Using Overlapped Volume Histogram (OVH) as a parameter for representing distance distribution was proposed as a new data-driven strategy for optimal Intensity-Modulated Radiation Therapy (IMRT) planning in JHH Radiation Oncology Department. Coding modules based on Matlab and SQL has been developed, there has been existing thoracic therapy planning based on python code. There is a practical need for modifying python code for head and neck radiation therapy, and integrating existing matlab code and SQL queries to promote optimized searching strategy for IMRT planning.

• Student: Yang Wuyang MD Division of Health Science Informatics
• Mentors: Dr. Todd McNutt; Dr. Harold Lehmann; Dr. Russ Taylor

Background and Significance Intensity-Modulated Radiation Therapy(IMRT) is a radiation based therapy used mainly for oncological diseases, the therapy itself use a set of beams with the feature that the radiation intensity of each beam can be modified to achieve different distribution of strength within each beam: According to the Dose- Volume Histogram (DVH) chosen by the planners, the intensity distribution of each beam is different to satisfy the radiation need for the Planning Target Volume (PTV), and at the same time, spare the Organ At Risk (OAR).

Former approaches to plan an IMRT takes time and needs individual planner experiences, there are, however, multiple methods based on statistical analysis to improve the planning process. However, such analytical results are based on a large number of sample and only focus on “average planning”, which may not apply to individual cases, and the uncertainty of success is still high; in addition, such attempts take no respect to dynamic accumulation of experiences, usually most analyses take a certain time period as a point, get the results, and the cycle ends; for up-to-date results, it would usually take years as large trials or data analysis takes time and is costly. In this scenario, we propose a new method on a data cycling basis that the data of every new patient has a tendency to affect the treatment plan of the next patient:

The goal of optimal planning using statistical modeling methods are extremely difficult because of the shape variation of organs in different human body, and also because of the random movement of the target. Currently, most planners have to manually do the planning of each beam based on experience and simulations, which usually takes a long time and may not come out with an optimized plan. A new method of planning based on a parameter of Overlap Volume Histogram(OVH) and data driven approach is therefore introduced for safety and efficiency issues. OVH is a parameter introduced to represent the practical distance distribution from the PTV and the OAR, it is a volume calculation of the OAR based on the measurement of distance extended from the PTV. On the same distance extended, the larger the OVH, the nearer the OAR is to the PTV. We can therefore create a diagram of OVH with the distance of extension of PTV as the X axis and Overlapped volume at the Y axis. For each volume point of OVH, we can then have different volume of OAR and map the respective volume to dose in a DVH. So now we can get the data as for every volume point of OAR, the respective dose it would get in the plan:

A logical assumption underlying is that the larger the distance between the OAR and PTV, the easier the OAR can be spared. In order to find the optimal plan, we first create an OVH and DVH diagram of the current plan of current patient, see if there are former patients that have a larger OVH at the same distance, find all of those patients, map all diagrams to DVH, and search for the minimum dose that can be achieved. If there are patients that have a larger OVH at the same distance, but have a lower dose at the same volume, then at least such a dose level at the specific volume can be achieved because the former patients are harder to plan than the current, but receives a smaller dose at the same volume. Each specific intensity of target PTV is searched, and each beam is modified, the search can then generate a optimized plan for the patient, and the new plan of the current patient is then stored into the database as a new piece of information. In this way, we can make use of the data of every patient dynamically, make the planning process faster, and achieve a more and more optimized (safety and quality) plan as the database grows. Searching codes based on python have been created for different parts of the body. We need a planning package for head and neck therapies. Such code have already been created on matlab, but to consider the practical need of the radiation oncology department, we need to create a package based on python.

Our specific aims is:

Create a data driven IMRT planning package based on python and SQL.  

There are three major tasks in this project:

1. Modify existing python codes that are used on thoracic therapy for head and neck therapy.
2. Integrate existing matlab code into the modified python code.
3. Integrate SQL search strategies into the modified python code to promote faster search.

1. Milestone name: MileStone_1: Familiarity with Existing Python and Matlab Code
   • Planned Date: End of Febuary
   • Expected Date: First Week of March
   • Status: Done
2. Milestone name: MileStone_2: Python package that can pull data out of database(minimum deliverable)
   • Planned Date: First week of April, before checkpoint
   • Expected Date: First week of April
   • Status: In Progress(On schedule)
   • 3.14-3.18 progress: 1. Get connection access to database(Finished); 2. Phone call with BinBin to clarify uncleared matlab codes(Expecting to make more phone calls in the progress); 3. Appointment on Wednesday with the group to report progress(coding phase started); 4. Appointment with Kim to set up demonstration connections(Thursday); 5. Try to set up a regular working time at the Radiation Onco Dept. with Dr. McNutt to do coding and testing on site; 6. Decided on where to modify on python package
   • 3.21-3.27 progress: The python package that can pull data and calculate minimum achievable dose correctly is developed, a single optimization run needs 1 hour.
   • Proposed progress: 1. By 3.22 – Start on site working in a regular pattern; 2. By 3.31 – Get prototype of the package done; 3. By first week of April – Testing Done;
3. Milestone name: MileStone_3: Python package optimization(Data calculation optimization, SQL for faster minimum dosage data pulling) (expected deliverable)
   • Proposed progress: By the third week of April, the optimization run time will be reduced to an acceptable level.
   • Actual Progress: Implemented before 4/20
4. Milestone name: MileStone_4: Clinical Implementation(automation of Pinnacle 3 script generation) (maximum deliverable)
   • Proposed progress: By the first week of May, the package should be automatically generating pinnacle 3 script, which means that the package should be including the hotspot and coldspot killing function.
   • Actual progress 1: Script generating function half done(5/1/2011), continue to work out this function for the next week.
   • Actual progress 2: Hot spot and cold spot killing function aborted

• bulletized summary of important changes to plan, unresolved dependencies, etc.

