Taramalı elektron mikroskobu görüntülerinde mitokondrilerin otomatik olarak bölütlenmesi

Many studies have shown that shape of mitochondria indicates the occurrence of diseases. Scanning Electron Microscopy (SEM) enables to obtain image of internal structures of the cell and mitochondria. Automatic segmentation of mitochondria contributes to the decision of diseases by specialists. There is limited study about automatic segmentation of mitochondria in Serial Block-Face Scanning Electron Microscopy (SFBSEM) images. SBFSEM imaging technique provides full automation, well registered images, less time and less effort for data acquisition. Therefore, SBFSEM imaging technique is selected for this study. Recently, deep learning methods have been implemented for image processing of SEM datasets. However, due to requirement of huge datasets, much effort and powerful computers for preparing testing and training data, energy based model is implemented for this study. The algorithms used in this thesis are primarily the algorithms developed by Tasel et al for mitochondria segmentation in TEM images. The method includes preprocessing, ridge detection, energy mapping, curve fitting, snake-based shape extraction, validation and post-processing steps. In this thesis, these algorithms are adapted and refined for SBFSEM images to obtain optimum performance. Evaluations are made by using Dice Similarity Coefficient (DSC), precision, recall and F-Score metrics.