影像切割之醫學細胞追蹤與分析

Abstract

近年來在醫學研究上，追蹤細胞軌跡、判定細胞死亡、細胞分裂等研究越來越受重視，但在做細胞追蹤和分析前，細胞切割將是會先面臨的問題，如果沒有準確的細胞切割，在之後的分析上很容易產生誤差並累積影響判斷，所以好的切割對於細胞追蹤和分析是很重要的一環。 本篇論文將詳細說明我們提出的細胞切割方法。在做細胞切割時，可能會面臨不同的細胞形狀、背景干擾和影像品質等問題，在本篇研究中，解決了以上問題並提供穩定且適應性強的細胞切割方法，首先利用提出的自動適應性的二分法將影像分成細胞和背景區域，接著利用改良的最短路徑切割法，將需要再分割的細胞區域進行進一步的分割，最後，我們利用二維所得到的細胞切割結果建立出三維的細胞切割。實驗結果顯示我們提出來的方法可以將大部分的細胞影像都切割出來，而且表現還勝過現今較新穎的方法。

Study of living cells like cell movement, cell death, and cell division are more and more popular these days. Before cell tracking and analysis, cell segmentation should first be performed. Without accurate cell segmentation, the later biological analysis will have large error due to accumulation of preprocessing error. As a result, good segmentation is an important step for cell tracking and analysis. This thesis describes the methods of cell segmentation that we proposed. In cell segmentation, we have to deal with problems such as different kinds of shapes of the cells, background interference, the quality of the image, etc. We solve those issues and propose a robust cell segmentation method. First, we apply automatic adaptive thresholding to separate images into cell region and background. Second, applying improved shortest path segmentation to cell regions which need to be further segmented. Finally, we construct the fundamental 3D cell segmentation by applying 3D cell labeling to the result of 2D cell segmentation. Simulations show that our proposed method segments most of cell images efficiently and outperforms state-of-the-art methods.