Quantitative Histopathological Analysis of CIN Sections

Quantitative histopathological analysis of CIN sections To the Editor, Cervical histological and cone biopsies constitute a major portion of the work load in a surgical pathology department (estimated 3% of all histological specimens). Post-diagnostic treatment of patients with a histologically confirmed cervical intraepithelial neo-plasia (CIN) largely depends on CIN grade, which is prognostic as to the risk of subsequent progression [14]. However, inter-observer reproducibility of normal , reactive, regenerative, metaplastic, and different CIN grades is notoriously low [8,17]. This explains the attempts, during the past three decades, to find quantitative histopathological sample classifiers for normal, koilocytotic and CIN grades. The system they are developing allows surgical pathologists to mouse-select a certain diagnostic area in a section. Subsequent digital image-supported sample classification of this selected image is then performed in seconds, giving pathologists the adjunct information requested with a minimum of delay in the flow of routine work. This is an important factor in the acceptance of a new diagnostic or therapeutic laboratory test [3,18]. However, before arriving at the point of daily implementation of these methods, many hurdles have to be crossed. A recent work by Guillaud et al. [4] discusses methodological issues of Quantitative histopathologi-cal analysis of cervical intra-epithelial neoplasia sections. For this study, they used digitized images from 280 samples, classified by consensus expert gynaeco-logical pathologists as normal (n = 199), koilocy-A cyto-technician delineated the basal membrane and superficial surface to select a particular epithelial (The intra-and inter-observer variation in number of cells selected is significant). This is the only interactive part of the analysis, the rest is automated. For densitomet-ric features, normalization is performed by the system. The cells and samples are then classified using geometric (size, shape, others), densitometric (darkness, etc.) and texture (chromatin distribution) features of the nu-clei. Using different feature combinations and groupings of the original 5 subclasses of cases, they arrive at a correct classification of roughly 85–95% of individual cells and 80% of the samples (65% for the CIN lesions). They concluded that (1) The use of intensity normalization from a subset of imaged non-overlapping intermediate layer cells works as well or better than any of the other methods tested and provides significant time saving; and (2) Although this result must be tested in a larger data set, the exclusive use of intermediate layer cells may be acceptable when using quantitative histopathology. In spite of the significant amount of work done and …