耳鼻喉醫師在用喉內視鏡檢查病患喉部時，利用幾何特徵、顏色及紋理作為臨床判斷腫瘤之依據，本研究將之結合醫學影像處理技術，能自動偵測動態影片當中腫瘤的位置，並同時進行圈選提示，能提供醫生更多病理資訊。
喉內視鏡拍攝的影片，除正常清晰影像，部分影像易受口腔唾液、舌根振動及食物殘渣影響，造成影像模糊，本研究利用K均值演算法（K-Means）、計算模糊影像的平滑度及熵值，並以懸壅垂遮罩（Uvula Mask）作為排除喉內視鏡拍攝時碰觸懸壅垂的影像，進行正常影像篩選；且聲門及會厭影像具有明顯幾何及明亮度特徵可進行影像分割，並依此兩區域相關位置，將喉部分為舌根、聲門、會厭及下咽四個區域，並進行腫瘤圈選。
舌根區的腫瘤，利用其邊緣具有影像灰階明顯之變化，進行索貝爾邊緣檢測（Sobel Edge Detection）運算形成梯度影像，並結合分水嶺（Watershed）演算法，有效圈選腫瘤輪廓；下咽及會厭區的腫瘤，利用L*a*b*色彩空間進行分割，轉換為灰階影像、進行二值化並結合形態學找出腫瘤部位、利用主動輪廓法得到腫瘤輪廓，計算腫瘤的平滑度、相關性及熵值三種特徵，進行下咽與會厭兩區域的分類；聲門區的腫瘤，依聲帶周圍組織顏色特徵，以模糊C均值演算法（Fuzzy C-Means, FCM）進行影像分割，得出聲門與聲帶腫瘤之影像，藉除去聲門影像，有效圈選聲帶腫瘤輪廓，完成喉部腫瘤圈選。
喉內視鏡拍攝影片為每秒30張影像，利用經已偵測之喉部腫瘤系統處理影像，將腫瘤灰階具有幾何形狀尖銳邊緣做索貝爾邊緣檢測運算，形成梯度影像得角點（Corner）特徵，可一次處理10張影像進行圈選腫瘤，相較於傳統處理方式一次只處理一張影像，能快速準確達成圈選腫瘤目標。
本研究以三軍總醫院耳鼻喉頭頸外科部所提供70個病患之喉內視鏡所拍攝動態影片，經與醫師進行本自動偵測喉部腫瘤系統臨床測試驗證，準確辨識率達95.7%，若以病患喉內視鏡所拍攝影片時間為15秒、450張影像為例，本系統處理時間約3分鐘，此結果經臨床測試，確可有效輔助醫生提高診斷效率，減少人為誤判率。

When Otorhinolaryngology doctor uses Laryngoscope to diagnose the patient's larynx, the detection of tumor is based on the geometric features, color and texture. It is combined with medical image processing techniques in this study to detect the location of tumor in dynamic video automatically, and it is circled as prompt, providing the doctor with more information.
Besides normal sharp images, some images taken by laryngoscope are likely to be influenced by saliva, radix linguae vibration and food residue, resulting in image blurring. This study uses K-Means Algorithm to calculate the smoothness and entropy of fuzzy image, and uses Uvula Mask pattern to remove the images touching uvula when laryn-goscope is shooting to screen normal images; and the glottis and epiglottis images have apparent geometric and brightness features for image segmentation. According to the relevant positions of the two areas, the larynx is divided into radix linguae, glottis, epi-glottis and hypopharynx areas.
For the tumor in the radix linguae area, the apparent change in image gray level of the edge is used for Sobel Edge Detection Algorithm operation to form the gradient image, combined with Watershed Algorithm to circle the tumor contour effectively. The tumor in the hypopharynx and epiglottis areas is segmented by using L*a*b* color space, con-verted into gray image, binarized and combined with morphology to find the tumor po-sition. The tumor contour is obtained by Active Contour Algorithm, the smoothness, correlation and entropy of tumor are calculated. The hypopharynx and epiglottis areas are classified. The tumor in the vocal area is segmented by Fuzzy C-Means (FCM) ac-cording to the color feature of tissues around the vocal fold, the images of glottis and vocal fold tumors are obtained, the glottis image is removed to circle the vocal fold tumor contour effectively.
The Laryngoscope can take 30 pictures per second, which are then processed by the verified laryngeal tumor system. Those tumors that are shown to have sharp geometric edges according to the gray level are subject to the Sobel Edge Detection Algorithm. A gradient image is generated in the process, hence the Corner characteristics. In this way, it is possible to process ten images at a time so that tumors could be picked out. Compared with the traditional process in which only one image is processed at a time, the target tumor could be identified in a more rapid and accurate manner.
