Geometric Machine Learning for Patient-Specific 3D Cardiac Anatomy Reconstruction

This is a hybrid event. Please find the Teams link in the abstract.

Brief Bio:
Dr Abhirup Banerjee is a Royal Society University Research Fellow (URF), Full Member of Faculty, and PI in the Department of Engineering Science, University of Oxford. He leads the Multimodal Medical Data Integration & Analysis (MultiMeDIA) Lab in the Institute of Biomedical Engineering (IBME), University of Oxford. Dr Banerjee received the BSc (Hons) and Master degrees in Statistics and the PhD degree in Computer Science in March 2017. He joined the University of Oxford as Postdoctoral Researcher in the Division of Cardiovascular Medicine in August 2017, started as the URF and Faculty Member in the Department of Engineering Science in October 2022, and officially started the MultiMeDIA Lab in March 2023. His research interest spans Biomedical Engineering, Computer Science, and classical Statistics, focusing on a range of topics including Biomedical Image Analysis, Machine Learning, AI, Geometric Deep Learning, Image Processing, etc. Dr Banerjee received the Young Scientist Award from the Indian Science Congress Association in the year 2016-2017.

Teams link:

Cardiac magnetic resonance imaging (MRI) is one of the most important imaging modalities for the diagnosis and characterisation of cardiovascular diseases, due to its non-invasive identification of abnormalities in structure and function of the myocardium without ionising radiation. However, in current clinical practice, it is commonly acquired as a collection of separated and independent 2D image planes, thus limiting its accuracy in 3D analysis.

In order to generate patient-specific 3D heart meshes from the 2D MRI, we have developed completely end-to-end automated pipeline, correcting for the sparsity and misalignment due to motion artifacts between slices. Our development of novel geometric deep learning in particular point cloud-based approaches has enabled the 3D cardiac anatomy reconstruction in real-time and made possible the population-level analyses of anatomy and functions including virtual population cohorts generation for in silico trials, cardiac motion modelling, combined modelling of anatomy and electrophysiology, risk prediction, etc. The reconstructed high-resolution 3D cardiac meshes have been utilised for in silico experiments to simulate the activation patterns. The effectiveness of the novel geometric deep learning-based approaches has been extensively investigated over large (>10K) UK population and have opened up the possibility of large virtual in silico trials.