Fine-grained and multi-pattern anti-nuclear antibody recognition: A new dataset and framework

Area of Science:

• Immunology
• Medical Imaging
• Artificial Intelligence

Background:

• Indirect immunofluorescence (IIF) staining of human epithelial cell (HEp-2) is crucial for anti-nuclear antibodies (ANA) screening.
• Current deep neural network approaches for ANA pattern recognition primarily focus on single, coarse-grained patterns, neglecting the complexity of multiple concurrent patterns defined by the International Consensus on Antinuclear Antibody Patterns (ICAP).

Purpose of the Study:

• To develop a large-scale dataset and a robust deep learning framework for fine-grained, multi-pattern recognition of ANA in IIF images.
• To establish an automated system for ANA pattern reading that aligns with ICAP standards and improves diagnostic accuracy.

Main Methods:

• Creation of a large-scale dataset with 40,000 high-resolution IIF images and precise annotations for 17 basic ICAP patterns and 5 pattern combinations.
• Development of a neural network framework comprising a mitotic detection network and a pattern classification network.
• Implementation of strategies such as dataset resampling, automatic image augmentation, and prediction aggregation to enhance model performance and robustness.

Main Results:

• The developed model achieved a high F-score of 0.883 on the test dataset, surpassing human technologists' performance.
• Assistance from the AI model led to significant and consistent improvements in human expert performance.
• The framework enables scalable, automated, and multi-pattern ANA reading in accordance with ICAP guidelines.

Conclusions:

• The proposed AI framework and dataset significantly advance automated ANA pattern recognition in IIF imaging.
• This technology offers a scalable and accurate solution for multi-pattern ANA detection, aligning with international diagnostic standards.
• The findings suggest a promising future for AI in improving the efficiency and reliability of autoimmune disease diagnostics.