Diagnosing PD can be challenging, especially in its early stages. Healthcare professionals employ several tests and tools in the diagnostic process, including comprehensive physical examinations and the use of rating scales like the Unified Parkinson's Disease Rating Scale (UPDRS) and the Hoehn and Yahr Scale (H&Y), but they have some limitations such as subjectivity bias and insensitivity in detecting early-stage changes. The study aims to validate the use of a Smart Watch as a viable tool for continuous monitoring of pathological tremors. Currently, quantitative diagnosis of tremors often involves expensive and specialized inertial measurement unit (IMU) sensors, which present limitations in daily life activities. In contrast, the Smart Watch is an affordable, user-friendly and widely accessible option which allows for the continuous monitoring of tremor development throughout the day without interfering with daily activities. This is especially crucial in neurodegenerative conditions like Parkinson's disease, where symptoms may be mild and sporadic in the early stages, making periodic clinical visits less effective in capturing fluctuations. Data collection was conducted at the Rehabilitation Department of the Federal University of Health Sciences of Porto Alegre (Brazil). Participants included 27 healthy subjects and 28 subjects with Parkinson's disease. Each participant performed a predetermined sequence of movements while wearing a Smart Watch and a G-sensor with an IMU on their wrist to simultaneously record movement data. Data pre-processing was carried out in Matlab and includes the removal of missing data, outliers and noise cleaning. A standard low-pass Butterworth digital filter was applied to preserve the tremor information. The signals from the smart watch and G-sensor were time-synchronized. Various features were extracted from the signals to characterize and validate the Smart Watch sensors. In the time domain, mean, standard deviation and root mean square error were calculated to assess accuracy. In the frequency domain, power spectral density ratio between frequency ranges was analyzed. The study compared the mean values of recordings for each variable measured by GS and SW in both the Control Group (CG) and the Parkinson's Disease (PD) group. In the CG, there was a moderate to strong correlation for accelerations along all three axes and only for angular velocity along the z-axis. The highest Pearson's correlation coefficient and the lowest RMSE were observed in the z-acceleration variable for the CG, resulting in an accuracy of over 50\% for all time-domain features. In the PD group, the SW values showed little correlation with the GS, except for a high Pearson's correlation coefficient observed in the z-acceleration. Similarly, only the z-acceleration showed the lowest RMSE and the highest significant Pearson's coefficient, with an accuracy greater than 50%. Moreover Bland-Altman plot were produced and a Statistical Parametric Mapping for better waveform analysis was also carried-out.
La diagnosi della malattia di Parkinson può essere complessa, soprattutto nelle fasi iniziali. Gli operatori sanitari utilizzano diversi test e strumenti nel processo di diagnosi, comp rese esami fisici approfonditi e l’uso di scale di valutazione come la Unified Parkinson’s Disease Rating Scale (UPDRS) e la Hoehn and Yahr Scale (H&Y), ma presentano alcune limitazioni, come il bias di soggettività e la scarsa sensibilità nel rilevare cambiamenti precoci. Lo studio mira a convalidare l’uso di un Smart Watch come strumento valido per il monitoraggio continuo dei tremori patologici. Attualmente, la diagnosi quantitativa dei tremori spesso coinvolge costosi e specializzati sensori di misurazione inerziale (IMU), che presentano limitazioni nelle attività quotidiane. Al contrario, lo Smart Watch offre diversi vantaggi, inclusi il monitoraggio continuo e confortevole dello sviluppo e della progressione dei tremori durante tutto il giorno. Questo è particolarmente cruciale in condizioni neurodegenerative come il morbo di Parkinson, dove i sintomi possono essere lievi e sporadici nelle fasi iniziali, rendendo le visite cliniche periodiche meno efficaci nel catturare le fluttuazioni. Gli Smart Watch sono accessibili dal punto di vista economico, facili da usare e ampiamente disponibili, consentendo il monitoraggio continuo dei sintomi senza interferire con le attività quotidiane. Lo studio ha effettuato la raccolta dei dati presso il Dipartimento di Riabilitazione dell’ Università Federale delle Scienze della Salute di Porto Alegre. I partecipanti includono 27 soggetti sani e 28 soggetti affetti da Parkinson. Ciascun partecipante ha eseguito una sequenza predeterminata di movimenti indossando uno Smart Watch e un G-sensor al polso per registrare contemporaneamente i dati di movimento. La pre-elaborazione dei dati è stata eseguita in Matlab e include la rimozione dei dati mancanti, la pulizia del rumore e la rimozione dei punti anomali. È stato applicato un filtro passa basso di Butterworth per preservare le informazioni sui tremori. I segnali dello Smart Watch e del