Bousselham Abdelmajid, Bouattane Omar, Youssfi Mohamed, Raihani Abdelhadi
Laboratory SSDIA, ENSET Mohammedia, University Hassan 2, Casablanca, Morocco.
Laboratory SSDIA, ENSET Mohammedia, University Hassan 2, Casablanca, Morocco.
J Therm Biol. 2018 Jan;71:52-61. doi: 10.1016/j.jtherbio.2017.10.014. Epub 2017 Oct 26.
The aim of this paper is to present a GPU parallel algorithm for brain tumor detection to estimate its size and location from surface temperature distribution obtained by thermography. The normal brain tissue is modeled as a rectangular cube including spherical tumor. The temperature distribution is calculated using forward three dimensional Pennes bioheat transfer equation, it's solved using massively parallel Finite Difference Method (FDM) and implemented on Graphics Processing Unit (GPU). Genetic Algorithm (GA) was used to solve the inverse problem and estimate the tumor size and location by minimizing an objective function involving measured temperature on the surface to those obtained by numerical simulation. The parallel implementation of Finite Difference Method reduces significantly the time of bioheat transfer and greatly accelerates the inverse identification of brain tumor thermophysical and geometrical properties. Experimental results show significant gains in the computational speed on GPU and achieve a speedup of around 41 compared to the CPU. The analysis performance of the estimation based on tumor size inside brain tissue also presented.
本文旨在提出一种用于脑肿瘤检测的GPU并行算法，以便根据热成像获得的表面温度分布来估计其大小和位置。正常脑组织被建模为包含球形肿瘤的长方体。温度分布使用正向三维彭尼斯生物热传递方程进行计算，通过大规模并行有限差分法（FDM）求解，并在图形处理单元（GPU）上实现。遗传算法（GA）用于解决逆问题，并通过最小化一个目标函数来估计肿瘤大小和位置，该目标函数涉及表面测量温度与数值模拟获得的温度。有限差分法的并行实现显著减少了生物热传递的时间，并极大地加速了脑肿瘤热物理和几何特性的逆识别。实验结果表明，与CPU相比，GPU上的计算速度有显著提高，加速比约为41。还展示了基于脑组织内肿瘤大小的估计分析性能。
