Aggregation of classifiers for staining pattern recognition in antinuclear autoantibodies analysis

Area of Science:

• Medical diagnostics
• Computational pathology
• Immunofluorescence assays

Background:

• Indirect immunofluorescence is the standard for antinuclear autoantibody (ANA) detection.
• Accurate assessment requires evaluating fluorescence intensity and staining patterns.
• Limited resources and expertise necessitate automated diagnostic support tools.

Purpose of the Study:

• To develop a computer-aided diagnosis (CAD) system for classifying ANA staining patterns.
• To support physician decision-making in interpreting immunofluorescence results.
• To introduce a novel reliability measure for classification output.

Main Methods:

• A multiple expert system (MES) using a one-per-class approach for single-cell pattern labeling.
• Hybrid architecture combining ensembles of classifiers with a fusion rule.
• Feature extraction from statistical and spectral measurements, including a new reliability parameter.

Main Results:

• The system achieved a low overall error rate (2.7%-5.8%) on 573 cells from 37 wells.
• The developed reliability parameter enabled a reject option, reducing classification errors.
• Performance was below the observed intralaboratory variability, indicating high accuracy.

Conclusions:

• The proposed CAD system effectively classifies ANA staining patterns.
• The novel reliability measure enhances classification accuracy by allowing error reduction.
• This automated approach offers a valuable tool for ANA diagnosis, especially where resources are limited.