Contrastive Learning With Transformer to Predict the Chronicity of Children With Immune Thrombocytopenia

Area of Science:

• Pediatric Hematology
• Immunology
• Computational Biology
• Machine Learning

Background:

• Immune thrombocytopenia (ITP) is an immune-mediated bleeding disorder in children, with approximately 20% progressing to chronic disease.
• Chronic ITP in children reduces quality of life and increases treatment burden, necessitating early prediction for personalized management.
• Existing challenges in predicting ITP chronicity include small and imbalanced pediatric patient datasets, hindering effective deep learning model training.

Purpose of the Study:

• To develop a novel deep learning method capable of accurately predicting chronic immune thrombocytopenia (ITP) in children.
• To address the challenges of small data size and class imbalance inherent in pediatric ITP datasets.
• To enable early, personalized treatment planning for children diagnosed with ITP.

Main Methods:

• Proposed a novel method integrating contrastive learning with the FT-Transformer architecture to process heterogeneous tabular data.
• Employed random masking for data amplification and oversampling to balance the imbalanced dataset.
• Constructed contrastive pairs using latent representations from the FT-Transformer encoder to leverage synthetic data.

Main Results:

• The proposed method demonstrated superior performance compared to state-of-the-art approaches on real-world pediatric ITP data.
• The approach effectively handled insufficient and imbalanced data issues, crucial for pediatric ITP research.
• Experimental results validated the significant advantages of the integrated contrastive learning and Transformer model.

Conclusions:

• The developed deep learning method offers a promising solution for predicting ITP chronicity in children.
• This approach effectively overcomes common data limitations in pediatric hematological studies.
• Accurate prediction of ITP chronicity can lead to improved clinical decision-making and patient outcomes.