Automatic detection of neurons in NeuN-stained histological images of human brain

Abstract In this paper we propose a new method for automatic detection of neurons in histological sections of the human brain cortex, based on anisotropic diffusion. The anisotropic diffusion is modeled using a partial differential equation (PDE) and is applied to high resolution microscopy images of the brain in order to detect neurons. We also present a novel approach for PDE-model parameter optimization. Due to the issue of inter-observer variability, three human experts have manually annotated neurons in the image dataset on which the proposed method was trained. The average correlation in neuron detection between the human experts was 86 . 88 % , while the average correlation between the proposed method and the human experts is 88 . 79 % , which shows that the proposed method’s performance is equal to that of human experts. Moreover, the proposed automatic method provides consistent and reproducible results on all sections and is much faster than human raters or other automatic methods. Additionally, the proposed method’s output was verified by a human expert and has correctly distinguished 95 . 41 % of neurons in the test images.