中文摘要 超音波具有非侵入性、高解析度、即時性掃描、非放射性及使用方便等優點,因此常被使用於掃描生物體內的組織器官;因而成為臨床上最普遍使用的診斷工具。藉由超音波影像,有助於臨床醫師了解各種組織構造,並且可以發現壞損或病變的部份。 然而因為物理特性的關係,超音波在介質內傳遞時,無法避免地會有不同相位的現象發生,其造成的建設性干涉與破壞性干涉,使得影像中產生大量明暗相間的斑點(speckle),這些斑點會影響醫師們的診斷,並降低影像的品質。 為了消除超音波影像中的斑點,本論文提出一種高階統計學的演算法-獨立成份分析法(Independent Component Analysis, ICA),試圖利用獨立成份分析法濾除不規則分佈的超音波斑點雜訊;我們進一步針對影像中有意義的區塊,使用主動輪廓模型(Active Contour Model, ACM)來對異物進行影像切割。 本研究的實驗結果証實,獨立成份分析法確實可以有效的濾除超音波斑點雜訊,配合主動輪廓的影像切割,將來可幫助臨床醫師進行超音波影像掃瞄時,提供腫瘤組織偵測的辨識度。 Abstract Medical ultrasound systems have the advantages of non-invasion, high spatial resolution, real-time scanning, low-radiation dose, and convenience for use, so that ultrasound has been used as powerful diagnosis tool and widely applied to different clinical applications. Especially, the benefit of high spatial resolution of ultrasound image allows clinical physicians can easily identify tumors or malignant tissues based on their spatial morphology or tissue characteristics. Nevertheless, the presence of speckle pattern in ultrasound image, generated by mutual interference of many diffraction waves with different phases, can degrade the quality of ultrasound image, even results in poor recognizability of small tissues. This dissertation aims to develop a speckle suppression technique based on independent component analysis (ICA), which is a multivariate high-order statistical method. Based on the independency between tissue image and ultrasound speckles, we decompose an ultrasound image into several sub-components. Only those sub-components which are not related to speckle noise are chosen for reconstructing speckle-suppressed ultrasound image. An Active Contour Model (ACM) is then applied in the following to segment the interested region from background image. In our study, we found the utilization of ICA can effectively suppress the unwanted speckle noise. The ICA-based speckle-suppression method has been incorporated with active contour model to label important information on an ultrasound image. This combined method might provide an efficacious way to improve the recogniability and sensitivity of malignant tissues in ultrasound image scanning.