Empirical Mode Decomposition on Biomedical Signals and Images

Biomedical data provides great convenience to doctors thankfully to automations that are used by them. It has been very common to extract and analyze structural features of these data by implementing various signal, image processing and classification algorithms. This wide usage in biomedical area accompanies some problems, too. In contrast to artificial signals which are created in computer environment, irregular and noisy structure of biomedical signals such as ECG, EEG reasons lack about adaptation of signal processing algorithms. Same case is valid for histopathology images which are 2-dimensional signals. Uncertainties and noises in these structures prevent to extract features that will be used in classification. In this thesis study, Empirical Mode Decomposition (EMD), which works adaptively with data and doesn't need for a specific formulation, has been used. In the first chapter of this study which is consisted of two chapters, ECG signals have been analyzed, a performance comparison has been made with classical signal decomposition methods and it has been observed that EMD is more successful about feature extraction. This performance value has been acquired by classifying with Support Vector Machines (SVM). In the second chapter, histopathology images has been analyzed, the features extracted from EMD have been optimized various morphological tools and it has been observed that a progress has been made at classification success obtained by using Random Forest (RF) method comparing with the extracted feature sets of the original pixel values. The acquired information from this study demonstrates that extracted features from biomedical signals with EMD is more identifier according to extracted features from classical methods about increasing classification performance by using suitable denoise tools.