A setup for ultrasound based assessment of the aortic root geometry

Deep transfer learning for aortic root dilation identification in 3D ultrasound images

Pre-operative planning of valve-sparing aortic root reconstruction relies on the automatic discrimination of healthy and pathologically dilated aortic roots. The basis of this classification are features extracted from 3D ultrasound images. In previously published approaches, handcrafted features showed a limited classification accuracy. However, feature learning is insufficient due to the small data sets available for this specific problem. In this work, we propose transfer learning to use deep learning on these small data sets. For this purpose, we used the convolutional layers of the pretrained deep neural network VGG16 as a feature extractor. To simplify the problem, we only took two prominent horizontal slices throgh the aortic root, the coaptation plane and the commissure plane, into account by stitching the features of both images together and training a Random Forest classifier on the resulting feature vectors. We evaluated this method on a data set of 48 images (24 healthy, 24 dilated) using 10-fold cross validation. Using the deep learned features we could reach a classification accuracy of 84 %, which clearly outperformed the handcrafted features (71 % accuracy). Even though the VGG16 network was trained on RGB photos and for different classification tasks, the learned features are still relevant for ultrasound image analysis of aortic root pathology identification. Hence, transfer learning makes deep learning possible even on very small ultrasound data sets.

Prediction of individual prosthesis size for valve-sparing aortic root reconstruction based on geometric features

Valve-sparing aortic root reconstruction is an up- and-coming approach for patients suffering from aortic valve insufficiencies which promises to significantly reduce complications. However, the success of the treatment strongly depends on the challenging task of choosing the correct size of the prosthesis, for which, up to now, surgeons solely have to rely on their experience. Here, we present a novel machine learning based approach, which might make it possible to predict the size of the prosthesis from pre-operatively acquired ultrasound images. We utilize support vector regression to train a prediction model on three geometric features extracted from the ultrasound data. In order to evaluate the accuracy and robustness of our approach we created a large data base of porcine aortic root geometries in a healthy state and an artificially dilated state. Our results indicate that prediction of correct prosthesis sizes is feasible. Furthermore, they suggest that it is crucial that the training data set faithfully represents the diversity of aortic root geometries.

A machine learning approach for planning valve-sparing aortic root reconstruction

Choosing the optimal prosthesis size and shape is a difficult task during surgical valve-sparing aortic root reconstruction. Hence, there is a need for surgery planning tools. Common surgery planning approaches try to model the mechanical behaviour of the aortic valve and its leaflets. However, these approaches suffer from inaccuracies due to unknown biomechanical properties and from a high computational complexity. In this paper, we present a new approach based on machine learning that avoids these problems. The valve geometry is described by geometrical features obtained from ultrasound images. We interpret the surgery planning as a learning problem, in which the features of the healthy valve are predicted from these of the dilated valve using support vector regression (SVR). Our first results indicate that a machine learning based surgery planning can be possible.