VI seminar #24 - Marie Curie Intelligent Ultrasound: Deep Learning for Improving the Interpretation of Echocardiography Data

The program will be available shortly. Please check back later.

Marie Curie Intelligent Ultrasound: Deep Learning for Improving the Interpretation of Echocardiography Data

Presenters: Mujde Akdeniz and Cristiana Ferreira Tiago,  PhD students at UiO in collaboration with GE Vingmed Ultrasound.

Mujde Akdeniz and Cristiana Ferreira

Abstract: In Marcius Project, our aim is to develop tools to automate and improve the interpretation of echocardiography data and to help the clinicians with fast and accurate diagnosis and outcome prediction of the heart disease. Two of the six individual projects are going to be introduced in this presentation.

Deformation (strain) curves of the left ventricle that are measured during the heart cycle have an important role in assessing the left ventricle (LV) function and predicting some of the possible abnormalities. Different supervised neural network architectures were trained on the strain traces of a virtual patient cohort to classify the severity of different LV abnormalities and to predict the existence of scar in 18 LV segments. Results show that the segmental strain traces can be used for detecting and localizing scar in LV segments and for defining the severity of abnormal delay of mechanical activation, global contractility and scar simultaneously.

Due to a limited amount of publicly available 3D echocardiography labeled datasets, an image generation pipeline was created to synthesize such images with corresponding ground truth labels, alleviating the need for data collection and laborious and error-prone labeling of images for Deep Learning (DL) tasks. The obtained results show that GANs can be used, as a data augmentation tool, to generate 3D echocardiography images and subsequently use these to train DL models.

In compliance with GDPR consent requirements, presentations given in a Visual Intelligence context may be recorded with the consent of the speaker. All recordings are edited to remove all faces, names and voices of other participants. Questions and comments by the audience will hence be removed and will not appear in the recording.  With the freely given consent from the speaker, recorded presentation may be posted on the Visual Intelligence YouTube channel.

This seminar is open for members of the consortium. If you want to participate as a guest please sign up.

Sign up here