以超解析技術輔助NBI內視鏡影像強化之探討

本研究是以超解析技術輔助窄頻內視鏡影像強化之探討。窄頻光譜成像為現今內視鏡常使用的影像增強技術，本研究採用的是超解析度的方法，用以提升由窄頻內視鏡影像得到的影像質量，並開發出適合窄頻內視鏡影像放大的技術，將窄頻內視鏡影像品質提升。此方法可以提供醫生放大且更高品質的消化道窄頻內視鏡影像，優化病症的判定，以利早期癌症診斷。本研究使用10張NBI影像，經高斯降低採樣率得128*128的縮小四倍低解析度影像做放大處理。將此低解析度影像分成高頻與低頻影像進行處理。先藉由找出影像邊緣及邊界骨架，將其細線化，再把經雙立方內插放大後的低頻影像與高頻影像做結合。藉由調整Alpha值和RGB值，控制影像的邊緣骨架和色調，讓影像因放大而模糊的輪廓以及邊緣部分更清晰，更能凸顯病灶。並使用GPU輔助圖形運算，來達到縮短運算時間的目的。在人眼主觀判斷下，本研究結果因經過細線化邊緣骨架的疊加，輪廓以及邊界部分相對於其他方法來的更明顯且清楚。客觀數據方面，256*256的放大後影像與原影像相減後像素值平方相加開根號後數據上，本研究的值2581.672為各方法最低，表示與原圖像素值差距最小。以未銳化影像與其他方法做PSNR與SNR比較上，本研究分別為31.292dB和23.99dB，可看出對於NBI影像，放大效果在各方法中相對高。同一影像做十次處理做時間平均紀錄，本研究為5.83秒，相對於其他方法時間已大幅縮短，GPU輔助運算下為5.252秒，時間雖縮短不多，但若處理大量影像，時間縮短效果則變明顯。通過問卷統計，本研究所得分數也是各方法中最高。主客觀比較下來，不論是影像品質和處理時間，顯示出本研究結果是相對好的。本研究不僅是提供醫生更高品質的NBI影像，更能透過GPU輔助運算，在處理大量影像上，得到更短的處理時間。

關鍵字

窄頻影像； NBI內視鏡影像增強技術；內視鏡影像分析；超解析； GPU輔助圖形運算

並列摘要

This is study of Super-resolution method for Narrow-band imaging of endoscopy image enhancement. Narrow Band Imaging (NBI) is the key of enhancement methods in the color enhancement systems for diagnosis field. In this study, we develop an image enlarger method by using super-resolution methods to enhance the Narrow-Band Image. This method can not only provide doctor a new image that with high quality image resolution for the early cancerous diagnosis, but also given clinical image data analysis. In this study, 10 NBI image by Gaussian reduce the sampling rate was reduced four times to transform to 128*128 low-resolution image. This low-resolution image is divided into high and low image processing. First, find the edge of the image and the boundary skeleton, which was thinning. Then low-frequency images by bi-cubic interpolation was magnify to make binding with the high-frequency image. By adjusting the values of Alpha and RGB values to control edge and tone of the image. Let the vague contour and edge of the image to be clearer. And use the GPU to support graphics computing. Subjective judgment under eye, the contour and the boundary of this study after thinning edge skeleton superimposed is more obvious and clear than other methods. In the objective evaluation, the root mean the square error was calculated between magnified image and original image, and the value of the study is least, 2581.672. It represents have the smallest gap of pixel value with the original image. Comparison on PSNR and SNR, this study was 31.292dB and 23.99dB, also show the magnified effect is relatively high than other methods. Moreover, the average processing time for ten images with/without GPU assisted is 5.252 seconds and 5.83 seconds, respectively. Through the questionnaire to count, our method obtained the highest score among all methods. Based on the result of subjective and objective evaluation, this enhancement method is acceptable. This study is not only to provide higher quality doctors NBI images, but also assisted by GPU computing to get the shorter processing time .

並列關鍵字

NBI ； Endoscopic image analysis ； NBI endoscopic image enhancement technology ； Super-Resolution ； GPU support computing

參考文獻

[9] 張孝慈, “協助上消化道診斷之超解析度放大技術之開發”, 中原大學生物醫學工程學系碩士論文, 中壢, 2014。

[21] 孫善群, “利用頻譜分析胃腸道系統影像之輔助診斷系統”, 中原大學生物醫學工程學系碩士論文,中壢, 2013。

[1] C.C. Cauberg Evelyne, Jean J.M.C.H. de la Rosette, and Theo M. de Reijke,”Emerging optical techniques in advanced cystoscopy for bladder cancer diagnosis: A review of the current literature”, Indian J Urol. 2011 Apr-Jun; 27(2): 245–251.

[2] Muto M, Katada C, Sano Y, Yoshida S.” Narrow Band Imaging: A New Diagnostic Approach to Visualize Angiogenesis in Superficial Neoplasia”. Clinical Gastroenterology and Hepatology, Volume 3, Issue 7, Supplement 1, July 2005, Pages S16–S20.

[3] H. Machida, Y. Sano, Y. Hamaoto, M. Muto, T. Kozu, H.Tajiri, and S. Yoshida, “Narrow-band imaging in the diagnosis of colorectal mucosal lesions: a pilot study“.Endoscopy, 36(12):1094-1098. 2004.

國際替代計量

以超解析技術輔助NBI內視鏡影像強化之探討

全文下載

主題瀏覽