Publication: BAYESIAN ESTIMATION OF A SUBJECT-SPECIFIC MODEL OF VOICE PRODUCTION FOR THE CLINICAL ASSESSMENT OF VOCAL FUNCTION
Loading...
Date
2017
Authors
GALINDO FLORES, GABRIEL EDUARDO
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Los problemas de voz son conocidos por afectar negativamente a la comunicación, la interacción social, a las funciones laborales y a la calidad de vida. Se cree que los problemas vocalesson condiciones recurrentes o crónicas que están relacionadas a patrones perjudiciales, lo que esconocido como hyperfuncionalidad vocal. Se cree que las principales causas biomecanicas de laformación de lesiones benignas en las cuerdas vocales (ej. nódulos y pólipos) y de las disfoniatenso-musculares son consecuencia de la hyperfuncionalidad vocal. A pesar de la prevalencia deestos trastornos, se conoce muy poco acerca de los mecanismos subyacentes de la hiperfunciónvocal, los cuales argumentamos que pueden describirse cualitativamente a trav´es del modeladonumérico basado en la física de la función vocal. Exploramos esta idea desarrollando un modelomejorado de producción de voz y un marco probabilista para estimar sus parámetros. Si bien seha prestado cada vez más atención a los modelos de producción de voz que pretenden describirel comportamiento fisiológico general, se ha prestado poca atención a los modelos de un sujetoespecifico. Se espera que el desarrollo del modelado sujeto-específico mejore el análisis clínico dela función vocal. En la primera parte de esta tesis, se propone un modelo triangular de coberturay cuerpo de la cuerdas vocales para la producción de voz, permitiendo captura las característicasclave de la hiperfunción vocal. Se consideraron importantes características anatómicas paramejorar la relación modelo-fisiología del modelo estándar body-cover model. El modelo propuestodescribe eficazmente el cierre glotal incompleto membranoso y cartilaginoso. La relevancia fisiológica del modelo triangular propuesto se explora en el contexto de la hiperfunción vocal, enla que los mecanismos compensatorios y la retroalimentación auditiva aumentan la presión decontacto de los pliegues vocales junto con las medidas clínicas relacionadas (por ejemplo, flujoinestable y tasa de declinación de flujo máximo).Utilizando el modelo propuesto, se desarrolló y probó un método de estimación Bayesianomultimodal utilizando tanto datos clínicos como datos generados sintéticamente. El uso de dichomecanismo de identificación de sistemas permite construir un modelo de sujeto específico basadoen un marco probabilístico, extendiendo as´i esfuerzos previos que sólo producen estimacionesdeterministas. El estimador propuesto permitió con éxito la identificación de los parámetros delmodelo y sus intervalos de credibilidad. Además, dado que el método pose una variación natural en el tiempo, también se obtuvieron señales relacionadas con el modelo que no son observablesen la configuración clínica (por ejemplo, presión de contacto de los pliegues vocales).Para la evaluación de los datos clínicos, se utilizó como fuente de información la videoendoscopiade alta velocidad y las mediciones de flujo glotal. La predicción y el contraste se hicieronusando una señal de micrófono adquirida en sincronía con las otras mediciones. La señal demicrófono estimada asemejó la medida clínica dentro de sus bandas de credibilidad. Por lo tanto,se ilustra el potencial del método propuesto para construir y extraer información adicional relacionadacon el modelo de las configuraciones clínicas. Las estimaciones de la presión de contactoobtenida con la técnica propuesta se ajustaron estrechamente a estudios previos. Así, el marcode modelaje propuesto permite investigar los vínculos causales entre la patología y/o compensación del paciente y la hiperfunción vocal, tanto por elucidar algunos de los mecanismos físicossubyacentes como también por proporcionar nuevas medidas de la función vocal, las cuales sondifíciles, si no imposibles, de obtener en ambientes clínicos.
