Hearing disability affects nearly half a billion people worldwide, despite this it is rarely con- sidered a handicap and older people tend to turn to professionals when the disability is already in the chronic phase. The test considered as the gold standard is pure tone audiometry, how- ever a series of precautions necessary for the test prevent it from spreading on a large scale. For this reason, in recent years the Politecnico di Milano in collaboration with Consiglio Nazionale delle Ricerche have developed a test that aims to promote awareness and early as- sessment of hearing conditions. The test uses the speech-in-noise form through the delivery of nonsense vowel-consonant-vowel stimuli. A multivariate analysis proved to be the right strategy for combining the outputs of the speech-in-noise test in order to be able to classify the subject according to the criteria issued by the WHO. This study aims to update the platform already created in order to make it able to collect information about the user that could help determine the health of the hearing system. For this reason, a closed-ended questionnaire able to collect information about risk factors for the hearing system was integrated to the platform. The data collection led to the formation of a new single data-set that was analyzed in order to classify the observations according to the audiometric examination. The attributes considered to train the decision tree model were cho- sen on the basis of the correlation criterion. The metrics showed similar results for subgroups of different features, demonstrating how a smaller set of features is sufficiently accurate. The analysis of new attributes led to the estimation of the psychometric curves referred to each subject and each stimulus, capable of leading to the distinction between normal hearing and hypoacusis observations. These are therefore eligible features to be considered for future integration for the purpose of audiological screening. In order to obtain more consistent re- sults on the dependence of the target variable with the considered attributes and psychometric curves, further measurements are needed on a larger population and on a greater number of people with hearing loss so as to obtain more reliable information.
La disabilita` uditiva colpisce quasi mezzo miliardo di persone nel mondo, nonostante cio` e` raramente considerata un handicap e le persone anziane tendono a rivolgersi a profes- sionisti quando la disabilita` e` gia` in fase cronica. Il test considerato come gold standard e` l’audiometria tonale pura, tuttavia una serie di precauzioni necessarie per il test ne im- pediscono la diffusione su larga scala. Per questo negli ultimi anni il Politecnico di Milano in collaborazione con il Consiglio Nazionale delle Ricerche hanno messo a punto un test che mira a promuovere la consapevolezza e la valutazione precoce delle condizioni uditive. Il test utilizza la forma speech-in-noise attraverso l’erogazione di stimoli nosense vocale- consonante-vocale. Un’analisi multivariata si e` rivelata la giusta strategia per combinare i risultati del test speech-in-noise al fine di poter classificare il soggetto secondo i criteri ema- nati dall’OMS. Questo studio mira ad aggiornare la piattaforma gia` creata per renderla in grado di raccogliere informazioni sull’utente che potrebbero aiutare a determinare lo stato di salute dell’apparato uditivo. Per questo motivo e` stato integrato un questionario a risposta chiusa in grado di raccogliere informazioni sui fattori di rischio per l’apparato uditivo. La raccolta dei dati ha portato alla formazione di un data-set unico che e` stato analizzato al fine di classificare le os- servazioni seguendo l’esame audiometrico. Gli attributi considerati per addestrare il modello dell’albero decisionale sono stati scelti sulla base del criterio di correlazione. Le metriche hanno mostrato risultati simili per sottogruppi di caratteristiche diverse, dimostrando come un set di variabili pia` piccolo sia sufficientemente accurato. L’analisi di nuovi attributi alla stime delle curve psicometriche riferite a ciascun soggetto e a ogni stimolo, in grado di con- durre alla distinzione tra osservazione normale e con deficit. Si tratta quindi di caratteristiche idonee da considerare per future integrazioni ai fini dello screening audiologico. Per ottenere risultati piu` coerenti sulla dipendenza della variabile target con gli attributi e le curve psico- metriche considerate, sono necessarie ulteriori misurazioni su una popolazione piu` ampia e su un numero maggiore di persone con ipoacusia in modo ottenere dati piu` consistenti.
A novel multivariate approach to adult hearing screening using a combination of speech-in-noise testing and self-admistreted questionnaire
PAOLINI, MARTINA
2020/2021
Abstract
Hearing disability affects nearly half a billion people worldwide, despite this it is rarely con- sidered a handicap and older people tend to turn to professionals when the disability is already in the chronic phase. The test considered as the gold standard is pure tone audiometry, how- ever a series of precautions necessary for the test prevent it from spreading on a large scale. For this reason, in recent years the Politecnico di Milano in collaboration with Consiglio Nazionale delle Ricerche have developed a test that aims to promote awareness and early as- sessment of hearing conditions. The test uses the speech-in-noise form through the delivery of nonsense vowel-consonant-vowel stimuli. A multivariate analysis proved to be the right strategy for combining the outputs of the speech-in-noise test in order to be able to classify the subject according to the criteria issued by the WHO. This study aims to update the platform already created in order to make it able to collect information about the user that could help determine the health of the hearing system. For this reason, a closed-ended questionnaire able to collect information about risk factors for the hearing system was integrated to the platform. The data collection led to the formation of a new single data-set that was analyzed in order to classify the observations according to the audiometric examination. The attributes considered to train the decision tree model were cho- sen on the basis of the correlation criterion. The metrics showed similar results for subgroups of different features, demonstrating how a smaller set of features is sufficiently accurate. The analysis of new attributes led to the estimation of the psychometric curves referred to each subject and each stimulus, capable of leading to the distinction between normal hearing and hypoacusis observations. These are therefore eligible features to be considered for future integration for the purpose of audiological screening. In order to obtain more consistent re- sults on the dependence of the target variable with the considered attributes and psychometric curves, further measurements are needed on a larger population and on a greater number of people with hearing loss so as to obtain more reliable information.File | Dimensione | Formato | |
---|---|---|---|
2021_12_Paolini.pdf
non accessibile
Descrizione: Executive Summary e Tesi
Dimensione
8.4 MB
Formato
Adobe PDF
|
8.4 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/182662