DOS 523 - Week 4 Discussion
Initial Post: Dose Volume Histogram Visualization
When dose is calculated for a patient's plan, it consists of a 3 dimensional intensity map that is overlaid onto the planning image, much like the way a PET scan can be overlaid onto an image to show the intensity of tracer uptake at any point in space. The intensity value recorded at each voxel position represents the amount of dose that will be delivered to that location during treatment. This spatial representation is extremely useful for highlighting the general distribution of the dose and showing where warm and cold spots are, but it is limited in its ability to provide quantitative information. Most planning systems have the ability to display the dose value at any point in space that a user selects, and isodose lines can be drawn to "connect the dots" along the levels of equal dose value, showing a stair-stepped visualization of intensity values. This type of visualization does not make it easy to quickly determine the maximum or minimum dose delivered to a particular structure without scrolling through multiple times and hunting for hot and cold spots with the outline on every slice. This visualization also makes it virtually impossible to answer the question of what percentage of a structure is covered by a certain dose level such as the prescription dose.
To solve these problems, the dose intensity grid can be examined voxel by voxel and sorted. One way to think about this is to picture a process of chopping it up voxel by voxel and throwing the voxels into appropriate bins that are labeled with dose ranges. A voxel that has an intensity value of 2031 cGy could end up thrown into the "2000-2050 cGy" bin or "2000-2100 cGy" bin if the planning system happened to be using bin widths of 50 cGy or 100 cGy. The choice of bin size can be somewhat arbitrary, but the more bins there are, the smoother the data representation. Here is an example from biology of measurements of peak sizes of finches that shows the relationship between number of bins and smoothness of display.1 Note that is is possible to have too fine of a bin size, which starts to break up the smoothness of display.
Turning back to radiation oncology, as the 3D dose map is chopped up and binned, the spatial information of where the voxels came from is completely discarded, and instead, trends showing how often certain dose values appear start to emerge. If information about a particular piece of anatomy such as a target volume is desired, the 3D dose dataset can be rebinned while discarding voxels that are not included in that anatomic region, leaving just the voxels that lie inside that contour. After binning, the maximum and minimum values will be clear, because no bins below a certain value range will have any voxels assigned to them, and no bins above a certain value range will have any voxels assigned to them. Advanced statistics for an anatomic structure such as mean, median, standard deviation, and others can be extracted by analyzing the values of all of the binned voxels.
The most effective way to visually represent this binned data is in the form of a histogram, which plots the bin values along the X axis and the frequency of occurrence (number of voxels in that bin) on the Y axis. This most basic representation of binned dose data is called a differential DVH, and it shows how many voxels in the dataset received a particular range of doses represented by the bin range. The frequency on the Y axis can be displayed as a raw volume, measured in cc or ml, or it can be normalized to the total volume of that structure and displayed as a percentage. The dose values on the X axis can be displayed as raw dose values or normalized to percentages of a chosen value such as the prescription dose. Differential DVHs are useful for looking at homogeneity of dose distribution. Even coverage will result in a sharp peak, with most of the voxels getting binned into a few closely clustered bins.
Differential DVHs are not particularly useful for judging aggregate information such as "how much of the lung received 20 Gy"? To answer that kind of question, the visualization of the bins can be switched to a cumulative DVH. This type of display shows the volume or percent of volume that exists in the current bin or a larger bin, starting from the lowest value. As the bins are tallied from left to right, each voxel knocks the height of the curve down a bit more, because there are that many fewer voxels still to be counted in higher bins. The result of this approach is that the curve always starts in the upper left corner of the chart and slopes progressively downwards to zero as X increases. This display format makes it easy to answer the question of how much of a structure received a particular dose by simply looking for that value on the X axis, and looking up the Y value for that point in the curve. Values on the Y axis can also be checked, such as "how much dose went to 95% of the volume"?
DVHs can also be stacked on top of each other to compare plan variants. Here is an example of two plans being compared, with one rendered as solid lines for each structure, and one as dashed lines.2
Looking at the rectum curve on this DVH, we see that the solid line has a lower maximum dose than the dashed line, but the tradeoff is that more voxels are receiving midrange doses. This kind of analysis, combined with spatial information from a 3D display, can give dosimetrists a much more complete understanding of dose distribution with a patient's plan, and can help identify areas where coverage is appropriate, and where adjustments need to be made.
- Robbins KA. Lesson 10: histograms questions. University of Texas San Antonio Website. http://vip.cs.utsa.edu/classes/cs1173/cs1173s2011/lessons/Lesson10HistogramsQuestions/Lesson10HistogramsQuestions.html. Updated December 31, 2010. Accessed March 11, 2015.
- Zarepisheh M, Shakourifar M, Trigila G, et al. A moment-based approach for DVH-guided radiotherapy treatment plan optimization. Phys Med Biol. 2013;58(6):1869-87. http://dx.doi.org/10.1088/0031-9155/58/6/1869