The dynamic video of 70 patients by laryngoscope provided by Department of Otorhinolaryngology - Head and Neck Surgery, Tri-Service General Hospital is used for clinical testing verification of this automatic larynx tumor detection system, the accurate recognition rate is 95.7%. If the video length of patient Laryngoscope is 15 seconds, 450 images. The processing time of this system is about 3 minutes. This result is tested clin-ically, actually assisting the doctors to increase the diagnosis efficiency and to reduce human misrecognition rate.

[1] 衛生福利部，「中華民國103年版衛生福利年報」，衛生福利部，第15-16頁，臺灣台北，2014。
[2] Study Blue, https://goo.gl/Q1YQqH, 2014。
[3] 王棨德，「圖說聲音障礙常見疾病」，健康世界，第37頁，臺灣台北，2009。
[4] Johns, M. M., “Update on the etiology, diagnosis, and treatment of vocal fold nodules, polyps, and cysts,” Current Opinion in Oto-laryngology and Head and Neck Surgery, Vol. 11, Issue 6, pp. 456-461, 2003.
[5] Wallis, L., Jackson, M. C., Holland, W. and Giraldo, A., “Vocal fold nodule vs. vocal fold polyp: answer from surgical pathologist and voice pathologist point of view,” Journal of Voice, Vol. 18, Issue 1, pp. 125-129, 2004.
[6] Ananthapadmanabha, T. V. and Yegnanarayana, B., “Epoch ex-traction from linear prediction residual for identification of closed glottis interval,” Acoustics, Speech and Signal Processing, Vol. 27, Issue 4, pp. 309-319, 1979.
[7] Ilgner, J. F., Palm, C., Schütz, A. G., Spitzer, K., Westhofen, M. and Lehmann, T. M., “Colour texture analysis for quantitative laryngoscopy,” Acta Oto-Laryngologica, Vol. 123, Issue 6, pp. 730-734, 2003.
[8] Mendez, A., Garcia, B., Ruiz, I. and Iturricha, I., “Glottal area segmentation without initialization using Gabor filters,” In Signal Processing and Information Technology, Sarajevo, pp. 18-22, December 16-19, 2008.
[9] Zhang, Y., Bieging, E., Tsui, H. and Jiang, J. J., “Efficient and ef-fective extraction of vocal fold vibratory patterns from high-speed digital imaging,” Journal of Voice, Vol. 24, Issue 1, pp. 21-29, 2010.
[10] Mendez, A., El, Z. N., Zapirain, B. G. and Elmaghraby, A., “Pathological vocal folds features extraction using a modified ac-tive contour segmentation,” Majlesi Journal of Electrical Engi-neering, Vol. 4, Issue 4, pp. 55-61, 2010.
[11] Verikas, A., Gelzinis, A., Valincius, D., Bacauskiene, M. and Uloza, V., “Multiple feature sets based categorization of laryngeal images,” Computer Methods and Programs in Biomedicine, Vol. 85, Issue 3, pp. 257-266, 2007.
[12] 游崇善、黃聖方、江國賢，「應用腹部電腦斷層掃瞄影像之肝臟腫瘤自動輔助診斷系統」，台灣國際醫學資訊研討會，臺灣台北，2009。
[13] Hwang, S. and Celebi, M. E., “Polyp detection in wireless capsule endoscopy videos based on image segmentation and geometric feature,” IEEE International Conference on In Acoustics Speech and Signal Processing (ICASSP), Dallas, TX, pp. 678-681, March 14-19, 2010.
[14] Yu, L., Yuen, P. C. and Lai, J., “Ulcer detection in wireless capsule endoscopy images,” In Pattern Recognition (ICPR) 2012 21st In-ternational Conference on, Tsukuba, pp. 45-48, November 11-15, 2012.
[15] Li, B. and Meng, M. H., “Tumor recognition in wireless capsule endoscopy images using textural features and SVM-based feature selection,” IEEE Transactions on Information Technology in Bio-medicine, Vol. 16, Issue 3, pp. 323-329, 2012.
[16] Song, C. I., Ryu, C. H., Choi, S. H., Roh, J. L., Nam, S. Y. and Kim, S. Y., “Quantitative evaluation of vocal‐fold mucosal irreg-ularities using GLCM‐based texture analysis,” The Laryngoscope, Vol. 123, Issue 11, pp. 45-50, 2013.
[17] Szczypiński, P., Klepaczko, A., Pazurek, M. and Daniel P., “Tex-ture and color based image segmentation and pathology detection in capsule endoscopy videos,” Computer Methods and Programs in Biomedicine, Vol. 113, Issue 1, pp. 396-411, 2014.
[18] Turkmen, H. I., Karsligil, M. E. and Kocak, I., “Classification of laryngeal disorders based on shape and vascular defects of vocal folds,” Computers in Biology and Medicine, Vol. 62, pp. 76-85, 2015.
[19] Masson, N., Nageotte, F., Zanne, P. and Mathelin, M., “In vivo comparison of real-time tracking algorithms for interventional flexible endoscopy,” In Biomedical Imaging: From Nano to Macro, 2009. ISBI'09. IEEE International Symposium on, Boston, MA, pp. 1350-1353, June 28, 2009.
