Application of image processing in neurobiology: Detection of low signals with high spatial resolution and a non-uniform variance

Abstract

Optical imaging of neuronal activity is a recent method for studying information processing by neuronal networks. It is known that neuronal activity was generated by the activity of membrane proteins, called “ion channels”.

To see the spatial action of channels on a single neuron, we use an imaging system combining a CCD camera mounted on an inverted microscope and fluorescence voltage sensitive probe and taking images of fluorescence intensity with high spatial resolution and low noise.

To visualize the difference between an excitated and a quiet state of a neuron we must compute the relative variation in fluorescence between both images representing both states. We obtained a very noisy resulting image of low signals. We have shown that its variance is non uniform and in inverse ratio to the square of raw data intensity and that the highest variances are located on the non-biological background.

We have developed a nonlinear filtering based on local image segmentation. The algorithm filters all the noise in the non-biological background and outlines the shape of the cell body. Tests of significance between groups of intensities have shown that the response of the neuron is patchy and biological experiments show that the patchyness is bound to electrical activities of the neuron.