Digital image analysis of the larynx

Objective quantification of normal and pathological laryngeal dimensions using videolaryngostroboscopic images has been an important clinical issue for speech-language pathologists. It is necessary to have a tool designed to objectively document changes in laryngeal structure with treatment. Documenting changes in laryngeal structure as a function of a particular surgical procedure or medication would assist physicians and speech-language pathologists in documenting the outcomes of their therapy regimens. The current and most widely used method for documenting the status of the laryngeal structure is endoscopy, with video recording of digital imagery for archival purposes. Unfortunately, the in vivo determination of laryngeal anatomy and dimensions has been seriously hampered by difficulties with image distortion from the endoscopic technique. Currently, researchers and clinicians are using commercially available products to assist in image analysis of the larynx. Unfortunately, the commercially available imaging software packages (for example NIH Image and Adobe PhotoShop) have serious inherent artifacts that remain unresolved, such as the lack of automated glottal gap tracing, the lack of ability to account for color and illumination distortion, and an inability to account for the problems of a multi-pixel boundary gradient. This mixed design repeated-measures study examined three imaging software packages for their usefulness in laryngeal image analysis. Specifically, Adobe PhotoShop, Gimp, and Scion Image were examined in terms of inter-program and inter-analyst consistency in the measurement of glottal gap area (GGA), and total vocal fold length (TVFL). Three endoscopic image frames of a normal subject's larynx and three endoscopic image frames of a patient with the appropriate diagnosis of adductor vocal fold paralysis during voicing were used as the samples for analysis. One condition was analyzed: maximum opening during voicing. Each image frame of the subjects' vocal folds was measured ten times by three analysts, two experts and one naïve.; The findings of this study for GGA area measurement suggest that the program used and the condition have an effect on the consistency of the results. Also, the combination of analyst and condition has an effect on such measurement. Measurement results for TVFL suggest that there were combined effects of analyst and condition, program and condition, and program and analyst on the consistency of the resulting measurements.; This study recommends investigating Scion Image thoroughly as a potential platform for the development of the new image analysis tool. The design of the new tool should analyze the effects of image noise as well as computational and systematic error on the accuracy of measured results.