Comparison of artificial intelligence assessment and clinician assessment in perceptual assessment of voice disorders

The combined use of acoustic, aerodynamic, and perceptual analyses in the assessment of voice disorders enables accurate diagnosis and effective treatment planning by addressing the multidimensional structure of the voice. With advancements in artificial intelligence, studies are being conducted to standardize and streamline the diagnosis and therapy process. For this purpose, the AI-supported GRBASZero application scores the GRBAS scale used in perceptual voice assessment. This study aimed to compare the GRBAS scores of individuals with voice disorders as assessed by clinicians with the AI-supported GRBASZero application. Sixty dysphonic and 15 healthy individuals participated in the study. Healthy individuals were found to influence the correlation, and 60 dysphonic participants were included in the analysis. Participants' voices were scored with the GRBASZero application, and the Voice Handicap Index-10, Voice-Related Quality of Life Scale, Turkish Vocal Fatigue Scale and Vocal Tract Disorder Scales were also administered. Participants were sampled for maximum /a/, /i/, /s/, /z/ phonations and reading passages were analyzed using Praat software. The recordings were evaluated using the GRBAS scale by 13 Speech and Language Pathologists. The study found a significant correlation with a moderate correlation between clinician and GRBASZero assessments for the G, R, B, and S parameters. A significant correlation was found for the A parameter with a low correlation. When acoustic and aerodynamic analyses were compared with GRBAS parameters, no significant differences were found between clinician and AI assessments. When self-perceptual scales were compared with GRBAS parameters, significant correlations were generally found in clinician assessments, but rarely in AI assessments. Consequently, a near-moderate level of significance was found between clinician and AI assessments. This provides hope for the development of AI in diagnosis and intervention. Further studies are needed in the future.