Estimation of End-Diastole in Cardiac Spectral Doppler Using Deep Learning.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Cardiology

Background:

• Electrocardiogram (ECG) and spectral Doppler ultrasound are used to identify cardiac cycles.
• ECG signals are not always available, necessitating manual estimation of cardiac cycles from ultrasound data.
• Accurate end-diastole detection is crucial for cardiac cycle analysis.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for automatic end-diastole detection in spectral Doppler spectrograms.
• To provide an automated solution for cardiac cycle estimation when ECG is not recorded.
• To assess the accuracy and reliability of the proposed method across different Doppler modalities.

Main Methods:

• A deep learning model combining a Convolutional Neural Network (CNN) for feature extraction and a Recurrent Neural Network (RNN) for temporal modeling was developed.
• The model was trained and tested on spectral Doppler spectrograms from continuous wave, pulsed wave, and tissue velocity Doppler modalities.
• Performance was evaluated on 643 spectrograms from external hospitals, defining true detections as within 60 ms of the reference value.

Main Results:

• The deep learning method achieved 97.7% true detections with a mean error of 14 ms.
• The method demonstrated a low false detection rate of 2.5% after rejecting low-confidence predictions (1.9% rejection rate).
• The model performed effectively across all three Doppler spectrogram modalities.

Conclusions:

• The proposed deep learning method offers an accurate and automated approach for end-diastole detection in spectral Doppler ultrasound.
• This technique can reliably estimate cardiac cycles in the absence of ECG signals, enhancing clinical workflow.
• The model's high performance across various modalities suggests its broad applicability in echocardiography.