Deep learning for the automated feature labelling of 3-Dimensional Imaged Placenta

Abstract

3-D visualisation of cellular structures within the placenta is important for advancing research into the factors determining fetal growth, which are linked to chronic disease risk and quality of lifelong health. 2-D analysis can be challenging, and spatial interaction between structures can be easily missed, but obtaining 3-D structural images is extremely labour-intensive due to the high level of rigorous manual processing required. Deep neural networks are used to automate this previously manual process to construct fast and accurate 3-D structural images, which can be used for 3-D image analysis. The deep networks described in this chapter are trained to label both single cell, a fibroblast and a pericyte, and multicellular, endothelial, structures from within serial block-face scanning electron microscopy placental imaging. Automated labels are equivalent, pixel-to-pixel, to manual labels by over 98% on average over all cell structures and network architectures, and are able to successfully label unseen regions and stacks.

More information

Identifiers

Catalogue record

Export record

ASCII CitationAtomBibTeXData Cite XMLDublin CoreDublin CoreEP3 XMLEndNoteHTML CitationHTML CitationHTML ListJSONMETSMODSMPEG-21 DIDLOpenURL ContextObjectOpenURL ContextObject in SpanRDF+N-TriplesRDF+N3RDF+XMLRIOXX2 XMLReferReference ManagerSimple Metadata

Altmetrics