A Novel Dual-Modality Dual-View Hybrid Deep Learning-Machine Learning Framework for the Prediction of Carotid Plaque...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Disease

Background:

• Ultrasound imaging is crucial for carotid plaque screening to identify stroke risk.
• Existing studies lack AI-enabled auto-classification of carotid plaque vulnerability using multi-modal, multi-view ultrasound.
• This research addresses the need for advanced AI in carotid plaque analysis.

Purpose of the Study:

• To develop and validate an effective AI model for carotid plaque vulnerability classification.
• To leverage dual-modal (B-Mode, CEUS) and dual-view (longitudinal, cross-sectional) ultrasound settings.
• To maximize the utility and potential of ultrasound imaging in stroke risk stratification.

Main Methods:

• Employed hybrid deep learning (DL) and machine learning (ML) methods for discriminability and interpretability.
• Retrospectively analyzed B-Mode ultrasound (BMUS) and contrast-enhanced ultrasound (CEUS) images from 241 patients.
• Utilized proposed hybrid DL-ML variants for carotid plaque classification.

Main Results:

• The hybrid VGG-RF model, using dual-modal dual-view settings, outperformed other configurations.
• Achieved optimal performance with an AUC of 0.908, precision of 0.765, recall of 0.929, specificity of 0.886, and F1 score of 0.839.
• Identified long-axis views of BMUS and CEUS images as key features for discriminating vulnerable plaques.

Conclusions:

• AI models developed from dual-modal dual-view ultrasound settings are effective for carotid plaque analysis.
• The hybrid VGG-RF model demonstrated superior performance among the studied DL-ML variants.
• Further prospective studies with larger cohorts are recommended to validate these findings.