Send to

Choose Destination
Cytometry A. 2015 Jun;87(6):524-40. doi: 10.1002/cyto.a.22632. Epub 2015 Jan 28.

Large-scale tracking and classification for automatic analysis of cell migration and proliferation, and experimental optimization of high-throughput screens of neuroblastoma cells.

Author information

Department of Bioinformatics and Functional Genomics, Biomedical Computer Vision Group, BioQuant and Institute of Pharmacy and Molecular Biotechnology (IPMB), Heidelberg University, 69120, Heidelberg, Germany.
Division of Theoretical Bioinformatics, German Cancer Research Center (DKFZ), 69120, Heidelberg, Germany.
Division of Neuroblastoma Genomics, German Cancer Research Center (DKFZ), 69120, Heidelberg, Germany.
Integrated Research and Treatment Center, Center for Sepsis Control and Care (CSCC), Jena University Hospital, 07747, Jena, Germany.
Network Modeling, Leibniz Institute for Natural Product Research and Infection Biology - Hans Knöll Institute Jena, 07745, Jena, Germany.


Computational approaches for automatic analysis of image-based high-throughput and high-content screens are gaining increased importance to cope with the large amounts of data generated by automated microscopy systems. Typically, automatic image analysis is used to extract phenotypic information once all images of a screen have been acquired. However, also in earlier stages of large-scale experiments image analysis is important, in particular, to support and accelerate the tedious and time-consuming optimization of the experimental conditions and technical settings. We here present a novel approach for automatic, large-scale analysis and experimental optimization with application to a screen on neuroblastoma cell lines. Our approach consists of cell segmentation, tracking, feature extraction, classification, and model-based error correction. The approach can be used for experimental optimization by extracting quantitative information which allows experimentalists to optimally choose and to verify the experimental parameters. This involves systematically studying the global cell movement and proliferation behavior. Moreover, we performed a comprehensive phenotypic analysis of a large-scale neuroblastoma screen including the detection of rare division events such as multi-polar divisions. Major challenges of the analyzed high-throughput data are the relatively low spatio-temporal resolution in conjunction with densely growing cells as well as the high variability of the data. To account for the data variability we optimized feature extraction and classification, and introduced a gray value normalization technique as well as a novel approach for automatic model-based correction of classification errors. In total, we analyzed 4,400 real image sequences, covering observation periods of around 120 h each. We performed an extensive quantitative evaluation, which showed that our approach yields high accuracies of 92.2% for segmentation, 98.2% for tracking, and 86.5% for classification.


cell classification; cell tracking; experimental optimization; high-throughput screening; model-based error correction

[Indexed for MEDLINE]
Free full text

Supplemental Content

Full text links

Icon for Wiley
Loading ...
Support Center