- Start date: 1 April 2019
- End date: 30 September 2020
- Funder: EPSRC
- Value: £135,402
- Primary investigator: 01039087
There are pieces of evidence that a large category of cardiovascular diseases (CADs) can be diagnosed by examining ventricular motion abnormalities when differences due to age, gender, weight, etc. (patient’s metadata) are removed. Unlike conventional motion atlases which explicitly ignore the impact of metadata, we significantly go beyond the state-of-the-art and propose a scalable probabilistic atlas to accurately evaluate bi-ventricular motion abnormalities by integrating both imaging and metadata from big populations.
The motion is modelled as the spatiotemporal (3D+t) sequence of the heart shapes across the full cardiac cycle, extracted from cine CMR images. Fundamentally different from existing models, the atlas is a recurrent deep model that, given a cine sequence, will predict a probabilistic distribution function (pdf) for the next status of the heart.
More importantly, the pdf will be conditional on the patient metadata. Thus, by measuring the spatial deviations from the expected shape at each phase, it will allow accurate and personalized quantification of functional abnormality maps.
The success of this project is underpinned by:
- Bi-ventricular shape modelling and Bayesian methods under the evident success of deep models;
- Using UK Biobank (UKB) Cardiac Magnetic Resonance (CMR) database providing richness and reliability of the proposed atlas.
It is the world’s largest population of the kind (planning to scan n>100,000).
In alignment with EPSRC’s strategic plan, this project contributes to the timely identification of patients with heart functional abnormalities. To the best of our knowledge, BIANDA would be the first framework to derive a statistical atlas from joint motion and metadata from a large scale population imaging data.
The atlas will offer an unprecedented statistical power thanks to the sheer number of the patient data to be analyzed. The atlas, along with all the software and tools developed during BIANDA, will be returned to UKB for further and globalized access by other researchers.
Furthermore, the theoretical developments will be of fundamental relevance for a wider community of machine learning, fostering the benefits of the Bayesian deep models (handling uncertainties and boosting the prediction performance). The tools developed within BIANDA will be immediately beneficial in all such non-medical frameworks.