Publicação
High-tech aid tool to monitor postural stability in Parkinson’s Disease
| Resumo: | Parkinson’s disease (PD) is a neurodegenerative disease that affects around 1% of the population over 65 and has increased in prevalence in recent years. One of the most disabling motor symptoms and a major contributor to falls is postural instability, which threatens the independence and well-being of people with PD. Usually, physicians assess this symptom with a traditional clinical examination named pull test, which, although easy to administer without requiring any instruments, it is a difficult test to standardize and lacks sensitivity to small but significant changes. Thus, other approaches based on high technologies have emerged to provide objective metrics and long-term data on postural stability, complementing clinical assessment. Wearable sensors appeared as a promising tech-based solution to better capture postural instability and eliminate the subjectivity of postural-associated clinical examinations. This dissertation proposes the design, development and validation of a postural assessment tool to perform more objective evaluations of postural instability during basic dynamic day-to-day activities. To achieve this goal, the following steps were accomplished: (i) create a dataset based on 3D motion data of PD patients performing the pull test and dynamic activities using an inertial measurement unit (IMU); (ii) extract relevant features from the data collected, conduct an extensive statistical search, and find correlations to clinical scales; (iii) implement a tool based in artificial intelligence (AI) to classify the level of postural instability through the data collected. Different deep learning models were designed and several combinations of data input were considered in order to find the best model to predict the pull test score. Overall, satisfactory results were achieved as the statistical analysis revealed that many features were considered relevant to distinguish between the scores of the pull test, for diagnostic purposes and also to differentiate the several stages of the disease and levels of motor disability. Regarding the AI-based tool, the results suggest that the combination of IMU-based data with deep learning may be a promising solution for postural instability assessment. The model that achieved the best performance in the testing phase with unseen data presented an accuracy, precision, recall and F1-score of approximately 0.86. The results also show that when fewer daily activities are included in the dataset, the complexity of the model reduces, making it more efficient. Despite the promising results, more data should be collected to assess the actual performance of the model as a postural assessment tool. |
|---|---|
| Autores principais: | Cardoso, Marta Sofia Guimarães |
| Assunto: | Artificial Intelligence Inertial measurement unit Parkinson’s Disease Postural instability Pull test score Doença de Parkinson Escala do Pull test Inteligência Artificial Instabilidade postural Unidade de Medição Inercia |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade do Minho |
| Idioma: | inglês |
| Origem: | RepositóriUM - Universidade do Minho |
| Resumo: | Parkinson’s disease (PD) is a neurodegenerative disease that affects around 1% of the population over 65 and has increased in prevalence in recent years. One of the most disabling motor symptoms and a major contributor to falls is postural instability, which threatens the independence and well-being of people with PD. Usually, physicians assess this symptom with a traditional clinical examination named pull test, which, although easy to administer without requiring any instruments, it is a difficult test to standardize and lacks sensitivity to small but significant changes. Thus, other approaches based on high technologies have emerged to provide objective metrics and long-term data on postural stability, complementing clinical assessment. Wearable sensors appeared as a promising tech-based solution to better capture postural instability and eliminate the subjectivity of postural-associated clinical examinations. This dissertation proposes the design, development and validation of a postural assessment tool to perform more objective evaluations of postural instability during basic dynamic day-to-day activities. To achieve this goal, the following steps were accomplished: (i) create a dataset based on 3D motion data of PD patients performing the pull test and dynamic activities using an inertial measurement unit (IMU); (ii) extract relevant features from the data collected, conduct an extensive statistical search, and find correlations to clinical scales; (iii) implement a tool based in artificial intelligence (AI) to classify the level of postural instability through the data collected. Different deep learning models were designed and several combinations of data input were considered in order to find the best model to predict the pull test score. Overall, satisfactory results were achieved as the statistical analysis revealed that many features were considered relevant to distinguish between the scores of the pull test, for diagnostic purposes and also to differentiate the several stages of the disease and levels of motor disability. Regarding the AI-based tool, the results suggest that the combination of IMU-based data with deep learning may be a promising solution for postural instability assessment. The model that achieved the best performance in the testing phase with unseen data presented an accuracy, precision, recall and F1-score of approximately 0.86. The results also show that when fewer daily activities are included in the dataset, the complexity of the model reduces, making it more efficient. Despite the promising results, more data should be collected to assess the actual performance of the model as a postural assessment tool. |
|---|