Gpu Enhanced Simulation of Angiogenesis

Abstract

In the paper we present the use of graphic processor units to accelerate the most time-consuming stages of a simulation of angiogenesis and tumor growth. By the use of advanced CUDA mechanisms such as shared memory, textures and atomic operations, we managed to speed up the CUDA kernels by a factor of 57x. However, in our simulation we used the GPU as a co-processor and data from CPU was copied back and forth in each phase. It decreased the speedup of rewritten stages by 40%. We showed that the performance of the entire simulation can be improved by a factor of 10 up to 20

Efficient model of tumor dynamics simulated in multi-GPU environment

“…We demonstrate its high efficiency for both the standard setup representing a homogeneous tissue, and the setup, which is more topologically complex and mimics the skin structure. We show that our implementation is superior over other GPU implementations of discrete/continuous tumor models (Wang et al, 2015a, 2015b; Wcisło et al, 2013; Worecki and Wcisło, 2012; Zhang et al, 2011) and multithreading CPU models based on more sophisticated numerical engines (Łoś et al, 2017). In the “Concluding remarks” section, we summarize our results and we briefly present future research directions.…”

“…The most difficult problem in heterogeneous continuous/discrete and discrete/continuous models (Dzwinel et al, 2016a, 2018) is the problem of coupling their continuous and discrete components. In case of tumor dynamics, we show in Dzwinel et al (2016a) and Worecki and Wcisło (2012) that the vessels remodeling is the most computationally demanding part of the particle-based discrete/continuous PAM model. Moreover, it is the model component, which drastically decreases the overall gain in efficiency expected in case of its implementation in GPU/CUDA environment.…”

“…The GPU implementations of discrete/continuous models, such as the particle automata model (PAM) (Panuszewska et al, 2018; Wcisło et al, 2013; Worecki and Wcisło, 2012) and agent-based models (Jagiella et al, 2017; Wang et al 2015a, 2015b; Zhang et al 2011), where the tissue is represented by the cell–cell interactions while oxygen, tumor angiogenic factors (TAF), and nutrients concentrations are calculated by integrating reaction–diffusion equations, allow us for simulating rather small tumors in a mesoscopic scale, that is, a few millimeters in diameter. As shown in Jagiella et al (2017), the evolution of N = 10 6 cells (tumor and healthy ones) in 3-D involves 3–4 days of CPU time (5–6 min on HPC cluster with 1000 CPU cores (Jagiella et al, 2017)).…”

3D Visualisation of Tumour-Induced Angiogenesis Using the CUDA Programming Model and OpenGL Interoperability

“…With such a predictive model it is much easier and faster to perform in silico experiments to test hypotheses and predictions than running time consuming and costly laboratory experiments. More recently, the advantages of supercomputing and parallel processing techniques has highlighted the speedup, amongst other benefits, from the numerical solution of complex mathematical models of tumour dynamics [6][7][8][9][10]. In a previous paper, the authors developed a 3D parallel algorithm based on a time-stepping finite difference method (FDM) to solve a hybrid continuous-discrete model of tumour-induced angiogenesis [10].…”