A deep learning model for carotid plaques detection based on CTA images: a two stepwise early-stage clinical...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiovascular Diagnostics

Background:

• Carotid atherosclerotic plaques are a significant risk factor for stroke.
• Accurate and efficient detection of carotid plaques is crucial for patient management.
• Current diagnostic methods may have limitations in speed and accuracy.

Purpose of the Study:

• To develop a deep learning (DL) model for carotid plaque detection using CTA images.
• To evaluate the clinical feasibility and value of the developed DL model.
• To enhance plaque segmentation using a combination of ResUNet and PSPNet.

Main Methods:

• Retrospective collection of CTA data from 647 patients (October 2020 - October 2022).
• Development of a DL model combining ResUNet and Pyramid Scene Parsing Network (PSPNet).
• Evaluation using lesion-level recall, patient-level sensitivity, and precision, with clinical validation studies.

Main Results:

• The DL model achieved high diagnostic performance (e.g., test set: 78.37% precision, 91.86% sensitivity, 84.58% recall).
• The model demonstrated accuracy across different plaque locations, morphologies, and types.
• Clinical validation showed the model outperformed 4 out of 6 radiologists and improved their sensitivity, while significantly reducing diagnosis time (6s).

Conclusions:

• The AI model shows strong clinical potential for carotid plaque detection.
• It improves clinician diagnostic performance and significantly shortens detection time.
• The model is practical for implementation in real-world clinical scenarios.