A carregar...
Publicação
Wearable technology and machine learning algorithms to monitor upper-limb use in brain injury survivors
| Nome: | Descrição: | Tamanho: | Formato: | |
|---|---|---|---|---|
| 37.4 MB | Adobe PDF |
Orientador(es)
Resumo(s)
There is a high incidence of upper-limb motor changes following an acquired brain injury (ABI). Those impairments negatively impact the patient’s ability to perform activities of daily living (ADL), especially reach-to-grasp of objects, a goal-directed action. This research aims to assess and monitor those changes (i.e., rehabilitation outcomes) in response to clinical intervention. Traditionally the outcomes are clinician-reported, making the assessment time-consuming and observer-dependent. Also, the assessment is not done on a regular basis (i.e., most of the time, assessments happen only at two time points: pre and post-intervention), and it is desirable to assess over the cycle of care. Thus, this will be addressed by deriving estimates of clinical-based outcome measures using wearable sensor data and Machine Learning algorithms (ML).
The study was conducted on a heterogeneous sample of thirty-seven ABI individuals with upperlimb emiparesis. Sixteen of these patients were stroke survivors, while twenty-one were traumatic brain injury survivors. Literature review shows that there is potential to use wearable technology to have a more precise assessment and collect data with different setups that will enable the performance of less constrained movements. Additionally, accelerometers are the most used devices for this purpose, meeting the user requirements (e.g., accuracy, comfort, low cost, and ease of use), and having relevant wearability factors (e.g., placement, material, weight, and sizing). Furthermore, these devices enable researchers and clinicians to collect data towards evaluating motor recovery at multiple time points, and thus, the recovery progress can be followed more closely. Consequently, to enable the assessment of the effects of the upper-limb rehabilitation interventions, subjects were instrumented with a total of seven wireless inertial measurement units (IMU). The subjects were guided to perform a series of eight standardized tasks specifically chosen to reproduce motor patterns of ADL. Dedicated ML algorithms were developed to derive firstly the Functional Ability Scale (FAS) concerning the patient’s quality of movement during the performance of activities of daily living and secondly the Fugl-Meyer Assessment (FMA) for upper extremity (UE) assessment regarding the severity of motor impairments, where the knowledge learned from the preceding FAS estimations was employed.
Broadly translated, our findings reveal that for the FAS, a dedicated Hierarchical Random Forestbased model with a Binary Classifier and Regression on the second level is the best solution to derive scores with a particular focus on high motor functioning patients. The final FAS model achieved excellent performance with a coefficient of determination, R2, of 0.91. The main contributions of this research pertaining to the FMA highlight that a subset of tasks emerges as the optimal to derive the FMA. Moreover, adding the FAS scores into the final FMA model for both single tasks and combined predictions was key to improving the performance by providing quality movement knowledge. As a result, the final FMA model achieved good performance with a R2 of 0.79. Therefore, our analysis is encouraging, having demonstrated the feasibility of incorporating wearable sensor data acquired in tasks explicitly selected and adapted to reproduce ADL motor patterns and taking advantage of the knowledge derived from estimating the FAS to improve the final FMA model predictions.
Altogether, the methods herein presented have the potential to empower healthcare professionals to
perform precision rehabilitation, an intervention adjusted according to patient-specific responsiveness,
maximizing ABI patients’ motor gains. Likewise, the findings of this analysis pave the way to transitioning from clinician-reported outcomes to sensor-based outcome estimates that, among others, can be assessed in home and community settings.
Descrição
Tese de Mestrado Integrado, Engenharia Biomédica e Biofísica (Engenharia Clínica e Instrumentação Médica), 2022, Universidade de Lisboa, Faculdade de Ciências
Palavras-chave
Lesão Cerebral Adquirida Aprendizagem Automática Reabilitação de Precisão Tecnologia Vestível Tecnologias Portáteis e Móveis para Reabilitação Teses de mestrado - 2023