Automatic Classification and Visualization of Text Data on Rare Diseases.

Area of Science:

• Medical Informatics
• Computational Biology
• Rare Diseases

Background:

• Over 7000 rare diseases affect 400 million people globally, presenting significant research and healthcare challenges.
• Precision medicine and artificial intelligence (AI) offer potential solutions for rare disease management.
• Accurate classification of medical information is crucial for advancing research and patient care.

Purpose of the Study:

• To develop and evaluate an AI-powered classifier for distinguishing between rare and non-rare disease research and news articles.
• To enhance the categorization of medical literature using established ontologies.
• To improve the identification of information relevant to rare diseases, including neurodevelopmental disorders.

Main Methods:

• Extracted 709 rare disease terms from Mondo and Medical Subject Headings (MeSH) to build a robust classifier.
• Evaluated the classifier on PubMed/MEDLINE abstracts and an expert-annotated news dataset.
• Focused analysis on four rare neurodevelopmental disorders (NDDs) within the broader dataset.

Main Results:

• Achieved an F1 score of 85% for classifying research abstracts.
• Attained an F1 score of 71% for classifying news articles.
• Demonstrated classifier robustness across diverse datasets, highlighting AI's potential in disease classification.

Conclusions:

• The developed AI classifier effectively distinguishes rare disease content, improving medical information categorization.
• Results underscore the need for continued refinement in handling data heterogeneity for AI models.
• The classifier supports advancements in rare disease research, information accessibility, policy development, and personalized medicine.