Machine learning to predict abnormal myocardial perfusion from pre-test features.

Area of Science:

• Cardiology
• Medical Imaging
• Machine Learning in Healthcare

Background:

• Accurate prediction of abnormal myocardial perfusion (MPI) is crucial for guiding non-invasive test selection in clinical practice.
• Existing clinical models have limitations in predicting MPI based on pre-test information.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting abnormal MPI using pre-test clinical features.
• To compare the performance of the ML model against established clinical prediction models.

Main Methods:

• An extreme gradient boosting ML model was trained on 20,418 patients from a multi-center international registry undergoing SPECT MPI.
• The model utilized 30 pre-test features to predict abnormal myocardial perfusion.
• External validation was performed on 9,019 patients from two separate sites.

Main Results:

• The ML model demonstrated superior prediction performance for abnormal MPI in external testing, achieving an AUC of 0.762.
• The ML model outperformed the clinical CAD consortium (AUC 0.689), basic CAD consortium (AUC 0.657), and Diamond-Forrester models (AUC 0.658).
• The ML model also showed superior calibration, with a Brier score of 0.149 compared to other models (0.165–0.198).

Conclusions:

• Machine learning models can effectively predict abnormal myocardial perfusion using readily available pre-test clinical information.
• This ML model has the potential to aid physicians in making informed decisions regarding non-invasive test selection for patients.
• The findings support the integration of ML tools into clinical workflows for improved diagnostic accuracy and patient management.