The objective of this project is to research and analyse the use of deep learning technology to develop rules that can accurately generate clinical reports with the same (or better) consistency as a microbiologist and in a manner that is explainable and transparent. Liverpool Clinical Laboratories conducts hundreds of tests per day on body fluid and tissue samples obtained from individuals by clinical teams throughout the Liverpool area. The standard procedure is to grow bacteria in the sample and test its sensitivity to a number of antibiotics, then use the test results to recommend at least two effective antibiotics to the clinical team.
There is a set of standard operating procedures to provide guidance on how to interpret test results and generate the clinical recommendations. However, these can be interpreted in different ways between microbiologists, depending on custom and practice. The focus of the project will be a large data set covering a decade of urine sample test results and recommendations. The data is in CSV format and shows the test results for a range of antibiotics along with the recommendation of the microbiologist at the time. A successful project outcome would be a rule-based software model that, when applied to this data set, will make the same recommendations.
By its very nature, the inner workings of deep machine learning cannot be fully explained. A topic of significant current research is explainable AI, which involves the development and application of AI techniques such that the results of the solution can be understood by human experts. Transparent solutions are especially important in domains such as medical diagnosis, where trust in the technology is a key consideration in adopting and acting on the results. There has been some recent progress in explaining how AI works in specific use cases. A major aspect of this project is to develop a machine learning tool that can generate consistent results in a transparent manner.
Some machine learning techniques learn undesirable tricks that optimally solve the problem with training data but do not reflect the objectives of the designers, often appearing to cheat or find an unintended shortcut to the end goal. An explainable AI system allows a human to audit the rules to determine how accurate it is likely to be with real, unseen data beyond the training set. Thus, a key aspect of this project is to show that the derived rules are correct and verify that they will work consistently on real data. The model will be evaluated against new testing data and microbiologist reports to determine the consistency and accuracy of its recommendations.
The University of Liverpool Doctoral Network in Artificial Intelligence (AI) for Future Digital Health aims to create and maintain a community of AI health care professionals who can develop and apply AI research to medical problems. See https://www.liverpool.ac.uk/study/postgraduate-research/doctoral-training-programmes/ai-for-future-digital-health
. The vision is to provide high-quality doctoral training within the broad domain of AI (including Machine Learning, Data Science and Statistics) for medical applications from health care to drug design. Weekly 3-hour training sessions include various topics from Statistics and Linear Algebra to guest lectures on AI and healthcare (see http://kurlin.org/doctoral-network.php#training
). New students starting in October 2020 will join our first cohort of 8 PhD students who started in October 2019.
Each PhD project has been carefully co-created in collaboration with a health care provider and/or a commercial partner working with medical data so that the outcomes of the PhD research will have immediate benefit. The non-academic supervisor in this project is a doctor from the microbiology team at the Royal Liverpool University Hospital. The academic supervision team has expertise in machine learning. The network will provide students with regular training and internship opportunities at industry partners.
Applications are welcome from enthusiastic candidates with a first class degree in Computer Science, Engineering, Mathematics, Biostatistics or in a similar area close to the proposed PhD research. A masters degree with a focus on the theory and practice of machine learning would be an advantage. Knowledge of Python and machine learning software libraries would also be beneficial. The available funding is strictly limited to UK citizens.
To apply for this opportunity, please visit: https://www.liverpool.ac.uk/study/postgraduate-research/how-to-apply/
’Applications should be made to a PhD in Computer Science.