[20] Dong, Z. and Wang, Z., “Screening of capsule endoscopy redun-dant image based on image motion model by block matching,” In Electrical and Control Engineering (ICECE), 2011 International Conference on, Yichang, pp. 5436-5439, September 16-18, 2011.
[21] Yanagawa, Y., Echigo, T., Vu, H., Okazaki, H., Fujiwara, Y., Arakawa, T. and Yagi, Y., “Tracking abnormalities in video cap-sule endoscopy using surrounding features with a triangular con-straint,” In Biomedical Imaging (ISBI), 2012 9th IEEE Interna-tional Symposium on, Barcelona, pp. 578-581, May 2-5, 2012.
[22] Cai, Z., Hu, C., Lin, W. and Yang, W. A., “A feature point extrac-tion and matching algorithm for wireless capsule endoscope im-age,” Information and Automation (ICIA), 2013 IEEE Interna-tional Conference on, Yinchuan, pp. 608-613, August 16-28, 2013.
[23] E Doctor Online, http://goo.gl/k1WBVT, 2014。
[24] Glottic Stenosis, http://goo.gl/TWUl9G, 2011。
[25] Michor, F., Iwasa, Y. and Nowak, M. A., “Dynamics of cancer progression,” Nature Reviews Cancer, Vol. 4, Issue 3, pp. 197-205, 2004.
[26] 閔華慶，「家庭實用醫學叢書4-腫瘤」，第2-11頁，牛頓出版公司，臺灣台北，1989。
[27] 余達德，「家庭實用醫學叢書6-耳鼻咽喉科」，第68-81頁，牛頓出版公司，臺灣台北，1989。
[28] 戴新民，「最新腫瘤防治」，第257-262頁，啟業書局，臺灣台北，1987。
[29] National Cancer Institute, http://www.cancer.gov/, 2015。
[30] 蒼州醫學高等專科學校，http://goo.gl/NG1qJH，2007。
[31] Edge, S. B. and Compton, C. C., “The American Joint Committee on Cancer: the 7th edition of the AJCC cancer staging manual and the future of TNM,” Annals of Surgical Oncology, Vol. 17, Issue 6, pp. 1471-1474, 2010.
[32] Pott, L. M. and Murray, W. B., “Review of video laryngoscopy and rigid fiberoptic laryngoscopy,” Current Opinion in Anesthesi-ology, Vol. 21, Issue 6, pp. 750-758, 2008.
[33] Wei, J., Wang, D., Yanwei, W., Jin, L., Wang, L. and Hong, L., “Industrial X-ray image enhancement algorithm based on AH and MSR,” Engineering, Vol. 3, Issue 10, pp. 1040-1044, 2011.
[34] Abdallah, R., Galway, U., You, J., Kurz, A., Sessler, D. I. and Doyle, D. J., “A randomized comparison between the Pentax AWS video laryngoscope and the Macintosh laryngoscope in morbidly obese patients,” Anesthesia and Analgesia, Vol. 113, Issue 5, pp. 1082-1087, 2011.
[35] Reza, A. M., “Realization of the Contrast Limited Adaptive His-togram Equalization (CLAHE) for real-time image enhancement,” Journal of VLSI Signal Processing Systems for Signal, Image and Video Technology, Vol. 38, Issue 1, pp. 35-44, 2004.
[36] Gao, W., Zhang, X., Yang, L. and Liu, H., “An improved Sobel edge detection,” Computer Science and Information Technology (ICCSIT), 2010 3rd IEEE International Conference on, Chengdu, pp. 67-71, July 9-11, 2010.
[37] Otsu, N., “A threshold selection method from gray-scale histo-gram,” IEEE Transactions on Systems, Man, and Cybernetics, Vol. 9, Issue 1, pp. 62-66, 1979.
[38] Kass, M., Witkin, A. and Terzopoulos, D., “Snakes: active contour models,” International Journal of Computer Vision, Vol. 1, Issue 4, pp. 321-331, 1988.
[39] Vincent, L. and Soille, P., “Watersheds in digital spaces: an effi-cient algorithm based on immersion simulations,” IEEE Transac-tions on Pattern Analysis and Machine Intelligence, Vol. 13, Issue 6, pp. 583-598, 1991.
[40] Forgy, E. W., “Cluster analysis of multivariate data: efficiency versus interpretability of classifications,” Biometrics, Vol. 21, pp. 768-769, 1965.
[41] Dunn, J. C., “A fuzzy relative of the ISODATA process and its use in detecting compact well-separated clusters,” Journal of Cyber-netics, Vol 3, Issue 3, pp 32-57, 1973.
[42] Bezdek, J. C., Ehrlich, R. and Full, W., “FCM: the fuzzy c-means clustering algorithm,” Computers and Geosciences, Vol. 10, Issue 2, pp. 191-203, 1984.