Complication Prediction after Esophagectomy with Machine Learning.

Esophageal cancer can be treated effectively with esophagectomy; however, the postoperative complication rate is high. In this paper, we study to what extent machine learning methods can predict anastomotic leakage and pneumonia up to two days in advance. We use a dataset with 417 patients who underwent esophagectomy between 2011 and 2021. The dataset contains multimodal temporal information, specifically, laboratory results, vital signs, thorax images, and preoperative patient characteristics. The best models scored mean test set AUROCs of 0.87 and 0.82 for leakage 1 and 2 days ahead, respectively. For pneumonia, this was 0.74 and 0.61 for 1 and 2 days ahead, respectively. We conclude that machine learning models can effectively predict anastomotic leakage and pneumonia after esophagectomy.

Development of an in vitro setup for flow studies in a stented carotid artery bifurcation.

To investigate flow conditions in a double-layered carotid artery stent, a bench-top in vitro flow setup including a bifurcation phantom was designed and fabricated. The geometry of the tissue-mimicking phantom was based on healthy individuals. Two identical phantoms were created using 3D-printing techniques and molding with PVA-gel. In one of them, a clinically available CGuard double-layer stent was inserted. Measurements were performed using both continuous and pulsatile flow conditions. Blood flow studies were performed using echoPIV: a novel ultrasound-based technique combined with particle image velocimetry. A maximum deviation of 3% was visible between desired and measured flow patterns. The echoPIV measurements showed promising results on visualization and quantification of blood flow in and downstream the stent. Further research could demonstrate the effects of a double-layered stent on blood flow patterns in a carotid bifurcation in detail.

Aneurysm-on-a-Chip: Setting Flow Parameters for Microfluidic Endothelial Cultures Based on Computational Fluid Dynamics Modeling of Intracranial Aneurysms.

Intracranial aneurysms are pouch-like extrusions from the vessels at the base of the brain which can rupture and cause a subarachnoid hemorrhage. The pathophysiological mechanism of aneurysm formation is thought to be a consequence of blood flow (hemodynamic) induced changes on the endothelium. In this study, the results of a personalized aneurysm-on-a-chip model using patient-specific flow parameters and patient-specific cells are presented. CT imaging was used to calculate CFD parameters using an immersed boundary method. A microfluidic device either cultured with human umbilical vein endothelial cells (HUVECs) or human induced pluripotent stem cell-derived endothelial cells (hiPSC-EC) was used. Both types of endothelial cells were exposed for 24 h to either 0.03 Pa or 1.5 Pa shear stress, corresponding to regions of low shear and high shear in the computational aneurysm model, respectively. As a control, both cell types were also cultured under static conditions for 24 h as a control. Both HUVEC and hiPSC-EC cultures presented as confluent monolayers with no particular cell alignment in static or low shear conditions. Under high shear conditions HUVEC elongated and aligned in the direction of the flow. HiPSC-EC exhibited reduced cell numbers, monolayer gap formation and cells with aberrant, spread-out morphology. Future research should focus on hiPSC-EC stabilization to allow personalized intracranial aneurysm models.

In vitro performance of echoPIV for assessment of laminar flow profiles in a carotid artery stent.

Purpose: Detailed blood flow studies may contribute to improvements in carotid artery stenting. High-frame-rate contrast-enhanced ultrasound followed by particle image velocimetry (PIV), also called echoPIV, is a technique to study blood flow patterns in detail. The performance of echoPIV in presence of a stent has not yet been studied extensively. We compared the performance of echoPIV in stented and nonstented regions in an in vitro flow setup. Approach: A carotid artery stent was deployed in a vessel-mimicking phantom. High-frame-rate contrast-enhanced ultrasound images were acquired with various settings. Signal intensities of the contrast agent, velocity values, and flow profiles were calculated. Results: The results showed decreased signal intensities and correlation coefficients inside the stent, however, PIV analysis in the stent still resulted in plausible flow vectors. Conclusions: Velocity values and laminar flow profiles can be measured in vitro in stented arteries using echoPIV.

Development and application of a volume penalization immersed boundary method for the computation of blood flow and shear stresses in cerebral vessels and aneurysms.

A volume-penalizing immersed boundary method is presented for the simulation of laminar incompressible flow inside geometrically complex blood vessels in the human brain. We concentrate on cerebral aneurysms and compute flow in curved brain vessels with and without spherical aneurysm cavities attached. We approximate blood as an incompressible Newtonian fluid and simulate the flow with the use of a skew-symmetric finite-volume discretization and explicit time-stepping. A key element of the immersed boundary method is the so-called masking function. This is a binary function with which we identify at any location in the domain whether it is 'solid' or 'fluid', allowing to represent objects immersed in a Cartesian grid. We compare three definitions of the masking function for geometries that are non-aligned with the grid. In each case a 'staircase' representation is used in which a grid cell is either 'solid' or 'fluid'. Reliable findings are obtained with our immersed boundary method, even at fairly coarse meshes with about 16 grid cells across a velocity profile. The validation of the immersed boundary method is provided on the basis of classical Poiseuille flow in a cylindrical pipe. We obtain first order convergence for the velocity and the shear stress, reflecting the fact that in our approach the solid-fluid interface is localized with an accuracy on the order of a grid cell. Simulations for curved vessels and aneurysms are done for different flow regimes, characterized by different values of the Reynolds number (Re). The validation is performed for laminar flow at Re = 250, while the flow in more complex geometries is studied at Re = 100 and Re = 250, as suggested by physiological conditions pertaining to flow of blood in the circle of Willis.