Spatial Peak SAR Evaluation

16.1 Spatial Peak SAR for 1 g and 10 g

The DASY5 software includes all numerical procedures necessary to evaluate the spatial peak SAR values.

Based on the IEEE 1528.1 draft standard, a new algorithm has been implemented. The spatial peak SAR can be computed over any required mass.

The base for the evaluation is a ”cube” measurement in a volume of 30 (7x7x7 points). The measured volume must include the 1 g and 10 g cubes with the highest averaged SAR values. For that purpose, the center of the measured volume is aligned to the interpolated peak SAR value of a previously performed area scan. If the 10g cube or both cubes are not entirely inside the measured volumes, the system issues a warning regarding the evaluated spatial peak values within the Postprocessing engine (SEMCAD X). This means that if the measured volume is shifted, higher values might be possible. To get the correct values you can use a finer measurement grid for the area scan. In complicated field distributions, a large grid spacing for the area scan might miss some details and give an incorrectly interpolated peak location.

The entire evaluation of the spatial peak values is performed within the Postprocessing engine (SEMCAD X). The system always gives the maximum values for the 1 g and 10 g cubes. The algorithm to find the cube with highest averaged SAR is divided into the following stages:

1. extraction of the measured data (grid and values) from the Zoom Scan

2. calculation of the SAR value at every measurement point based on all stored data (A/D values and measurement parameters)

3. generation of a high-resolution mesh within the measured volume

4. interpolation of all measured values from the measurement grid to the high-resolution grid

5. extrapolation of the entire 3-D field distribution to the phantom surface over the distance from sensor to surface

6. calculation of the averaged SAR within masses of 1 g and 10 g

The significant parts are outlined in more detail within the following sections.

Interpolation, Extrapolation and Detection of Maxima

The probe is calibrated at the center of the dipole sensors which is located 1 to 2.7 mm away from the probe tip. During measurements, the probe stops shortly above the phantom surface, depending on the probe and the surface detecting system. Both distances are included as parameters in the probe configuration file. The software always knows exactly how far away the measured point is from the surface. As the probe cannot directly measure at the surface, the values between the deepest measured point and the surface must be extrapolated.

In DASY5, the choice of the coordinate system defining the location of the measurement points has no influence on the uncertainty of the interpolation, Maxima Search and extrapolation routines. The interpolation, extrapolation and maximum search routines are all based on the modified Quadratic Shepard’s method [1].

Thereby, the interpolation scheme combines a least-square fitted function method and a weighted average method which are the two basic types of computational interpolation and approximation. The DASY5 routines construct a once-continuously differentiable function that interpolates the measurement values as follows:

• For each measurement point a trivariate (3-D) / bivariate (2-D) quadratic is computed. It interpolates the measurement values at the data point and forms a least-square fit to neighboring measurement values.

• the spatial location of the quadratic with respect to the measurement values is attenuated by an inverse distance weighting. This is performed since the calculated quadratic will fit measurement values at nearby points more accurate than at points located further away.

• After the quadratics are calculated for at all measurement points, the interpolating function is calculated as a weighted average of the quadratics.

There are two control parameters that govern the behavior of the interpolation method . One specifies the number of measurement points to be used in computing the least-square fits for the local quadratics. These measurement points are the ones nearest the input point for which the quadratic is being computed. The second parameter specifies the number of measurement points that will be used in calculating the weights for the quadratics to produce the final function. The input data points used there are the ones nearest the point at which the interpolation is desired. Appropriate defaults are chosen for each of the control parameters

The trivariate quadratics that have been previously computed for the 3-D interpolation and whose input data are at the closest distance from the phantom surface, are used in order to extrapolate the fields to the surface of the phantom.

In order to determine all the field maxima in 2-D (Area Scan) and 3-D (Zoom Scan), the measurement grid is refined by a default factor of 10 and the interpolation function is used to evaluate all field values between corresponding measurement points. Subsequently, a linear search is applied to find all the candidate maxima. In a last step, non physical maxima are removed and only those maxima which are within 2 dB of the global maximum value are retained.

