Microscopic image segmentation and tracking

The image processing step plays a crucial role in determining the quality of a microscopic imaging scenario. It is during this stage that biologically relevant data is extracted from the image, which ultimately determines how the cells will be categorized (by phenotype, by infection rate, etc.).

Image segmentation

The problem of automatically inferring which regions in the image correspond to biologically interesting object(s) (e.g. cell phenotypes, 3D sub-cellular structures) is a difficult problem. Like most inference problems, it can be framed in probabilistic setting. To obtain a solution, one attempts to maximize probabilities, which corresponds to minimizing their negative logarithm in an energy minimization approach (Horvath 2006, 2009). We are interested in segmentation approaches that combine shape models and classic energy minimization techniques and incorporate not only a priori shape information into image segmentation, but intensity and textural cues as well.


We are interested in developing methods for identifying and tracking cells or sub-cellular structures on live cell images. We have developed a software, the CellTracker, which corrects illumination problems, finds alignments, as well as automatically and manually tracks cells, mainly on phase contrast images. 

Microscopic image correction techniques

For quantitative measurements based on light microscopy and especially fluorescent intensities, it is essential to normalize the image data to correct for aberrations inherent in the acquisition process. One common source of error is the result of a non-ideal illumination field produced by the objective. Our novel algorithms, described in (Smith, et al. , Piccinini et al.), addresses these issues using energy minimization. The corrected field resulting from our approach is extremely flat, and we can achieve this level of quality without requiring a calibrated reference sample. 

Machine learning methods

The multi-parametric analysis step is concerned with interpreting the variety of collected information by identifying known patterns and discovering outliers. We are interested in the application of machine learning techniques for tasks such as phenotype identification and population analysis. We are developing the Advanced Cell Classifier (Horvath et. al. 2011; Piccinini et al. 2017), which provides an interface and machine learning software for large scale microscopic imaging scenarios.