Automated Segmentation of Microtubules in Cryo-EM Images with Excessive White Noise

Abstract

Microtubules are essential part of the cell structure. Cryo-Electron Microscopy (Cryo-EM) is employed to collect visual images of the microtubules in atomic resolution. However the electron microscopic images suffer from excessive white noise and incoherent cell structures. The Signal-to-Noise Ratio (SNR) of the images is rather low (less than 0.1). Automated segmentation of the microtubule region from high-noise electron micrographs is of significance, otherwise scientists have to manually locate and extract microtubules from a large amount of noisy data. Here, we proposed a new composite algorithm based on Chan-Vese (CV) model, which consists of three steps: (1) Remove the contaminant area in the micrograph, where the Otsu algorithm and the morphological closing operation are utilized. (2) Enhance the microtubules image by using improved diffusion filtering algorithm. (3) Use the adapted CV model to segment the microtubules. Our test results show that the new algorithm is quite effective for segmentation and extraction of microtubules from highly noisy images, with few errors and less time-cost.