使用最小成本路徑搜尋演算法的血管擷取與分割

Abstract

從醫學影像中擷取線型結構，如血管、神經，在臨床診斷與研究上扮演重要的角色。本論文提出一個半自動，在不調整參數的情況下能適用於種類不同之醫學影像的管狀結構偵測演算法。我們經由梯度向量的方向尋找可能剖面，再使用多尺度空間中的Hessian矩陣分析，驗證剖面的真確性。最後再以最小成本路徑搜尋演算法擷取出線形結構的中線與邊界。實驗結果顯示本文提出之演算法在冠狀動脈血管造影，視網膜血管造影及共軛焦顯微鏡神經影像均有良好的效果。

Extraction of tubular structures in medical images, such as vessels, neuron, plays an important role in clinical diagnosis and research. A new semiautomatic approach for tubular structure extraction is proposed in this thesis which can be applied to various types of images without parameter tuning. We use gradient vectors to find possible profiles, which are then verified by analysis of Hessian matrix in multiscale space. Finally a center line and boundaries are extracted using minimum cost path finding algorithms. Experimental results show that the proposed algorithm is feasible for coronary artery angiograms, retinal angiograms, as well as confocal microscope images of neural fibers.