To determine the boundaries of objects in an image, binarization is used – a process in which an image in grayscale is converted to binary image consisting only of black and white pixels, which allows you to separate the objects of interest from the background for subsequent analysis. In the task of determining the numerical values of the dendritic spines morphological parameters in two-dimensional images, it is necessary to designate a clear boundary of the dendritic membrane protrusions, which will determine the area of calculations. Dendritic spines in confocal images are extremely small objects and vary significantly in the brightness of the pixels that form them. The author compares global and adaptive binarization method on a sample of two-dimensional projections of hippocampal neurons in vitro confocal images series after deconvolution. Hippocampal neurons were transfected with a plasmid encoding the mCherry fluorescent protein on the 7th DIV and fixed with a 4% solution of paraformaldehyde in phosphate-buffered saline on the 14-15th DIV. Among the global binarization methods presented in ImageJ software, the best result was demonstrated by the Lee segmentation method based on minimizing the relative (cross) entropy. In local binarization methods, the threshold value is determined for each pixel separately, depending on the intensity of the pixels surrounding it. The value of the gray level in a pixel was calculated as the average of the minimum and maximum values of the gray level inside the sphere, the center of which is located in the considered pixel, with an arbitrarily chosen radius r. For a comparative assessment of global and local binarization, a multi-scale structural similarity index (MS_MSSI) between the confocal and binary image was calculated. The average MS_MSSI for 10 different fragments was 0.81 ± 0.02 for Lee global binarization method and 0.91 ± 0.01 for local binarization method (p <0.001). According to the MS_MSSI index maps, local binarization performes significantly better then global in cases where the dendritic spine is characterized by an inhomogeneity of intensity, the dendritic spine is small in size and two spines are located in close proximity to each other. Thus, the application of the local method in the task of binarization of images containg such small objects as dendritic spines allows one to achieve better results than global methods.
Пчицкая Е. И.