Search

High-speed videoendoscopy is a major tool for quantitative laryngology. Glottis segmentation and glottal midline detection are crucial for computing vocal fold-specific, quantitative parameters. However, fully automated solutions show limited clinical applicability. Especially unbiased glottal midline detection remains a challenging problem. We developed a multitask deep neural network for glottis segmentation and glottal midline detection. We used techniques from pose estimation to estimate the anterior and posterior points in endoscopy images. Neural networks were set up in TensorFlow/Keras and trained and evaluated with the BAGLS dataset. We found that a dual decoder deep neural network termed GlottisNetV2 outperforms the previously proposed GlottisNet in terms of MAPE on the test dataset (1.85% to 6.3%) while converging faster. Using various hyperparameter tunings, we allow fast and directed training. Using temporal variant data on an additional data set designed for this task, we can improve the median prediction accuracy from 2.1% to 1.76% when using 12 consecutive frames and additional temporal filtering. We found that temporal glottal midline detection using a dual decoder architecture together with keypoint estimation allows accurate midline prediction. We show that our proposed architecture allows stable and reliable glottal midline predictions ready for clinical use and analysis of symmetry measures.

Background: Various voice assessment tools, such as questionnaires and aerodynamic voice characteristics, can be used to assess vocal function of individuals. However, not much is known about the best combinations of these parameters in identification of functional dysphonia in clinical settings., Methods: This study investigated six scores from clinically commonly used questionnaires and seven acoustic parameters. 514 females and 277 males were analyzed. The subjects were divided into three groups: one healthy group (N 01 ) (49 females, 50 males) and two disordered groups with perceptually hoarse (FD 23 ) (220 females, 96 males) and perceptually not hoarse (FD 01 ) (245 females, 131 males) sounding voices. A tree stumps Adaboost approach was applied to find the subset of parameters that best separates the groups. Subsequently, it was determined if this parameter subset reflects treatment outcome for 120 female and 51 male patients by pairwise pre- and post-treatment comparisons of parameters., Results: The questionnaire "Voice-related-quality-of-Life" and three objective parameters ("maximum fundamental frequency", "maximum Intensity" and "Jitter Percent") were sufficient to separate the groups (accuracy ranging from 0.690 (FD 01 vs. FD 23 , females) to 0.961 (N 01 vs. FD 23 , females)). Our study suggests that a reduced parameter subset (4 out of 13) is sufficient to separate these three groups. All parameters reflected treatment outcome for patients with hoarse voices, Voice-related-quality-of-Life showed improvement for the not hoarse group (FD 01 )., Conclusion: Results show that single parameters are insufficient to separate voice disorders but a set of several well-chosen parameters is. These findings will help to optimize and reduce clinical assessment time.