Wearable Computing for Dementia Patients

Abstract

The number of dementia patients is increasing dramatically. As the dementia cannot be cured, only drug treatment can temporarily improve the symptoms. The doctor needs to adjust the medication according to the patients’ symptoms. Therefore, the caregiver is required to give information about the patients’ behavioral and psychological symptoms thoroughly in order for the doctor to adjust the medication efficiently. However, the caregiver has many daily tasks to complete. Thus, he might not know the behavioral and psychological symptoms in detail. This research proposes a wearable computing that monitor patient activities. This research is done based on the user centered design. Therefore, after the interview with doctors and caregivers, we finalized the activities that the doctors need to know in order to adjust the medication efficiently. The activities are stand-sit, shaking, walking, sitting and standing. Moreover, the best position of the wearable was concluded to be on the back of the patients. Then according to the previous works in fall detection systems, the models that we chose to compare their performance are feedforward neural network, support vector machine, decision tree and random forest. The selected features are Mean, Standard Deviation (STD), Peak counts, Zero crossing rate (ZCR), Spectral Energy and Spectral Entropy of x, y, z axis data from accelerometer and gyroscope. The result shows that the feedforward neural network achieved the highest accuracy.