Ulrich Eysholdt 1 , Michael Doellinger 1 , Joerg Lohscheller 1

1 University Hospital, Dept. Phoniatrics, Erlangen

The source of voice sound is the vibration of the vocal folds. In case of any voice disorder the examination of the sound source is mandatory. While the most wide-spread stroboscopy is very useful in normal or near-normal voice, i.e. harmonic, periodic vibrations, it fails more and more with increasing vibration pathology. High speed recordings (HSR) are able to visualize irregular vibrations with real time resolution. However, due to expenditure of time HSR are difficult to handle in a clinical environment. Even when replayed at slow motion high speed recordings are less suggestive for the examiner, as motion is less obvious to human perception than static object features are. Under the name phonovibrogram (PVG) we developed an image processing procedure to make HSR clinically more applicable. The PVG concentrates the motion of a complete recording into one single image at the cost of loss of the direct anatomical information. PVG shows a remarkable discriminatory power in voice pathology. It works like a fingerprint of the vibration and - at the first time - allows to automatically classify irregular vibrations with very high reliability from endoscopic HSR.

Leonardo Bocchi 1 , Marcello Calisti 1 , Claudia Manfredi 1 , Lorenzo Mirabile 2 , Giorgio Peretti 3

1 Università degli Studi di Firenze, Electronics and Telecommunications, Firenze

2 Children Hospital A. Meyer, , Firenze

3 Spedali Civili, , Brescia

Videokymography (VKG) is a recently developed but emerging imaging technique for high speed visualisation of vocal folds vibration (8000 frames/s). Thanks to its high-resolution and low-cost characteristics, it has gained more and more interest in the clinical field, especially in case of irregular movement of the vocal folds. This growing interest has motivated efforts in the development of specific imaging techniques, capable to extract relevant objective information from VKG recordings, such as time and amplitude measures of asymmetry of vibration.

Recently, a new VKG device has been developed, with improved image quality with respect to the previous ones, that allows the visualisation of an uninterrupted VKG strip instead of a sequence of images of very short duration (15-20 ms) separated by black rows. Moreover, the new device is suitable for usage also on a flexible endoscope, thus allowing performing VKG exams on a wider range of patients, children included.

The aim of this work is to present a new tool (named VKG-Analyser) devoted to VKG image analysis, which has been provided with a user-friendly interface that makes VKG-Analyser usable also by non-experts in the daily routine. The tool has been developed in strict co-operation with clinicians that routinely use videokymography in the hospital. Its flexible structure allows for images retrieval and analysis, and storage of results in a very simple way.

A new feature consist in an automatic selection of VKG frames from the whole recording, based on Fast Fourier Transforming (FFT) each column of the frames, to point out those frames that contain any periodicity. The procedure gives a "spectral map" that allows for selecting the VKG frames from the whole sequence. This greatly speeds up the whole computational time.

Due to possible strong irregularities of pathological vocal cycles, the analysis of videokimograms presented significant challenges for computer analysis. By applying robust contour detection algorithms, suitably modified to deal with VKG images, first reliable results have been obtained and tested on simulated images. The underlying image analysis techniques require low computer time, allowing for an almost real-time analysis and tracking relevant parameters.

Experiments have been performed both with rigid and flexible endoscopes. Specifically, flexible endoscope has been used at the Children Hospital A. Meyer, in Firenze, Italy, even on 1-2 years old children. Though its optimal usage requires a suitable training period, first results are promising.

Thomas Woellner 1 , Andreas Krenmayr 2 , Nicola Supper 1 , Patrick Zorowka 1

1 Medical University Innsbruck, Clinic for Hearing, Voice and Speech Disorders, Innsbruck

2 University of Innsbruck, C. Doppler Laboratory for Active Implantable Systems, Institute of Ion Physics and Applied Physics, Innsbruck

OBJECTIVES:

Analysis algorithms for digital high-speed videolaryngoscopy (HSV) offer a range of techniques to depict functional properties of the vocal fold vibration in single images. The present study evaluates an algorithm based on Fourier analysis, which depicts phase relations by means of hue within grayscale snapshots of the vocal folds. These false color Fourier Images (FI) are suitable for displaying glottal insufficiencies and phase asymmetries between parts of the vocal folds. The findings in normophonic, impaired and professional voices are compared to videolaryngostroboscopy (VLS) and acoustic voice assessment.

METHODS:

Recordings of the vocal fold vibration in adults with healthy, impaired and professional voices were made with a standard VLS unit (Stroboskop 5052, Richard Wolf GmbH) and a digital HSV unit (HRES-Endocam 5562, Richard Wolf GmbH), capable of capturing 4000 color images per second during a recording time of 2 seconds. Additionally a Speech and Language Therapist assessed the voices by means of acoustic voice analysis and the Voice Disorder Index 12 (VDI-12) questionnaire. For analyzing the HSV datasets a tracking algorithm tracked the vocal folds' movements within the single images, serving as a basis for an endoscope motion compensation algorithm. Subsequently kymograms and Fourier Images were generated and functional parameters like the open quotient, the glottal area modulation coefficient were calculated and the vibration of amplitude, the phase difference along and between the vocal folds, the glottal insufficiency and the anterior-posterior modes (ap-modes) were assessed.

RESULTS:

The findings of the VLS and the acoustic voice analysis served as a reference to confirm that the voices examined in the first group were normophonic. The Fourier Images proved to be able to display even those phase differences which are below the detection limit with VLS. Both dorsoventral (ap-modes) as well as transversal phase differences were found and quantified for healthy voices in demarcation to pathological voices.

CONCLUSIONS:

Combined with kymograms the Fourier Image seems to constitute a useful technique to visualize and quantify functional properties of the vocal fold oscillations. The method is especially sensible to swing on and off phase and to phase differences, either transversal (between both vocal folds) or dorsoventral in the case of ap-modes. The problem with former local reflections leading to artefacts in the fourier image could solved by new software. The data gathered in this work can serve as a reference for the amount of phase asymmetry which has to be expected in the normophonic condition. The investigation shows the transition from the physiological to the pathological range in the above mentioned parameters (on trial new defined reference values).

Keywords: stroboscopy, high-speed videolaryngoscopy, vocal fold vibration, fourier analysis