Unsupervised Machine Learning for Vascular Mesh Compression.

Abstract

Machine learning (ML) models are becoming increasingly valuable for cardiovascular prediction and simulation, offering critical support for medical decision-making. These models are particularly useful for predicting disease progression and evaluating potential treatments. A major challenge in these models is to preserve the geometric fidelity of meshes while optimizing parameter efficiency to reduce memory usage, computational resources and execution time. In this paper, we present innovative approaches to abdominal aortic aneurysm (AAA) mesh compression, utilizing both statistical and deep learning models, with a focus on unsupervised learning techniques. We explore principal component analysis (PCA) as a statistical method and compare it with several deep learning models, including a simple autoencoder, an enhanced autoencoder based on PCA, a convolutional neural network (CNN), and a graph neural network (GNN). Human aortas are compressed using different statistical and deep learning methods to get the most relevant features. The mesh is reconstructed using the computed features and the error of the reconstructed meshes is compared. Our results indicate that PCA, using 64 principal components, outperforms deep learning models with a comparable latent space of 64, achieving the best overall performance. Among the deep learning approaches, the PCA-based autoencoder demonstrates the highest effectiveness.