###Artificial Intelligence/Machine Learning for the detection and stratification of Necrotising Enterocolitis in premature born infants
Project ID: 2228bd1012 (You will need this ID for your application)
Research Theme: Healthcare Technologies
UCL Lead department: Medical Physics and Biomedical Engineering
Lead Supervisor: Evangelos Mazomenos
Project Summary:
Why this research is important: Necrotising Enterocolitis (NEC) is a severe neonatal condition with significant morbidity and mortality. Nearly 12% of infants born weighing less than 1500 g will develop NEC, with mortality between 19-30% and poor neurodevelopmental outcomes. Complications include inflammatory strictures and bowel obstruction with severe cases requiring emergency surgery to manage the disease. The economic and societal impact associated with NEC is considerable (est. $5-6B / year in US).
Early diagnosis from abdominal X-ray (AXR) imaging and subsequent referral are vital as delays result in worse outcomes for patients. However, correct diagnosis presents challenges due to confounding radiological signs and variability in presentation, resulting in NEC patients being investigated late as specialised radiological personnel is required.
What you will be doing: Extending the state-of-the-art in AI technology for diagnosis in X-ray imaging, the project will prototype methodologies to identify and ultimately stratify NEC cases according to their severity, providing novel diagnostic tools for clinicians. The research challenge is to develop AI models to identify the characteristic patterns of NEC in radiographs and distinguish them from confounders. The long-term goal is to identify patients that require urgent surgical intervention as opposed to cases that can be managed with medical treatment alone (conservative management). Further focus will be on Explainable AI methods to rationalise machine understanding with respect to human perception.
Who you will be working with: Early developments from an ongoing collaboration between UCL WEISS and UCL GOS Institute of Child Health, indicate that data-driven, learning models based on convolution units and spatial attention mechanisms (e.g. transformers) can successfully identify the subtle NEC radiological signs in AXRs. Severity stratification will be based on multimodal AI with specialised architectures combining heterogenous data sources (imaging, physiological data, lab tests). This may even allow us to better understand the different disease stages, through novel multimodal features. Ethical approval for retrospective and prospective anonymised datasets is secured (IRAS: 21DS17) and a dataset of 800 cases is already collected and curated.
Who we are looking for: This project is ideal for a dynamic and motivated engineer/scientist graduate, who is interested in developing novel AI methodologies for, i) accelerating detection and diagnosis of NEC; ii) streamlining patient management through stratification and iii) decisioning on the optimal course of treatment between surgery and conservative (non-surgical) approaches. The EPSRC DTP candidate will join the UCL WEISS centre and collaborate with a clinical team from the UCL GOS Institute of Child Health.