Exploring Fever of Unknown Origin Intelligent Diagnosis Based on Clinical Data: Model Development and Validation

Area of Science:

• Artificial Intelligence in Medicine
• Clinical Decision Support Systems
• Machine Learning for Healthcare

Background:

• Fever of unknown origin (FUO) presents diagnostic challenges due to heterogeneous causes and delayed diagnoses.
• Current FUO research often focuses on statistical analysis, with treatments varying significantly by category.
• Intelligent categorization of FUO is crucial for accurate and timely diagnosis.

Purpose of the Study:

• To develop a machine learning model for automatic prediction of FUO cause categories.
• To enhance diagnostic accuracy for FUO by fusing diverse medical data.
• To assist clinicians in diagnosing FUO more precisely.

Main Methods:

• Developed the FUO intelligent diagnosis (FID) model.
• Classified FUO cases into four categories: infections, immune diseases, tumors, and others.
• Utilized bidirectional encoder representations from transformers (BERT) for electronic medical record (EMR) data structuring.
• Trained the FID model using LightGBM on cleaned basic and laboratory data.

Main Results:

• The FID model achieved 81.68% precision for top 1 classification diagnosis.
• The model demonstrated 96.17% precision for top 2 classification diagnosis.
• Performance surpassed that of comparative methods in extensive experiments.

Conclusions:

• The FID model shows high efficacy in diagnosing FUO.
• It serves as a valuable tool for clinicians to improve diagnostic precision.
• The model has the potential to reduce misdiagnosis rates in FUO cases.