Semi-automatic/user-guided segmentation of mitochondria on transmission electron microscopy images

Çöçelli, Mustafa
The purpose of the study is to develop and implement some user-guided/semi-automatic methods to segment mitochondria on transmission electron microscopy images. Earlier automatic active contour based detection and segmentation algorithm applied on slices of mitochondria dataset are subject to some errors. Therefore, semi-automatic methods are necessary to adjust structure of wrongly detected and segmented mitochondria. There are six semi-automatic methods introduced in this thesis, namely, splitting, merging, deletion, initiation of automatic segmentation algorithm at intended position, selection of low scored mitochondria and manual drawing. In summary, firstly, splitting is separation of wrongly detected mitochondrion into two pieces with the help of generic surface defined by the user. Merging is the combination of two mitochondria which results in formation of one valid mitochondrion. Deletion allows removing false detections in non-mitochondrial region. Initiation of automatic algorithm at userspecified position is done by softening validation phase parameters of automatic segmentation algorithm. Lastly, manual drawing is to form 3D shape of a mitochondrion from user-defined points. Prior to the implementation of these semi-automatic methods, mitochondrial images and the output of the automatic algorithm were visualized to ensure user evaluation on mitochondrial images.