Automated three-dimensional image analysis of thick and overlapped clusters in cytologic preparations. Application to cytologic smears.

Methods are presented for automated analysis of thick and heavily overlapped regions of cytologic preparations, such as cervical/vaginal smears. Current systems are unable to process these regions although they contain diagnostically valuable information. We argue that analysis of such regions is inherently a three-dimensional (3-D) problem that cannot be solved reliably with conventional two-dimensional methods. Furthermore, this issue cannot be side-stepped by special thin preparation methods. Even with 3-D imaging, analysis of these regions is complicated by the high variability in the image gray level and textural features resulting from the uncontrollable cell overlaps and folds and large computational requirements. A novel approach based on 3-D imaging and adaptive 3-D analysis algorithms based on the principles of localization, adaptive data reduction and clustering theory is presented. It was successful in detecting and separating deeply embedded and overlapping nuclei, cytoplasmic folds and creases in thick and overlapped regions of conventional smears and special thin preparations.