Vision-based estimation of volume status in ultrasound

Smith, Jordan (2020) Vision-based estimation of volume status in ultrasound. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (4MB)

Abstract

This thesis provides a proof-of-concept approach to the analysis of ultrasound imagery using machine learning and computer vision for the purposes of tracking relative changes in apparent circulating blood volume. Data for the models was collected from a simulation which involved having healthy subjects recline at angles between 0 and 90 degrees to induce changes in the size of the internal jugular vein (IJV) resulting from gravity. Ultrasound video clips were then captured of the IJV. The clips were segmented, followed by feature generation, feature selection and training of predictive models to determine the angle of inclination. This research provides insight into the feasibility of using automated analysis techniques to enhance portable ultrasound as a monitoring tool. In a dataset of 34 subjects the angle was predicted within 11 degrees. An accuracy of 89% was achieved for high/low classification.

Actions (login required)

View Item