Important: To be processable by the interpolation/extrapolation scheme, the Area Scan requires at least 6 measurement points. The Zoom Scan requires at least 10 measurement points to allow the application of these algorithms.

In the Area Scan, the gradient of the interpolation function is evaluated to find all the extrema of the SAR distribution. The uncertainty on the locations of the extrema is less than 1/20 of the grid size. Only local maxima within 2 dB of the global maximum are searched and passed for the Zoom Scan measurement. 

In the Zoom Scan, the interpolation function is used to extrapolate the Peak SAR from the lowest measurement points to the inner phantom surface (the extrapolation distance). The uncertainty increases with the extrapolation distance. To keep the uncertainty within 1 % for the 1 g and 10 g cubes, the extrapolation distance should not be larger than 5mm.

Averaging and Determination of Spatial Peak SAR

The interpolated data is used to average the SAR over the 1g and 10g cubes by spatially discretizing the entire measured volume. The resolution of this spatial grid used to calculate the averaged SAR is 1 mm or about 42875 interpolated points. The resulting volumes are defined as cubical volumes containing the appropriate tissue parameters that are centered at the location. The location is defined as the center of the incremental volume (voxel).

The spatial-peak SAR must be evaluated in cubical volumes containing a mass that is within 5% of the required mass. The cubical volume centered at each location, as defined above, should be expanded in all directions until the desired value for the mass is reached, with no surface boundaries of the averaging volume extending beyond the outermost surface of the considered region. In addition, the cubical volume should not consist of more than 10% of air. If these conditions are not satisfied then the center of the averaging volume is moved to the next location. Otherwise, the exact size of the final sampling cube is found using an inverse polynomial approximation algorithm, leading to results with improved accuracy. If one boundary of the averaging volume reaches the boundary of the measured volume during its expansion, it will not be evaluated at all. Reference is kept of all locations used and those not used for averaging the SAR. All average SAR values are finally assigned to the centered location in each valid averaging volume.

All locations included in an averaging volume are marked to indicate that they have been used at least once. If a location has been marked as used, but has never been assigned to the center of a cube, the highest averaged SAR value of all other cubical volumes which have used this location for averaging, is assigned to this location. Only those locations that are not part of any valid averaging volume should be marked as unused. For the case of an unused location, a new averaging volume must be constructed which will have the unused location centered at one surface of the cube. The remaining five surfaces are expanded evenly in all directions until the required mass is enclosed, regardless of the amount of included air. Of the six possible cubes with one surface centered on the unused location, the smallest cube is used, which still contains the required mass.

If the final cube containing the highest averaged SAR touches the surface of the measured volume, an appropriate warning is issued within the Postprocessing engine.

16.2 Evaluation Errors

Cube shape

The mentioned procedures search for the maximum averaged 1g and 10g volumes of cubical shape according to the ANSI and ICNIRP standard. A density of 1000 kg/m3 is used to represent the head tissue density and not the tissue simulating liquid density.

For the extrapolation the distance must be specified in the Area Scan and Zoom Scan Jobs (see Section 7.2 Jobs). The distance is defined as the distance between the probe sensor center and the phantom surface! The recommended distance is 2 to 5 mm depending on the probe type used.

Boundary effects

The dosimetric probes are calibrated in a gradient field with energy flow and decay in direction of the probe axis. During calibration the probe tip is completely surrounded by the simulating solution. If the probe is used in the immediate vicinity of a media boundary, the field in the probe is altered due to interaction with the field in the boundary and the probe sensitivity changes. The influence of the boundary effect depends on the probe construction, the media parameters and the probe orientation with respect to the boundary. It disappears at a distance of 1 mm (E1D-probe) to 5 mm (ET3D-probes) between the probe tip and the boundary. The boundary effect must be considered in the extrapolation to the surface.

