Multimodal Artificial Intelligence for Predicting the Risk of Postoperative Morbidity

1) Brief description of the context of the research including potential impact

The development of complications after major surgery is common (16% - 50%) and is of great public health importance, due to outcomes such as decreased length and quality of life, increase in hospital costs, and readmission. Specific risks for complications relate to loss of muscle mass and quality. This project focuses mainly on sarcopenia, which is a type of muscle loss characterised by the degenerative loss of mass, quality, and strength of skeletal muscle with age.

Sarcopenia is associated with reduced physical function, cancer, and frailty, amongst other conditions.

This research will aim to produce multimodal AI-powered models to assess body composition in surgical patients and estimate the risk of specific adverse postoperative complications, related to the loss or reduction in quality of muscle mass. This will be achieved by combining information from heterogeneous datasets, namely, CT, ultrasound (US), CPET, and body composition datasets.

These models will allow for greater insight into the specific needs and risks associated to each patient in the perioperative period, allowing for improved quality of patient care, and reduced strain on healthcare services. 2) Aims and Objectives

The aim of this project is to develop and optimise multimodal AI models to assess body composition in surgical patients and estimate the risk of specific adverse postoperative complications related to the loss or reduction in quality of muscle mass. The specific objectives are to:

- (MRes) Develop and optimise AI models to analyse CT scans and automatically characterise the skeletal muscle quality/ body composition of patients in the preoperative period.

- Analyse datasets relating information about body composition and CPET examinations to operative outcome to accurately predict adverse perioperative outcomes.

- Investigate the relationship between preoperative body composition and longitudinal changes in the metabolism during the perioperative period.

- Develop and optimise multimodal AI models to combine information from CT, US, CPET, and body composition datasets to accurately predict morbidity and mortality post-surgery.

Preliminary CT data from 500 patents will support the initial stages of development, while the full dataset of clinical cases is continuously integrated from UCLH/CPOM. Adequate volumes of data with granular labelling ensure that the candidate can advance the project objectives, expanding the state-of-the-art on multi-modal AI over the project duration.

3) Novelty of Research Methodology

The student will develop and optimise multi-modal AI models to accurately predict the patientspecific risk of post-operative morbidity, using muscular information gathered from CT and US scans, along side information from CPET and body composition test. The use of multi-modal AI models to do this is novel, and the potential of the models to provide clinicians with accurate information pertaining to specific patient risk is unique.

4) Alignment to EPSRC's strategies and research areas

This research is most closely aligned with EPSRC's healthcare technologies theme as it aims to provide novel, state-of-the-art healthcare AI models. The specific research area pertaining to this research is Medical Imaging. The research utilises advanced artificial intelligence technologies, where multimodal deep learning models will be developed and employed for predicting patient surgical risk and outcome.

5) Any companies or collaborators involved

Technical support for clinical data sharing (XNAT servers) and AI development (GPU systems) are available through the WEISS Centre. Further funding (consumables, dissemination) is available through a departmental starting fund. BRC Perioperative Medicine theme funding (research nurse support). IARS Mentored Research Grant (Disposables).