图9：不同优化策略的内存和SM利用率通过GPU指标可以看出（图8，图9）量化策略通过提升IO同时降低计算的方式提高整体计算性能；循环展开通过大幅度降低I/O同时提高计算密度的方式提高计算性能。计算精度团队统计了加速前与加速后的结果误差，仿真的膜电位V的时程差别<2 ms (0.6%)，模电位平均误差为0.72mV(0.4%)，均满足生理准确度要求。优化前后主要离子通道的仿真曲线吻合（如图10所示）。
图10：仿真前后细胞主要离子通道电流与胞内离子浓度在一心律节拍间的变化
总结
智源研究院从心脏模型的解剖结构、心肌细胞电生理的计算特点及计算系统的硬件架构出发，设计了心脏仿真系统的数据结构和优化策略，以提高计算效率。团队采用先进的并行处理方法，充分利用现代GPU设备的强大计算能力，优化数据传输和通讯方式，以减少延迟并提高数据吞吐量。通过这些策略，不仅提升了仿真系统的计算速度，还保证了在可接受误差范围内的计算精度，最终成功实现了心脏仿真的实时计算目标，达到超实时计算结果。这一成果为进一步研究心律失常产生的离子通道与分子机制等关键医学问题，也为手术规划如房颤射频消融方案等临床应用，以及新药研发与其心脏安全性筛选奠定了坚实基础，同时也为其它超大复杂物理系统的实时仿真提供坚实基础。参考资料：1. Hodgkin A L, Huxley A F. A quantitative description of membrane current and its application to conduction and excitation in nerve[J]. The Journal of Physiology, 1952, 117(4): 500-544.2. Noble D. Cardiac action and pacemaker potentials based on the Hodgkin-Huxley equations.[J]. Nature, 1960, 188(4749): 495-497.3. Crampin E J , Halstead M , Hunter P , et al. Computational physiology and the physiome project[J]. Experimental Physiology, 2004, 89.4. Smaill B, Hunter P. Structure and function of the diastolic heart: material properties of passive myocardium[M]//Theory of heart. Springer New York, 1991: 1-29.5. Alday E A P, Colman M A, Langley P, Zhang H. Novel non-invasive algorithm to identify the origins of re-entry and ectopic foci in the atria from 64-lead ECGs: A computational study[J]. PLoS Computational Biolog. 2017, 13(3): e1005270.6. Boyett M R, Li J, Inada S, et al. Imaging the heart: computer three-dimensional anatomical models of the heart[J]. Journal of Electrocardiology, 2005.7. Nordsletten D, Niederer S, Nash M P, et al. Coupling multi-physics models to cardiac mechanics[J]. Progress in Biophysics & Molecular Biology, 2011, 104(1): 77-88.8. Colman M A, Aslanidi O V, Kharche S, et al. Pro‐arrhythmogenic effects of atrial fibrillation‐induced electrical remodelling: insights from the three‐dimensional virtual human atria[J]. The Journal of physiology, 2013, 591(17): 4249-4272.9. Wang W, Xu L, Cavazos J, Huang HH, Kay M. Fast acceleration of 2D wave propagation simulations using modern computational accelerators. PLoS One. 2014;9(1):e86484.10. Kaboudian A, Cherry EM, Fenton FH. Real-time interactive simulations of large-scale systems on personal computers and cell phones: Toward patient-specific heart modeling and other applications. Sci Adv. 2019;5(3):eaav6019.11. Garcia-Molla VM, Liberos A, Vidal A, Guillem MS, Millet J, Gonzalez A, et al. Adaptive step ODE algorithms for the 3D simulation of electric heart activity with graphics processing units. Comput Biol Med. 2014;44:15-26.12.Sachetto Oliveira R, Martins Rocha B, Burgarelli D, Meira W, Jr., Constantinides C, Weber Dos Santos R. Performance evaluation of GPU parallelization, space-time adaptive algorithms, and their combination for simulating cardiac electrophysiology. Int J Numer Method Biomed Eng. 2018;34(2).
