A class discovery and class prediction approach to histopathological classification of mammographic screen detected columnar cell lesions of the breast

Columnar cell lesions (CCLs) of the breast have been increasingly recognised in biopsies taken to investigate mammographic screen detected microcalcification. The aim of this study was to identify distinct CCL subtypes by systematic analysis of histopathology. Hierarchical cluster analysis was performed based on the profile of histopathological features in 102 screen detected CCLs. Features assessed included nuclear morphology, acinar dilatation, epithelial cell hyperplasia, cell crowding, apical snout formation and intraluminal secretion. The stability of this classification was tested in an independent cohort of 32 cases. The histopathology of screen detected CCLs was extremely variable. Hierarchical cluster analysis identified two subclasses: Class 1 (34/102, 33%) characterised by absence of nuclear atypia and less pronounced hyperplasia; and Class 2 (68/102, 67%) that were generally more atypical. Ki-67 scores were significantly lower for Class 1 CCLs (p