Methodology of supervised and unsupervised learning in clinical AS assessment and management. (A) Workflow illustration of supervised (classification/regression) and unsupervised (segmentation) learning tasks in the context of AS assessment and management. Each step in the workflow varies based on the learning task. Input data, representing training material for the model, is used for learning. Classification/regression tasks involve algorithms like random forest and support vector machine, while segmentation tasks often employ convolutional neural networks (CNNs) for encode-decode frameworks. Model output is compared iteratively to the “ground truth”, represented either by labels/categories or annotations, depending on the task. Loss function convergence is achieved by minimizing the gap between output and ground truth. The AS severity grading scheme [30] serves as ground truth for classification/regression, while manual delineations in images such as CT scans [83] serve as gold standards for image segmentation. (B) Unsupervised learning, exemplified by cluster analysis, utilizes similar input data. Algorithms include model-based clustering (a) [18], topological data analysis (b) [19], and agglomerative hierarchical clustering (c) [17], with results visualized using heatmaps and dendrograms. Both supervised and unsupervised models undergo internal and external validation processes. * Convergence of Loss Function refers to the point where the loss can no longer be reduced under current training settings. AVC, aortic valve calcium; AVA, aortic valve area; PG, peak gradient; RHC, right heart catheterization; 4D, 4 dimensional; CT, computed tomography; CMR, cardiovascular magnetic resonance; ECHO, echocardiography; ECG, electrocardiography; PCG, phonocardiogram; RF, random forest; SVM, support vector machine; CNN, convolutional neural network; AS, aortic stenosis.