Lohscheller Jörg, Toy Hikmet, Rosanowski Frank, Eysholdt Ulrich, Döllinger Michael
Department of Phoniatrics and Pediatric Audiology, Bohlenplatz 21, 91054 Erlangen, Germany.
Med Image Anal. 2007 Aug;11(4):400-13. doi: 10.1016/j.media.2007.04.005. Epub 2007 Apr 29.
Investigation of voice disorders requires the examination of vocal fold vibrations. State of the art is the recording of endoscopic high-speed movies which capture vocal fold vibrations in real-time. It enables investigating the interrelation between disturbances of vocal fold vibrations and voice disorders. However, the lack of clinical studies and of a standardized procedure to reconstruct vocal fold vibrations from high-speed videos constrain the clinical acceptance of the high-speed technique. An image processing approach is presented that extracts the vibrating vocal fold edges from digital high-speed movies. The initial segmentation is principally based on a seeded region-growing algorithm. Even in movies with low image quality the algorithm segments successfully the glottal area by an introduced two-dimensional threshold matrix. Following segmentation, the vocal fold edges are reconstructed from the computed time-varying glottal area. The performance of the procedure was objectively evaluated within a study comprising 372 high-speed recordings. The accuracy of vocal fold reconstruction exceeds manual segmentation results obtained by clinical experts. The algorithm reaches an information flow-rate of up to 98 images per second. The robustness and high accuracy of the procedure makes it suitable for the application in clinical routine. It enables an objective and highly accurate description of vocal fold vibrations which is essential to realize extensive clinical studies which focus on the classification of voice disorders.
嗓音障碍的研究需要检查声带振动。目前的技术水平是记录内镜高速电影，它能实时捕捉声带振动。这使得研究声带振动紊乱与嗓音障碍之间的相互关系成为可能。然而，缺乏临床研究以及从高速视频重建声带振动的标准化程序限制了高速技术在临床上的应用。本文提出了一种图像处理方法，可从数字高速电影中提取振动的声带边缘。初始分割主要基于种子区域生长算法。即使在图像质量较低的电影中，该算法也能通过引入的二维阈值矩阵成功分割声门区域。分割之后，根据计算出的随时间变化的声门区域重建声带边缘。在一项包含372个高速记录的研究中，对该程序的性能进行了客观评估。声带重建的准确性超过了临床专家手动分割的结果。该算法的信息流速率高达每秒98帧图像。该程序的稳健性和高精度使其适用于临床常规应用。它能够对声带振动进行客观且高度准确的描述，这对于开展专注于嗓音障碍分类的广泛临床研究至关重要。
