Details of the participants are presented in Table 1. The study included patients with a wide spectrum of voice disorders, encompassing functional, neurologic, and morphological lesions such as muscle tension dysphonia, vocal fold palsy, tremor, spasmodic dysphonia, presbyphonia, nodules, polyps, cysts, and Reinke’s edema.

Among patients who visited to Pusan National University Yangsan Hospital, 102 patients with voice disorders diagnosed by laryngoscopy and pathological examination were registered (female=51, male=51). The distribution of age was 24 to 74 years for female, with an average age of 52.5 years, and 32 to 85 years old for male, with an average age of 62.8 years. This study was approved by the Institutional Review Board of Pusan National University Yangsan Hospital (IRB No. 55-2023-050).

Details of the diagnostic categories are summarized in Table 1. Functional and neurologic disorders accounted for 67.6% of cases, whereas morphological or structural lesions accounted for 32.4%. The clinical diagnosis established at the time of otolaryngology consultation was regarded as the reference standard for subsequent analysis.

An HSV system (USC-700MFC, U-medical) was used to assess vocal fold vibration. A 9.2 mm diameter, 70° rigid laryngoscope (8706 CA, Karl Storz), a zoom coupler (f=16-34 mm, MGB), and a 300-watt xenon light source (NOVA 300, Storz) were used for examination. Images were recorded at the rate of 1500-2000 frames per second, with a spatial resolution of 208×304 pixels. The total recording time was 10 s, and the total encoding time was less than 2 min. The size of the acquired image file was approximately 70 MB. The laryngoscope connected to HSV camera was inserted through the oral cavity to examine the vocal folds. All patients were instructed to phonate a sustained vowel /e/ or /i/ with a comfortable pitch for 3 s, followed by a high pitch for 2 s and a low pitch for 2 s. Subsequently, the patients were directed to produce 3 to 4 consecutive shouts. The patients’ voices were recorded simultaneously during the HSV examination. The microphone for voice recording was positioned approximately 10-15 cm away from the patient. Caution was exercised to avoid contact with the oral and oropharyngeal mucosa because the distal end of the laryngoscope was hot. After the HSV examination, a post-processing procedure was conducted to obtain a multi-line scanning DKG. We developed software for the automatic scrolling DKG with synchronized voice playback for the viewer program. Multi-slice DKG imaging consists of still image files in JPEG format, with the time information for each frame stored in a separate file. The DKG files in the JPEG format had a size of 30328×1880 pixels. Using this system, the maximum recording time is 10 s, and up to 10 scan lines can be displayed simultaneously.

To generate a multi-slice digital kymograph, the obtained HSV image was loaded and 10 lines were drawn perpendicular to the glottal axis. Automatic scrolling of the DKG files made it possible to view all files within 10 s. At the same time, automated DKG with synchronized voice playback was implemented by dubbing the prerecorded waveform audio format (WAV) files. This software was developed as a Windows desktop application using C# language. The user interface of the developed automated DKG with synchronized voice playback is shown in Fig. 1 (Supplementary Videos 1-7).

To facilitate diagnostic interpretation, representative multiline kymographic characteristics of functional dysphonia are illustrated in Fig. 2. These images presented as reference materials to the raters prior to independent evaluation.

From each HSV recording, three sets of multi-line DKG materials were generated: 1) DKG without voice playback, 2) DKG with non-synchronized voice playback, and 3) DKG with synchronized voice playback. Except for the presence or absence of audio, all visual parameters and display settings were identical across the three conditions. For diagnostic assessment, raters were instructed to provide the diagnostic label in an open-ended manner to avoid bias and to reflect real- world clinical decision-making.

A Z-test was performed to determine the differences in diagnostic accuracy between the methods (Table 2). Statistical analyses of Cohen’s kappa, Fleiss’ kappa, and the Z-test were performed using SPSS software (version 22.0; IBM Corp.).

Intra-rater reliability was calculated using Cohen’s kappa based on test-retest results obtained at 3-week intervals. For Rater 1, the coefficients were 0.719 for DKG, 0.720 for DKG with non-synchronized playback, and 0.832 for DKG with synchronized playback, showing a statistically significant improvement under the synchronized condition. Rater 2 recorded coefficients of 0.722, 0.696, and 0.856, respectively, also indicating a significant advantage for the synchronized condition. Rater 3 obtained coefficients of 0.832, 0.856, and 0.759, with significant differences across conditions, although the direction of change varied. Overall, these results suggest substantial intra-rater reliability (Table 3).

Inter-rater reliability was assessed using Fleiss’ kappa. The coefficients were 0.650 for DKG, 0.624 for DKG with non-synchronized playback, and 0.658 for DKG with synchronized playback, indicating moderate to substantial agreement across raters (Table 4).

As shown in Table 2, diagnostic accuracy was highest for DKG with synchronized voice playback (76.1%), followed by DKG with non-synchronized playback (58.1%) and conventional DKG (41.5%). Pairwise comparisons showed that synchronized playback significantly outperformed both conventional (p=0.0319) and non-synchronized DKG (p=0.0001), while non-synchronized playback also exceeded conventional DKG (p=0.0425).

Notably, diagnostic accuracy under the synchronized condition was almost twice that of conventional DKG, underscoring the added value of auditory feedback in complementing visual analysis. Improvements were particularly evident in functional disorders such as spasmodic dysphonia and tremor, where irregular closure or periodic fluctuations could be more reliably identified when the acoustic breaks were simultaneously heard.

Reliability analyses further supported these findings (Tables 3 and 4). For intra-rater reliability, Cohen’s kappa coefficients ranged from 0.696 to 0.856, with the highest values under the synchronized condition (0.832, 0.856, and 0.759 for Raters 1-3, respectively). These results suggest that synchronized playback reduces subjective variability and enhances consistency in interpretation, as intra-rater agreement reached the “almost perfect” range. Inter-rater reliability assessed with Fleiss’ kappa was moderate to substantial across conditions (0.650 for conventional DKG, 0.624 for non-synchronized playback, and 0.658 for synchronized playback).

In addition, representative multi-line DKG patterns were observed for different disorders (Supplementary Videos 2-7). In presbyphonia, the left and right vocal folds vibrated in phase, but glottic insufficiency due to vocal muscle atrophy was apparent. Tremor cases exhibited irregular vibration patterns with fluctuations in amplitude and timing, sometimes failing to complete opening or closure. Spasmodic dysphonia showed intermittent spasms leading to irregular amplitude and closure timing, with asymmetry between upper/lower and left/right folds. In vocal fold paralysis, marked phase differences, asymmetric mucosal waves, and incomplete closure were observed. Reinke’s edema resulted in thickened folds with incomplete or slow closure and reduced amplitude, usually symmetrical but sometimes slightly asymmetric depending on severity. Vocal fold cysts caused phase asymmetry and reduced amplitude on the affected side, along with disrupted mucosal waves. These representative findings demonstrate that multi-line DKG provides visual cues that support the diagnosis of various voice disorders.