Healthcare Analysis: Cardiovascular Diseases, Drowsiness, Stress and Distraction


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date18 Jan 2018


The current model of healthcare in the world is subject to unprecedented challenges. An ageing population, inadequacies of medical personnel, the impact of lifestyle factors and increasing expenditure mean that the existing approaches may become unsustainable. This gives an impetus to promote medical automation presents an opportunity for change in healthcare. With the rapid increase in computation power and availability of health data, healthcare analysis can be achieved by machine learning. Health data contain valuable information which reflects the nature and properties of the object itself. In this thesis, health data classification is performed in two important aspects: cardiovascular diseases classification and safe driving.

Cardiovascular disease has remained the leading cause of global death for more than 100 years. Firstly, cardiovascular diseases classifier (CDC) has been designed using multiple criteria decision making (MCDM). CDC is able to detect four types of cardiovascular diseases, Dysrhythmia, Heart Failure, Myocardial Infarction and Bundle Branch Block. CDC is designed in two structures, binary classifier (BC) and multi-class classifier (MC). The former classifies between health and unhealthy candidates, whereas the latter gives the exact type of cardiovascular diseases. Six parameters as of criteria, sensitivity, specificity, overall accuracy, feature vector dimensions, overall training and testing time of CDC and a newly proposed average confidence index. Analytic hierarch process (AHP) is applied with these criteria for MCDM, to determine the optimal BC and MC.

Most of the traffic accidents are attributable to human negligence. Drowsy driving, stressed driving and distracted driving are main reasons for traffic deaths and injuries. To relieve these issues, this thesis proposed a full directional driver monitoring solution with triple classification and triple protection using drowsiness detector, stress level detector and distraction detection which monitor the drivers’ vigilance and provide an instant warning. Even more important, it will initialize the auto-navigator of the vehicle (if available) and give suitable advice to the drivers for some critical cases and at the same time, the proposed system would inform the control centre of a public transport company so that they can send another driver to take over the vehicle. This system helps minimize road traffic accidents. It is envisaged that the diminishing accidents and casualties will yield an inexpensive third party motor insurance cost which benefits public transport companies, drivers and passengers.

An accurate driver drowsiness detection scheme (D3S) has been proposed using electrocardiogram (ECG). ECG is adopted instead of the existing method using heart rate variability. A precomputed kernel is designed with cross-correlation, convolution and weighting factors. It captures both symmetric and asymmetric information between ECG signals so that different status of drivers can be detected. Performance evaluation demonstrates that the proposed D3S achieves significant improvement compared to traditional kernels and existing biometric-signal based method, vehicle-based method and image-based method.

The second safe driving application is a driver stress detection algorithm which is developed using multiobjective genetic algorithm based fuzzy c-means clustering (MOGA-FCM). This algorithm is effective to detect the stress level of driver. It takes the advantage of the ability for an ECG, as representative health data input, to provide simultaneous detection for drowsy driving and stressed driving which minimizes the complexity for data acquisition. The analysis shows that the proposed MOGA-FCM yields better accuracy than traditional clustering algorithms and related works.

A driver distraction detection system has been proposed. Such effective detection of distraction in drivers may help to prevent accidents, thus ensuring public transport safety. The entire system consists of distraction detection module which processes a video stream and computes a motion coefficient to reinforce identification of distraction in drivers. Experimental analysis shows that five pre-defined scenarios that include nodding, head shaking, falling asleep, blinking and of a person walking behind can be detected.

While in this thesis we focused on cardiovascular diseases and safe driving issues, the above algorithms can be applied to other applications related to healthcare.