Predicting rapid progression phases in glaucoma using a soft voting ensemble classifier exploiting Kalman filtering.

Area of Science:

• Biomedical Engineering
• Data Science in Healthcare
• Ophthalmology

Background:

• Accurate disease phase tracking is crucial for managing chronic conditions like open-angle glaucoma (OAG).
• Medical tests with high residual variability can hinder precise disease phase identification.
• Existing methods struggle with dynamic changes and variability in patient metrics.

Purpose of the Study:

• To present a novel framework for estimating true patient disease metrics despite moderate to high residual variability.
• To dynamically adapt to changes in disease metrics over time, accommodating rapid and non-rapid phases.
• To improve the classification accuracy of rapid disease progression in OAG patients.

Main Methods:

• Integration of interacting multiple model Kalman filtering with supervised learning classification.
• Utilizing Kalman filtering outputs as features for a supervised learning model.
• Application to classify the likelihood of rapid progression in OAG patients within 2-3 years.

Main Results:

• The integrated framework demonstrated improved performance in classifying OAG progression.
• Area Under the Curve (AUC) increased by approximately 7% (from 0.752 to 0.819) with the inclusion of Kalman filtering results.
• The methodology shows significant benefits in combining filtering and statistical learning for clinical health.

Conclusions:

• The developed framework effectively handles residual variability in medical tests for chronic disease management.
• Combining Kalman filtering with supervised learning offers substantial advantages for clinical health applications.
• The methodology is broadly applicable to various chronic conditions beyond OAG.