應用模糊能量之區域化形變模型於超音波影像中無形狀限制自動化血管偵測

Abstract

超音波影像中之血管偵測可廣泛地應用於電腦輔助診斷系統，影像導引治療及不同醫學造影系統之間之影像對位。由於超音波影像常帶有過多的雜訊、不均質性和低對比度等干擾影像處理的因素，以往的血管偵測方法無法兼顧自動化及無形狀限制等要求，對於帶有模糊輪廓之血管也無法有效的分割出來。本論文提出一套自動化且強健之演算法，透過快速區域化前置分割處理來克服傳統形變模型無法處理不均質影像及收斂速度過慢之缺點，並利用其可以擷取帶有模糊邊緣物體之優點，進行局部血管輪廓之定位。不同於以往提出之方法，此新方法不需人為介入且沒有任何血管形狀上之限制。本論文研究結果顯示一張影像之平均處理時間約為0.197秒，血管偵測成功率達到89.4%。

Vessel detection from ultrasound images could be widely applied to computer aided diagnosis, image-guided online treatments and image registration between different imaging modalities. However, since the image quality of ultrasound images is often degraded by speckle noise, intensity inhomogeneity and low contrast, previous vessel detection approaches could not achieve the process of vessel detection automatically along with high accuracy and without certain shape constraints. Besides, they are not able to detect vessels with ambiguous boundary. In this thesis, a novel approach for detecting vessels automatically and robustly is proposed. Hence we propose a fast localized preliminary segmentation approach combined with fuzzy energy-based active contour so that it can deal with intensity inhomogeneity and it is able to segment objects with ambiguous boundary in one iteration. Apart from previous approaches, our approach does not require manual intervention and does not need a prior knowledge on the shape of vessels. The result of this study shows that we could process one frame with average processing time of 0.197 seconds and the overall accuracy of vessel detection is 89.4%.