利用K-mean與混合模型的方法對微正子斷層掃描的動態影像做影像分割

Abstract

在活體追蹤和認識基因活動的過程中，動態microPET影像處理及分割是其中很重要的一部份。 為了能夠觀測到微觀的基因活動， 使用高精密度及較少假影的重建技術來探討真實的活度是必要的。 因此，我們使用最大概似函數估計並經由EM演算法來重建microPET 圖像，然後利用k-mean 和混合模型的統計方法到重建的圖像。在本研究中使用模擬資料與實驗數據，所得到的結果證實這些新方法是可行的。

The segmentation of dynamic microPET images is an important issue in tracing and recognizing the gene activity in vivo. In order to discover the gene activity near the resolution of molecular level, reconstruction techniques with high precision and less artifacts are necessary to recover the genuine activity. Hence, we will apply the maximum likelihood estimate by the EM algorithms to reconstruct microPET images with improved resolution. Then, advanced statistical techniques based on k-mean and mixture models are developed to segment the reconstructed images. Simulation and empirical studies are carried out to evaluate the performance of new methods proposed in this report. The results show that these new methods are promising.