A Decision Support System for StressDiagnosis using ECG Sensor

Sammanfattning: Diagnosis of stress is important because it can cause many diseases e.g., heart disease, headache, migraine, sleep problems, irritability etc. Diagnosis of stress in patients often involves acquisition of biological signals for example heart rate, finger temperature, electrocardiogram (ECG), electromyography signal (EMG), skin conductance signal (SC) etc. followed up by a careful analysis of the acquired signals. The accuracy is totally dependent on the experience of an expert. Again the number of such experts is also very limited. Heart rate is considered as an important parameter in determining stress. It reflects status of the autonomic nervous system (ANS) and thus is very effective in monitoring any imbalance in patient’s stress level. Therefore, a computer-aided system is useful to determine stress level based on various features that can be extracted from a patient’s heart rate signals. Stress diagnosis using biomedical signals is difficult and since the biomedical signals are too complex to generate any rule an experienced person or expert is needed to determine stress levels. Also, it is not feasible to use all the features that are available or possible to extract from the signal. So, relevant features should be chosen from the extracted features that are capable to diagnose stress. Again, ECG signal is frequently contaminated by outliers produced by the loose conduction of the electrode due to sneezing, itching etcetera that hampers the value of the features. A Case-Based Reasoning (CBR) System is helpful when it is really hard to formulate rule and the knowledge on the domain is also weak. A CBR system is developed to evaluate how closely it can diagnose stress levels compare to an expert. A study is done to find out mostly used features to reduce the number of features used in the system and in case library. A software prototype is developed that can collect ECG signal from a patient through ECG sensor and calculate Inter Beat Interval (IBI) signal and features from it. Instead of doing manual visual inspection a new way to remove outliers from the IBI signal is also proposed and implemented here. The case base has been initiated with 22 reference cases classified by an expert. A performance analysis has been done and the result considering how close the system can perform compare to the expert is presented. On the basis of the evaluations an accuracy of 86% is obtained compare to an expert. However, the correctly classified case for stressed group (Sensitivity) was 57% and it is quite important to increase as it is related to the safety issue of health. The reasons of relatively lower sensitivity and possible ways to improve it are also investigated and explained.