Cardiovascular diseases (CVDs) are a group of disorders of the heart and blood vessels.
CVDs are the leading cause of death worldwide. To diagnose and treat CVDs, clinicians and cardiologists use multiple noninvasive imaging techniques. These scans are used to segment certain structures of the heart. Deep learning-based cardiac segmentation on short-axis cardiac magnetic resonance images (CMRI) has gained popularity over the past few years because of its generalisability and accuracy. This has exponentially reduced contouring times for clinicians. The development of such deep learning techniques has seen a common trend. In order to accommodate learning for larger cardiac datasets, the depth and size of segmentation networks have been increased. Unfortunately, the environmental impact of exploding such networks is not taken into account. One solution to mitigate having computationally expensive networks is to incorporate anatomical knowledge in the form of shape priors. The Gridnet and UNet with a shape prior are computationally efficient networks that are used to evaluate segmentation performance on a large and varied cardiac dataset (Combination of the Automated Cardiac Diagnosis Challenge - ACDC and Multi-Centre, Multi-Vendor and Multi-Disease Cardiac Segmentation challenge - M&M datasets). On average, these networks segment CMRIs with an average dice score of 0.87 and a Hausdorff distance of 11.7mm. In parallel, one of the major issues in cardiac technology is the under-representation of women in cardiac datasets. Purposefully curated cardiac datasets such as ACDC and M&M try and maintain equal representation. In real-world scenarios, this might not always be the case. Clinical trials to collect such data often report female representation as low as 25%. Evaluation of segmentation performance between a balanced and skewed dataset is conducted. This is to address if bias in such cardiac training datasets affects the performance of segmentation networks between male and female test patients.

This thesis investigates the ability of Bayesian EUCLID to retrieve a predictive approximate material model for the myocardium in the presence of heterogeneous deformation fields due to simulated biaxial stretch tests. The Holzapfel-Ogden material model is used as the ground-truth material model of the simulation, since it is capable of describing the orthotropic and hyperelastic behavior of the myocardium. In total, four material models were retrieved from Bayesian-EUCLID by considering two fiber orientations within the sample and two displacement measurement techniques. The two fiber directions included one in which the sheet-like structure of the sample lies within the test plane and one in which it lies perpendicular to the test plane. The displacement field measurements considered were a full displacement field measurement and an approximation thereof based on a stereo digital image correlation method. The results show that when the sheet-like structure of the myocardium lies within the test plane of the biaxial stretch test and the full displacement field is available, Bayesian EUCLID is capable of finding orthotropic and hyperelastic material models that correspond well to the ground truth. This shows that a biaxial stretch test can adequately characterize the material behavior of the myocardium.

A congenital heart defect (CHD) is an anomaly in the structure of the heart that is present at birth. In the last 15 years, a CHD is present in 9 per 1,000 live births, making it the most prevalent birth defect (Linde et al., 2011).
CHD’s prevalence coupled with its inherent complexity culminates into situations that are both complicated and unique. This makes the treatment of patients with CHDs highly patient-specific. Patient-specific physics-based computational models can aid in selecting the optimal course of treatment for the patient and serve as a predictive tool for different surgical plans (Capelli et al., 2011). The creation of patient-specific geometric heart models of children with CHDs will be the first crucial step towards a patient-specific approach in ameliorating this challenge. Approaches have to be found to face challenges including: a lack of high quality imaging data of the target population; slice misalignment due to breathing in a breath-hold procedure; and a lack of protocols in image segmentation.

To accomplish the objective, seven biventricular geometric models and meshes are created for healthy hearts, tetralogy of Fallot hearts and Fontan hearts. Slice misalignment is corrected using contours of the papillary muscle and the epicardium as a reference. Lastly, ground truth geometries are constructed by stacking disks that originate directly from the MRI segmentation.
To validate the geometric models, global and local approaches are used. The global validation includes ventricular mass and volume analyses with a relative error (RE) ranging from 0.00036 to 0.79. The Dice similarity coefficient (DSC) for the local validation is between 0.711 +/- 0.138 and 0.968 +/- 0.018. The slice misalignment correction is validated using qualitative and quantitative measures. A quantitative measure is the 4-chamber long-axis local validation which has a DSC of 0.877 for the epicardium and 0.916 for the papillary muscle correction method. Global validation in the ground truth exercise has a mean RE of 0.010 +/- 0.0066. Results of the local validation show a minimum DSC of 0.960 +/- 0.008.

Large differences in RE can be explained by the small number of MRI slices available, segmentation variability, the complex geometry of the right ventricle, and the methods of volume and mass computations. The DSC computed in the local validation of the models shows good to excellent agreement for all models. According to the long-axis local validation, the papillary muscle is the best reference method proposed for the slice misalignment correction. The ground truth exercise reveals close correlation to the Medis Suite measurements, possibly leading to underestimation of the computed volumes.

Patient-specific biventricular geometric models are created and validated in this thesis. The models are representative of a range of patients including healthy hearts and hearts with complex CHDs. A patient-specific model of the heart would be beneficial for clinicians to understand complex pathological cases, as the ones described in this work. Herein, a patient-specific biventricular geometric model is presented. By including a healthy population as well as patients affected by complex congenital heart defects, this work can be used as a starting point for the development of more patient-specific computational models of the heart.