Fast random walker for neutrophil cell segmentation in 3D

The random walker method [1] has many nice characteristics for 3D image segmentation. However, it is computational expensive and slow due to a massive linear system to be solved. In this work, we take advantage of the probabilistic output from the random walker method applied to a small downsampled image. By using the original image and seeds information, a novel edge-preserving method is introduced to upsample the probability, which is used for final segmentation. The running time is significantly reduced, taking less than one second to segment a 180×283×12 image stack on a normal laptop. Furthermore, the proposed method performs better when dealing with noisy boundaries than the original random walker method and the level set method.