Extraction of Peak Velocity Profiles from Doppler Echocardiography Using Image Processing

Machine learning methods to predict transvalvular gradient waveform post-transcatheter aortic valve replacement using preprocedural echocardiogram

OBJECTIVE

Time-varying transvalvular pressure gradient after transcatheter aortic valve replacement indicates the effectiveness of the therapy. The objective was to develop a novel machine learning method enhanced by generative artificial intelligence and smart data selection strategies to predict the post-transcatheter aortic valve replacement gradient waveform using preprocedural Doppler echocardiogram.

METHODS

A total of 110 patients undergoing transcatheter aortic valve replacement (mean age 78.2 ± 9.0 years, 52.5% female) were included for pressure gradient collection. A deep machine learning model was trained and tested to predict postprocedural pressure gradient waveform from preprocedural pressure gradient waveform based on the proposed generative active learning framework.

RESULTS

The trained model demonstrated an average prediction accuracy of 84.85% across the 10 test patients measured from the relative mean absolute error between the predicted gradient waveform and the ground truth. The generative method improved prediction accuracy by 3.11%, whereas the data selection strategy increased it by 16.03% compared with the baseline experimental group using plain machine learning. Additionally, Bland-Altman analysis demonstrated a strong agreement between the proposed method and clinical measurements for both mean and peak pressure gradient predictions.

CONCLUSIONS

A deep, generative, active machine learning model was developed to output the prediction of post-transcatheter aortic valve replacement time-varying pressure gradient from the preprocedural time-varying gradient obtained from Doppler echocardiogram. Such a predictive method may help guide decision-making for the prevention of various post-transcatheter aortic valve replacement complications. Further studies are necessary to investigate the gradient change of other valve types.

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

Electrocardiogram (ECG) is often used together with a spectral Doppler ultrasound to separate heart cycles by determining the end-diastole locations. However, the ECG signal is not always recorded. In such cases, the cardiac cycles can be estimated manually from the ultrasound data retrospectively. We present a deep learning-based method for automatic detection of the end-diastoles in spectral Doppler spectrograms. The method uses a combination of a convolutional neural network (CNN) for extracting features and a recurrent neural network (RNN) for modeling temporal relations. In echocardiography, there are three Doppler spectrogram modalities, continuous wave, pulsed wave, and tissue velocity Doppler. Both the training and test data sets include all three modalities. The model was tested on 643 spectrograms coming from different hospitals than in the training data set. For the purposes described in this work, a valid end-diastole detection is defined as a prediction being closer than 60 ms to the reference value. We will refer to these as true detections. Similarly, a prediction farther away is defined as nonvalid or false detections. The method automatically rejects spectrograms where the detection of an end-diastole has low confidence. When setting the algorithm to reject 1.9%, the method achieved 97.7% true detections with a mean error of 14 ms and had 2.5% false detections on the remaining spectrograms.