Learning to Detect Different Types of Cells under Phase Contrast Microscopy

Abstract

We propose a learning-based algorithm to detect various types of cells in phase contrast microscopy images. The algorithm automatically adapts to different cell types by learning variations in cell appearances and shapes, while staying discriminative against non-cell distractions in the background. Benefiting from a rich set of carefully designed feature descriptors, the proposed algorithm is able to detect considerably different types of cells with high accuracy. With experimental results achieving an average detection F-measure of 90%, the statistical learning framework proposed in this paper has proven to be promising in evolving into a highly adaptive and robust cell detection system.