CDT Postgraduate Studentship in Healthcare Innovation (RCUK Digital Economy Programme grant number EP/G036861/1)
Ultrasound segmentation techniques and their application to assess fetal nutrition and development
Prior research has involved the quantification of adipose tissue in the fetal arm and leg from 2D ultrasound images, by developing an interactive image segmentation technique. The developed method built on the Live Wire framework by introducing two new sets of Live Wire costs; namely a Feature Asymmetry (FA) cost to localise edges and a weak shape constraint cost to aid the selection of appropriate boundaries in the presence of missing information or artefacts. The resulting semi-automatic segmentation method follows edges based on structural relevance rather than intensity gradients, adapting the method to ultrasound images, where the object boundaries are normally fuzzy and regions are often occluded. We conducted a study using 48 cross-sectional ultrasound images of fetal arm across gestation, presenting significant variability in size and appearance, comparing our method to manual clinical segmentations, intensity-gradient Live Wire and an automated technique involving Feature Asymmetry.
Current research is concerning the automatic detection and quantification of the fetal liver from 2D ultrasound images. Initial investigations have involved a machine learning method, taking into account the distances of image pixels from the stomach and umbilical vein (two landmarks which are far easier to delineate than the liver). A study into previous attempts to produce growth curves for the fetal liver has been used to try to standardise these distances, and allow for training data from a broad range of gestational ages to be used. Currently training and validation has been conducted on a small set of data, using images from fetuses of gestational age between 16 and 36 weeks. Further to this, texture analysis is being used to differentiate between liver and non-liver tissue.