A Pathological Diagnosis Method for Fever of Unknown Origin Based on Multipath Hierarchical Classification: Model...

Area of Science:

• Medical informatics
• Machine learning applications
• Diagnostic algorithms

Background:

• Fever of unknown origin (FUO) presents diagnostic challenges due to diverse causes and potential misdiagnosis.
• Machine learning offers a promising approach to analyze complex patient data for FUO diagnosis.

Purpose of the Study:

• To develop a multipath hierarchical classification algorithm for diagnosing FUO.
• To enhance diagnostic performance through an integrated feature selection mechanism.

Main Methods:

• Utilized a dataset of FUO patients from 2011-2020.
• Designed a 3-layer hierarchical classification structure based on etiology.
• Incorporated a feature selection mechanism and conducted ablation and comparison experiments.

Main Results:

• The optimal model achieved 76.08% accuracy with 3% middle paths and 25% feature selection.
• The proposed model outperformed existing methods in both feature selection and hierarchical classification.
• Specific diagnoses like brucellosis (100%), liver abscess, viral infection, and lymphoma (>80%) showed high accuracy.

Conclusions:

• A novel multipath feature selection and hierarchical classification model for FUO diagnosis was successfully designed and evaluated.
• This machine learning model aids physicians by improving the accuracy and efficiency of FUO diagnosis.