A Dose-Volume Histogram Analysis Software in Radiation Therapy Research S. Jang 1, A. Pyakuryal *,2,3, W. K. Myint 4, M.

Gopalakrishnan2, B. Curran1, E. Sternick1, and B. Mittal2,4 (1) Rhode Island Hospital/ Brown University Medical School, Providence, Rhode Island.

(2) Northwestern Memorial Hospital, Chicago, Illinois. (3) University of Illinois at Chicago, Chicago, Illinois (4) Northwestern University, Chicago, Illinois. *Presenting Author 50th Annual Meeting of American Association of Physicists in Medicine (2008), Houston, TX

 Abstract •

Purpose: A dose-volume histogram (DVH) analysis software, HART (Histogram Analysis in Radiation Therapy), was developed to increase the accuracy and efficiency of processing large quantities of treatment planning system dose data. This software was designed specifically for an NIH funded head and neck IMRT study ( B.B.Mittal,J.A.Logemann) and it can be used at any other treatment planning sites.

•

Method and Materials: IMRT patient data to be analyzed were exported into RTOG format files from the TPS. RTOG files, with differential DVH information, were read and transformed into cumulative DVH data. Matlab ( The Mathworks Inc, Natick, MA) based codes were developed to identify all the target and normal structure volumes, and treatment planning parameters in RTOG formats. The DVH analysis was based on a cosine interpolation numerical analysis technique; and the uncertainty in data interpolation was controlled by using piecewise polynomial fittings in DVH curves. The accuracy of the software was validated by comparing DVH output with the data from the ADAC pinnacle treatment planning system (TPS) ( Philips Healthcare, version 7.6c ).

•

Results: For each patient, over 4000 data points from DVH curves were measured. HART processed this data in approx. 10 to 15 minutes. Reading this data directly from the TPS DVH curves required 4 to 6 hours and included significant human error. The normalized root mean square deviation (NRMSD) for all HART DVH curves was less than 1% or within 0.5% DTA (distance-to-agreement).

•

Conclusion: A DVH analysis software system has been developed that can be efficiently used for research requiring the handling of a large number of structures or patient data with high precision results. More user-friendly features of dose and volume selections, expansion to other TPSs, and statistical indices are under development.

 Objective 

To develop HART (Histogram Analysis in Radiation Therapy), a dose volume histogram (DVH) analysis software tool used to increase the accuracy and efficiency of extracting large quantities of head and neck patient data from IMRT treatment plans.

Head and neck general anatomy

 Introduction: 

An in-house built automated software in Matlab codes, HART, is introduced for precise DVH analysis in 3D conformal and IMRT based research.



HART was designed for an NIH-funded IMRT study for over 100 head & neck cancer patients. Each IMRT patient plan data involved 29 structures, 4 trials and over 40 DVH parameters for a total of over 4000 data points measured.

 Manual extraction of large quantities of TPS DVH data for

IMRT plans is prone to human error and is time consuming.

 Methods and Materials: Treatment 



 

Process

Patients were treated with an Elekta SLI series clinical accelerator with 80 MLCs (1cm width at isocenter, 7.5 cm depth) at a nominal beam energy of 6 MV. Adac Pinnacle3 TPS ( Philips Healthcare, version 7.6c). Mathworks Inc., Natick, MA ( Vsn: 7.5.0.342 (R2007b) ). Computer speed: 1.8 GHz and 1024 MB RAM.

 Head and neck IMRT treatment plans :



Prescribed doses (PD) for Planning target volumes (PTVs): 3900 cGy ( PTV1 ); 1200-2100 cGy ( PTV3 );

1200-1500 cGy ( PTV2) ; 6300-7500 cGy (COMPOSITE);

 Dose per fraction : 150 cGy.  Critical organs ( protocol): Larynx, Optics, Optic nerves, Chiasm, Parotid, Mandible,Hyoid, BOT, Cochlea, Spinal Cord, Carotid, Pitutary, Brainstem, Brain, Constrictor, Lips, Oral Cavity, Oropharynx, Esophagus.

Methods and Materials:Differential & cumulative DVH  In a 3D plan, DVH represents a frequency distribution of average dose values over a 3D matrix of voxels constituting a defined volume of PTV or critical structure in a patient anatomy.  HART computes the fractional volumes of target receiving an assigned dose using the differential DVH matrices which is converted into cumulative DVHs. Cumulative plot shows that all of the volume receives at least no dose (0 cGy).

Differential DVHs

 The ideal target differential DVHs would be infinitely narrow peaks at target dose (PD) for PTV and 0 cGy for critical structures. But several peaks in a real DVHs indicate that different parts of an organ receive different doses.

Cumulative DVHs

 Methods and Materials: Numerical Analysis Method

 Differential DVH dose matrices from RTOG files exported from TPS, were read and transformed into cumulative DVH by software.  A cosine interpolation technique with weight factor (h)=0.5,was implemented as a smoother transition function amongst adjacent dose bins.

 Piecewise 2nddegree polynomial fitting was employed to estimate the volume of interest for arbitrary dose provided in cumulative curve.

