Assessing the efficacy of low-level image content descriptors for computer-based fluorescence microscopy image analysis.

The increasing prevalence of automated image acquisition systems and state-of-the-art information technology has enabled new types of microscopy experiments based on automatic processing of massive image data sets, and numerous methods of high-content screening using machine vision and pattern recognition methods have been proposed. However, as a relatively young discipline, it is important to validate these methods and ensure that the machine vision and pattern recognition techniques reliably reflect the actual morphology, and can be effectively used for finding and validating scientific discoveries. In this report we show that some of the previously reported experimental results using automatic microscopy image analysis might be biased, and discuss practices and methods that can be used to obtain objective and reliable automatic analysis of microscopy images.