Aortic dissection is a serious life-threatening condition affecting the aorta, the largest artery in the body [1]. Approximately 5-30 cases per million people are reported each year [2]. It is characterized by a tear in the intimal layer of the aorta, creating a new channel for the blood to enter the abnormal aortic wall (dissection). This abnormal channel is called the false lumen; the original lumen of the aorta is called the true lumen. There are two major classifications based on where the primary tear occurs: 1) type-A refers to dissections that occur either at the aortic root or in the ascending aorta, and 2) type-B refers to all other dissections that occur from the arch to the descending aorta. This work focuses on type-B aortic dissections (TBAD).

Initially uncomplicated TBAD has an unfavorable long-term prognosis [3]. Morphological features from computed tomography (CT) images may help identify patients who benefit from preventive thoracic endovascular aortic repair, to improve long-term survival [4]. Precise segmentation of the true and false lumen is a prerequisite, but manual segmentation methods are labor-intensive, time-consuming, and poorly reproducible. Fully-automated deep-learning models have shown promise in segmenting the true and false lumen in aortic dissection [5], [6]; however, these models require large expert-annotated datasets for training and validation. Consequently, we investigated Meta AI's off-the-shelf zero-shot learning 'Segment Anything Model' (SAM) [7]. This model is freely available and requires no prior training using large amounts of expert labeled data; thus, making it a versatile tool for many segmentation tasks.