WARNING: the interactive features of this website use CSS3, which your browser does not support. To use the full features of this website, please update your browser.
Convolutional neural networks (CNNs) with Bayesian inference are a category of artificial neural networks which model the uncertainty of the network output. This study presents an automated framework for tissue characterisation from native shortened modified Look-Locker inversion recovery ShMOLLI T1 mapping at 1.5 T using a Probabilistic Hierarchical Segmentation (PHiSeg) network (PHCUMIS 119 127, 2019). In addition, we use the uncertainty information provided by the PHiSeg network in a novel automated quality control (QC) step to identify uncertain T1 values. The PHiSeg network and QC were validated against manual analysis on a cohort of the UK Biobank containing healthy subjects and chronic cardiomyopathy patients (N=100 for the PHiSeg network and N=700 for the QC). We used the proposed method to obtain reference T1 ranges for the left ventricular (LV) myocardium in healthy subjects as well as common clinical cardiac conditions.
Motion-based diagnosis and characterisation of heart disease
The motion and shape of the human heart are closely tied with cardiac health, and vary greatly across the population. We aim to develop a cardiac magnetic resonance (CMR)-derived left-ventricle (LV) atlas from UK Biobank data for characterising the landscape of population LV shape and motion. We would use machine learning to analyse the rich spatio-temporal information embedded in the data to uncover new relationships between LV motion/shape with available health indicators. In this way we aim to derive new metrics of cardiac health to enhance diagnosis and potentially reduce time and cost in the clinic. The aim of our research is to characterise the landscape of LV motions and shapes across a large population. We aim to derive from our LV motion atlas new clinically relevant metrics of cardiac health by applying recent methods from machine learning. These methods will allow relationships to be drawn between LV motion and other factors which have been recorded for the subject cohort. Benefits of using this approach include enhancing diagnosis; disease staging; reducing the number of required diagnostic tests; and monitoring/predicting treatment response. Additionally the motion atlas can be used to enhance ultrasound imaging of cardiac motion. The LV myocardium is delineated for each subject from an end-diastolic CMR volume. LV motion is tracked over the course of the cardiac cycle using CINE/tagged CMR. Each patient's heart motion is normalised to an unbiased LV geometry so that motion can be directly compared between patients. Important characteristics of LV motion and shape can then be identified, including the mean and variance of the LV shape and motion patterns across the subject cohort. Machine learning can be used to identify correlations with health indicators and to enhance cardiac ultrasound by building a joint MR-US model.
The data which would be useful for our research includes all subject data for whom CINE and/or tagged CMR images have been acquired. The predictive power of machine learning improves with larger datasets as does the richness of the LV shape and motion characterisation, so all available subject data with the appropriate CMR images would benefit this project. Additionally, any meta-data, for example related to volunteer health, would be useful (e.g. ECG recording), and could be used to draw insight into their relationship with LV motion using machine learning.