16.3 References

[1] Robert J. Renka, ”Multivariate Interpolation Of Large Sets Of Scattered Data”, University of North Texas ACM Transactions on Mathematical Software, vol. 14, no. 2, June 1988 , pp. 139-148.

空间峰值SAR评估

16.1  1g 和 10g 的空间峰值 SAR

DASY5 软件包括评估空间峰值 SAR 值所需的所有数值程序。

基于IEEE 1528.1标准草案，实施了一种新的算法。可以在任何所需质量上计算空间峰值 SAR。

评估的基础是体积为 30立方毫米（7x7x7 点）的“立方体”测量值。测量的体积必须包括平均 SAR 值最高的 1 g 和 10 g 立方体。为此，测量体积的中心与先前执行的区域扫描的插值峰值 SAR 值对齐。如果 10g 立方体或两个立方体不完全在测量体积内，系统会发出有关后处理引擎 （SEMCAD X） 中评估的空间峰值的警告。这意味着，如果测量的体积发生变化，则可能会有更高的值。要获得正确的值，您可以使用更精细的测量网格进行区域扫描。在复杂的场分布中，区域扫描的大网格间距可能会遗漏一些细节，并给出不正确的插值峰位置。

空间峰值的整个评估是在后处理引擎 （SEMCAD X） 中执行的。系统始终给出 1 g 和 10 g 立方体的最大值。查找平均SAR最高的立方体的算法分为以下几个阶段：

1. 从变焦扫描中提取测量数据（网格和值）

2. 根据所有存储的数据（A/D值和测量参数）计算每个测量点的SAR值

3. 在测量体积内生成高分辨率网格

4. 将所有测量值从测量网格插值到高分辨率网格

5. 在从传感器到表面的距离内将整个 3D 场分布外推到Phantom模型表面

6. 计算质量为 1 g 和 10 g 的平均 SAR

以下各节将更详细地概述重要部分。

极大值的插值、外推和检测

探头在偶极子传感器的中心进行校准，该传感器位于 1 至 2.7 mm 处探头尖端。在测量过程中，根据探头和表面检测系统的不同，探头会很快停止在幻影表面上方。两个距离都作为参数包含在探测配置文件。该软件总是准确地知道测量点离Phantom表面有多远。由于探针不能直接在Phantom表面测量，因此必须外推最深测点与Phantom表面之间的值。

在DASY5中，选择定义测量的点坐标系位置对插值、Maxima Search和外推例程的不确定性没有影响。插值、外推和最大搜索例程都基于修改后的二次谢泼德算法[1]。
因此，插值方案结合了最小二乘拟合函数方法和加权平均法，这是计算插值和近似的两种基本类型。
DASY5 例程构造一个一次性连续可微函数，该函数插值
测量值如下：
• 对于每个测量点，计算三变量 （3-D） / 二变量 （2-D） 二次函数。
它在数据点对测量值进行插值，并与相邻测量值形成最小二乘拟合。• 二次方相对于测量值的空间位置通过反距离加权衰减。之所以这样做，是因为计算出的二次方将比距离较远的点更准确地拟合附近点的测量值。
• 计算所有测量点的二次函数后，插值函数计算为二次函数的加权平均值。

有两个控制参数控制插值法的行为。一种是指定用于计算局部二次曲面的最小二乘拟合度的测量点的数量。这些测量点是距离要计算二次曲线的输入点最近的测量点。第二个参数指定测量点的数量。它将用于计算二次曲面的权重，以产生最终函数。在那里使用的输入数据点是距离所需插值点最近的数据点。为每个控制参数选择适当的缺省值。

使用先前已经为3-D内插计算的并且其输入数据与模体表面最近距离的三变量二次曲面，以便将场外推到模体表面。

为了确定二维(区域扫描)和三维(缩放扫描)中的所有场极大值，测量网格被默认因子10细化，并且使用内插函数来评估相应测量点之间的所有场值。随后，应用线性搜索来寻找所有候选极大值。在最后一步中，去除非物理极大值，并且仅保留那些在全局最大值的2分贝以内的极大值。

