Computational challenges in intensity Computational challenges in - PowerPoint PPT Presentation

Computational challenges in intensity Computational challenges in intensity modulated radiation therapy treatment modulated radiation therapy treatment planning planning Joe Deasy, PhD, Division of Joe Deasy, PhD, Division of Bioinformatics and Outcomes Bioinformatics and Outcomes Research Research

Collaborators Collaborators • Jeff Bradley, M.D. • Issam El Naqa, PhD • Jeff Bradley, M.D. • Issam El Naqa, PhD • Wade Thorstad, M.D. • Patricia Lindsay, PhD • Wade Thorstad, M.D. • Patricia Lindsay, PhD • Cliff Chao, M.D. • Jan Wilkens, PhD • Cliff Chao, M.D. • Jan Wilkens, PhD • Angel Blanco, M.D. • James Alaly, B.S. • Angel Blanco, M.D. • James Alaly, B.S. • Andrew Hope, M.D. • Andrew Hope, M.D. • Eva Lee, PhD • Eva Lee, PhD • Jing Cui, PhD. • And many others… • Jing Cui, PhD. • And many others… Supported by grants from the NIH (R01s, R29, F32), NIH funding, and commercial funding by: CMS, ADAC, Varian, Sun Nuclear, and Tomotherapy, Inc.

Preface: the current paradigm Preface: the current paradigm

A Pencil beam or beamlet Source Fluence of i ’th Beamlet, denoted b i Port or ‘beam’ of 8 beamlets

Optimization of beamlet fluence weights results in a ‘fluence map’ for each treatment head position Fluence map example (a map of the b i ’s) (From: Chui et al. , Medical Physics (2001) 28 :2441-2449.)

An IMRT dose distribution is constructed from a superposition of open static fields of variable fluence Beam’s Eye View of target volume First delivered field “segment” Second segment. (From: Kung and Chen, Medical Physics (2000) 27 :1617-1622.)

Basic IMRTP approaches Basic IMRTP approaches ����������������� �������� ���������������� ���������������� ����������� ������������������ ��������� (most common by far)

The ‘objective function’ The ‘objective function’ • Typically, the objective function is a sum of • Typically, the objective function is a sum of terms, some of which represent normal terms, some of which represent normal tissue structures and one or more terms tissue structures and one or more terms represents the target. represents the target. – This is called a ‘linear sum objective function’ – This is called a ‘linear sum objective function’ – The different terms have different multiplying – The different terms have different multiplying weights (constants) in front, representing weights (constants) in front, representing relative importance relative importance

Linearly weighted objective functions Linearly weighted objective functions • Individual terms (or goal • Individual terms (or goal functions) are added to functions) are added to comprise the objective comprise the objective function. function. • Typically, each anatomy • Typically, each anatomy structure of importance has structure of importance has one or more goal terms. one or more goal terms. • Goals are evaluated for each • Goals are evaluated for each voxel contained in a structure. voxel contained in a structure. Graph of cost per voxel vs. dose n m � � 2 2 F = w ( D − 0 ) + w ( D − 64 ) OAR i Target j i = 1 j = 1 Objective for an Objective for a OAR of n voxels target of m voxels

Iterative solution • Start with a set of initial Start beamlet weights. • Search along a series of Calculate Cost directions in beamlet weight space. Yes • Stop when Convergence Finish Criterion Met? – cost is zero – cost not improved No – fixed number of iterations Select Search exceeded Direction • When done, beamlet weights are ‘optimized.’ Do Line Search

A ‘state of the art’ IMRT treatment A ‘state of the art’ IMRT treatment planning system... planning system... • Accepts constraints • Accepts constraints – Max dose – Max dose – Min dose – Min dose – Dose-volume constraints: no more than x% of an organ – Dose-volume constraints: no more than x% of an organ can receive y% dose (e.g., “V20 can be no larger can receive y% dose (e.g., “V20 can be no larger than...”). than...”). • Tries to match or exceed goal DVH parameters • Tries to match or exceed goal DVH parameters – for target volumes – for target volumes – for normal tissues – for normal tissues

The CMS XiO Prescription Page The CMS XiO Prescription Page

The weight paradox: hard-to-control tradeoffs The weight paradox: hard-to-control tradeoffs and the lack of clear priorities and the lack of clear priorities • Normal tissue weights should be large enough so • Normal tissue weights should be large enough so the mathematical engine tries to reduce dose to the mathematical engine tries to reduce dose to those structures those structures • Target weights should be much larger than normal • Target weights should be much larger than normal tissue weights so that good target coverage is not tissue weights so that good target coverage is not compromised...but... compromised...but... • There is no perfect compromise • There is no perfect compromise – Very high target weights: engine neglects normal – Very high target weights: engine neglects normal tissues tissues – Not very high target weights: engine does not preserve – Not very high target weights: engine does not preserve target dose characteristics target dose characteristics

