Advanced systems for head scratch detection

Itching is a condition that affects a substantial group of people. This condition may be caused by different conditions, such as scabies, atopic dermatitis, or kidney failure; it can also be a symptom of a malignant condition, such as lymphoma. So far, a scratch was being detected by manually counting the occurances or using a bone-conducting microphone, which is uncomfortably set up. Thus, there is a need for a next-generation system that allows detecting scratches on multiple people simultaneously without invading patients’ lives. Wearable sensors allow the ability to directly collect the data asynchronously from many people and detect activities by applying machine learning algorithms. In this thesis, we propose using multimodal wearable sensors and combining the data from Inertial Measurement Units (IMU), Electric Potential Sensor (EPS) and a microphone using machine learning-based fusion for scalp scratch detection. In this thesis, we describe the results on three problems: (1) the impact of fusing EPS and IMU for scratch detection, (2) the ambient microphone’s ability to detect scratch, (3) the future direction for next-generation scratch detection system. We evaluated the fusion of EPS and IMU on a constrained dataset that mimics an office worker’s daily activities, which we collected in the Wearable Technologies Lab at the University of Sussex. We showed that multimodal fusion is superior to using a wrist-worn IMU solely. For the (2) objective, we collected a small dataset from four people showing that an ambient microphone can be a powerful modality for scratch detection. Finally, we propose a clear direction for future research that involves a wide-scale dataset collection, novel hardware, and powerful Deep Learning algorithms to power the next generation scratch detection system.