重要提示：要通过内插/外推方案进行处理，区域扫描需要至少6个测量点。缩放扫描需要至少10个测量点才能应用这些算法。

在区域扫描中，计算内插函数的梯度，以求出SAR分布的所有极值。极值位置的不确定性不到网格大小的1/20。只有全局最大值2分贝内的局部最大值被搜索并通过，以用于变焦扫描测量。

在变焦扫描中，插值函数用于将峰值SAR从最低测量点外推至内部体模表面（外推距离）。 不确定性随着外推距离的增加而增加。为了将1克和10克立方体的不确定性保持在1 %以内，外推距离不应大于5毫米。

空间峰值SAR的平均和确定

通过对整个测量体积进行空间离散化，插值数据用于计算1g和10g立方体的SAR平均值。 用于计算平均SAR的该空间网格的分辨率为1 mm或大约42875个插值点。 得到的体积被定义为包含以该位置为中心的适当组织参数的立方体积。 该位置被定义为增量体积（体素）的中心。

空间峰值SAR必须在体积体积中进行评估，体积质量在所需质量的5%以内。 如上所述，以每个位置为中心的体积应该向所有方向扩展，直到达到质量的期望值，并且平均体积的表面边界不会延伸到所考虑区域的最外表面之外。 此外，体积中的空气含量不应超过10%。 如果不满足这些条件，则平均体积的中心移动到下一个位置。 否则，使用逆多项式近似算法找到最终采样立方体的确切大小，从而提高结果的准确性。 如果平均体积的一个边界在其扩展期间达到测量体积的边界，则根本不会对其进行评估。 保留所有用于平均SAR的位置和未用于平均SAR的位置的参考。 所有平均SAR值最终被分配到每个有效平均体积的中心位置。

平均量中包含的所有位置都被标记，以表明它们至少被使用过一次。 如果某个位置已标记为已使用，但从未分配给立方体的中心，则将使用该位置进行平均的所有其他立方体体积的最高平均SAR值分配给该位置。 只有那些不属于任何有效平均体积的位置才应标记为未使用。 对于未使用位置的情况，必须构建一个新的平均体积，该平均体积将使未使用位置以立方体的一个表面为中心。 其余五个表面在所有方向上均匀膨胀，直到所需的质量被封闭，而不管包含的空气量如何。 在六个可能的立方体中，有一个面以未使用的位置为中心，使用最小的立方体，它仍然包含所需的质量。

如果包含最高平均SAR的最终立方体接触到测量体积的表面，则在后处理引擎中发出适当的警告。

16.2评估错误

立方体形状

上述程序根据ANSI和ICNIRP标准搜索立方体形状的最大平均1g和10g体积。 1000 kg/m3的密度用于表示头部组织密度，而不是模拟液体密度的组织。 施密德与合作伙伴工程公司，DASY5手册第52版，2011年11月16-3日 参考书目应用笔记

推断

对于外推，必须在区域扫描和缩放扫描作业中指定距离（参见第7.2节作业）。 该距离定义为探头传感器中心和模型表面之间的距离！根据所使用的探头类型，建议距离为2至5 mm。

边界效应

剂量测量探头在梯度场中校准，能量沿探头轴方向流动和衰减。 校准期间，探针尖端完全被模拟溶液包围。 如果在介质边界附近使用探头，探头中的场会因与边界中的场相互作用而改变，探头灵敏度也会改变。 边界效应的影响取决于探头结构、介质参数和探头相对于边界的方向。 它在探针尖端和边界之间1mm（E1D探针）到5mm（ET3D探针）的距离处消失。外推地表时必须考虑边界效应。

16.3参考

【1】Robert j . Renka，“大型分散数据集的多元插值”，北德克萨斯大学ACM数学软件学报，第14卷第2期，1988年6月，第139-148页。