探討二維與三維形態結構運算子對濾波反投影之核子醫學影像進行雜訊處理

Investigation of noise reduction by 2D and 3D-morphological structure operation elements from filtered back-projection of nuclear medicine images

邱駿逸(Chun-Yi Chiu)

指導教授 ： 黃詠暉

共同指導教授 ： 陳泰賓(Tai-Been Chen)

義守大學/電機資訊學院/資訊工程學系/碩士(2016年)

摘要

濾波反投影法(Filtered Back-Projection, FBP)為核子醫學中常見之影像重建演算法，並且能在較少光子計數與較短時間內重建出正子斷層(Positron Emission Tomography, PET)影像。其比較疊代式演算法能提供較高之影像對比度。但重建結果容易於PET影像中產生星狀假影(Strike Artifacts)。本研究分別探討FBP影像使用二維與三維之形態結構運算(Morphological Structure Operator, MSO)後，在不減少影像品質的情況下星狀假影去除之效益，同時找出其形態結構運算子組合。
本研究利用Deluxe Jaszczak假體、線射源與大鼠之18F-Fluorine PET影像並使用FBP演算法重建影像。其重建影像分別進行二維與三維MSO雜訊處理並計算其影像背景標準差、訊雜比(Signal to Noise Ratio, SNR)與半高全寬(Full Width at Half Maximum, FWHM)，以評估經雜訊處理後背景雜訊、影像品質與影像解析度之效益，並找出其形態結構運算子組合。利用Deluxe Jaszczak假體、線射源與大鼠影像進行3x3(二維)MSO雜訊處理後其背景雜訊分別降低約83%、37%及30%；影像品質分別提升約24%、22%及7%。另外，3x3x3(三維)MSO雜訊處理矩陣之組合數量龐大，於運算過程中將造成硬體設備不足而無法運作，或需要花費較長的計算時間。故執行三維MSO雜訊處理前使用最佳反應曲線(Optimal Response Curve, ORC)之方法，以有效篩選其有貢獻之運算子矩陣組合，並減少MSO之運算時間。於三維MSO雜訊處理與最佳反應曲線之搭配應用下，其影像背景雜訊分別降低約87%、43%及55%；影像品質分別提升約28%、24%及12%。
　　透過二維及三維MSO處理後所量測Deluxe Jaszczak假體影像之FWHM為31.25及37.50 mm。於Deluxe Jaszczak假體之三維MSO處理上顯示有過度校正的情況發生，造成影像解析度不佳；二維MSO處理有較佳的影像解析度。經二維及三維MSO處理後所量測線射源影像之FWHM為2.46及2.21 mm。於線射源之三維MSO處理上顯示有較佳的影像解析度。導致此結果之原因可能為影像非均質性之關係，使於假體研究上有過度校正的情況。利用二維MSO雜訊處理其運算時間為1925秒、3689秒與3446秒；利用三維MSO雜訊處理與最佳反應曲線之搭配應用下其運算時間為219秒、101秒與100秒。然二維MSO之計算時間較長，其主要差別在於二維之逐張處理的步驟過程。
　　三維MSO與最佳反應曲線的搭配使用能有效減少其影像背景雜訊與提升影像品質。儘管於本研究中有部分影像因非均質性之關係影響而造成解析度的降低。期望於未來研究可找尋新的篩選資料之方法並與三維MSO搭配應用以解決影像解析度降低之問題。

關鍵字

濾波反投影 ； 正子斷層 ； 星狀假影 ； 形態結構運算 ； 最佳反應曲線

並列摘要

FBP algorithm is usually used to reconstruct PET images under low count rates and fast running time for diagnostic purposes. The FBP algorithm also provides higher contrast of images compared to iterative reconstruction algorithms. However, reconstruction of PET images by FBP algorithm often yields background streak artifacts because of the limited number of projections. This study use morphological structure operation (MSO) to removing strike artifacts of noise from nuclear medicine images.
　　This study reconstructed Deluxe Jaszczak phantom, line source and rat of 18F-Fluorine of PET images by FBP algorithm. The 2D and 3D MSO was used to remove strike artifacts from PET images by FBP algorithm. The background standard deviation, SNR and FWHM were computed with a specific background ROI for assessed the image quality and resolution. The background standard deviation from Deluxe Jaszczak phantom, line source and rat image was decreased 83%, 37% and 30% for 2D (3x3) MSO, and image quality was increased 24%, 22% and 7%. Moreover, MSO could not be complied by loss computer equipment, or computing time be costed much for the matrix combination of 3D (3x3x3) is numerous. The useful matrix of operation elements was selected effectively and MSO of computing time be reduced for 3D MSO with optimal response curve (ORC). The background standard deviation from Deluxe Jaszczak phantom, line source and rat image was decreased 87%, 43% and 55% for 3D MSO, and image quality was increased 28%, 24% and 12%.
　　The FWHM of Deluxe Jaszczak phantom was 31.25 and 37.50 mm for 2D and 3D MSO. Deluxe Jaszczak phantom of images were loss image resolution for over correction by 3D MSO with ORC and compared 2D MSO. The FWHM of line source were 2.46 and 2.21 mm for 2D and 3D MSO. Line source of images were better image resolution by 3D MSO with ORC. The result of phantom study had been over correction for inhomogeneous image. The running costs are 1925, 3689 and 3446 seconds for 2D MSO, and 219, 101 and 100 seconds for 3D MSO with ORC. The result of computing time was long by 2D MSO due to slices was estimated each during process of 2D MSO.
　　The 3D MSO with ORC was used to decrease noise and provides better image quality effectively. Although these study of part results were loss image resolution for inhomogeneous image. In the future, new selection of matrix combination method has been discovered and resolved problem of loss image resolution with 3D MSO.

並列關鍵字

FBP ； PET ； strike ； artifacts ； MSO ； ORC

參考文獻

[1] Fowler JS, Ido T, 2002. Initial and subsequent approach for the synthesis of 18FDG. Semin Nucl Med 32 (1): 6–12.

[2] Yu S, 2006. Review of 18F-FDG synthesis and quality control. Biomedical Imaging and Intervention Journal 2(4), e57.

[3] Simon G, 2015. Roles of facilitative glucose transporter GLUT1 in [18F] FDG positron emission tomography (PET) imaging of human diseases. Journal of Diagnostic Imaging in Therapy. 2(1), 30-102.

[4] Norman EB, Remo G, Johrnee’ W, Bradley RN, 2009. PET/MRI: The Blended-Modality Choice of the Future? J. Nucl. Med. Technol 37(2), 63-71.

[5] Bélanger MJ, Mann JJ, Parsey RV, 2004. OS-EM and FBP reconstruction satlow count rates: effecton 3D PET studies of [11C] WAY-100635. Neuro Image 21, 244–250.