Cell Tracking with Deep Learning and the Viterbi Algorithm

Abstract

We present a cell tracking pipeline that combines deep cell segmentation with a Viterbi algorithm tracker to accurately detect and track cells in microscopy videos. Our pipeline handles large illumination shifts, large appearance variability in the cells, and heavy occlusion from other cells and debris. We first train a Fully Convolutional Network (FCN) to detect the cells, then track the cells across frames using a tracker based on the Viterbi algorithm. We evaluate our algorithm on a dataset featuring Escherichia coli (E. coli) where the experimental goal is to immobilize the E. coli using blue light, thus making the dataset especially challenging due to large illumination shifts. Our results demonstrate that despite these challenges, our pipeline is able to accurately detect and track the cells.