複合式模糊平均數群聚演算法應用於電腦斷層影像分割

背景：結合正子斷層掃描和電腦斷層掃描的(PET/CT, Positron Emission Tomography/Computer Tomography)中，CT影像除了可以提供解剖定位外，亦可以提供PET定量分析所需之衰減係數資訊。本研究的目的在於提出一套自動化CT影像分割方法，藉由對於不同組織之分割，以期有助於增加PFT衰減修正(attenuation correction)的準確性，與臨床上PET/CT定量分析的可行性。方法：本研究的方法延伸1973年由Bezdek首先提出的模糊平均數群聚演算法(Fuzzy C-Means Clustering, FCM)為架構，直接探討影像屬性(features)在影像中的分布關係，整體地評估區域的分布特性。本研究所提出之複合式FCM演算法(Mixed Fuzzy K-Means Clustering, Mixed FCM)，結合了兩種影像屬性：影像値的區域變化和影像值的大小，整合群聚分析中各種影像點的資訊。結果與結論：由臨床CT影像的分割結果可以顯示，利用複合式FCM演算法能明顯地提高骨骼和空氣區域的分群效果，凸顯骨骼和空氣在解剖空間上的位置。此外，自動化的分割策略，更能增加衰減修正整體的效率，提高臨床上PET/CT掃描定量分析的可行性。

關鍵字

PET/CT ；衰減修正；影像分割；模糊平均數群聚演算法

並列摘要

Objective: PET/CT scanner provides ”hardware” image fusion capabilities which will make interpretation of PET images much easier due to the anatomical landmarks offered by CT scans. In addition, the CT data can be used to correct PET scans for photon attenuation. Methods: In this work, a mixed fuzzy c-means (FCM) clustering algorithm is proposed for the automatic CT image segmentation aiming at the possibility for the improvement of PET/CT image quantitation. The proposed algorithm combines two metrics, pixel intensity and local variance, as image features. Under the fuzzy c-means framework of Bezdek, a CT image can be automatically segmented into various tissue types: bone, soft tissues, lung, and air. Results: The experiments of clinical CT images have shown the improved segmentation accuracy compared to the conventional c-means clustering algorithm. Particularly, the proposed algorithm can improve segmentation accuracy around air and bone areas. Bone will not contribute any activities of PET scan, but will introduce photon attenuation and scattering. Conclusion: Therefore, better segmentation result in bone and air areas should provide more accurate attenuation information and could lead to potential improvement in PET quantitation.

並列關鍵字

PET/CT ； attenuation correction ； image segmentation ； Fuzzy C-Means Clustering

國際替代計量

複合式模糊平均數群聚演算法應用於電腦斷層影像分割

全文下載

主題瀏覽