br Fig Graphical effect of cell overlapping on reconstructio
Fig. 8. Graphical effect of cell overlapping on reconstruction accuracy: left – impact of Necrostatin 1 overlapped by other cells, right - impact of a cell overlapped by stroma; the blue color represents the obtained accuracy and the red color indicates the trend line.
Numerical results of experiments presenting range of cell overlap-ping level.
of the image the cells identified using the watershed method and then reconstructed by applying the presented procedure. The to-tal reconstructed cell area (RecArea) and area of the original cell (OrigArea) could be characterized by the parameter known as over-lap_level:
The numerical results of the experiment are displayed in Table 2, in which the area of the cells is expressed in μm2 (the area of the pixel is equal 0.0703 μm2), and overlap_level in %. A low value of the overlap_level provides evidence that the cell reconstructed from the distorted original prototype should be included in the FISH analysis procedure, while a high value indi-cates that it should be omitted. For the analyzed set of cells, the overlap_level ranged from 27% to 62%, with a standard deviation of 9.11%. This confirms that the percentage of distorted cells obtained in the watershed method was relatively high and reconstruction was urgently required.
The results of all performed experiments demonstrated strong applicability of the presented methods in proper segmentation of the FISH images representing breast cancer. The method enabled a proper set of cells to be obtained in the analyzed field of view, which was compatible with the expert decisions and achieved high accuracy of the segmented cell area compared to the true one, irrespective of the deformation of cells. However, the numer-ical parameterization of the overlapping phenomena, obtained as a side effect of the method, is another important factor for medi-cal experts, because it allows for deciding in an objective manner whether a given cell should be considered or omitted in the anal-ysis.
The paper has presented an innovative method for cell recon-struction in breast cancer images based on the comparison of pat-terns by applying the PatchMatch approach. The developed method was verified on the cell nuclei in microscopic images obtained in FISH cytogenetic medical tests. The method enables accurate re-construction of cells incorrectly segmented by the watershed algo-rithm.
The experiments performed confirmed the high effectiveness of the reconstruction method on the tested images. It can properly re-construct deformed cells with a large variation in shape and size,
and it can deliver correct results in cell image segmentation. It can effectively deal with the distortions caused by the stroma or other types of cell overlapping. Moreover, it allows for assessing the pre-cise level of cell overlapping, which provides significant hints for medical experts in image analyses.
The presented reconstruction technique is of general use and can be applied in the reconstruction of different cell types, and even in other staining technologies. The analysis can be performed in a timely and cost-effective manner. Therefore, the proposed au-tomatic segmentation method offers considerable promise for the automating analysis. The computer program developed for apply-ing the presented reconstruction procedure can substitute human medical experts in this tedious and error-prone work, delivering results that are acceptable in medical practice within a substan-tially shorter time.
Future research in this field will be directed toward the follow-ing tasks:
• The proposed method will be implemented in hospital practice. Additional experiments on a higher number of images repre-senting the enlarged population of patients will be required.
• The set of typical cell patterns corresponding to the true cell types will be expanded, based on which reconstruction will be performed.
• Future investigations will focus on applying the method to other staining technologies, such as Ki-67 or ER/PR staining, which are used in different types of pathology.
• Additional efforts will be directed at accelerating the computa-tion by implementing parallel processing of the data.
Declaration of interests
The authors declare that they have no known competing finan-cial interests or personal relationships that could have appeared to influence the work reported in this paper.
Credit authorship contribution statement
T. Les: Conceptualization, Formal analysis, Investigation, Methodology, Software, Validation, Writing - original draft. T. Markiewicz: Conceptualization, Methodology, Project administra-tion, Resources, Supervision. S. Osowski: Methodology, Writing - original draft, Writing - review & editing. M. Jesiotr: Data curation, Resources, Validation.