Hybrid artificial intelligence outcome prediction using features extraction from stress perfusion cardiac magnetic...

Area of Science:

• Cardiology
• Medical Imaging
• Artificial Intelligence

Background:

• Clinical risk factors traditionally predict coronary artery disease (CAD) outcomes.
• The link between imaging features and patient outcomes remains unclear.
• This study explores AI's potential to connect image features with mortality in CAD.

Purpose of the Study:

• To develop and evaluate an AI model for predicting all-cause mortality in patients with suspected or known CAD.
• To investigate the utility of stress perfusion cardiac magnetic resonance (SP-CMR) imaging features in outcome prediction.
• To compare the performance of AI models with traditional clinical risk factor models.

Main Methods:

• Retrospective analysis of 1,286 patients undergoing SP-CMR (2011-2021).
• Utilized convolutional neural networks (CNN) for image feature extraction and multilayer perceptron (MLP) for electronic health record (EHR) data.
• Developed a hybrid neural network (HNN) combining CNN and MLP, and an image-only CNN model for mortality prediction.

Main Results:

• The HNN model achieved the highest performance with an Area Under the Curve (AUC) of 82% and an F1 score of 43%.
• The image CNN model showed an AUC of 72% and an F1 score of 38%.
• The linear clinical model had an AUC of 80% and an F1 score of 37%, with no significant difference compared to HNN (P=0.15).

Conclusions:

• Mortality in CAD patients can be predicted directly from SP-CMR images.
• A hybrid approach integrating SP-CMR imaging and clinical data via HNN significantly enhances prediction accuracy.
• AI-driven analysis of SP-CMR offers a promising tool for risk stratification in CAD.