 Methods and Materials: Statistical Definitions o Min dose: Smallest dose of all non-zero voxel elements in target .

o Max dose: Largest dose of all non-zero voxel elements in target .

o Prescribed dose: Dose prescribed to isodose line covering 95 % of PTV.

o Median dose: Dose separating 50 % of dose distribution spectrum.

o Mean dose: [Σi (V(i)*D(i)) ]./N for N voxels.

o V(X) : Volume coverage above X cGy of prescription dose. Slice overview

 Results

and

Discussion: HART Flow Chart ADAC Pinnacle 3 TPS Clock = 0 min Differential DVH (4 Plans) Manual Extraction

Matlab codes

Clock = 4-6 hrs

HART SOFTWARE DVH Analysis Statistical Analysis Prostate

Head & Neck

Patient Folder → Mode → Prescribed Dose → User Defined Quantity → Output Location DVH Output in Spread Sheet Clock = 10-15 mins

 Results and Discussion : HART Features  Extraction of IMRT plan RTOG data files for all types of plans.  Portable software in Unix and Windows platforms.  User friendly set up.  Trial specific operations.

IMRT plan target volumes ( PTVs)

 Structure specific analysis.  Polynomial DVH modelings.  Flexible output (.xls) formats.

Sample structures contouring

 Results and Discussion : HART Features

 Results and Discussion : HART Features

 Results and Discussion : Error Analysis  Normalised Root Mean Square Difference (NRMSD):

NRMSD = [ Σ

N

2 – V ( V ) N ] Vtarget; i=1 i,exp i,model

NRMSD tested for absolute and normalized volumes at 6 different points relative to prescription in targets and 4 different points relative to prescription in critical structures for 5 head and neck patients.



Distance to Agreement Analysis (DTA): DTA = ((Dn, expt – Dn,model)/PD).100 % ; DTA test was performed in high dose gradient region specially when NRMSD > 1% .

 Accuracy of DVH analysis was compared with TPS produced cumulative DVHs at 4 mm voxel resolution.

 Results and Discussion: Error Analysis 90

80Normalised Root Mean Square 70Difference in Critical

DTA Tests for Planning Target Volumes

Structures

60 RT PAROTID

50 0.7 40 0.5

BRAINSTEM

NRMSD 30 0.3 Error 20 0.1 10 -0.1 0

SPINAL CORD

PTV1 (DTA)

0.25

PTV2 (DTA)

DTA 0.2 PTV3 (DTA)

RT. MANDIBLE

Critical Structures 1st Qtr

2nd Qtr

PHARYNGEAL 3rd CONSTRICTOR 4th

Qtr

Qtr

0.15 Planning target volumes

NRMSD and DTA test performed on 5 head and neck patients

 Results and Discussion : HART outcome Normal structure

 5 to 10 human errors introduced during manual data extraction were not present using the HART software.

 The data precision fell with in an agreement of 1% NRMS difference with TPS or 0.5% of DTA agreement normalized with the prescription dose.

cc

0.04

0.02

00

500

1000

1500

2000

2500

Dose (cGy) 6

Expt ABS. VOL (cc)

 Execution time for HART- DVH analysis for each patient was typically 10-15 minutes for over 4000 data points per with a cpu clock speed of 1.8 GHz and 1GB RAM as compared to the manual extraction time of 4-6 hrs per patient.

0.06

Poly fit 4

2

0 0

500

1000 1500 Dose (cGy)

2000

2500

Differential (top) and cumulative (bottom) DVHs.

 Conclusion : 

An economical and efficient approach in design and automation of HART in Matlab software has been presented.

 HART is an user friendly and portable tool designed for DVH analysis, simulation and modeling for target and critical structures in 3D Conformal and IMRT Plans.  We demonstrated the accuracy and efficiency of a new automated multimode operational DVH analysis software in 3D conformal and IMRT based research.  Feasible future tasks: • Statistical analysis automation. • Treatment plan indices evaluation. • Surface dose histogram analysis.

 References 1) M.Miften, S.Das et.al,”A dose volume based tool for evaluating and ranking IMRT treatment plans,”Journal of Applied Clin.Med.Phys. Vol.5 (4), pp.1-14,(2004). 2) R.E.Drzymala, R.Mohan et.al,“Dose Volume Histograms,”Int.J.Rad.Onc.Biol.Phys. Vol.21, pp.71-78 (1991). 3) J.Deasy et.al,“CERR: A computational environment for radiotherapy research,” Med.Phys.30(5),pp.979-985 (2003). 4) J.Deasy, A.I.Blanco and V.H.Clark et.al,“A collaboratory for radiation therapy treatment planning optimisation research,”Ann.Oper.Res.148,pp 55-63 (2006). 5) J.Bradley,J.O.Deasy et.al,“Dosimetric correlates for acute esophagitis in patients treated with radiotherapy for Lung carcinoma,” Int.J.Rad.Onc.Biol.Phys..58,pp.110613(2004).

 Acknowledgement This work is supported by Northwestern Memorial Foundation.