Lewis Aiden Vincent, Fang Qianqian
Northeastern University, Department of Bioengineering, 360 Huntington Avenue, Boston, USA, 02115.
Northeastern University, Department of EECS, 360 Huntington Avenue, Boston, USA, 02115.
bioRxiv. 2024 Dec 9:2024.08.20.608859. doi: 10.1101/2024.08.20.608859.
The diffusion approximation (DA) is used in functional near-infrared spectroscopy (fNIRS) studies despite its known limitations due to the presence of cerebrospinal fluid (CSF). Many of these studies rely on a set of empirical CSF optical properties, recommended by a previous simulation study, that were not selected for the purpose of minimizing DA modeling errors.
We aim to directly quantify the accuracy of DA solutions in brain models by comparing those with the gold-standard solutions produced by the mesh-based Monte Carlo (MMC), based on which we derive updated recommendations.
For both a 5-layer head and Colin27 atlas models, we obtain DA solutions by independently sweeping the CSF absorption and reduced scattering coefficients. Using an MMC solution with literature CSF optical properties as reference, we compute the errors for surface fluence, total brain sensitivity and brain energy-deposition, and identify the optimized settings where the such error is minimized.
Our results suggest that previously recommended CSF properties can cause significant errors (8.7% to 52%) in multiple tested metrics. By simultaneously sweeping and , we can identify infinite numbers of solutions that can exactly match DA with MMC solutions for any single tested metric. Furthermore, it is also possible to simultaneously minimize multiple metrics at multiple source/detector separations, leading to our new recommendation of setting while maintaining physiological for CSF in DA simulations.
Our new recommendation of CSF equivalent optical properties can greatly reduce the model mismatches between DA and MMC solutions at multiple metrics without sacrificing computational speed. We also show that it is possible to eliminate such a mismatch for a single or a pair of metrics of interest.
尽管扩散近似法（DA）因脑脊液（CSF）的存在存在已知局限性，但仍用于功能近红外光谱（fNIRS）研究。许多此类研究依赖于先前模拟研究推荐的一组经验性脑脊液光学特性，这些特性并非为最小化DA建模误差而选择。
我们旨在通过将DA解决方案与基于网格的蒙特卡罗（MMC）产生的金标准解决方案进行比较，直接量化大脑模型中DA解决方案的准确性，并据此得出更新的建议。
对于五层头部模型和Colin27图谱模型，我们通过独立扫描脑脊液吸收系数和约化散射系数来获得DA解决方案。以具有文献脑脊液光学特性的MMC解决方案为参考，我们计算表面注量、全脑灵敏度和脑能量沉积的误差，并确定使此类误差最小化的优化设置。
我们的结果表明，先前推荐的脑脊液特性在多个测试指标中可能导致显著误差（8.7%至52%）。通过同时扫描吸收系数和约化散射系数，我们可以找到无数组解决方案，这些方案可以使DA与MMC解决方案在任何单个测试指标上精确匹配。此外，在多个源/探测器间距下同时最小化多个指标也是可能的，这就得出了我们在DA模拟中设置吸收系数的新建议，同时保持脑脊液生理状态下的约化散射系数。
我们关于脑脊液等效光学特性的新建议可以在不牺牲计算速度的情况下，大大减少DA和MMC解决方案在多个指标上的模型不匹配。我们还表明，对于单个或一对感兴趣的指标，消除这种不匹配是可能的。
