An Automatic Quantitative Evaluation of Ladder Pattern Presented in Intercellular Spaces in Histopathological Images of Reflux Disease Specimen Stained with Desmoglein-3 Antibody

Following paper presents automatic method for quantitative evaluation of ladder pattern, which can appear in microscopic images of the esophageal epithelium tissue in a course of gastroesophageal reflux disease (GERD). Studies carried on the electron microscopy biopsy samples, have shown relation between dilated intercellular spaces (DIS) and treatment response. According to the studies, DIS can be classified as "bubbles" and "ladders". Performing electron microscopy diagnosis generates high cost for hospitals, although immunohistochemistry (ihc) offers another cheaper possibility of DIS visualization. Using ihc technique causes problem with less noticeable, therefore harder to evaluate structures. This results in high interobserver verbality, and questionable manual evaluation. Aim of our paper was to design an algorithm, which would allow automated morphometric study of ladder patterns that were visualized by ihc staining techniques. For automatic intercellular binding detection, we designed context based method. Neighborhood mask is divided in two areas: foreground, which should contain line structures and background, where we should not find them. Detection of intercellular binding is done by evaluating location of pixel's collection in respect to mask's background and foreground. Intercellular binding has a shape of curved lines; therefore, mask areas have line characteristic as well. Identified structures were divided on the longest line segment possible, which ware evaluated using morphological covariance. As estimated distance value, we took distance between peaks seen on morphological covariance plot. Our proposed method could grade 77% of segments evaluated by expert with relevant mean squared error of 10%, therefore, showing possibility of automatic quantitative evaluation of ladder pattern.