Development of automatic FISH probe counting in CTC

Sjoerd T. Ligthart, Joost F. Swennenhuis, Jan Greve, Leon W.M.M. Terstappen

Research output: Contribution to conferencePosterOther research output


Introduction: Presence of Circulating Tumor Cells (CTC) in blood of patients with metastatic carcinomas has been associated with poor progression free and overall survival. Characterization of CTC can be performed by Fluorescence In Situ Hybridization (FISH), however counting of FISH dots by human reviewers can be tiring and subjective and thus likely produces variable outcomes. We investigated whether automated counting of FISH dots in CTC is comparable to the counts obtained by expert reviewers.

Material and Methods: Samples processed on the CellSearchTM system for CTC counting were hybridized with fluorescent DNA probes targeting the HER2/neu gene region and the centromeric region of chromosome 17 or the centromeric regions of chromosome 1,7,8,17. (1) For optimization of the algorithm 492 Z-stacks from leukocytes carried over through the CellSearch procedure were recorded and a maximum intensity profile (MIP) was created. Five reviewers counted FISH dots in the MIP data set to create a ground truth. The automatic counting algorithm was validated in a set of stored images of CTC probed for chromosome 1,7,8,17 from castration resistant prostate cancer patients (CRPC).(1)

Results: The data set with carried-over leukocytes was counted reliably by the algorithm: 97.8% of the HER2/neu FISH dots and 97.5% of the centromeric 17 dots were counted equally by the PC and the reviewers, regarding only the subset of images for which all the reviewers agreed. The mean intra-reviewer agreement was 96.5%. In the validation set copy number of chromosome 1 in carried-over leukocytes scored by an expert agreed in 50.8% with the automated count, for chromosome 7 34.4% for chromosome 8 22.8.% and for chromosome 17 55.0%. For CTC in the validation set copy number of chromosome 1 scored by an expert agreed in 71.6% with the automated count, for chromosome 7 56.2% for chromosome 8 64.8% and for chromosome 17 41.0%. For copy numbers larger than 6 both the expert and automated count were recorded as larger than 6. Agreement between the expert count and automated count did not significantly alter with the detected copy number. Over-count in the validation set was largely due to clustering of nuclei that were counted as one cell.

Conclusions: Automatic FISH dot counting in CTC images is feasible for images where reviewers agree upon. The intra-reviewer variation of 3.5% shows that reviewers have ambiguous rules and are not reliable. This variation is closely related to the heterogeneity -size and shape- of the FISH dots. The PC has a reproducibility of 100% and is thus a good replacement for human reviewers.
Original languageEnglish
Publication statusPublished - 19 Apr 2010
Event101st AACR Annual Meeting 2010 - Washington, United States
Duration: 17 Apr 201021 Apr 2010
Conference number: 101


Conference101st AACR Annual Meeting 2010
Country/TerritoryUnited States


  • METIS-269639


Dive into the research topics of 'Development of automatic FISH probe counting in CTC'. Together they form a unique fingerprint.

Cite this