Decision support methods for finding phenotype--disorder associations in the bone dysplasia domain

Area of Science:

• Medical Informatics
• Artificial Intelligence in Medicine
• Genetics

Background:

• Diagnosing skeletal dysplasias, a group of rare genetic disorders, is challenging due to limited domain knowledge and established guidelines.
• Machine learning (ML) offers potential for objective medical decision support but struggles with rare diseases due to sparse knowledge bases.

Purpose of the Study:

• To develop a decision support model for medical domains with sparse knowledge bases, specifically for diagnosing rare genetic disorders.
• To enhance the accuracy and efficiency of diagnosing skeletal dysplasias.

Main Methods:

• A hybrid approach combining association rule mining with Dempster-Shafer theory (DST) was developed.
• This method computes probabilistic associations between clinical features and rare disorders to support medical decision-making.
• The model was evaluated on small datasets with sparse distributions.

Main Results:

• The proposed approach demonstrated meaningful outcomes even with small and sparse datasets.
• It outperformed other ML techniques in diagnostic accuracy for skeletal dysplasias.
• The method showed slightly better performance than an initial clinical diagnosis by a physician.

Conclusions:

• The combined association rule mining and DST approach is effective for developing decision support models in data-scarce medical domains.
• This method offers a promising solution for improving the diagnosis of rare genetic disorders like skeletal dysplasias.
• The findings suggest a valuable tool for clinicians in diagnosing complex rare diseases.