Aortic stenosis (AS) is one of the most common valvulopathies, particularly prevalent in the elderly population due to cellular degeneration. Early and accurate detection of AS is crucial to prevent serious complications. Transthoracic echocardiography (TTE) is the modality of first choice among cardiologists for the evaluation of AS, as it offers a non-radiating and cost-effective solution. However, its accuracy is often limited by dependence on the expertise of the cardiologist and by difficulties in obtaining accurate measurements, particularly of left ventricular outflow tract diameter (LVOTD), a key parameter in assessing the severity of AS To handle these limitations, our study proposes a new approach based on semi-supervised learning (SSL) on multi-view TTE images for automatic detection of aortic stenosis severity, with the aim of improving diagnostic accuracy and reducing reliance on manual interpretation.