finally, the assessment of reverse cardiac remodeling by CMR requires reproducibility. AI has been proven to outperform humans in both precision and accuracy, and therefore has great potential for the comprehensive evaluation of longitudinal structural changes in AS following AVR. We aim to analyze mid-term reverse cardiac remodeling in patients with AS using novel AI technology.

Aims

With significant previous contributions in cardiac imaging and valvular heart disease being made by our research group, we aim to provide automated, precise, and time-saving algorithms to identify patients at risk post-AVR by addressing the following:

• Association of AI-measured LV and RV structural and functional markers on CMR prior to AVR with mid-term clinical outcomes at 24-months following AVR.
• Reverse cardiac remodeling, as determined by CMR-AI parameters, at baseline versus 24-months after AVR.

Methods

This project is designed as a large-scale international, prospective, multi-center, longitudinal-observational cohort study aimed at identifying predictors of structural and functional recovery in patients with severe AS undergoing clinically indicated AVR. Participants were previously enrolled from seven university-affiliated tertiary care centers in Continental Europe, the UK, and Asia between January 2020 and August 2024.

Baseline evaluation consisted of comprehensive pre-operative cardiac phenotyping including quality of life assessment, blood tests, electrocardiogram (ECG), and imaging (CMR and echocardiography). For this proposed project, reverse cardiac remodeling and mid-term clinical outcomes will be evaluated 24-months post-AVR through repeat baseline investigations.

Inclusion criteria

• Written informed consent
• Severe AS scheduled for Heart Team decision

Exclusion criteria

• Inability or unwillingness to perform any of the diagnostic tests
• Inability or unwillingness to participate in follow-up visits
• Metal implants, e.g. cochlear implants and pacemakers
• Chronic kidney failure (GFR < 30 ml/min/1.73m2)