Abstract
Synthetic data generation plays a crucial role in medical research bymitigating privacy concerns and enabling large-scale patient data analysis.This study presents a beta-Variational Autoencoder Graph Convolutional NeuralNetwork framework for generating synthetic Abdominal Aorta Aneurysms (AAA).Using a small real-world dataset, our approach extracts key anatomical featuresand captures complex statistical relationships within a compact disentangledlatent space. To address data limitations, low-impact data augmentation basedon Procrustes analysis was employed, preserving anatomical integrity. Thegeneration strategies, both deterministic and stochastic, manage to enhancedata diversity while ensuring realism. Compared to PCA-based approaches, ourmodel performs more robustly on unseen data by capturing complex, nonlinearanatomical variations. This enables more comprehensive clinical and statisticalanalyses than the original dataset alone. The resulting synthetic AAA datasetpreserves patient privacy while providing a scalable foundation for medicalresearch, device testing, and computational modeling.