Aims of the Project:
The aims of the project include:
- Extracting and generating geometric representations (meshes or graphs) of patient-specific cardiac motion patterns from cardiac MRI
- Developing geometric deep learning approaches to predict clinically relevant information from these cardiac motion patterns.
The vision for this PhD project is to explore novel approaches for the diagnosis of cardiovascular diseases (CVD) from Magnetic Resonance Images (MRI). Our proposed approach will exploit recent developments in the field of Artificial Intelligence (AI) and Machine Learning (ML) to derive new biomarkers of CVD by assessing both the structure and function of the heart. This will enable a comprehensive diagnostic assessment and interpretation of the patient’s cardiovascular health/disease, enabling optimal treatment decisions for the best patient outcome.
Many of the recent advanced in AI and ML rely on so-called deep learning approaches that exploit convolutional neural networks (CNNs) to extract and interpret information derived from images that are represented as 2D, 3D or 4D regularly spaced arrays of voxels. However, in cardiac MRI, the imaging data is often acquired as stacks of 2D images with irregular configurations. Moreover, in clinical practice, the acquired imaging data is often sparse in 3D. More recently geometric deep learning approaches have shown to be highly successful in extracting and interpreting information derived from data which is represented in graph form. In this project, we will develop geometric deep learning approaches that use graphs in form of epi- and endocardial surfaces of the heart to derive new biomarkers of cardiac structure and function from cardiac MRI.
The ideal PhD student for this project is proficient in AI and ML and has very good programming and software engineering skills. Good communication skills and a desire to work in an interdisciplinary team are also required. Previous experience in medical imaging is not required as the MRes will cover these aspects.