Establishment of biomedical decision support systems through deep learning techniques

This thesis provides a comprehensive examination of the effective applications of deep learning models and advanced image processing techniques in biomedical signal and image analysis through two distinctive studies. The section dedicated to biological signals in the thesis focused on the examination of biomedical signals such as EOG, EEG, and EKG. To record the EOG and EEG signal data in a computerized environment, two distinct devices were employed. One of these devices is the 14-channel EMOTIV EPOC+, while the other is the OPENBCI Ganglion biosensing signal recording board. The first study focuses on determining the emotional state while listening to music using electroencephalography (EEG) signals. Here, pre-trained deep learning models such as AlexNet and VGG16 are deployed, revealing their potential in solving the problem of recognizing human emotional states. The highest classification result is achieved at a 73.28% accuracy using VGG16 and Beta frequency band spectrograms. These findings advocate the usage of EEG datasets prepared with music data and existing datasets with different deep network models in the problem of human emotional state recognition. The second study is geared towards the early detection of the COVID-19 disease. The contrast of X-ray images is enhanced with an advanced image enhancement technology, MOCSCLAHE, and this enhanced data is classified with convolutional neural networks. The study results show high accuracy rates of the method in classifying COVID-19, normal, and pneumonia cases. This thesis presents compelling insights into the significant value and applicability of using deep learning models and image processing techniques in various application areas such as emotion recognition over EEG signals and COVID-19 detection through X-ray image analysis.