Projects using AI in evaluation and prognosis

The objective of this study within the Tumour Evolution and Cell Identity Laboratory is to develop an organ-omics approach for characterising tumour organoids and identifying biomarkers of CRC drug response. Organ-omics refers to the application of multi-omics technologies to study and characterise organoids with high-throughput assays to profile their cellular features to identify biomarkers of function, disease, and drug response. This would be used to leverage the predictive power of organoid models to accelerate drug discovery, clinical trials, personalised medicine, and regenerative therapies for a wide range of diseases. We aim to perform comprehensive front line drug sensitivity profiling of patient-derived CRC tumour organoids, with multi-omics analysis and clinical outcomes, to identify potential biomarkers of drug sensitivity and resistance using bioinformatic analysis and machine learning, which in turn can be used to improve treatment.

Mr Simon Knight and team within the Centre for Evidence in Transplantation are collaborating with the Computational Health Informatics Lab, developing machine learning models to predict patient outcomes if an organ offer is accepted, or declined to wait for another offer. We hope to integrate this into our clinical workflow to test a prototype clinical decision tool in a Phase 1 Clinical Trial. 

Around 2,500 deceased donor kidney transplants are performed in the UK each year. At any time, there are around 5,000 patients on the kidney transplant waiting list. The shortage of organs available for transplant means that some patients become too unwell for surgery or die whilst waiting. Because of this, doctors often consider kidneys from donors who are less ideal due to age or other medical problems.

These decisions are made by the transplant doctors based upon the information available at the time of offer. Details are rarely discussed with the patient. Doctors use their clinical experience, but do not have tools available to help them predict what would happen if they chose to accept or decline an offer and wait for the next one. Previous research suggests that it is often better to accept an offer rather than waiting longer for something better.

This project will use real-world data from twenty years of previous transplant offers and outcomes in the UK to train Artificial Intelligence (AI) models that will allow us to predict the outcome if an offer is accepted and transplanted or declined and wait for another offer. We plan to develop a web-based tool that presents this information in a simple fashion that both the doctors and patients can use when making a decision. We will then test this tool in a real-life setting to assess the impact on patients, clinicians and transplant outcomes. 

The initial work has been funded by the NIHR AI in Health and Care Awards and the John Fell Fund.

Professor Peter McCulloch and team are working on clinical evaluation of decision support systems driven by artificial intelligence where they have proposed a new set of guidelines called DECIDE AI, and the team are currently following up on this work and are involved in several projects testing AI in the clinic.

As an increasing number of clinical decision-support systems driven by artificial intelligence progress from development to implementation, better guidance on the reporting of human factors and early-stage clinical evaluation is needed.

The objective of DECIDE-AI is to improve reporting of clinical AI studies along four main axes:

• the performance of the AI systems when first used with humans in small-scale, actual clinical settings,
• the safety profile of the AI systems prior to large-scale utilisation,
• the human factors (ergonomic) evaluation of the AI systems,
• the preparatory steps towards large-scale (randomised controlled) clinical trials. 

Professor McCulloch and team have also been part of a successful consortium bid for a Horizon 2020 grant about AI and associated technology in healthcare called ASSESS-DHT.