VMAC: overlapping cervical cell segmentation from label-free quantitative microscopy images.

Cervical cancer is one of the most common cancers in women. Non-invasive cytopathological methods are becoming increasingly important in the routine surveillance and early diagnosis of cervical cancer. However, reliable approaches are usually based on an extended staining process with subjective interpretation. Early detection of cervical cancer increases survival rates. Using unstained Differential Interference Contrast (DIC) images, the current study reveals the morphological changes in cervical cells towards cancer progression. To overcome the challenges of overlapping cells, the Voronoi-based mixed-breed active contour (VMAC) method was applied to segment overlapping cells in cytopathological images. In VMAC-based segmentation, a Voronoi diagram divides the preprocessed input image. In addition to that, the energy associated with the information stored by each individual cell is reduced through the use of the Voronoi diagram and the active contour model. Although region-based active contour model is well suited for segmenting images with imprecise edges, their applications to images with variation in intensities lead to an objective interpretation. An experimental result shows that the validation of the proposed method provides accurate results for 2 to 10 cells, with an average false negative rate of 0.1. The proposed model has an accuracy and true positive rate of 0.97 and 0.96, respectively. [ABSTRACT FROM AUTHOR]