電腦輔助腎臟腫瘤分割之檢測;The segmentation of renal tumor using Computer assisted detection

[[abstract]]由於影像分割方法大多皆需給予初始位置與輪廓，而影像分割方法的收斂與擴張又與初始位置與輪廓有著極大的相關。本研究提出一種創新的方法，結合卷積神經網路和水平集方法，用於電腦斷層掃描的腹部影像的半自動分割腎臟腫瘤，利用卷積神經網路在目標識別方面的優異表現，獲得腎臟腫瘤初始位置與輪廓，以解決因為人工手動圈選初始位置與輪廓的誤判而導致分割結果的錯誤，進而節省了進行影像判讀的時間與精力。本研究詳細描述了卷積神經網路和水平集算法，並用於電腦斷層掃描的腹部影像之腎臟腫瘤分割。經實驗結果證實，本研究得到了相當高準確的結果，且本研究將所提出的方法與過去文獻所提出的方法做比較，結果顯示本研究提出的方法在靈敏度的部分高於過去文獻的結果。 Due to the image segmentation methods need to be given the initial position and contour, and the convergence and expansion of image segmentation methods also has a strong association with the initial position and contour, this study propose an innovative method. We combine convolutional neural network and level set for semi-automatic segmentation of renal tumors in computer tomography images. Using the advantage of CNN in target recognition to solve the misjudgment caused by manually circle the initial position and contour, and then save the time and energy of image interpretation.In this study, we describe the convolutional neural network and the level set algorithm in detail, and applied them to renal tumor segmentations in abdominal images. The results show that this study has great accuracy. And compared with the literature in the past, the proposed method has higher sensitivity