於PET/CT上的肝臟自動化分割

Abstract

PET/CT是核子醫學近年發展的造影技術，其結合正子斷層造影(PET, positron emission tomography)與電腦斷層造影(CT, computed tomography)，可同時提供醫師功能性與解剖性的影像。目前臨床最常使用的18F-FDG PET/CT以18F-FDG (2-[Fluorine-18]fluoro-2-deoxy-D-glucose)作為追蹤劑(tracer)可用於多種癌症的診斷與分期。 本論文目的為實作PET/CT上的自動化肝臟分割，除了可以幫助臨床醫師選擇ROI(region of interest)外亦可幫助診斷與評估，分割後的結果並可用於後續其他研究使用。我們共分析37組PET/CT資料，其中包含17組正常肝臟與20組肝癌(HCC, hepatocellular carcinoma)資料。將結果與手動人工分割的標準作比較，正常肝臟敏感度(sensitivity)與特異性(septicity)分別為89.82%與99.89%，而肝癌資料的敏感度與特異性分別為84.36%與99.87%。與人工分割比較，敏感度為86.61%，特異性為99.88%。本論文成功提供一套自動化肝臟分割法，並期待將其進一步用於電腦輔助診斷系統(CAD, computer-aided diagnosis)。

PET/CT is newly developing technology in nuclear medicine. It combined PET(positron emission tomography) images and CT(computed tomography) images, which can provide functional and anatomical information to medical doctor. Now, most widely used 18-F-FDG PET/CT use 2-[Fluorine-18]fluoro-2-deoxy-D-glucose as tracer, it is well known by the value of diagnosis and staging in many kinds of caners. This search tries to provide an automatic liver segmentation system in PET/CT, which can help medical doctor to choose ROI (region of interest). The system can be used for diagnosis and assessment; the result also can use in advance research. We analysis 37 PET/CT data sets, include 17 sets of normal liver and 20 sets of HCC(hepatocellular carcinoma) cases. Compare with manually segmentation, sensitivity and septicity in normal data sets are 89.82% and 99.89%, 84.36% and 99.87% in HCC cases. Totally are 86.61% and 99.88%.