Motor deterioration in neurocognitive disorder (NCD) patients is common, which affects overall Quality of Life (QoL). Parkinson’s Disease (PD) is one of the prominent NCD, which typically appears after the age of 60. PD disease most commonly begins with a tremor in one hand but can also cause limb stiffness or slowness of movement without tremor. Or, perhaps, someone else may notice that PD patient is not swinging his arm normally as he walks. PD also changes the gait cycle and subject may face freezing of gait or stride rate variability. Clinical evaluation of NCDs is based on rating scales, which is totally observer based, time consuming and inaccurate. The aim of this thesis is to provide neurologists with trustworthy wearable devices for objective evaluating motor deficits in persons with PD and NCDs. That is why a novel framework is developed for assessment of motor deficits from upper extremities of NCDs patients. The suggested system is based on wearable technology and cloud computing. The proposed approach is an integration of signal processing, resampling methods and deep learning classifiers. The recommended system addresses numerous challenges by developing a small form-factor wearable bracelet named A-WEAR which provides movement related data continuously connected with smart phone using BLE, mobile application as user interface, a network that enables wirelessly to MS Azure cloud, which provides the ServiceNow platform as a web application and builds a database utilizing the movement information from the A-WEAR. In order to avoid classifier bias because of imbalance data distribution resampling is performed which consequently improved tremor severity estimation with accuracy 96%, IBA=97%, F1-score=97%, G-mean=97% and AUC=99%. Randomly sampling worked better than under and hybrid sampling. The XGBoost and CatBoost classifiers provide the best performance to evaluate tremor severity while patients are on OFF and ON state, according to the performance comparison of several classifiers. Furthermore, stride rate variability in elderly and NCDs patients is also analyzed the Detrended Fluctuation Analysis (DFA) on data collected using wearable insoles. A specific parameter is considered as a measure of a degree to which one stride time is compared with the previous and the consecutive intervals over a various time span. The scaling exponent α is exigently curtailed in older adults as compared to young healthy participants. The scaling exponent α is also lower in the subjects with PD compared with the disease-free participants. Moreover, α seems linearly correlated to the degree of functional impairment in subjects with PD. These findings demonstrate that stride time fluctuations are more arbitrary in elderly subjects and in subjects with PD. Abnormal fluctuations in the fractal properties of lower limb dynamics are clearly related to functioning in central nervous system control. In this thesis various hybrid machine learning techniques in combination with wearables are discussed, which brings the innovative solutions for healthcare problems
Il deterioramento motorio nei pazienti affetti da disturbi neurocognitivi (NCD) è comune e influisce sulla qualità di vita complessiva (QoL). Il morbo di Parkinson (PD) è uno dei principali NCD, che compare in genere dopo i 60 anni. La malattia di Parkinson inizia più comunemente con un tremore a una mano, ma può anche causare rigidità degli arti o lentezza nei movimenti senza tremore. Oppure, magari, qualcun altro può notare che il paziente affetto da PD non oscilla normalmente il braccio mentre cammina. La PD modifica anche il ciclo dell'andatura e il soggetto può trovarsi di fronte al congelamento dell'andatura o alla variabilità della falcata. La valutazione clinica dei NCD si basa su scale di valutazione, che sono totalmente basate sull'osservatore, richiedono tempo e sono imprecise. L'obiettivo di questa tesi è fornire ai neurologi dispositivi indossabili affidabili per la valutazione oggettiva dei deficit motori nelle persone con PD e NCD. Per questo motivo, è stato sviluppato un nuovo framework per la valutazione dei deficit motori degli arti superiori dei pazienti affetti da NCD. Il sistema proposto si basa sulla tecnologia indossabile e sul cloud computing. L'approccio proposto è un'integrazione di elaborazione del segnale, metodi di ricampionamento e classificatori di deep learning. Il sistema proposto affronta numerose sfide sviluppando un braccialetto indossabile di piccole dimensioni, denominato A-WEAR, che fornisce dati relativi al movimento in modo continuo, collegato a uno smartphone tramite BLE, un'applicazione mobile come interfaccia utente, una rete che consente di collegarsi in modalità wireless al cloud MS Azure, che fornisce la piattaforma ServiceNow come applicazione web e costruisce un database utilizzando le informazioni sul movimento provenienti dall'A-WEAR. Per evitare distorsioni del classificatore dovute allo squilibrio nella distribuzione dei dati, è stato eseguito un ricampionamento che ha migliorato la stima della gravità del tremore con un'accuratezza del 96%, IBA=97%, F1-score=97%, G-mean=97% e