Detection of protein spots in real gel images by a two-phase marker-controlled watershed method

Protein spot detection is an important step in gel image analysis. The results of spot detection may substantially influence gel image analysis. However, accurate automatic spot detection is still difficult due to spots overlapping, background variation and various inevitable artifacts. In this paper, we propose a robust spot detection algorithm which is capable of detection of spots in real gel image. It is achieved by a two-phase process. Firstly, a standard marker-controlled watershed algorithm was applied to roughly segment gel image. Secondly, we identify the segmented regions as normal or complex region by the region shape factor. For those complex region contained multiple overlapping spots, we estimate spot centroids through Euclidean distance transform. Finally, the estimated centroids are taken as spot markers to do a fine marker-controlled watershed segmentation. Experimental results show that the proposed method can steadily obtain high detection performance in real gel images, and can accurately identify overlapping spots in complex region, when compared to other popular spot detection methods.