A New Method of Modeling the Multi-stage Decision-Making Process of CRT Using Machine Learning with Uncertainty...

Area of Science:

• Cardiology
• Artificial Intelligence
• Medical Imaging

Background:

• Machine learning models for predicting patient outcomes often use all available data, disregarding acquisition costs.
• Predicting response to cardiac resynchronization therapy (CRT) in heart failure (HF) patients is crucial for effective treatment.
• Current methods may not optimally balance predictive performance with the cost and time of data acquisition.

Purpose of the Study:

• To develop a multi-stage machine learning (ML) model for predicting CRT response in HF patients.
• To integrate uncertainty quantification to guide the selective acquisition of single-photon emission computed tomography myocardial perfusion imaging (SPECT MPI) data.
• To reduce data acquisition costs without compromising predictive accuracy.

Main Methods:

• A multi-stage ML model was constructed by combining two ensemble models.
• Ensemble 1 utilized clinical variables and electrocardiogram (ECG) data.
• Ensemble 2 incorporated SPECT MPI features, with uncertainty quantification from Ensemble 1 dictating the need for SPECT MPI data.

Main Results:

• The multi-stage model achieved performance comparable to a model using all SPECT MPI data (AUC 0.75 vs. 0.77).
• It required SPECT MPI data for only 52.7% of patients, significantly reducing data acquisition needs.
• Key performance metrics included accuracy of 0.71 and sensitivity of 0.70 for the multi-stage model.

Conclusions:

• The developed multi-stage ML model effectively predicts CRT response in HF patients.
• Uncertainty quantification enables significant reduction in SPECT MPI data acquisition without substantial performance loss.
• This approach offers a cost-effective strategy for personalized HF treatment prediction.