A Novel Spot-Enhancement Anisotropic Diffusion Method for the Improvement of Segmentation in Two-dimensional Gel Electrophoresis Images, Based on the Watershed Transform Algorithm

Two-dimensional gel electrophoresis (2DGE) is a powerful technique in proteomics for protein separation. In this technique, spot segmentation is an essential stage, which can be challenging due to problems such as overlapping spots, streaks, artifacts and noise. Watershed transform is one of the common methods for image segmentation. Nevertheless, in 2DGE image segmentation, the noise and artifacts of images cause over-segmentation in the watershed algorithm. In this study, we proposed a novel spot-enhancement anisotropic diffusion (SEAD) method, based on multi-scale second-order derivatives and eigensystemto enhance the spots and remove noise and artifacts. The proposed SEAD algorithm was plugged to a watershed transform in order to improve the performance of watershed segmentation algorithm. The performance of the proposed SEAD method was evaluated on synthetic and real 2DGE images. The proposed algorithm was compared with other segmentation methodsin terms of different criteria including efficiency, precision and true positive rate. The performance of the methods were evaluated in the presence of noise and the results were evaluated by t-test. According to the count of detected spots, precision and efficiency of the proposed method were 0.82 and 0.67 respectively. The precision and efficiency values of the comparative methods were as follows: 0.65 and 0.42 for MCW algorithm, 0.40 and 0.37 for BWT method, 0.74 and 0.53 for the method proposed by Kostopoulou and 0.76 and 0.55 for the method proposed by Mylona. The comparison of the proposed method with four other conventional methods revealed the superiority and effectiveness of the proposed SEAD method.