State-of-the-art workflow: “Are we State-of-the-art workflow: “Are we finished yet?” finished yet?” Physician: “ Here is what I’d like .” Physician: “ Here is what I’d like .” Later....Dosimetrist: “ I tried it, and tried to Later....Dosimetrist: “ I tried it, and tried to fix it. Here it is .” fix it. Here it is .” Physician thinks “ Is that the best they can Physician thinks “ Is that the best they can do? ” Says: “ How busy are you? Can you do? ” Says: “ How busy are you? Can you try to improve this part? ” try to improve this part? ” Dosimetrist: “ Pretty busy. But I’ll try if you Dosimetrist: “ Pretty busy. But I’ll try if you want me to .” want me to .”

Thus, current IMRT systems are highly Thus, current IMRT systems are highly inefficient, and lead to planning inefficient, and lead to planning iterations with no clear guidelines for iterations with no clear guidelines for establishing that a ‘clinically superior’ establishing that a ‘clinically superior’ plan cannot be achieved. plan cannot be achieved.

IMRT planning challenges IMRT planning challenges 1. Lack of scientific comparisons 1. Lack of scientific comparisons 2. Incorporating accurate dose calculations 2. Incorporating accurate dose calculations 3. Mastering the ‘data-glut’ 3. Mastering the ‘data-glut’ 4. Controlling dose distribution 4. Controlling dose distribution characteristics & tradeoffs characteristics & tradeoffs 5. Making tradeoffs responsive to outcomes 5. Making tradeoffs responsive to outcomes models models

Challenge #1: Lack of scientific Challenge #1: Lack of scientific comparisons comparisons

IMRT optimization and operations IMRT optimization and operations research: facilitating operations research: facilitating operations research approaches in IMRT research approaches in IMRT J Deasy* 1 , E Lee 2 , M Langer 3 , T Bortfeld 4 , Y Zhang 5 , H Liu 6 , J Deasy* 1 , E Lee 2 , M Langer 3 , T Bortfeld 4 , Y Zhang 5 , H Liu 6 , R Mohan 6 , R Ahuja 7 , J Dempsey 7 , A Pollack 8 , J Rosenman 9 , A R Mohan 6 , R Ahuja 7 , J Dempsey 7 , A Pollack 8 , J Rosenman 9 , A Eisbruch 10 , R Rardin 11 , J Purdy 1 , K Zakarian 1 , J Alaly 1 Eisbruch 10 , R Rardin 11 , J Purdy 1 , K Zakarian 1 , J Alaly 1 (1) Washington Univ, Saint Louis, MO, (2) Georgia Inst Tech and Emory (1) Washington Univ, Saint Louis, MO, (2) Georgia Inst Tech and Emory Univ, Atlanta, GA, (3) Indiana Univ, Indianapolis, IN, (4) Massachusetts Univ, Atlanta, GA, (3) Indiana Univ, Indianapolis, IN, (4) Massachusetts General Hospital, Boston, MA, (5) Rice University, Houston, TX, (6) UT General Hospital, Boston, MA, (5) Rice University, Houston, TX, (6) UT M.D. Anderson Cancer Center, Houston, TX, (7) University of Florida, M.D. Anderson Cancer Center, Houston, TX, (7) University of Florida, Gainesville, FL, (8) Fox Chase Cancer Center, Philadelphia, PA, (9) Univ Gainesville, FL, (8) Fox Chase Cancer Center, Philadelphia, PA, (9) Univ of North Carolina, Chapel Hill, NC, (10) Univ Michigan, Ann Arbor, MI, of North Carolina, Chapel Hill, NC, (10) Univ Michigan, Ann Arbor, MI, (11) Purdue Univ, W. Lafayette, IN, (11) Purdue Univ, W. Lafayette, IN, (Deasy et al., Annals Op Res, In press)

Motivation I Motivation I • Many IMRT treatment planning algorithms, • Many IMRT treatment planning algorithms, but… but… • Few comparisons • Few comparisons • Tools for comparison and common data • Tools for comparison and common data access are missing access are missing • Common datasets are missing • Common datasets are missing • Few (no?) comparisons of techniques. • Few (no?) comparisons of techniques.