Clustering Results Interpretation of Continuous Variables Using Bayesian Inference

Area of Science:

• Artificial Intelligence
• Medical Informatics
• Bayesian Statistics

Background:

• Interpretable AI is crucial for medical applications.
• Previous work focused on binary features for clinical pathway interpretation using Bayesian inference.
• The prior method's limitation was its inability to handle continuous data.

Purpose of the Study:

• To extend interpretable AI methods for analyzing continuous features in medicine.
• To adapt Bayesian inference for modeling posterior distributions of continuous clinical data.
• To compare the enhanced method's interpretation with medical expert consensus.

Main Methods:

• Applied the Bayesian Estimation (BEST) algorithm for Bayesian t-testing.
• Utilized the No-U-Turn Sampler (NUTS) algorithm for posterior sampling.
• Developed an approach for interpretable clustering of continuous clinical features.

Main Results:

• Successfully adapted Bayesian inference for continuous feature interpretation in clinical pathways.
• The enhanced method demonstrated robust posterior distribution modeling.
• Algorithm's interpretations for both binary and continuous features were benchmarked against two medical experts.

Conclusions:

• The expanded interpretable AI approach effectively models continuous features in medical data.
• This advancement allows for more nuanced clinical pathway interpretation.
• The method shows promise in supporting clinical decision-making through AI.