I am a PhD student at the University of Cambridge working on machine learning problems for healthcare, supervised by Pietro Liò. I am part of the AI group in the computer science and technology department, but I am originally from a medical background (my medical studies are paused between the 4th and 5th years to allow me to complete this PhD).
I have previously completed parts IA and IB (years 1 and 2) of the pre-clinical medical and veterinary sciences tripos, part IIA (intercalation year) of the engineering tripos and one year of clinical medicine (on the wards in the hospital setting).
Download my CV.
PhD in Deep Learning (in progress), 2021
University of Cambridge
MB BChir (in progress), 2017
University of Cambridge
BA in Engineering/Preclinical Medicine, 2016
University of Cambridge
My work has focused on predicting patient outcomes in the Intensive Care Unit (ICU). When designing my deep learning models, I am often inspired by my knowledge of clinical decision making.
For example, for time series processing in Electronic Health Records (EHRs), I use temporal and pointwise convolution to efficiently extract patient trajectories over time – a method inspired by clinicians.
I am also working on using graph neural networks to link the experiences of similar patients. The rationale is that when clinicians make decisions they will typically lean on their past experience, especially if they are dealing with a rare disease.
If you’re interested in the kind of stuff I do, follow me on twitter!
Our model, LSTM-GNN, is designed to take advantage of similarity between patients in the EHR (established using the diagnoses). First, it processes the time series data for each patient with the LSTM component, before sharing information within the neighbourhood of patients via the GNN. This is an alternative way of presenting diagnoses information (the common approach is to use an encoder in the late stages of a model). We found that using both methods together gains the best performance.
Our model, Temporal Pointwise Convolution (TPC), is specifically designed to mitigate common challenges with Electronic Health Records, such as skewness, irregular sampling and missing data. We have achieved significant performance benefits of 18-68% over the Long-Short Term Memory (LSTM) network, and the Transformer.
*My ML4H reviews were explicitly recognised as excellent by metareviewers