Model-based automated segmentation of kinetochore microtubule from electron tomography

The segmentation of kinetochore microtubules from electron tomography is challenging due to the poor quality of the acquired data and the cluttered cellular surroundings. We propose to automate the microtubule segmentation by extending the active shape model (ASM) in two aspects. First, we develop a higher order boundary model obtained by 3-D local surface estimation that characterizes the microtubule boundary better than the gray level appearance model in the 2-D microtubule cross section. We then incorporate this model into the weight matrix of the fitting error measurement to increase the influence of salient features. Second, we integrate the ASM with Kalman filtering to utilize the shape information along the longitudinal direction of the microtubules. The ASM modified in this way is robust against missing data and outliers frequently present in the kinetochore tomography volume. Experimental results demonstrate that our automated method outperforms manual process but using only a fraction of the time of the latter.