RESUMEN
Trials for therapies after an upper limb amputation (ULA) require a focus on the real-world use of the upper limb prosthesis. In this paper, we extend a novel method for identifying upper extremity functional and nonfunctional use to a new patient population: upper limb amputees. We videotaped five amputees and 10 controls performing a series of minimally structured activities while wearing sensors on both wrists that measured linear acceleration and angular velocity. The video data was annotated to provide ground truth for annotating the sensor data. Two different analysis methods were used: one that used fixed-size data chunks to create features to train a Random Forest classifier and one that used variable-size data chunks. For the amputees, the fixed-size data chunk method yielded good results, with 82.7% median accuracy (range of 79.3-85.8) on the 10-fold cross-validation intra-subject test and 69.8% in the leave-one-out inter-subject test (range of 61.4-72.8). The variable-size data method did not improve classifier accuracy compared to the fixed-size method. Our method shows promise for inexpensive and objective quantification of functional upper extremity (UE) use in amputees and furthers the case for use of this method in assessing the impact of UE rehabilitative treatments.
Asunto(s)
Miembros Artificiales , Dispositivos Electrónicos Vestibles , Humanos , Actividades Cotidianas , Extremidad Superior/cirugía , Aprendizaje AutomáticoRESUMEN
BACKGROUND AND PURPOSE: Trials of restorative therapies after stroke and clinical rehabilitation require relevant and objective efficacy end points; real-world upper extremity (UE) functional use is an attractive candidate. We present a novel, inexpensive, and feasible method for separating UE functional use from nonfunctional movement after stroke using a single wrist-worn accelerometer. METHODS: Ten controls and 10 individuals with stroke performed a series of minimally structured activities while simultaneously being videotaped and wearing a sensor on each wrist that captured the linear acceleration and angular velocity of their UEs. Video data provided ground truth to annotate sensor data as functional or nonfunctional limb use. Using the annotated sensor data, we trained a machine learning tool, a Random Forest model. We then assessed the accuracy of that classification. RESULTS: In intrasubject test trials, our method correctly classified sensor data with an average of 94.80% in controls and 88.38% in stroke subjects. In leave-one-out intersubject testing and training, correct classification averaged 91.53% for controls and 70.18% in stroke subjects. CONCLUSIONS: Our method shows promise for inexpensive and objective quantification of functional UE use in hemiparesis, and for assessing the impact of UE treatments. Training a classifier on raw sensor data is feasible, and determination of whether patients functionally use their UE can thus be done remotely. For the restorative treatment trial setting, an intrasubject test/train approach would be especially accurate. This method presents a potentially precise, cost-effective, and objective measurement of UE use outside the clinical or laboratory environment.
Asunto(s)
Actigrafía/instrumentación , Actividades Cotidianas , Monitores de Ejercicio , Aprendizaje Automático , Movimiento , Procesamiento de Señales Asistido por Computador , Accidente Cerebrovascular/diagnóstico , Extremidad Superior/inervación , Aceleración , Adulto , Anciano , Fenómenos Biomecánicos , Estudios de Casos y Controles , Diseño de Equipo , Estudios de Factibilidad , Femenino , Estado de Salud , Humanos , Masculino , Persona de Mediana Edad , Valor Predictivo de las Pruebas , Reproducibilidad de los Resultados , Accidente Cerebrovascular/fisiopatología , Factores de Tiempo , Grabación en VideoRESUMEN
OBJECTIVE: To improve measurement of upper extremity (UE) use in the community by evaluating the feasibility of using body-worn sensor data and machine learning models to distinguish productive prehensile and bimanual UE activity use from extraneous movements associated with walking. DESIGN: Comparison of machine learning classification models with criterion standard of manually scored videos of performance in UE prosthesis users. SETTING: Rehabilitation hospital training apartment. PARTICIPANTS: Convenience sample of UE prosthesis users (n=5) and controls (n=13) similar in age and hand dominance (N=18). INTERVENTIONS: Participants were filmed executing a series of functional activities; a trained observer annotated each frame to indicate either UE movement directed at functional activity or walking. Synchronized data from an inertial sensor attached to the dominant wrist were similarly classified as indicating either a functional use or walking. These data were used to train 3 classification models to predict the functional versus walking state given the associated sensor information. Models were trained over 4 trials: on UE amputees and controls and both within subject and across subject. Model performance was also examined with and without preprocessing (centering) in the across-subject trials. MAIN OUTCOME MEASURE: Percent correct classification. RESULTS: With the exception of the amputee/across-subject trial, at least 1 model classified >95% of test data correctly for all trial types. The top performer in the amputee/across-subject trial classified 85% of test examples correctly. CONCLUSIONS: We have demonstrated that computationally lightweight classification models can use inertial data collected from wrist-worn sensors to reliably distinguish prosthetic UE movements during functional use from walking-associated movement. This approach has promise in objectively measuring real-world UE use of prosthetic limbs and may be helpful in clinical trials and in measuring response to treatment of other UE pathologies.