Healthcare analytics for chronic disease management : a deep learning approach

Abstract

The remarkable advancements in information technology taking place in healthcare have contributed to the improvement of the quality of life and the reduction of the healthcare burden on government and medical professionals. With a new renaissance of artificial intelligence, it is able for machines to perform healthcare tasks intelligently and conduct massively parallel computing on rich health data. Mobile and wearable technologies aid the remote and continuous monitoring of patients, and thus are used for preventive and diagnostic healthcare services. Meanwhile, with the data explosion on social media and digital management platforms, there leaves plenty of space to better utilize various medical data. Deep learning, as one of the most important approaches in artificial intelligence, offers the ability to analyze these data (e.g. textual, image, and sensor data) with fast speed and high precision. Although the application of deep learning in the healthcare domain has caused attention, there remains a lot of room for improvement. The advances in technology provide valuable opportunities and stimulate the data-driven trend. This dissertation presents two essays that adopt the design science approach to develop a series of IT artifacts based on deep learning to solve important healthcare issues. In the first study, I propose DK-BiLSTM (which stands for Domain Knowledge-enhanced bidirectional Long Short-Term Memory), a novel design based on deep learning to identify people with depression and emotional distress. Based on bidirectional Long Short-Term Memory (BiLSTM) networks, the proposed model incorporates both general knowledge and domain knowledge in the learning process through word embedding and parallel BiLSTM units. In the second study, I develop Deep Temporal Multimodal Learning (DTML) framework to assess Parkinson’s Disease (PD) progression. The proposed model learns shared representations from various mobile sensor data and incorporates temporal information. This dissertation has implications for both IS researchers and healthcare practitioners. It explores the application of IS in the healthcare domain by employing advanced deep learning methods and utilizing high-dimensional heterogeneous data. This dissertation provides an improved understanding of how technological advancements help decision making in the healthcare field. In addition, the results of two studies can inform healthcare practitioners and academicians on the health and social impacts of social media and mobile technology.