Hierarchical Adaptive Multi-task Learning Framework for Patient Diagnoses and Diagnostic Category Classification.

Area of Science:

• Medical Informatics
• Machine Learning
• Computational Biology

Background:

• Patient diagnoses are often organized in hierarchical structures.
• Machine learning models can predict diagnoses, but often treat each task independently.
• Leveraging diagnostic hierarchies could enhance classification accuracy.

Purpose of the Study:

• To develop and evaluate a hierarchical classification learning framework for patient diagnoses.
• To investigate whether incorporating diagnostic hierarchy relations improves classification performance compared to independent models.
• To assess the impact of the framework on diagnoses with low prior probability and limited training data.

Main Methods:

• Designed a novel hierarchical classification learning framework explicitly relating multiple diagnostic targets via hierarchy.
• Conducted experiments using the Medical Information Mart for Intensive Care III (MIMIC-III) database.
• Utilized the International Classification of Diseases, Ninth Revision (ICD-9) diagnosis hierarchy.

Main Results:

• The proposed hierarchical framework demonstrated improved classification performance on individual diagnostic tasks.
• Performance gains were more pronounced for diagnoses with a low prior probability.
• The framework showed greater benefits for diagnoses with fewer positive training examples.

Conclusions:

• Hierarchical classification learning frameworks can effectively leverage diagnostic structures.
• This approach offers a significant advantage over independent diagnostic models, particularly for challenging cases.
• The findings suggest a promising direction for improving automated diagnostic systems.