Процесирање биоелектрични и биоакустични сигнали за предвидување медицински состоjби

Abstract

According to the World Health Organization, 30% from all diseases are related to the cardiovascular system. Beside the hypertension condition, the chronic heart failure also presents a global pandemic. Both conditions could be prevented at least partially, if continuously observed, especially at the patients at risk. The focus of the preventive medicine, which has become a lifestyle, is to detect abnormalities, improve early diagnosis and avoid hospitalizations or even life-threatening situations. Recently, there is a technology trend of creating devices, biosensors, that are able to measure the physiological functions of the human body. Motivated by the statistics and the power of the biosensors, detailed researches in the field of bioelectrical and bioacoustical physiological signals, as well as their application in the medical protocols, are presented in the thesis. The basic information of the stability of the human system is provided by the four vital signs: heart rate, respiratory rate, blood pressure and oxygen saturation in the blood. However, there is still no single biosensor that is able to measure all the four parameters at once. Commonly, the biosensors provide information of the heart rate and the respiratory rate, whereas for the blood pressure and the oxygen saturation measurement, there are separate devices. There are many experiments provided in the literature aimed to eliminate the need of the additional devices to measure the blood pressure. However, most of them rely on two or more physiological signals, meaning there is a need of multiple biosensors which is again not a practical solution to the problem. The goal of this research is to prove that the blood pressure can be predicted by using only one physiological signal - ECG. In such case, the biosensors that measure ECG, usually provide information about the heart rate and the respiratory rate, meaning a single biosensor that is able to measure three vital parameters could be achieved. The methodology proposed in the paper analyses data from multiple biosensors in order to model the relationship between blood pressure and the ECG signal. It is based on complexity analysis of the signal and machine learning approach that leaded to results close to the requirements of the standards for blood pressure devices certification (tolerating an error of 58 mmHg). The achieved absolute errors of the proposed methodology are 7.93 8.16 for the systolic, 6.41 7.5 for the diastolic, and 5.72 6.69 mmHg for the mean arterial pressure. Another research provided in the thesis is the acoustical, PCG, signal analysis with the aim to detect chronic heart failures. The method relies on specific audio feature extractor and machine learning design. The evaluation of the methodology showed that the overall accuracy is 96%, i.e., 87% of all chronic failures were successfully detected. Eventually, a model that includes the biosensor technology into the existing protocols for medical response in massive incidents is proposed. The model shows the data flow among the different entities involved in the protocols. The aim is to enable automatic patients’ triage and the testing showed a sensitivity of 0.94/0.74 and positive predictivity of 0.92/0.93 for the estimation of the heart rate and the respiratory rate, respectively. The application of the researches presented in the thesis is multi-fold and reliable for both civil and military environments.