基於多階段主動輪廓模型之自動表皮組織影像分割

Abstract

周邊神經可收集訊息，如冷、熱、痛等之感覺反應，經傳達至中樞神經後，再由中樞神經作決策，下達指令，以完成動作。神經若因病變或其他外來因素，將使神經傳導失去功能，引起身體功能異常。醫生針對神經切片進行診斷與分析是一件費時費力的工作。本研究提出一個自動化表皮組織區域分割演算法，希望可以減輕醫生於臨床診斷以及研究之負擔。本論文利用CIELAB色彩空間、自動門檻化、最大連通分量、骨架化、簡單移動平均線來取得初始輪廓。因為表皮神經組織影像相當巨大，所以將主動輪廓模型分成三個步驟來減少分割時間，最後利用大津演算法修正最終結果。使用的表皮神經組織影像張數為50張，平均戴斯相似度達到 81.55%。在相同的實驗設定下，本論文提出的方法優於其他比較的分割演算法結果。本研究提出之自動表皮區域分割方法，可以輔助醫生進行表皮神經組織相關之研究。

Peripheral nerves gather information such as cold, heat, and pain, and transmit to the central nervous system, which then makes decisions and commits commands to perform actions. However, nerve conduction may be lost if there are nerve damage and other external factors, leading to abnormal bodily functions. Examining and analyzing nerve skin slices is a tedious task for physicians. This thesis proposes an automated algorithm that segments the epidermal tissue region to reduce the burden of physicians for clinical diagnosis and research. We employ various techniques, including the CIELAB color space, automatic thresholding, maximum connected component, skeletonization, and simple moving average, to obtain the initial contour. Because the epidermal tissue image is quite large, the proposed active contour model is divided into three stages to reduce the segmentation time. The Otsu method is used to amend the final results. There were 50 experimental epidermis nerve tissue images utilized in this work. The average Dice similarity coefficient was 81.55% by the proposed method, which performed better than other compared segmentation algorithms under the same experimental conditions. The proposed automatic epidermal region segmentation method can assist physicians in conducting research related to epidermal nerve tissue image segmentation.