Automated ultrasound doppler angle estimation using deep learning

The paper presents a deep learning approach for automated Doppler angle estimation in ultrasound imaging, utilizing a dataset of 2100 human carotid images with augmentation. Five pre-trained models were employed for feature extraction, feeding into a custom shallow network, achieving a mean absolute error (MAE) of 3.9° to 9.4°, with the best model performing below the clinically acceptable error threshold. This advancement could enhance the accuracy of Doppler blood velocity measurements in clinical settings, reducing the risk of misclassification in diagnosing vascular conditions.