Patient-level explainable machine learning to predict major adverse cardiovascular events from SPECT MPI and CCTA...

Area of Science:

• Cardiovascular Imaging and Diagnostics
• Machine Learning in Medicine
• Predictive Analytics for Cardiovascular Disease

Background:

• Machine learning (ML) shows potential for enhancing risk prediction in non-invasive cardiovascular imaging like SPECT MPI and coronary CT angiography (CCTA).
• Current ML algorithms often function as 'black boxes,' limiting clinical understanding of prediction mechanisms.
• Integrating diverse clinical data with CCTA and SPECT assessments is crucial for accurate patient risk stratification.

Purpose of the Study:

• To develop an explainable machine learning (ML) approach for predicting major adverse cardiovascular events (MACE).
• To utilize clinical, CCTA, and SPECT data for MACE prediction.
• To identify critical risk predictors at both global and patient levels.

Main Methods:

• An Automated Machine Learning (AutoML) approach was employed for MACE prediction in patients undergoing CCTA and SPECT MPI.
• Data included clinical information, CCTA, and SPECT findings from consecutive patients suspected of coronary artery disease (CAD).
• Explainable AI techniques were used to identify key risk predictors, with a 10-fold cross-validation for model evaluation.

Main Results:

• The study included 956 patients, with 11% experiencing MACE.
• The ML model achieved high predictive performance: 69.61% sensitivity, 99.77% specificity, and 96.54% accuracy for MACE.
• Top predictors included CCTA attributes (e.g., segment involvement score, plaque characteristics) and clinical factors (e.g., history of MI, smoking status).

Conclusions:

• Machine learning accurately predicts MACE risk in patients undergoing SPECT MPI and CCTA for suspected CAD.
• Explainable ML effectively identifies critical predictive features from both imaging and clinical data.
• This approach enhances risk stratification and clinical decision-making for cardiovascular patients.