PhD defence by Peidi Xu

Follow this link to participate on Zoom

Title

Virtual Kidney - Full-scale reconstruction and simulation of the renal nephron-vascular network

Abstract

The renal vasculature, functioning as a resource distribution network, plays an essential role in the kidney’s physiology and pathophysiology. The diagnosis of vascular diseases such as renal artery stenosis can usually be achieved through medical scanning of the large renal arteries. However, the renal filtration and autoregulation mechanisms take place in the smallest terminal arteries in the renal arterial tree, also known as afferent arterioles, which are beyond the resolution of most existing medical imaging modalities. This limitation of medical imaging and the fundamental function of afferent arterioles makes data-driven generative models an important tool for studies of the regulation of renal blood flow. To better understand the regulatory mechanisms of renal blood flow, this PhD thesis presents a novel hybrid approach to reconstruct a full-scale renal arterial network. The hybrid approach reconstructs small arteries in the fullscale arterial tree using a physiologically based optimization method, while also integrating vascular geometries extracted from medical images. Then, models of nephrons, where the renal filtration and the control renal hemodynamics takes place, are attached to each afferent arteriole of the reconstructed full-scale arterial tree. With the full-scale nephron-vascular model, autoregulatory mechanisms are simulated to define how topological and hemodynamic profiles of microcirculatory networks are optimized and adapted to pathological changes. Finally, although most of the blood vessels in the full-scale model are reconstructed physiologically, segmenting visible large blood vessels from medical scans is still a crucial step for guiding the physiological reconstruction of the remaining small vessels. Therefore, the PhD thesis also includes medical image segmentation, and the related deep learning approaches, which have become state-of-the-art methods for automatic segmentation tasks.

Supervisors

Principal Supervisor Sune Darkner, DIKU
Co-supervisor Kenny Erleben, DIKU
Co-supervisor Olga Sosnovtseva, BMI, SUND
Co-supervisor Niels-Henrik von Holstein-Rathlou BMI, SUND

Assessment Committee

Associate Professor Jens Petersen, DIKU
Research Professor Molly Maleckar, Simula
Associate professor Vedrana Andersen Dahl, DTU Compute

For an electronic copy of the thesis, please visit the PhD Programme page.