利用空間分叢演算法自動分割腹部磁振造影影像中的脂肪含量

Automatic Abdominal Fat Segmentation in Magnetic Resonance Images Using Spatial Clustering Algorithm

孫靜儀(Jing-Yi Sun)

指導教授 ： 吳杰

中臺科技大學/健康科學院/放射科學研究所/碩士(2011年)

摘要

高血壓、動脈硬化、第二型糖尿病等新陳代謝症候群與腹部脂肪含量有關，因此，正確的分析腹部脂肪含量為監測代謝性症候群之重要指標。本研究設計一套全自動化腹部脂肪分割演算法，應用於T2加權腹部磁振造影影像（magnetic resonance image, MRI）以分析腹部皮下脂肪與內臟脂肪含量。自動分割演算法依序執行下列步驟，首先利用新型模糊c平均(Modified Fuzzy c-means, MFCM)演算法校正不均勻磁場；接著使用邊緣函數Canny演算法製作體遮罩，藉由k平均叢集(k-means cluster)演算法取得脂肪組織與非脂肪組織；最後一個步驟利用體遮罩與脂肪組織做布林邏輯運算以分割腹部脂肪。利用線性迴歸分析（simple linear regression）與布蘭德-奧特曼（Bland-Altman）評估手動與自動分割方法的關聯性、一致性與相對百分誤差。SAT與VAT的分析結果顯示，手動與自動分割的皮爾森相關係數分別為0.999(p < 0.05)與0.997(p < 0.05)。除此之外，布蘭德-奧特曼分析顯示自動分割與手動分割的一致性。皮下脂肪之百分誤差為-1.7% ~ 2.1%以及內臟脂肪為-4.2 % ~ 5.2%。本研究所設計的自動分割演算法可有效測量腹部脂肪分佈並可減少手動分割的誤差，因此，此演算法應可作為脂肪計量分析的研究工具，進一步可提升新陳代謝相關疾病之診斷率。

關鍵字

脂肪 ； 影像分割 ； 磁振造影 ； k-means 分叢 ； 影像不均勻

並列摘要

Metabolic syndrome, such as hypertension, arteriosclerosis, and type-two diabetes, are related with the amount of abdominal fat. Thus, calculation of the abdominal fat provides an important index for preventing metabolic diseases. This research was to measure abdominal adipose tissues for T2-weighted magnetic resonance imaging (MRI) by applying an accurate unsupervised method. The proposed automatic procedures were divided into the following steps. First process was the image inhomogeneity correction using the Modified Fuzzy C-means (MFCM). Second step was to creation the body masks using Canny edge algorithm. Third, the adipose tissue and non-adipose tissue was detected by k-means cluster algorithm. Final process was the abdomen fat segmentation by the body mask and the non-adipose tissue mask. The simple linear regression and Blot-Altmen plot were used to analyze and compare the consistency and correlation between the manual and automatic segmentation methods. The Pearson correlation coefficients for SAT and VAT between manual and automatic segmentations were 0.999 (p < 0.05) and 0.997 (p < 0.05), respectively. The Blot-Altmen plot showed that manual and automatic segmentations were consistent for SAT and VAT. The percent error for SAT and VAT were -1.7% ~ 2.1% and -4.2% ~ 5.2%, respectively. The proposed algorithm can obtain the abdominal fat distribution efficiently and can lower the error of manual segmentation. Therefore, this approach could be used as an effective tool for quantitative research. Furthermore it could be used to prevent the metabolic related diseases.

並列關鍵字

magnetic resonance imaging ； segmentation ； abdominal fat

參考文獻

Adult Treatment Panel Ⅲ. Executive Summary of the Third Report of the National Cholesterol Education Program (NCEP)Expert Panel on Detection, Evaluation, and Treatment of High Blood cholesterol in Adults. The Journal of the American Medical Association. 2001;285:2486-2496.

Brožek J, Grande F, Anderson JT, Keys A (1963). "Densitometric analysis of body composition: revision of some quantitive assumptions". Ann N Y Acad Sci 101: 113–140.

C.S. Fox, J.M. Massaro, U. Hoffmann, K.M. Pou, M.H. Pal, C.Y. Liu, R.S. Vasan, J.M. Murabito, J.B. Meigs, L.A. Cupples, R.B. D’Agostino, C.J. O’Donnell. Abdominal visceral and subcutaneous adipose tissue compartments: association with metabolic risk factors in the framingham heart study. Circulation. 2007;116（1）:39-48.

C. Würslin, J. Machann, H. Rempp, C. Claussen, B. Yang, and F. Schick. Topography mapping of whole body adipose tissue using a fully automated and standardized procedure. Journal of Magnetic Resonance Imaging. 2010;31:430-439.

E.G. Kamel, G.M. Neill, T.S. Han, F.W. Smith, A. Avenell, L. Davidson, P. Tothill. Measurement of abdominal fat by magnetic resonance imaging, dual-energy X-ray absorptiometry and anthropometry in non-obese men and women. International Journal of Obesity. 1999;23（8）:686-692.