Cell detection in phase-contrast images used for alpha-particle track-etch dosimetry: a semi-automated approach.

A novel alpha-particle irradiator has recently been developed that provides the ability to characterize cell response. The irradiator is comprised of a collimated, planar alpha-particle source which, from below, irradiates cells cultured on a track-etch material. Cells are imaged using phase-contrast microscopy before and following irradiation to obtain geometric information and survival rates; these can be used with data from alpha-particle track images to assess cell response. A key step in this process is determining cell location within the pre-irradiation images. Although this can be done completely by a human observer, the number of images requiring analysis makes the process time-consuming and tedious. To reduce the potential human error and decrease user interaction time, a semi-automated, computer-aided method of cell detection has been developed. The method employs a two-level adaptive thresholding technique to obtain size and position information about potential cell cytoplasms and nuclei. Proximity and geometry-based thresholds are then used to mark structures as cells. False-positive detections from the automated algorithm are due mostly to imperfections in the track-etch background, camera effects and cellular residue. To correct for these, a human observer reviews all detected structures, discarding false positives. When analysing two randomly selected cell dish image databases, the semi-automated method detected 92-94% of all cells and 94-97% of cells with a well-defined cytoplasm and nucleus while reducing human workload by 32-83%.