Can convolutional neural networks identify external carotid artery calcifications?

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Radiology
• Cardiovascular Health

Background:

• External carotid artery calcifications (ECACs) are indicators of cardiovascular disease.
• Accurate detection of ECACs is crucial for risk assessment.
• Cone beam computed tomography (CBCT) is increasingly used in dental and maxillofacial imaging.

Purpose of the Study:

• To develop and evaluate a convolutional neural network (CNN) for detecting ECACs.
• To assess the accuracy and reliability of the CNN model compared to human evaluation.
• To investigate the utility of AI in analyzing CBCT scans for vascular calcifications.

Main Methods:

• A CNN model was developed using TensorFlow.
• Trained on 427 CBCT scans with an 80:20 training-to-validation split.
• Performance evaluated using k-fold cross-validation, F1 score, and Matthews Correlation Coefficient.

Main Results:

• The CNN achieved a k-fold cross-validation accuracy of 76%.
• Recall was 66% and precision was 79%, yielding an F1 score of 0.72.
• A Matthews correlation coefficient of 0.53 indicated a strong performance across all confusion matrix categories.

Conclusions:

• The developed CNN model demonstrates reliable performance in identifying ECACs.
• AI-powered analysis can aid in the detection of vascular calcifications in CBCT scans.
• This technology holds promise for improving cardiovascular risk assessment through routine imaging.