BIANDA: Bayesian Deep Atlases for Cardiac Motion Abnormality Assessment from Imaging and Metadata

Lead Research Organisation: University of Leeds
Department Name: Sch of Computing

Abstract

Cardiovascular diseases (CVDs) is the second biggest killer in the UK and currently, more than 7 million people are living with CVD in the country. Early identification of individuals with significant risk is critical to improve the patient quality of life and reduce the financial burden on the social and healthcare systems. A large number of CVDs lead to the shortage of blood supply to the heart muscle and abnormal motion, which can be diagnosed non-invasively by analysing the patient's dynamic cardiac imaging data. Manual assessment of these images is subjective, non-reproducible, limited to the left ventricle, and time-consuming. Statistical atlases, describing the 'average' pattern of the heart motion over a large healthy population, can be potentially useful to identify deviations from normality in individuals. However, the integration of the existing atlases into clinical practice is inhibited by three key limitations: (i) the derived motion statistics are often independent of the patient's age, gender, weight, etc. (metadata) that are essential for precise diagnosis, (ii) Being non-probabilistic, these atlases fail to provide a measure of certainty in the extracted motion abnormalities thus their clinical reliability is seriously hampered, (iii) they are often derived using a small number of data sets (n<1000), limiting their statistical power.

To alleviate these key limitations, this proposal aims, for the first time, to develop a full probabilistic atlas to accurately evaluate bi-ventricular motion abnormalities by holistically integrating imaging and metadata from a large population cardiac imaging study. BIANDA will be a novel Bayesian approach extending the recent developments in deep recurrent neural networks (RNNs). These networks provide a natural mechanism to model sequential data such as 2D video. Yet, using RNNs to model the complex dynamics of the heart motion is conceptually new and evidently powerful. The motion will be modelled as the spatiotemporal (3D+t) sequence of the heart shapes across the full cardiac cycle, extracted from cine Cardiac Magnetic Resonance (CMR) images. The atlas will be a recurrent model that, given a sequence, it will predict a probabilistic distribution function (pdf) for the next status of the heart. More importantly, the pdf will be conditioned on the patient's metadata. Thus by measuring the spatial deviations from the expected shape at each phase, the atlas will allow very accurate quantification of anatomical and functional cardiac abnormalities (and variances showing uncertainties) specific to the patient's age, gender, age, ethnicity, etc.

The PI has an extensive experience in developing Bayesian and non-Gaussian statistical atlases from shapes. However, the previous work (i) was not designed to analyse motion data, (ii) discarded the patient metadata (such as age, gender, ethnicity, etc.), and (iii) did not scale into large populations. Therefore, the atlas was not clinically deployable to study cardiac motion abnormalities, which are relevant to various CVDs. This proposal will significantly depart from the PI's previous by combining Bayesian models with deep neural networks. The former is required to handle uncertainties; the latter will significantly boost the prediction and computational efficiency (using GPUs), thus scalability.

The atlas will be derived from the UK Biobank CMR study aiming to scan n>100,000 patients by 2022. The training of the atlas will be pursued as the new releases of the data sets from the UK Biobank becomes available. The PI has established collaboration with the clinical advisor for this study and has full access to the CMR data sets. This is essential for the success of the proposal as the training of deep neural networks requires access to an ample of data sets, a possibility which has emerged only recently. In this regard, BIANDA is timely and promising.

Planned Impact

BIANDA will have a profound impact on the economy, health, and well-being of the society. An impact maximizing strategy has been implemented in the design of the research. By delivering a novel probabilistic cardiac motion atlas, normative with regard to patient's metadata, it responds to real clinical needs, i.e., achieving more patient-specific and personalized assessment of cardiac motion abnormality (and their confidence) maps. Thanks to the sheer data size, incorporation of the patient metadata, and the cutting edge computational algorithms employed in the project, the atlas will achieve an unprecedented accuracy in specifying deviations from a normal heart motion, hence early detection of a CVD status. Thus, the most obvious beneficiaries of the research are patients with various forms of Cardio Vascular Diseases (CVD), such as Coronary Heart Diseases (CHD), Pulmonary Hypertension, etc., which can result in abnormal heart motion.

By quantifying uncertainties in assessments of shape/motion abnormalities at each cardiac phase, BIANDA provides spatiotemporal confidence maps to clinicians showing the reliability of the extracted abnormalities. This new feature is clearly beyond the cutting edge in statistical atlases and will help the clinicians to make better-informed decisions. Thus, BIANDA will eventually lead into an improved patient quality of life, and reduced rate of unexpected heart failures and mortality due to CVD, supporting "Healthy Nation" as one of the EPSRC's four Prosperity Outcomes.