8. Results 8.1. Target Contours Successfully Loaded: After we input the pinnacle scripts generated from the package, we loaded all the constraints to the system. The system first defines the contour of targets for the new patient, the contours were manually drawn before. Figure 12 shows the contours of PTVs for the patients:

The Green Contour represents PTV58.1; Brown Contour represents PTV63; Purple Contour represents PTV 70

As we can see from the pictures, each section shows the coverage of different PTVs. The purple part, which is the PTV 70, generally represents the highest dose level received (70Gy), so it is also another way of saying where the tumor was in this patient.

8.2. Beams Successfully Setup: After target contours were defined, we now setup the beams and the intensity modules of each direction of beams to meet the requirements of each constraint. The Pinnacle3 system automatically computes a suggested continuous dose with the consideration of each dose level the planner wants to constrain. The intensity of the beams has to be modulated to meet the requirements. Figure 13 and Figure 14 shows the simulation of the beams in which direction should the beam radiate and in how much amount of radiation should the beam release. We can view iso-dose lines in figure 15, as we can see that the high doses are trying to avoid the critical structures and in the same time ensures PTV coverage. For this patient particularly, because the tumor has invaded the right parotid, so the right parotid is very hard to spare, but the dose requirements of other critical structures are basically met.

Figure 3D simulation image on how the beams should radiate

Figure Iso-dose lines of the first round computation. The result is reasonable as most of the PTVs have a satisfactory coverage, and OARs are very much spared. The red line is the highest dose line, as one can see, it is trying very hard to cover the tumor which is at the right parotid gland, but also stressing to spare as much as possible on esophagus and other OARs. The red arrow indicates where the tumor was, and the yellow arrow points at a dent in 70Gy isodose line to spare some of the esophagus.

8.3. DVH Result

DVH Result

As shown in figure , the DVH result shows a continuous dose curve for the organs (in this project, there are 13 organs, brain, brainstem, cord, mandible, right parotid, left parotid, right inner ear, left inner ear, larynx, oral mucosa and esophagus, please see appendices for dose point configurations) involved in the project. The arrows indicate a constraining objective defined by the planner which was read into the python package. Almost all objectives are met, except for the right parotid, which was invaded by the tumor. Compromise of the PTVs are very low, coverage objectives are basically met.

8.4. Planning Time: The package running time is 5 minutes, and a single round of Pinnacle3 running time is about 20 minutes. As the results is reasonably good enough, we can anticipate the general planning time to be greatly lesser than 2 weeks, although the actual planning time/clinical efficiency needs the package to be widely used in clinics and evaluate the general time used for planning.

• Project Plan
  • Project Proposalproject_proposal.pdf
  • Project Presentationproject_proposal_presentation.pdf

• Project Background Reading
  • Media link(s) to powerpoint, pdf of background presentationspaper_presentation.pdf
  • Media link(s) to chosen articlered_journal.pdf
• Project Checkpoint
  • Checkpoint Presentation checkpoint_presentation.pdf
  • Mini checkpointcheckpoint_project_11.pdf
• Project Final Presentation
  • Poster Presentation (Poster Teaser) poster_teaser_11.pptx
  • Poster project_poster2.pptx
• Project Final Report
  • Final Report project_final_report.docx

NOTE: In uploading media, upload them to your own name space or to private sub-namespace, depending on whether the general public should be able to see them.

[1] Wu B, Ricchetti F, Sanguineti G, Kazhdan M, Simari P, Jacques R, Taylor R, McNutt T. Data-driven approach to generating achievable dose-volume histogram objectives in intensity-modulated radiotherapy planning.Int J Radiat Oncol Biol Phys. 2011 Mar 15;79(4):1241-7. Epub 2010 Aug 26. red_journal.pdf

[2]Wu B, Ricchetti F, Sanguineti G, Kazhdan M, Simari P, Chuang M, Taylor R, Jacques R, McNutt T. Patient geometry-driven information retrieval for IMRT treatment plan quality control. Med Phys. 2009 Dec;36(12):5497-505. wu_mph005497.pdf

[3] Simari P, Wu B, Jacques R, King A, McNutt T, Taylor R, Kazhdan M. A statistical approach for achievable dose querying in IMRT planning. Med Image Comput Comput Assist Interv. 2010;13(Pt 3):521-8. simari_miccai.pdf

Here give list of other project files (e.g., source code) associated with the project. If these are online give a link to an appropriate external repository or to uploaded media files under this name space.