Deep Learning for Foetal Monitoring in Labour

The main aim of this project is to apply Machine Learning methods to a big medical dataset containing data of pregnant women (and their foetus/baby) before, during and after giving birth, for developing evidence-based diagnostics assessing the danger of foetal brain injury or even asphyxia.
The dataset contains continuous records (time series - every second) of Cardiotocograms (CTG's) from nearly 60,000 labours - a uniquely large birth cohort (all monitored mothers and babies in Oxford, spanning in more than 20 year period) and growing daily (a further 20,000 births have already been collected currently).
The project is to investigate, analyse, process and apply deep learning, employing autoassociative, convolutional neural networks (CNN), and long short-term neural networks (LSTM) for extracting vital information, trends and relationships from the dataset containing time series records (Cardiotocograms - CTG's), collected continuously from pregnant women (and their foetus) before, during, and after giving birth. The purpose is to develop evidence-based diagnostics in this clinical field that quantifies the CTG in the context of clinical risk factors and relates these to perinatal outcome in order to reduce the risk of fetal brain injury (as a result of oxygen deficiency) or even suffocation.

The aim of this project is to employ machine learning techniques (deep learning) on the Oxford extremely large data archives, utilising the enormous scope for improvement by extracting more diagnostic rules, capturing vital relationships and trends from the datasets to define other sub-groups of fetal compromise (one of the limitations of the recently developed OxSys (in Oxford) is that it employs only two diagnostic rules, which is quite restrictive considering the fact that the group of compromised babies during labour is highly heterogeneous, with numerous aetiologies to fetal brain injury).