Statistical shape modelling of patient-specific coronary arteries

Coronary arteries are susceptible to atherosclerosis, a chronic inflammatory condition of the arterial wall that gradually narrows the lumen due to plaque buildup and leading to coronary artery disease (CAD), ranking as the third leading cause of death globally. In this context, arterial anatomical features play a key role in shaping local hemodynamics and its harmful effects on the endothelium, which contribute to the onset and progression of CAD. Statistical shape models (SSMs) are increasingly being used in the cardiovascular field to investigate the complex anatomical variations of cardiovascular structures. By capturing the geometric variability within populations, SSMs might enable the identification of specific shape patterns and deviations associated with diseases. This work aims to develop two different SSMs to describe the geometric variability of diseased left anterior descending coronary arteries (LAD) without side branches. The two SSMs were built by considering two patient-specific diseased LADs cohorts and applying the same overall workflow. The first dataset included 69 LADs, with percentage area stenosis (%AS) ranging from 26% to 78%. The second dataset comprised 191 LADs, with %AS ranging from 50% to 70%. The three-dimensional (3D) vessel geometries previously reconstructed from coronary angiography were used to extract (using the statistical shape analysis tool Deformetrica) an anatomical template, representing the average anatomical shape, and the moment vectors mapping the template to each vessel-specific shape of the cohort. Principal component analysis (PCA) was then adopted to build the SSMs by identifying the largest shape variability in the two LAD cohorts. K-fold cross-validation with K=10 was employed to evaluate the stability of the final templates obtained from the statistical shape analysis. Technically, the cross-validation tested the stability of the template to the addition or removal of specific coronary artery shapes. For the first dataset, twelve, sixteen and twenty-seven shape modes explained 90%, 95% and 99% of total shape variability, respectively. The reconstruction error considering 12, 16 and 27 shape modes was evaluated through the mean and maximum Euclidean distance between the original and the reconstructed shapes, obtaining the following results: a mean distance of 0.37 mm, with a maximum distance of 2.42 mm, with 12 shape modes; a mean distance of 0.30 mm, with a maximum distance of 2.27 mm, with 16 shape modes; a mean distance of 0.26 mm, with a maximum distance of 2.27 mm with 27 shape modes. For the second dataset, fourteen, eighteen and thirty-four shape modes explained 90%, 95% and 99% of total shape variability, respectively. The reconstruction error considering 14, 18 and 34 shape modes was evaluated through the mean and maximum Euclidean distance between the original and the reconstructed shapes, obtaining: a mean distance of 0.36 mm, with a maximum distance of 3.38 mm, with 14 shape modes; a mean distance of 0.32 mm, with a maximum distance of 2.67 mm, with 18 shape modes; a mean distance of 0.29 mm, with a maximum distance of 2.57 mm with 34 shape modes. In conclusion, the approach presented in this Thesis project successfully enabled the derivation of a SSM of coronary arteries, capturing the main anatomical features of two different populations of LADs. The proposed approach could be exploited both to define the main morphometric features of coronary arteries and to generate virtual datasets from real-world patients.