- 學位論文
針對果蠅腦部影像蕈狀體之適應性切割系統
An Adaptive Mushroom-body Segmentation System for Drosophila Brain Images
摘要
摘要 影像處理技術對於醫學影像以及生物實驗影像的分析佔有很重要的地位。經過良好設計的影像處理系統不但可以對於實驗的影像做大量的分析，提高準確度，更能降低人為操縱所造成的疲勞以及所需花費的時間。本論文所提出的高適應性果蠅腦神經影像切割系統主要目的即為達成快速且準確的果蠅腦神經—蕈狀體(mushroom bodies)之切割。 在果蠅腦神經科學的研究中，一個極為重要的目標為:建立一個標準的三維腦模型以提供經由不同的實驗操作或是發育程度不同的果蠅腦所得到的影像的共同比較標準。而為了要建立一個標準化且具代表性的果蠅腦，首先必須要有足夠多的果蠅三維腦模型，並針對這些模型加以平均，這樣所得到的平均腦模型才具有代表性。在果蠅腦中，最具指標性的“地標”為蕈狀體(mushroom bodies)，因此建立標準果蠅腦三維模型的第一步即為建立此蕈狀體之三維模型。為了要幫助更快速建立足夠多的蕈狀體三維模型加以平均以便求得標準蕈狀體三維模型，我們設計一個快速且準確的果蠅腦影像切割系統。此切割系統大幅降低原本經由人工切割果蠅腦影像所需的時間，並且經由實驗結果分析比較之後，我們證明了這個系統的可靠性及準確度。 此系統所處理的影像為國立清華大學腦科學研究中心所提供之共軛焦顯微鏡影像，針對這些果蠅腦影像的特性，我們著重於兩個影像處理過程的設計--前處理(preprocessing procedure)去除影像雜訊以及利用可形變的邊界圈選模型 GVF snake mode來偵測蕈狀體的邊界並切割。 前處理過程分為兩個部份，分別針對果蠅腦影像中含有Calyxes的部份以及不含Calyxes的部份設計不同的方式去除非蕈狀體的腦神經以及取像過程中所造成的雜訊。對於帶有Calyxes的影像，由於蕈狀體的邊界較為模糊且難以將其與緊鄰的Kenyon cells分離，因此我們採用影像紋理分析的方式合併模糊理論(fuzzy C-means algorithm)來分辨屬於蕈狀體及非蕈狀體的區域。另一方面，針對不帶有Calyxes的影像，我們採取mathematical morphology的方式合併adaptive thresholding來除掉雜訊。 經由前處理以後的影像接著由GVF snake model來偵測蕈狀體的邊界，由於此邊界偵測模型之高度適應性加上影像大部分的雜訊已經去除，因此即使給予與真實邊界差異很大的初始邊界條件，我們仍然能得到準確的切割結果。本系統對於帶有不同雜訊的果蠅腦影像具有高度適應性且對於所需輸入的初始邊界無嚴格的要求，只需少許的人為介入來消除與Calyxes高度相似的模糊區域，即可一次處理一個果蠅腦實驗所產生的140張影像，並建立其三維模型。本論文所提出的影像切割系統亦可與特定的影像處理技術合併，達成果蠅腦研究中所需的其它神經元對位、建立標準座標系統的目標。
並列摘要
Abstract Image segmentation plays a crucial role in many medical imaging applications by automating or facilitating the delineation of anatomical structures. In the Drosophila brain research, the automatic segmentation of individual Drosophila brain neuropils from the large amount of experimental images is highly demanded because manual segmentation is a tedious and time consuming process. Mushroom bodies are the most important landmark in Drosophila brains. As a result, the automatic segmentation of mushroom bodies becomes a fundamental and significant step in establishing a standard landmark for the ensuing task of building a standard Drosophila brain model. In this thesis, we propose a system to accomplish the goal of automatic and accurate segmentation of mushroom bodies in given stacks of experimental images. The proposed system consists of two major steps—preprocessing and boundary extraction. On each image, a preprocessing procedure is performed, followed by the gradient vector flow snake (GVF snake) to define the final contours for segmenting. The preprocessing procedure includes the techniques of morphological smoothing, statistical thresholding, texture analysis and the fuzzy C-means algorithm. The goal of morphological smoothing and statistical thresholding methods is to remove the noise composed of the speckles near the mushroom bodies and the stained non-mushroom-body neuropils. The texture analysis and fuzzy c-means algorithm aim at sharpening the blurred boundaries of mushroom bodies especially around the calyx regions. We apply these techniques according to the contents of the processed images. The final contours of mushroom-body segmentation are extracted by the GVF snake algorithm. After the segmentation, most of the stained cells along with the noisy artifacts and neuropils other than mushroom bodies are cleaned away. The only human interaction required in our segmentation system is the selection of the real fore-four region among the segmented possible ones in the Calyxes region. Each slice containing images showing only the mushroom bodies can be used to construct a 3D model. From the experimental result, we have demonstrated that the proposed system can provide rapid segmentation and construction of individual 3D mushroom-body models. This system is proven instrumental in generating a more precise standard 3D Drosophila brain model, which is crucial in the Drosophila brain research.