A comparison of image deconvolution algorithms applied to the detection of endocytic vesicles in fluorescence images of neural proteins

In this work we present a comparative study of three image deconvolution methods applied to fluorescence images of neural proteins. The purpose of this work is to compare the efficiency of these methods, in order to establish which one performs better the restoration of this type of image. Moreover we show that image deconvolution improve not only image quality, but detection capabilities and thus the counting of endocytic vesicles. Image deconvolution was performed by Gold-Meinel (GM) and Lucy-Richardson Maximum likelihood (LRML) non-blind methods and by Lucy-Richardson Maximum likelihood blind method (LRMLB). These methods were tested in 120 images from two different experiments. Computed theoretical point spread function (psf) was used for non-blind deconcovolution methods. Twenty five iterations were performed to restore each image using GM and LRML algorithms. In the case of LRMLB, 10 cycles were performed with 15 psf iterations and 5 image iterations per cycle to deconvolve each image. Endocytic vessels' counting was manually made in deconvolved and non-deconvolved images by a trained observer. Results showed an increase of 22% and 24% in the detection of endocytic vessels using LRML and LRMLB methods respectively and a decrease of 6% using GM method, against detection with non deconvolved images.