Automated image analysis for electron microscopy specimen assessment

This paper presents an automatic image analysis process that emulates the decision of a microscopist in evaluating 2D crystallized proteins. The purpose of the process is to locate the interesting regions in each image and finally assess if the crystallization succeeded. A top-down approach decomposes the process into three steps, corresponding to levels of magnification of the transmission electron microscope. For the first step (low magnification), the automatic process efficiently evaluates the quality of the grid where the specimen relies. In the second step (medium magnification), the protein-embedded membranes are analyzed. The images are classified with a histogram-based analysis to reject uninteresting images and the remaining ones are processed with an edge-based algorithm to localize the membranes and select the potentially crystallized areas. In the last step (high magnification), an algorithm extracts automatically the peaks from the diffraction pattern of the previously selected areas. The tests show encouraging results.