A Benchmarking Platform for Mitotic Cell Classification of ANA IIF HEp-2 Images

Anti-nuclear antibody (ANA) indirect immunofluoresence (IIF) human epithelial type-2 (HEp-2) testing provides important clinical information for the diagnosis of systemic autoimmune rheumatic diseases (SARD). Recent developments in computer aided diagnosis (CAD) systems aim to improve the reliability and reproducibility of ANA IIF laboratory testing by providing ANA classification. The limited prior work into ANA IIF HEp-2 mitotic classification derives its protocol from clinically abstract performance metrics dependent on a single data set of FITC channel images. In this paper, we propose an extensive benchmarking platform considering both the DAPI and FITC channel images for mitotic classification. Furthermore, we argue that pedestrian detection metrics better define the mitotic classification problem. To support these propositions, we design a simple classifier on the DAPI channel and use meaningful performance metrics to demonstrate that it significantly outperforms the most recent state-of-the-art approach.