RF-CNN-F: random forest with convolutional neural network features for coronary artery disease diagnosis based on...

Area of Science:

• Cardiovascular Imaging and Machine Learning
• Artificial Intelligence in Medical Diagnosis

Background:

• Coronary artery disease (CAD) presents significant morbidity and mortality.
• Invasive coronary angiography, the gold standard for CAD diagnosis, is invasive and costly.
• Noninvasive cardiac magnetic resonance (CMR) imaging offers a safer alternative for CAD assessment.

Purpose of the Study:

• To develop and evaluate a novel, automated method for detecting CAD using CMR images.
• To leverage deep neural networks for feature extraction and random forests for classification.
• To improve the accuracy of CAD detection compared to existing methods.

Main Methods:

• Proposed a hybrid machine learning model named RF-CNN-F (Random Forest with CNN Features).
• Utilized convolutional neural networks (CNNs) for automated feature extraction from CMR images.
• Integrated extracted CNN features into a random forest classifier for CAD detection.

Main Results:

• The RF-CNN-F method achieved a high accuracy of 99.18% on a large, publicly accessible CMR dataset.
• This performance significantly outperformed a stand-alone CNN model, which achieved 93.92% accuracy.
• The Adam optimizer was employed for training the models.

Conclusions:

• The proposed RF-CNN-F method demonstrates superior performance for automated CAD detection from CMR images.
• This hybrid approach effectively combines the feature extraction capabilities of deep learning with the classification power of random forests.
• RF-CNN-F shows promise as a robust and accurate tool for noninvasive CAD diagnosis.