G-sensor sono stati sincronizzati temporalmente. Diverse caratteristiche sono state estratte dai segnali per caratterizzare e convalidare i sensori dello Smart Watch. Nel dominio temporale, sono stati calcolati media, deviazione standard e errore quadratico medio per valutare l’accuratezza. Nel dominio delle fre quenze, è stata analizzata la densità spettrale di potenza come rapporto tra due diversi range di frequenza. Lo studio ha confrontato i valori medi delle registrazioni per ciascuna variabile misurata da GS (G-sensor) e SW (Smart Watch) sia nel Gruppo di Controllo (CG) che nel gruppo di soggetti con morbo di Parkinson (PD). Nel CG, è stata osservata una correlazione mod erata a forte per le accelerazioni lungo tutti e tre gli assi e solo per la velocità angolare lungo l’asse z. Il coefficiente di correlazione di Pearson più alto e il minimo RMSE sono stati osservati nella variabile di accelerazione lungo l’asse z per il CG, con un’accuratezza superiore al 50% per tutte le caratteristiche nel dominio temporale. Nel gruppo PD, i valori di SW hanno mostrato poca correlazione con GS, ad eccezione di un elevato coef ficiente di correlazione di Pearson osservato nell’accelerazione lungo l’asse z. Allo stesso modo, solo l’accelerazione lungo l’asse z ha mostrato il minimo RMSE e il coefficiente di Pearson più alto e significativo, con un’accuratezza maggiore del 50%. Inoltre, sono stati prodotti grafici di Bland-Altman e sono stati condotti anche test di Statistical Parametric Mapping per una migliore analisi delle forme d’onda.
Validation of a smart watch sensor for tremor detection in mild Parkinson's disease patients
GARAGIOLA, BENEDETTA
2022/2023
Abstract
Diagnosing PD can be challenging, especially in its early stages. Healthcare professionals employ several tests and tools in the diagnostic process, including comprehensive physical examinations and the use of rating scales like the Unified Parkinson's Disease Rating Scale (UPDRS) and the Hoehn and Yahr Scale (H&Y), but they have some limitations such as subjectivity bias and insensitivity in detecting early-stage changes. The study aims to validate the use of a Smart Watch as a viable tool for continuous monitoring of pathological tremors. Currently, quantitative diagnosis of tremors often involves expensive and specialized inertial measurement unit (IMU) sensors, which present limitations in daily life activities. In contrast, the Smart Watch is an affordable, user-friendly and widely accessible option which allows for the continuous monitoring of tremor development throughout the day without interfering with daily activities. This is especially crucial in neurodegenerative conditions like Parkinson's disease, where symptoms may be mild and sporadic in the early stages, making periodic clinical visits less effective in capturing fluctuations. Data collection was conducted at the Rehabilitation Department of the Federal University of Health Sciences of Porto Alegre (Brazil). Participants included 27 healthy subjects and 28 subjects with Parkinson's disease. Each participant performed a predetermined sequence of movements while wearing a Smart Watch and a G-sensor with an IMU on their wrist to simultaneously record movement data. Data pre-processing was carried out in Matlab and includes the removal of missing data, outliers and noise cleaning. A standard low-pass Butterworth digital filter was applied to preserve the tremor information. The signals from the smart watch and G-sensor were time-synchronized. Various features were extracted from the signals to characterize and validate the Smart Watch sensors. In the time domain, mean, standard deviation and root mean square error were calculated to assess accuracy. In the frequency domain, power spectral density ratio between frequency ranges was analyzed. The study compared the mean values of recordings for each variable measured by GS and SW in both the Control Group (CG) and the Parkinson's Disease (PD) group. In the CG, there was a moderate to strong correlation for accelerations along all three axes and only for angular velocity along the z-axis. The highest Pearson's correlation coefficient and the lowest RMSE were observed in the z-acceleration variable for the CG, resulting in an accuracy of over 50\% for all time-domain features. In the PD group, the SW values showed little correlation with the GS, except for a high Pearson's correlation coefficient observed in the z-acceleration. Similarly, only the z-acceleration showed the lowest RMSE and the highest significant Pearson's coefficient, with an accuracy greater than 50%. Moreover Bland-Altman plot were produced and a Statistical Parametric Mapping for better waveform analysis was also carried-out.File | Dimensione | Formato | |
---|---|---|---|
2023_10_Garagiola_Tesi_01.pdf
non accessibile
Descrizione: testo tesi
Dimensione
3.02 MB
Formato
Adobe PDF
|
3.02 MB | Adobe PDF | Visualizza/Apri |
2023_10_Garagiola_ExecutiveSummary_02.pdf
non accessibile
Descrizione: testo executive summary
Dimensione
312.53 kB
Formato
Adobe PDF
|
312.53 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/10589/210070