ИСТИНА

We introduce a novel automated technique for segmentation of anatomical structures in the images of Nissl-stained histological slices of mouse brain. Segmentation method includes atlas-based supervised learning. An experimental mouse brain slice is preliminarily associated with the corresponding slice from the mouse brain atlas. Then a preprocessing procedure is performed in order to enhance the quality of the experimental image and make it as close to the corresponding atlas image as possible. An effective method ofluminance equalization, which is an extension to Retinex algorithm, is proposed. A supervised learning is performed on the atlas image associated with the experimental slice. A Random Forest is trained on a data derived from the atlas image along with its annotated map, and experimental image pixels are then classified into anatomical structures. The result is refined by Markov random field. Preprocessing and segmentation procedures have been tested and evaluated on real experimental Nissl-stained slices.