Cascaded Multi-Class Pairwise Classifier (CascaMPa) For Normal, Cancerous, And Cancer Confounder Classes In Prostate Histology

TitleCascaded Multi-Class Pairwise Classifier (CascaMPa) For Normal, Cancerous, And Cancer Confounder Classes In Prostate Histology
Publication TypeConference Paper
Year of Publication2011
AuthorsDoyle, S, Feldman MD, Tomaszewski JE, Shih N, Madabhushi A
Conference NameIEEE International Symposium on Biomedical Imaging (ISBI)
Date Published2011 Mar 30

Gleason grading of prostate cancer is complicated by cancer confounders, or benign tissues closely resembling malignant processes (e.g. atrophy), which account for as much as 34% of misdiagnoses. Thus, it is critical to correctly identify confounders in a computer-aided diagnosis system. In this work, we present a cascaded multi-class pairwise classifier (CascaMPa) to identify the class of regions of interest (ROIs) of prostate tissue biopsies. CascaMPa incorporates domain knowledge to partition the multi-class problem into several binary-class tasks, reducing the intra-class heterogeneity that causes errors in one-versus-all multi-class approaches. Nuclear centroids are detected automatically via a deconvolution and watershed algorithm, and a set of features are calculated from graphs (Voronoi, Delaunay, and minimum spanning tree graphs) constructed on the centroids. The cascaded multiclass algorithm identifies each feature vector as one of six tissue classes: Gleason patterns 3, 4, and 5, normal epithelium and stroma, and non-cancerous atrophy (a cancer confounder) using a decision tree classifier. Performance of CascaMPa is evaluated using positive predictive value (PPV) and area under the receiver operating characteristic curve (AUC). For a database of 50 image patches per class (300 ROIs) from 118 patient studies, the classifier achieves an average PPV of 0.760 and AUC of 0.762 across all classes, while the oneversus- all approach yields an average PPV of 0.633 and AUC of 0.444.

PDF Link

 *IEEE COPYRIGHT NOTICE: 1997 IEEE. * Personal use of this material is permitted. However, permission to reprint/ republish this material for advertising or promotional purposes or for creating new collective works for resale or redistribution to servers or lists, or to reuse any copyrighted component of this work in other works must be obtained from the IEEE.

*COPYRIGHT NOTICE:* These materials are presented to ensure timely dissemination of scholarly and technical work. Copyright and all rights therein are retained by authors or by other copyright holders. All persons copying this information are expected to adhere to the terms and constraints invoked by each author's copyright. In most cases, these works may not be reposted without the explicit permission of the copyright holder.