HEp-2 fluorescence pattern classification

• Comparison of different textural features for HEp-2 pattern classification.
• Analysis of experimental protocols used for evaluation of HEp-2 algorithms.
• Currently available data sets are insufficient.
• Cell-level classifier performance does not predict sample-level performance.
• Complete measurements from all cells are a better basis for sample decisions.

Automation of HEp-2 cell pattern classification would drastically improve the accuracy and throughput of diagnostic services for many auto-immune diseases, but it has proven difficult to reach a sufficient level of precision. Correct diagnosis relies on a subtle assessment of texture type in microscopic images of indirect immunofluorescence (IIF), which has, so far, eluded reliable replication through automated measurements. Following the recent HEp-2 Cells Classification contest held at ICPR 2012, we extend the scope of research in this field to develop a method of feature comparison that goes beyond the analysis of individual cells and majority-vote decisions to consider the full distribution of cell parameters within a patient sample. We demonstrate that this richer analysis is better able to predict the results of majority vote decisions than the cell-level performance analysed in all previous works.