基於結構統計分析的腦動脈虛擬患者群體生成

Abstract

三維腦動脈模型的可獲得性有限，對於依賴大量資料進行訓練的機器學習系統而言，造成了在生理模擬、手術規劃與治療優化等應用上的挑戰，特別是在涵蓋不同年齡層的結構多樣性方面更為不足。 為解決此問題，本研究提出一套虛擬腦動脈族群生成框架，結合血管自動分段、結構標註與統計建模方法。首先，透過血管結構與位置特徵進行自動分類與標註，接著運用主成分分析（PCA）進行降維，並引入隨機擾動與Dirichlet分布權重平均方式，生成具有年齡與性別分層特徵的虛擬個體模型。 最後，本研究採用基於導通係數的血流模擬模型，並透過差分演化（Differential Evolution）演算法調整出口壓力，使模擬流量符合生理統計範圍。我們從血管幾何與血流分布兩個層面，定量比較虛擬模型與原始資料，驗證其結構一致性與生理合理性。此方法提供了一種可擴展的虛擬病人生成機制，有助於支持個人化模擬與臨床決策研究。

The limited availability of 3D cerebral arterial models poses a major challenge to training data-hungry machine learning systems used for physiological simulations, surgical planning, and treatment optimization. These limitations hinder structural diversity and demographic coverage, especially across age groups. To address this, we present a framework for generating a virtual population of cerebral arteries by combining automatic vascular segmentation, structural labelling, and statistical modelling. Vascular segments are classified using structural and positional features, followed by dimensionality reduction via Principal Component Analysis (PCA). We then apply gaussian perturbations and age-weighted aggregation using Dirichlet distributions to synthesize new patient-specific models stratified by age and gender. Finally, we simulate cerebral blood flow using a conductance-based flow model and optimize outlet pressures via Differential Evolution to match target physiological flow ranges. The generated models were quantitatively evaluated against flow measurements reported in the literature, confirming physiological plausibility. This approach provides a scalable pathway for generating diverse, demographically representative virtual populations for personalized simulation studies.