Automated Analysis of Mitral Inflow Doppler Using Deep Neural Networks.

Doppler echocardiography is a widely applied modality for the functional assessment of heart valves, such as the mitral valve. Currently, Doppler echocardiography analysis is manually performed by human experts. This process is not only expensive and time-consuming, but often suffers from intra- and inter-observer variability. An automated analysis tool for non-invasive evaluation of cardiac hemodynamic has potential to improve accuracy, patient outcomes, and save valuable resources for health services. Here, a robust algorithm is presented for automatic Doppler Mitral Inflow peak velocity detection utilising state-of-the-art deep learning techniques. The proposed framework consists of a multi-stage convolutional neural network which can process Doppler images spanning arbitrary number of heartbeats, independent from the electrocardiogram signal and any human intervention. Automated measurements are compared to Ground-truth annotations obtained manually by human experts. Results show the proposed model can efficiently detect peak mitral inflow velocity achieving an average F1 score of 0.88 for both E- and A-peaks across the entire test set.