AUC=99%. Il campionamento casuale ha funzionato meglio del sottocampionamento e del campionamento ibrido. I classificatori XGBoost e CatBoost forniscono le migliori prestazioni per valutare la gravità del tremore quando i pazienti sono in stato OFF e ON, secondo il confronto delle prestazioni di diversi classificatori. Inoltre, la variabilità della frequenza del passo in pazienti anziani e affetti da NCD è stata analizzata con la Detrended Fluctuation Analysis (DFA) su dati raccolti utilizzando solette indossabili. Un parametro specifico viene considerato come una misura del grado di confronto tra un tempo di falcata e il precedente, nonché tra gli intervalli consecutivi in un determinato arco di tempo. L'esponente di scala α è fortemente ridotto negli adulti anziani rispetto ai giovani partecipanti sani. L'esponente di scala α è anche più basso nei soggetti con PD rispetto a quelli senza malattia. Inoltre, α sembra linearmente correlato al grado di compromissione funzionale nei soggetti con PD. Questi risultati dimostrano che le fluttuazioni del tempo di falcata sono più arbitrarie nei soggetti anziani e nei soggetti con PD. Le fluttuazioni anomale nelle proprietà frattali della dinamica degli arti inferiori sono chiaramente correlate al funzionamento del controllo del sistema nervoso centrale. In questa tesi vengono discusse varie tecniche ibride di apprendimento automatico in combinazione con i dispositivi indossabili, che portano a soluzioni innovative per i problemi sanitari
Contributions to development of medical werable-based applications for subjects with neurocognitive disorders / Channa, Asma. - (2023 Jan 13).
Contributions to development of medical werable-based applications for subjects with neurocognitive disorders
2023-01-13
Abstract
Motor deterioration in neurocognitive disorder (NCD) patients is common, which affects overall Quality of Life (QoL). Parkinson’s Disease (PD) is one of the prominent NCD, which typically appears after the age of 60. PD disease most commonly begins with a tremor in one hand but can also cause limb stiffness or slowness of movement without tremor. Or, perhaps, someone else may notice that PD patient is not swinging his arm normally as he walks. PD also changes the gait cycle and subject may face freezing of gait or stride rate variability. Clinical evaluation of NCDs is based on rating scales, which is totally observer based, time consuming and inaccurate. The aim of this thesis is to provide neurologists with trustworthy wearable devices for objective evaluating motor deficits in persons with PD and NCDs. That is why a novel framework is developed for assessment of motor deficits from upper extremities of NCDs patients. The suggested system is based on wearable technology and cloud computing. The proposed approach is an integration of signal processing, resampling methods and deep learning classifiers. The recommended system addresses numerous challenges by developing a small form-factor wearable bracelet named A-WEAR which provides movement related data continuously connected with smart phone using BLE, mobile application as user interface, a network that enables wirelessly to MS Azure cloud, which provides the ServiceNow platform as a web application and builds a database utilizing the movement information from the A-WEAR. In order to avoid classifier bias because of imbalance data distribution resampling is performed which consequently improved tremor severity estimation with accuracy 96%, IBA=97%, F1-score=97%, G-mean=97% and AUC=99%. Randomly sampling worked better than under and hybrid sampling. The XGBoost and CatBoost classifiers provide the best performance to evaluate tremor severity while patients are on OFF and ON state, according to the performance comparison of several classifiers. Furthermore, stride rate variability in elderly and NCDs patients is also analyzed the Detrended Fluctuation Analysis (DFA) on data collected using wearable insoles. A specific parameter is considered as a measure of a degree to which one stride time is compared with the previous and the consecutive intervals over a various time span. The scaling exponent α is exigently curtailed in older adults as compared to young healthy participants. The scaling exponent α is also lower in the subjects with PD compared with the disease-free participants. Moreover, α seems linearly correlated to the degree of functional impairment in subjects with PD. These findings demonstrate that stride time fluctuations are more arbitrary in elderly subjects and in subjects with PD. Abnormal fluctuations in the fractal properties of lower limb dynamics are clearly related to functioning in central nervous system control. In this thesis various hybrid machine learning techniques in combination with wearables are discussed, which brings the innovative solutions for healthcare problemsFile | Dimensione | Formato | |
---|---|---|---|
Channa Asma.pdf
Open Access dal 27/04/2024
Tipologia:
Tesi di dottorato
Licenza:
DRM non definito
Dimensione
10.01 MB
Formato
Adobe PDF
|
10.01 MB | Adobe PDF | Visualizza/Apri |
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.