Mitochondria is an important organelle of eukaryotic cells. Its intracellular segmentation presents a challenge as mitochondria differs qualitatively and quantitatively according to the cell in which they are located. The classic approach of mitochondria segmentation is manual, which is time consuming and prone to human error and interpretation. As a part of this diploma thesis, we are presenting a procedure that enables automatic segmentation of mitochondria over anisotropic volumetric data obtained with an electronic microscope (EM). We propose a procedure or a pipeline, which includes operations on volumetric data, and automatic segmentation using a convolutional neural network. We have evaluated the results of automatic segmentation using two different test sets. The evaluation showed very promising results – the classification accuracy was 99 % on both test sets.
|
