Hearing degeneration cases are increasingly frequent nowadays among the adult population; however, this condition often remains under-diagnosed, leading to long-term consequences such as declining auditory capabilities and cognitive issues like anxiety, depression, and attention deficits. This lack of adequate diagnosis frequently exacerbates overall health deterioration. An effective screening process can early identify risk factors and allow timely interventions, thereby mitigating the severity of negative outcomes. Screening tools such as those offered by the WHISPER (Widespread Hearing Impairment Screening and Prevention of Risk) platform developed by CNR-IEIIT and Politecnico di Milano, and the Virtual Hearing Clinic developed by Carl von Ossietzky University of Oldenburg, play a crucial role in this context. WHISPER provides a straightforward, effective, and clinician-independent screening test, which includes a Speech-in-noise test, a risk factor questionnaire, and a Digit Span Test (DST) to assess cognitive abilities. The collaboration among CNR-IEIIT, Politecnico di Milano, and the University of Oldenburg has facilitated the combined use of both platforms. The main objective of the project is to develop automated methods for multivariate analysis of audiological data using machine learning techniques, aimed at identifying and classifying auditory and cognitive issues in population screening contexts. The analyses conducted in this thesis provide valuable insights into the impact of various features on auditory and cognitive performances, and the effectiveness of different clustering and classification methods. Data analysis confirmed the expected age-related hearing loss trend, with key variables such as age, Pure Tone Average (PTA), number of stimuli, and total test time playing a crucial role in identifying homogeneous groups through clustering techniques. Despite the challenges posed by the small dataset size, models like Random Forest and SVM demonstrated notable robustness. In collaboration with the University of Oldenburg, a new dataset has been created to enable cross-validation of speech-in-noise tests. Future research should focus on expanding the dataset and adopting advanced techniques to enhance the reliability of clustering and classification results, thereby deepening our understanding of the complex interactions between hearing, aging, and cognitive performance. This thesis lays the groundwork for further research in the field, emphasizing the importance of comprehensive data collection and methodological rigor in advancing our understanding of the dynamics between hearing, aging, and cognitive performance.
I casi di degenerazione uditiva sono sempre più frequenti nella popolazione adulta, tuttavia questa patologia rimane spesso sotto-diagnosticata, con gravi conseguenze a lungo termine tra cui il declino delle capacità uditive e problemi cognitivi come ansia, depressione e deficit dell'attenzione. Questa mancanza di diagnosi adeguata spesso porta al peggioramento della salute complessiva dell'individuo. Un processo di screening efficace può individuare precocemente i fattori di rischio e permettere interventi tempestivi, mitigando così l'entità delle conseguenze negative. Gli strumenti di screening come quelli proposti dalle piattaforme WHISPER (Widespread Hearing Impairment Screening and PrEvention of Risk) sviluppata da CNR-IEIIT e Politecnico di Milano e Virtual Hearing Clinic sviluppata dalla Carl von Ossietzky University of Oldenburg giocano un ruolo cruciale in questo contesto. WHISPER offre un test di screening semplice, efficace e indipendente dal personale clinico, che include uno Speech-in-noise test, un questionario sui fattori di rischio e un Digit Span Test (DST) per valutare le abilità cognitive. La collaborazione tra CNR-IEIIT, Politecnico di Milano e l'Università di Oldenburg ha permesso l'utilizzo congiunto di entrambe le piattaforme. L'obiettivo principale del progetto è sviluppare metodi automatici per l'analisi multivariata dei dati audiologici tramite tecniche di machine learning, al fine di identificare e classificare problemi uditivi e cognitivi in contesti di screening di popolazione. Le analisi condotte in questa tesi forniscono preziose informazioni sull'impatto delle diverse caratteristiche sulle prestazioni uditive e cognitive, e sull'efficacia di vari metodi di clustering e classificazione. L'analisi dei dati ha confermato il trend atteso di perdita uditiva correlata all'età, con variabili chiave come età, Pure Tone Average (PTA), numero di stimoli e tempo totale del test che hanno giocato un ruolo fondamentale nell'identificare gruppi omogenei attraverso tecniche di clustering. Nonostante le sfide rappresentate dalle dimensioni ridotte del dataset, modelli come Random Forest e SVM hanno dimostrato una robustezza notevole. In collaborazione con l'Università di Oldenburg, è stato creato un nuovo dataset che permette la cross-validazione dei test di speech in noise. Le future ricerche dovrebbero concentrarsi sull'espansione del dataset e sull'adozione di tecniche avanzate per migliorare l'affidabilità dei risultati di clustering e classificazione, approfondendo così la nostra comprensione delle interazioni complesse tra udito, invecchiamento e performance cognitive. Questa tesi getta le basi per ulteriori ricerche nel campo, enfatizzando l'importanza di una raccolta dati esaustiva e di un rigore metodologico nel progredire nella comprensione delle dinamiche tra udito, invecchiamento e performance cognitive.
Multivariate analysis of audiological data from WHISPER and virtual hearing clinic platforms: a machine learning approach
STAIANO, ILARIA
2023/2024
Abstract
Hearing degeneration cases are increasingly frequent nowadays among the adult population; however, this condition often remains under-diagnosed, leading to long-term consequences such as declining auditory capabilities and cognitive issues like anxiety, depression, and attention deficits. This lack of adequate diagnosis frequently exacerbates overall health deterioration. An effective screening process can early identify risk factors and allow timely interventions, thereby mitigating the severity of negative outcomes. Screening tools such as those offered by the WHISPER (Widespread Hearing Impairment Screening and Prevention of Risk) platform developed by CNR-IEIIT and Politecnico di Milano, and the Virtual Hearing Clinic developed by Carl von Ossietzky University of Oldenburg, play a crucial role in this context. WHISPER provides a straightforward, effective, and clinician-independent screening test, which includes a Speech-in-noise test, a risk factor questionnaire, and a Digit Span Test (DST) to assess cognitive abilities. The collaboration among CNR-IEIIT, Politecnico di Milano, and the University of Oldenburg has facilitated the combined use of both platforms. The main objective of the project is to develop automated methods for multivariate analysis of audiological data using machine learning techniques, aimed at identifying and classifying auditory and cognitive issues in population screening contexts. The analyses conducted in this thesis provide valuable insights into the impact of various features on auditory and cognitive performances, and the effectiveness of different clustering and classification methods. Data analysis confirmed the expected age-related hearing loss trend, with key variables such as age, Pure Tone Average (PTA), number of stimuli, and total test time playing a crucial role in identifying homogeneous groups through clustering techniques. Despite the challenges posed by the small dataset size, models like Random Forest and SVM demonstrated notable robustness. In collaboration with the University of Oldenburg, a new dataset has been created to enable cross-validation of speech-in-noise tests. Future research should focus on expanding the dataset and adopting advanced techniques to enhance the reliability of clustering and classification results, thereby deepening our understanding of the complex interactions between hearing, aging, and cognitive performance. This thesis lays the groundwork for further research in the field, emphasizing the importance of comprehensive data collection and methodological rigor in advancing our understanding of the dynamics between hearing, aging, and cognitive performance.File | Dimensione | Formato | |
---|---|---|---|
2024_07_Staiano_Tesi.pdf
accessibile in internet per tutti
Descrizione: testo tesi
Dimensione
9.25 MB
Formato
Adobe PDF
|
9.25 MB | Adobe PDF | Visualizza/Apri |
2024_07_Staiano_Executive_Summary.pdf
accessibile in internet per tutti
Descrizione: testo executive summary
Dimensione
847.05 kB
Formato
Adobe PDF
|
847.05 kB | 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/10589/223450