Automatic Detection of Aortic Valve Events Using Deep Neural Networks on Cardiac Signals From Epicardially Placed Accelerometer.

BACKGROUND: Miniaturized accelerometers incorporated in pacing leads attached to the myocardium, are used to monitor cardiac function. For this purpose functional indices must be extracted from the acceleration signal. A method that automatically detects the time of aortic valve opening (AVO) and aortic valve closure (AVC) will be helpful for such extraction. We tested if deep learning can be used to detect these valve events from epicardially attached accelerometers, using high fidelity pressure measurements to establish ground truth for these valve events. METHOD: A deep neural network consisting of a CNN, an RNN, and a multi-head attention module was trained and tested on 130 recordings from 19 canines and 159 recordings from 27 porcines covering different interventions. Due to limited data, nested cross-validation was used to assess the accuracy of the method. RESULT: The correct detection rates were 98.9% and 97.1% for AVO and AVC in canines and 98.2% and 96.7% in porcines when defining a correct detection as a prediction closer than 40 ms to the ground truth. The incorrect detection rates were 0.7% and 2.3% for AVO and AVC in canines and 1.1% and 2.3% in porcines. The mean absolute error between correct detections and their ground truth was 8.4 ms and 7.2 ms for AVO and AVC in canines, and 8.9 ms and 10.1 ms in porcines. CONCLUSION: Deep neural networks can be used on signals from epicardially attached accelerometers for robust and accurate detection of the opening and closing of the aortic valve.

Mitral Annulus Segmentation Using Deep Learning in 3-D Transesophageal Echocardiography.

3D Transesophageal Echocardiography is an excellent tool for evaluating the mitral valve and is also well suited for guiding cardiac interventions. We introduce a fully automatic method for mitral annulus segmentation in 3D Transesophageal Echocardiography, which requires no manual input. One hundred eleven multi-frame 3D transesophageal echocardiography recordings were split into training, validation, and test sets. Each 3D recording was decomposed into a set of 2D planes, exploiting the symmetry around the centerline of the left ventricle. A deep 2D convolutional neural network was trained to predict the mitral annulus coordinates, and the predictions from neighboring planes were regularized by enforcing continuity around the annulus. Applying the final model and post-processing to the test set data gave a mean error of 2.0 mm - with a standard deviation of 1.9 mm. Fully automatic segmentation of the mitral annulus can alleviate the need for manual interaction in the quantification of an array of mitral annular parameters and has the potential to eliminate inter-observer variability.

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.

Real-Time Nonlocal Means-Based Despeckling.

In this paper, we propose a multiscale nonlocal means-based despeckling method for medical ultrasound. The multiscale approach leads to large computational savings and improves despeckling results over single-scale iterative approaches. We present two variants of the method. The first, denoted multiscale nonlocal means (MNLM), yields uniform robust filtering of speckle both in structured and homogeneous regions. The second, denoted unnormalized MNLM (UMNLM), is more conservative in regions of structure assuring minimal disruption of salient image details. Due to the popularity of anisotropic diffusion-based methods in the despeckling literature, we review the connection between anisotropic diffusion and iterative variants of NLM. These iterative variants in turn relate to our multiscale variant. As part of our evaluation, we conduct a simulation study making use of ground truth phantoms generated from clinical B-mode ultrasound images. We evaluate our method against a set of popular methods from the despeckling literature on both fine and coarse speckle noise. In terms of computational efficiency, our method outperforms the other considered methods. Quantitatively on simulations and on a tissue-mimicking phantom, our method is found to be competitive with the state-of-the-art. On clinical B-mode images, our method is found to effectively smooth speckle while preserving low-contrast and highly localized salient image detail.