Machine learning based identification of relevant parameters for functional voice disorders derived from endoscopic high-speed recordings

Abstract

In voice research and clinical assessment, many objective parameters are in use. However, there is no commonly used set of parameters that reflect certain voice disorders, such as functional dysphonia (FD); i.e. disorders with no visible anatomical changes. Hence, 358 high-speed videoendoscopy (HSV) recordings (159 normal females (N f), 101 FD females (FD f), 66 normal males (N M), 32 FD males (FD M)) were analyzed. We investigated 91 quantitative HSV parameters towards their significance. First, 25 highly correlated parameters were discarded. Second, further 54 parameters were discarded by using a LogitBoost decision stumps approach. This yielded a subset of 12 parameters sufficient to reflect functional dysphonia. These parameters separated groups N f vs. FD f and n M vs. FD M with fair accuracy of 0.745 or 0.768, respectively. Parameters solely computed from the changing glottal area waveform (1D-function called GAW) between the vocal folds were less important than parameters describing the oscillation characteristics along the vocal folds (2D-function called Phonovibrogram). Regularity of GAW phases and peak shape, harmonic structure and Phonovibrogram-based vocal fold open and closing angles were mainly important. This study showed the high degree of redundancy of HSV-voiceparameters but also affirms the need of multidimensional based assessment of clinical data.

Interdependencies between acoustic and high-speed videoendoscopy parameters

“…Further, subject age differed between healthy and disordered groups. Albeit we found no strong influence of subject age in a previous study [ 38 ], the influence of age on voice parameters is well documented in literature [ 66 – 68 ] and may have influenced the results.…”

“…This may then lead to changes in the shape of the parameter distribution in comparison to healthy subjects. In summary, single parameters are not suitable for differentiating healthy from FD subjects and multi-parametric approaches are needed as suggested before [ 38 , 59 , 60 ]. However, if not FD in general but subcategories of FD (e.g.…”

“…HSV recordings (N: 351) with simultaneously recorded acoustic signal (time-synchronized) were used for data evaluation. This data (without the acoustic recordings) was already used in a previous study applying machine learning approaches for classification purposes [ 38 ]. All 351 acoustic recordings were unanimously rated by three experts on ordinal scales (0 to 2) for signal noise and background noise: 0 was chosen as the best rating (no signal noise / background noise) and 2 as the worst (strong signal noise / background noise).…”

“…It is the next version of GAT-2018, and includes several bug fixes and an improved cycle detection algorithm. The process of segmentation and parameter calculation is illustrated in Fig 2 . For a detailed explanation of the segmentation process see [38]. GAWs describing the total glottal area (GAW T ) and the left and right half of this glottal area (GAW L and GAW R ) were extracted from HSV videos.…”

Maximum and Minimum Phonatory Glottal Area before and after Treatment for Vocal Nodules

“…Montage images are created using videostroboscopy, and the maximum and minimum GAs are calculated using Image J software. High-speed videoendoscopy is an alternative assessment technique commonly used in the field of laryngology [14][15][16]. However, videostroboscopy has several clinical advantages.…”