The overall annual cost of CVD on UK's economy is over £15 billion. The 'Five years Forward View', the NHS's report in 2014, strongly recommends that 'system must get serious about prevention', highlighting the need for extra work. Despite this, NHS funding in 2016 ended up with an aggregated deficit of £1.85 billion. With an aging UK population and the increased prevalence of the CVDs (esp. CHD) in older ages, the significance of prevention of cardiac adverse eves becomes even more critical. Therefore, there is an urgent need for cost-effective imaging-based markers. By early identification of individuals at risk, this project can contribute to reducing the overall CVD's cost and help prosperity of the national economy.

In the long term, we anticipate that the developed statistical cardiac motion atlas will be incorporated into clinical practice and may be adopted internationally. The clinical translation will be realized in coordination and communication with Prof Steffen Petersen (cardiologist in Queen Mary University of London), who will sit in the steering committee to provide clinical insight. This engagement will ensure that the overall the direction of the project will remain coherent with its clinical objectives, addressing real patient needs. The research meetings will create an opportunity for QMUL researchers to closely follow up the progress of the BIANDA and later advocate the use of the statistical atlas in a wider clinical community. This engagement will be key to increase the effectiveness of the research and promote its impact on clinical cardiology.

The project will yield to a fully trained research team (including the PDRA/PGR1-2) with an exceptional set of computational skills. The trained staff can be subsequently employed in UK healthcare organizations, further reducing overall the national cost of hiring skilled workers. The project will seek for active engagement of industry.

BIANDA will entail extensive use of high-performance GPUs. The PI has been awarded a GPU unit and received interest by NVIDIA. The company is keen to see BIANDA's wider impact on the industry with a provision that many of the computational developments achieved in this project will be general enough to promote similar research in other fields.

Publications

10 25 50
 
Title Probabilistic model of Encoder-Decoder to predict cardiac shapes and abnormality using Expectation-Maximization algorithm 
Description We used a probabilistic encoder- decoder deep model to predict the cardiac motions in a supervised manner. In addition, abnormality assessment was performed by proposing a novel implementation of EM method and estimation of posterior probability of abnormality in each time frame of cardiac motions. The EM algorithm is an approach for performing maximum likelihood estimation in the presence of latent variables. It does this by first estimating the values for the latent variables, then optimizing the model, then repeating these two steps until convergence. 
Type Of Material Model of mechanisms or symptoms - human 
Provided To Others? Yes  
Impact By implementing EM in a neural network architecture it is possible to fit and tune a pdf on observations automatically without analytical computations in the conventional EM algorithm. Hence, it is possible to apply that easily to model nonlinear behaviours in physiological mechanisms such as cardiac motions and predicting abnormality in data automatically. 
 
Title Expectation Maximization based parameter learning in recurrent probabilistic neural network architecture for cardiac motion prediction using Uk Biobank 
Description In this research we propose a novel deep probabilistic neural network architecture which its parameters are learned by EM algorithm. we use UK Biobank to evaluate our proposed method. UK Biobank is a major national and international health resource which includes data of a wide range of serious and life-threatening illnesses including heart diseases. More than 32k cardiac subjects will be analysed in our project. 
Type Of Material Computer model/algorithm 
Year Produced 2012 
Provided To Others? Yes  
Impact The developed model should be capable to predict the cardiac motions as well as the posterior probability of abnormality in a given data. Since UK Biobank is the most complete database with more than 32K cardiac subjects currently available to analyse, the evaluated model by this database could be considered as a reliable atlas for abnormality assessment. 
 
Title Python and Pytorch 
Description We use python software which is an interpreted, high-level, general-purpose programming language. In order to implement our proposed model, we used Pytorch library. PyTorch is an open source machine learning library based on the Torch library. We execute our model in a cloud- based environment called MULTI-X. Indeed, MULTI-X serves as middleware between storage and computing cloud providers (e.g., Amazon Web Services, GoogleCloud, and Microsoft Azure), workflow managers (e.g., Taverna and Nipype), data sources (e.g., UKB servers) and analytics tools providers. 
Type Of Technology Software 
Year Produced 2018 
Open Source License? Yes  
Impact Developing MULTI-X facilitates secure access and execution, component integration and interoperability (e.g., across different programming languages, frameworks, operating systems, and hardware), workflow execution, monitoring, and execution report generation.