Cytoplasm image classification based on Kolmogorov complexity

Cell image similarity measurement is a fundamental and common issue in a broad range of problems, especially in small molecule screening. Existing cell image similarity measures are based on nucleus image segmentation result, which is the basis of subsequent feature extraction and classification. However, the need to set constraints on the segmentation algorithms, such as maximum/minimum cell size, means segmentation errors when nuclear or cell changes; even cells located in close proximity to each other would lead to segmentation errors. To overcome these limitations, a Cytoplasm Image Classification (CDC) method is proposed. It is based on Kolmogorov complexity, which promises to be optimal in theory, and hence easy-to-use in practice. Compared with traditional approaches, the CDC method analyzes cytoplasm images directly, requires no nucleus image segmentation and subsequent feature extraction to do classification, thus no human intervention step is needed. In classifying two cell image datasets, the CDC method shows comparable results to conventional analysis.