Voice problems are known to negatively affect communication, social interaction, work-relatedfunctions and quality of life. Common voice disorders are believed to be chronic or recurrentconditions related to detrimental voice patterns, which are referred to as vocal hyperfunction.It is believed that the main biomechanical causes of benign vocal folds lesions formation (e.g.,nodules and polyps) and muscle tension dysphonia are a consequence of vocal hyperfunction. Inspite of the prevalence of these disorders, very little is known about the underlying mechanismsof vocal hyperfunction, which we argue that can be qualitatively described through physic-basednumerical modeling of vocal function. We explore this idea by developing an enhanced modelof voice production and a probabilistic framework to estimate its parameters. While increasingattention has been placed on voice production models that aim to describe general physiologicalbehavior, little attention has been placed on single-subject models. The development of subjectspecificmodeling is expected to enhance clinical analysis of vocal function. In the first part ofthis thesis, a triangular body-cover model of voice production that capture the key features ofvocal hyperfunction is proposed. Several important anatomical characteristics were considered toimprove the model-physiology relation of the standard body-cover model. The proposed modeleffectively describes membranous and cartilaginous incomplete glottal closure. The physiologicalrelevance of the proposed triangular model is explored in the context of vocal hyperfunction,where compensatory mechanisms and auditory feedback yield increased contact pressure of thevocal folds, and related clinical measures (e.g., unsteady flow, and maximum flow declinationrate).Using the proposed model, a multi-modal Bayesian estimation method was developed andtested using synthetically generated and clinical data. The use of such system-identification mechanismallows for constructing a subject-specific model based on a probabilistic framework, thusextending previous efforts that only produce deterministic point-based estimates. The proposedestimator successfully allowed for the identification of model parameters, and their associatedcredibility intervals. In addition, given the time-varying nature of the method, model-relatedsignals that are non-observable in clinical setup were also obtained (e.g., vocal folds contactpressure).For the assessment of clinical data, high-speed videoendoscopy and glottal flow measurementswere used as information source. Prediction and contrast was made using a separate microphonesignal acquired in synchrony with the other measurements. The estimated microphone signalmatched the clinical measurement within its credibility bands, illustrating the potential of theproposed method to construct and extract additional model-related information from clinicalsetups. Estimates of contact pressure obtained with the proposed technique closely matchedprevious studies. Thus, the proposed numerical modeling framework enables the investigation ofcausal links between patient pathology/compensation and vocal hyperfunction by both elucidatingsome of the underlying physical mechanisms and providing new measures of vocal functionthat are difficult, if not impossible, to obtain in clinical setups.
Voice problems are known to negatively affect communication, social interaction, work-relatedfunctions and quality of life. Common voice disorders are believed to be chronic or recurrentconditions related to detrimental voice patterns, which are referred to as vocal hyperfunction.It is believed that the main biomechanical causes of benign vocal folds lesions formation (e.g.,nodules and polyps) and muscle tension dysphonia are a consequence of vocal hyperfunction. Inspite of the prevalence of these disorders, very little is known about the underlying mechanismsof vocal hyperfunction, which we argue that can be qualitatively described through physic-basednumerical modeling of vocal function. We explore this idea by developing an enhanced modelof voice production and a probabilistic framework to estimate its parameters. While increasingattention has been placed on voice production models that aim to describe general physiologicalbehavior, little attention has been placed on single-subject models. The development of subjectspecificmodeling is expected to enhance clinical analysis of vocal function. In the first part ofthis thesis, a triangular body-cover model of voice production that capture the key features ofvocal hyperfunction is proposed. Several important anatomical characteristics were considered toimprove the model-physiology relation of the standard body-cover model. The proposed modeleffectively describes membranous and cartilaginous incomplete glottal closure. The physiologicalrelevance of the proposed triangular model is explored in the context of vocal hyperfunction,where compensatory mechanisms and auditory feedback yield increased contact pressure of thevocal folds, and related clinical measures (e.g., unsteady flow, and maximum flow declinationrate).Using the proposed model, a multi-modal Bayesian estimation method was developed andtested using synthetically generated and clinical data. The use of such system-identification mechanismallows for constructing a subject-specific model based on a probabilistic framework, thusextending previous efforts that only produce deterministic point-based estimates. The proposedestimator successfully allowed for the identification of model parameters, and their associatedcredibility intervals. In addition, given the time-varying nature of the method, model-relatedsignals that are non-observable in clinical setup were also obtained (e.g., vocal folds contactpressure).For the assessment of clinical data, high-speed videoendoscopy and glottal flow measurementswere used as information source. Prediction and contrast was made using a separate microphonesignal acquired in synchrony with the other measurements. The estimated microphone signalmatched the clinical measurement within its credibility bands, illustrating the potential of theproposed method to construct and extract additional model-related information from clinicalsetups. Estimates of contact pressure obtained with the proposed technique closely matchedprevious studies. Thus, the proposed numerical modeling framework enables the investigation ofcausal links between patient pathology/compensation and vocal hyperfunction by both elucidatingsome of the underlying physical mechanisms and providing new measures of vocal functionthat are difficult, if not impossible, to obtain in clinical setups.
Description
Catalogado desde la version PDF de la tesis.
Keywords
CUERDAS VOCALES , ESTIMACION BAYESIANA , HIPERFUNCIONALIDAD VOCAL , IDENTIFICACION DE SISTEMAS , PLIEGUES VOCALES