Sammarini, Gaia
(2025)
Measuring freezing of gait using wearable-based algorithms in persons with Parkinson's disease undergoing deep brain stimulation.
[Laurea magistrale], Università di Bologna, Corso di Studio in
Biomedical engineering [LM-DM270] - Cesena, Documento ad accesso riservato.
Documenti full-text disponibili:
Abstract
Parkinson's disease (PD) is a progressive neurodegenerative disease. One of the most debilitating symptoms of PD is freezing of gait (FOG), an episodic absence or reduction of forward progression in the feet despite the intention to walk. The most common treatments for PD are dopaminergic drugs, but Deep Brain Stimulation, a technological treatment that delivers electrical stimulation to affected brain regions, has become more common. Wearable sensor-based evaluation systems have been developed to complement clinical evaluations especially in every day real-life setting. In particular, algorithms have been proposed to automatically detect the occurrence and duration of FOG events based on the signals recorded by wearable sensors, where the reference is the video assessment from an expert clinician. In this dissertation we set out to analyze three open- source algorithms for detecting FOG. These algorithms were externally validated using a dataset recorded in the context of a research project carried out at Istituto delle Scienze Neurologiche, Bologna including a motor protocol designed for DBS screening and follow-up assessment. A complete evaluation was performed of the performance of the algorithms for the whole protocol, each single task and the pre-DBS and post-DBS conditions. Additionally, the performance of the algorithms was evaluated on an external dataset of healthy older subjects to test for false positives. Finally, one algorithm was modified to obtain a more accurate result, and one machine-learning model was retrained using our dataset. The best result on the DBS dataset was obtained using the modified threshold-based algorithm with an ICC of 0.69. On the healthy subjects dataset, the second machine-learning model achieved the best performance with a bias of 4.17%. The obtained performances are lower than the ones reported, which underlines the necessity of external validation for developing such algorithms, to obtain universally valuable systems.
Abstract
Parkinson's disease (PD) is a progressive neurodegenerative disease. One of the most debilitating symptoms of PD is freezing of gait (FOG), an episodic absence or reduction of forward progression in the feet despite the intention to walk. The most common treatments for PD are dopaminergic drugs, but Deep Brain Stimulation, a technological treatment that delivers electrical stimulation to affected brain regions, has become more common. Wearable sensor-based evaluation systems have been developed to complement clinical evaluations especially in every day real-life setting. In particular, algorithms have been proposed to automatically detect the occurrence and duration of FOG events based on the signals recorded by wearable sensors, where the reference is the video assessment from an expert clinician. In this dissertation we set out to analyze three open- source algorithms for detecting FOG. These algorithms were externally validated using a dataset recorded in the context of a research project carried out at Istituto delle Scienze Neurologiche, Bologna including a motor protocol designed for DBS screening and follow-up assessment. A complete evaluation was performed of the performance of the algorithms for the whole protocol, each single task and the pre-DBS and post-DBS conditions. Additionally, the performance of the algorithms was evaluated on an external dataset of healthy older subjects to test for false positives. Finally, one algorithm was modified to obtain a more accurate result, and one machine-learning model was retrained using our dataset. The best result on the DBS dataset was obtained using the modified threshold-based algorithm with an ICC of 0.69. On the healthy subjects dataset, the second machine-learning model achieved the best performance with a bias of 4.17%. The obtained performances are lower than the ones reported, which underlines the necessity of external validation for developing such algorithms, to obtain universally valuable systems.
Tipologia del documento
Tesi di laurea
(Laurea magistrale)
Autore della tesi
Sammarini, Gaia
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Indirizzo
CURRICULUM INNOVATIVE TECHNOLOGIES IN DIAGNOSTICS AND THERAPY
Ordinamento Cds
DM270
Parole chiave
Parkinson,Disease,Wearable,Sensors,Detection,Freezing,Gait,Deep,Brain,Stimulation
Data di discussione della Tesi
13 Marzo 2025
URI
Altri metadati
Tipologia del documento
Tesi di laurea
(NON SPECIFICATO)
Autore della tesi
Sammarini, Gaia
Relatore della tesi
Correlatore della tesi
Scuola
Corso di studio
Indirizzo
CURRICULUM INNOVATIVE TECHNOLOGIES IN DIAGNOSTICS AND THERAPY
Ordinamento Cds
DM270
Parole chiave
Parkinson,Disease,Wearable,Sensors,Detection,Freezing,Gait,Deep,Brain,Stimulation
Data di discussione della Tesi
13 Marzo 2025
URI
Gestione del documento: