AbstractInformation technology has virtually altered every aspect of human life in the present era. The application of informatics in the health sector is rapidly gaining prominence and the benefits of this innovative paradigm are being realized across the globe. This evolution produced large number of patients’ data that can be employed by computer technologies and machine learning techniques, and turned into useful information and knowledge. This data can be used to develop expert systems to help in diagnosing some life-threating diseases such as heart diseases, with less cost, processing time and improved diagnosis accuracy. Even though, modern medicine is generating huge amount of data every day, little has been done to use this available data to solve challenges faced in the successful diagnosis of heart diseases. Highlighting the need for more research into the usage of robust data mining techniques to help health care professionals in the diagnosis of heart diseases and other debilitating disease conditions.
Based on the foregoing, this thesis aims to develop a health informatics system for the classification of heart diseases using data mining techniques focusing on Radial Basis functions and emerging Neural Networks approach. The presented research involves three development stages; firstly, the development of a preliminary classification system for Coronary Artery Disease (CAD) using Radial Basis Function (RBF) neural networks. The research then deploys the deep learning approach to detect three different types of heart diseases i.e. Sleep Apnea, Arrhythmias and CAD by designing two novel classification systems; the first adopt a novel deep neural network method (with Rectified Linear unit activation) design as the second approach in this thesis and the other implements a novel multilayer kernel machine to mimic the behaviour of deep learning as the third approach. Additionally, this thesis uses a dataset obtained from patients, and employs normalization and feature extraction means to explore it in a unique way that facilitates its usage for training and validating different classification methods. This unique dataset is useful to researchers and practitioners working in heart disease treatment and diagnosis.
The findings from the study reveal that the proposed models have high classification performance that is comparable, or perhaps exceed in some cases, the existing automated and manual methods of heart disease diagnosis. Besides, the proposed deep-learning models provide better performance when applied on large data sets (e.g., in the case of Sleep Apnea), with reasonable performance with smaller data sets.
The proposed system for clinical diagnoses of heart diseases, contributes to the accurate detection of such disease, and could serve as an important tool in the area of clinic support system. The outcome of this study in form of implementation tool can be used by cardiologists to help them make more consistent diagnosis of heart diseases.
|Date of Award