Synthetic Vessels 0.1

This dataset consists of a set of synthetically generated 3D vessels, intended to support the development and benchmarking of representation learning models for continuous 3D vascular geometry, particularly for models capable of capturing fine-grain details like stenoses. The generation framework is based on physical principles designed to produce a wide variety of smooth, biologically-inspired vessel structures. Each vessel is represented as a generalized cylinder, stored as an array of shape (n, 4), where n is the number of points along the curvilinear centerline. The four coordinates represent the x, y, z position and the vessel radius (r) at that point. The 3D curvilinear centerlines are created by numerically integrating the trajectory of a particle under a set of dynamic forces. This produces a wide variety of smooth, tapering vessel paths. A radius profile is generated along the arc of each vessel. Simulated narrowings (localised stenoses) are added to 50% of the generated vessels. These stenoses are characterised by their position, size, and severity (occlusion), which are introduced by modifying the radius profile with a smooth bump function. More information, can be found in the associated GitHub repository: https://github.com/JamesBatten/SyntheticVessels