Automatic detection of neurons in high-content microscope images using machine learning approaches

The study of neuronal cell morphology and function in relation to neurological disease processes is of high importance for developing suitable drugs and therapies. To accelerate discovery, biological experiments for this purpose are increasingly scaled up using high-content screening, resulting in vast amounts of image data. For the analysis of these data fully automatic methods are needed. The first step in this process is the detection of neuron regions in the high-content images. In this paper we investigate the potential of two machine-learning based detection approaches based on different feature sets and classifiers and we compare their performance to an alternative method based on hysteresis thresholding. The experimental results indicate that with the right feature set and training procedure, machine-learning based methods may yield superior detection performance.