Evaluation of methods for detection of fluorescence labeled subcellular objects in microscope images

Abstract

Several algorithms have been proposed for detection of subcellular objects in fluorescence microscope images. Commonly the algorithms have been designed for specific tasks and validated with very limited image data. Extensive comparisons between different algorithms could provide useful information and confidence on the method selection, but such studies have gained minor attention. For broader understanding of algorithm performance under varying conditions, we have carried out a comparison study including eleven spot detection or segmentation algorithms from various application fields. We used microscope images from well plate experiments of human osteosarcoma cell line, frames from image stacks of yeast cells with different focal planes, and simulated microscope images for comparing the methods indirectly and validating against reference result. The study shows major differences in the performance of different algorithms, suggesting the selection of detection algorithm for image based screens should be done carefully taking into account different conditions, such as the possibility of acquiring empty images or images with very few spots.

Supplementary figures

The following figures present supplementary results for the manuscript. For more detailed figures, click on the thumbnail images.

Figure 1. Example of a well plate image and detection results by 11 algorithms.

Supplementary material

Parameter ranges used in grid-search are given in a supplementary table.

Detection accuracies for simulated images and yeast images as a function of parameters are given in Figures 2 and 3 of the article. However, visualizations were limited to two parameters per method in order to simplify illustrations. Here, the rest of the parameters are covered for the methods with more than two user-tunable parameters.
Simulated:

Yeast:

The algorithms are available as a CellProfiler module.

The well plate images and simulated images are available for download.

Contact

Pekka Ruusuvuori
Department of Signal Processing
Tampere University of Technology
PO Box 553
33101 Tampere
FINLAND
pekka.ruusuvuori@tut.fi