Navigation

• Main Page
• About

• Random Page
• Create a new Page
• All Pages
• Categories

• Administration
• File Management

• Login/Logout
• Language Selection
• Your Profile

Quick Search

Advanced Search »

Pheonix is an experimental inverse planning application for radiotherapy planning. The underlying algorithm is based on a novel optimization technique, and it possesses several interesting characteristics of a steerable inverse planning application. These include:

• Intuitive definition of clinical goals, using a set of target DVH curves.
• Predictable response: adjustments to target DVHs produce proportional changes to plan dose and DVHs.
• Efficient optimization: on standard PC hardware, the objective function can be evaluated with sufficient efficiency that users can interactively modify parameters.

PTV+Cord prescription

Edit

2D Prototype Results

The results of a 2D version of the algorithm are displayed above and below. This plan was based on a 7-field beam arrangement, with 15 MV beamlets computed at 4mm spacing. The corresponding DVH curves are displayed below.

The resulting plan was obtained using the following goal DVH curves (goal DVHs are also shown on the graph as dotted lines):

Prescription and OAR limits

Target / Organ-at-Risk	100% Dose (Gy)	0% Dose (Gy)
PTV	60	70
Cord	22	45
Right Parotid	8	35

PTV+Cord+RtParotid prescription

The three optimizations shown here are the result of increasingly complex stages of planning: first including only the target PTV, then including the target and cord OAR, and finally including the target, cord, and right parotid.

The multi-resolution optimization started at a pyramid level G3, with voxel dimensions of 3.2 cm³. Each optimization ran from an initial starting point with all beamlets weights set to zero. The number of iterations needed to converge at each level is given in the following table:

Iterations per Level

Prescription	Level G3	Level G2	Level G1	Level G0
PTV, Cord, Rt. Parotid	2	11	7	4
PTV, Cord	2	10	5	4
PTV only	2	8	5	4
Voxel Size (cm3)	3.2	1.6	0.8	0.4

As can be seen from the table, increasing the complexity of the plan did not drastically increase the number of iterations needed to converge. It is expected that this same pattern will be followed when the algorithm is applied to a 3-dimensional clinical case as well.

















Edit

Theory of Operation

The Pheonix algorithm iteratively improves the plan based on a comparison between the actual and goal DVH curves for all terms in the prescription. The comparison metric is the Kullback-Liebler Divergence (also known as relative entropy), which compares two probability distributions and determines how similar they are.

In this case, the probability distributions are the DVH curves in differential form. The curves are normalized so that the integral is equal to 1.0. Then the K-L divergence is calculated between the goal distribution p and the actual distribution q for the current iteration using the following formula:


As can be seen from the formula, two identical probability distributions will have a K-L divergence equal to zero. As the two distributions become more dissimilar, the K-L divergence becomes more positive. So the optimization algorithm attempts to minimize the K-L divergence for all terms in the prescription.

For more information see Pheonix Theory of Operation.

Edit

Beamlet Calculation

The individual beamlets were calculated for a 7-field beam arrangement. Each beam was decomposed in to 31 beamlets, each 4mm wide.

Dose was calculated on a 4mm x 4mm grid, using an energy deposition kernel for 15 MV photon fluence in a 60cm diameter ICRU521 water phantom. The kernel was calculated using the EDKnrc user code for the EGSnrc Monte Carlo particle transport simulation system.

Dose was calculated using a superposition algorithm, with inhomogeneity correction performed utilizing a piecewise linear CT-to-ED lookup table.

Edit

NLM's Insight Toolkit (ITK)

Brimstone uses the Insight Toolkit for its main image processing pipeline. ITK is produced by the National Library of Medicine in support of the Visible Human project.



Please check out the ITK website for more details about this powerful platform.