NA-MIC National Alliance for Medical Image Computing An Integrated System for Image-Guided Radiofrequency Ablation (RFA) of Liver Tumors Kevin Cleary, Ziv Yaniv, Georgetown University Noby Hata, Brigham and Women’s Hospital Enrique Campos-Nanez, George Washington University

National Alliance for Medical Image Computing Liver tumor RFA Liver cancer that cannot be resected due to extent and location of the disease or concurrent medical conditions. Introduce localized RF energy directly to tumor, typically through expanding metal tines within a small gauge insulated needle. LeVeen probe Before (left) and after (right) treatment (courtesy of Brad Wood, MD, NIH CC)

National Alliance for Medical Image Computing Overlapping Burns Large tumors can be treated with overlapping spherical zones Hard to visualize the overlapping areas Lack of real-time image guidance requires repeated insertions to hit the target lesion and establish sufficient margins From Dodd, Soulen et al. 2000

National Alliance for Medical Image Computing Project Overview Goal: Develop an open source workstation for liver RFA planning and treatment based on IGSTK and Slicer. Specific aims: 1.Develop and evaluate semi-automatic segmentation techniques for the liver, liver vasculature, and liver tumors. [Georgetown, BWH] 2.Develop a path planning module for evaluating alternative paths to the liver tumor and incorporating multiple overlapping placements as needed for larger tumors. [George Washington University] 3.Integrate the two capabilities developed above along with electromagnetic tracking of the RFA probe to provide a complete software environment for liver tumor planning, visualization, and execution. [Georgetown, BWH] 4.Validate the clinical feasibility of the system in a swine animal model. [Georgetown]

National Alliance for Medical Image Computing YearAim 1 Software architecture Aim 2 Segmentation Aim 3 Treatment planning Aim 4 Integration / phantom study Aim 5 Animal studies 1Requirements document complete, Prototype architecture complete Ribs, liver, tumor, vasculature - manual segmentation First stage algorithms complete 2Complete gold standard – database segmentation Second stage algorithms complete – port to open source 3Semi-automatic segmentation Evaluate on 10 data sets Integrated system complete Later years YearAim 1 Software architecture Aim 2 Segmentation Aim 3 Treatment planning Aim 4 Integration / phantom study Aim 5 Animal studies 1Requirements document complete Ribs, liver, tumor algorithms complete First stage algorithms complete 2Prototype architecture complete Vasculature completeSecond stage algorithms complete 3Evaluate on 10 data sets Evaluate on 10 data sets Integrated system complete Later years Proposed: Actual: Timeline

National Alliance for Medical Image Computing Segmentation Gold Standard Database Collected approximately 50 CT liver images from Georgetown University Hospital Followed Health Insurance Portability and Accountability Act (HIPAA) rules for anonymization of data Medical student segmented 32 livers using ITK-SNAP under supervision of attending interventional radiologist Next step is to segment tumors

National Alliance for Medical Image Computing Path Planning Pre-emptive goal programming approach based on integer programming techniques using a discrete set of data points representing the tumor (uniform sampling): 1.Minimize needle insertions (trajectories), minimizes the number of punctures to the liver capsule, and the number of needle insertions. 2.Minimizing Ablations Given needle trajectories. 3.Minimizing Damage to Healthy Tissue

National Alliance for Medical Image Computing Path Planning Status Currently implemented using commercial software Xpress-MP (Fair Issac dash optimization) Worked well in preliminary swine feasibility studies Now porting to open source Gnu Linear Programming (GLPK) package Initial results from porting are extremely slow compared to commercial package Requires further evaluation

National Alliance for Medical Image Computing Integration Initial implementation started at IGT Project Week in December 2008 in Boston Tracking data is broadcast from IGSTK components using OpenIGTLink protocol Program resides in IGSTK sandbox and supports FLTK GUI and command line mode

National Alliance for Medical Image Computing Integration (continued) Specify configuration using xml files Supports multiple tools and broadcasting data to multiple computers Supports the use of a Dynamic Reference Frame. If specified all broadcasted transformations will be relative to the DRF (DRF transformation is not sent) Will present at SPIE Medical Imaging 2009 in Orlando in February N 1 0 large reference …

National Alliance for Medical Image Computing Next steps Goal is to create a Slicer3 workflow based interface for navigated RFA (Nav-RFA) Requirements document developed by Georgetown and sent to BWH Slicer will be configured with the following modules –Data: manage scene graph –EM Segmenter : potentially used for automatic segmentation (maybe not in the initial setup) –Editor: manual segmentation and modification of automated segmentation. –Fiducials : marking the fiducials. –Linear registration: fiducial based registration. –OpenIGTLink: tracking This will create a skeleton end to end workflow where we can fill in individual pieces and improve as we go

National Alliance for Medical Image Computing Summary Goal is to develop an integrated system for RFA of liver tumors Progress is good but vigilance needs to continue Open source implementation of optimization is one concern