Personal Notes of Robert E. Zimmerman from the SNM mid-winter meeting Palm Springs CA Sat - Tues, 8-11 Feb 1997 Palm Springs is a wonderful place escape for a winter meeting. Warm sunny days, cool evenings and an interesting desert environment to explore when the meeting lulls. SNM conducts a lot of business at this meeting but the highlight for the nerds is the Computer & Instrumentation council's technical meeting. More on that a bit later. Saturday Academic Council The academic council is in turmoil. I, as the lone physicist member, am a quiet observer, picking up a free lunch. As a person long involved in resident training it hurts to see the dilemma imposed by the current rapid changes in the health care system. The number of residents wanting to do nuclear medicine seems to be determined by the number of radiology residents who need a holding pen until they can line up a job. There are few persons ready to dedicate themselves to a career in nuclear medicine. The academic council is trying to address the issue but it seems way bigger than this council. Nuclear medicine can count on few allies, not the ACR, not the chairmen of radiology programs (even if they are sympathetic to nuclear medicine), not the health care suits that have now gained the upper hand. Who would have ever thought one of our few friends would be the NRC? I saw no answers at this meeting only more problems. Sunday Ah Sunday! At these meetings Sunday is the day the SNM board meets and it is my day to go exploring. This year's exploration was in the lands of the Agua Caliente Indians where a delightful and exhilarating hike showed off the desert at its winter finest. Highly recommended. Computer & Instrumentation Council's Data Management in Nuclear Medicine The theme of this meeting was data, how to work with it and how to manage and store it. It was particularly relevant for me as we are presently installing (and designing and making work) a data archiving system in our nuclear medicine division at BWH which is to tie into our Radiology PACS system. I was eager to learn how other were approaching this important subject. Gary Hutchins and Jack Correia put together a fine program that held my interest as well as the 60 or so other attendees. It is something of a challenge for the Computer & instrumentation Council to come up with innovative and interesting programs to attract persons to their mid-winter meeting, but they always seem to pull it off, although I detected some disappointment in this years attendance. It is a tradition to be strongly encouraged. However, perhaps they could save some money by dropping the free lunches and selling us all a box lunch to take out in the wonderful sun to eat and have interchanges with the other participants. I would strongly encourage the meetings to continue, on a smaller scale if necessary. It is a wonderful opportunity to explore a limited number of topics in depth. The material that follows are my personal notes and are not meant to be a comprehensive review of all the presentations. I have necessarily concentrated my attention on things I was most interested in and somewhat ignored that which I was less interested in. This is not to be taken as a reflection against the individual speakers in any way. And the notes below are what I THOUGHT the speaker said, not necessarily what he DID say. Monday MULTIMODALITY IMAGE MANIPULATION AND INTEGRATION The meaning of image registration generally used by the speakers was the transformation of one image space into that of another. This could be a rigid body transformation (rotation and translation only), an affine transformation (allow non-uniform scaling), mapping to a standardized system such as Talairach coordinates or mapping to a standard atlas. Fusion was usually meant as the display of two registered datasets. In all of this cross-modality registration and fusion was considered as well as registration and fusion confined to nuclear medicine datasets (SPECT and PET). Techniques for Image Fusion and Registration in the Brain Nathaniel Alpert Massachusetts General Hospital, Boston The first day started out with a review on image registration. Dr. Nat Alpert gave an excellent review of registration in the brain in the opening presentation. Incidentally what I call registration others call superposition, fusion, merge, etc. There are sometimes distinctions between these terms by some authors, but I will not hold to these. The reasons for doing registrations vary and every kind of inter/intra subject and intra/inter modality has been tried and been found useful. Some want to mold all brains into a common framework to classify some measured physiological response. Evans (Montreal Neurological Institute) (http://www.mni.mcgill.ca/) has gone a long way in this work. Other want to apply a common coordinate system to all brains, e.g. the Talairach system (http://topaz.sensor.com/staff/shane/atlas/atlasintro.html).. Others have the more modest goal of visualizing corresponding anatomy in differing modalities with some confidence. Nat divided the methods into the following: 1. guess and try: Many groups have worked on variations of this as it is not difficult to program. Results of different groups are somewhat mixed and seem to depend mostly on the speed of display and updating (degree of interactiveness, to coin a phrase. It can be useful especially when there is not a great number of patients to register. 2. Matching of intrinsic and extrinsic fiducials 3. Principal axes transformation (Alpert) 4. Surface matching: This method was introduced by Pelizzari (http://www.radonc.uchicago.edu/~chuck/) and Chen in the late 80's. Many groups have had success with variations on this technique. It can work across modalities when corresponding surfaces can be identified. Some knowledge of the patient and supervision of the contour or surface drawing is important. 5. Volume-intensity: Automatic Image Registration (AIR) of Woods (http://topaz.sensor.com/work/ref/air/roger-woods.html) has currently gained much favor recently. Initially applied to intra-modality registration, more recently it has been extended to apply across different modalities. This method can be totally automatic and has documented registration accuracy of 0.5 to 2 mm in and across various modalities. Dr. Alpert emphasized the need for good visualization software to see the results of the registration. At his institution, Massachusetts General Hospital, Boston, they rely to a great extent on AVS with routines they have added specifically for this purpose. A major problem faced by nuclear medicine workers in registration is how to register PET and SPECT images across modalities and between patients if only part of the anatomy is visualized, e.g. dopamine receptors. In the near future Dr. Alpert predicted automatic segmentation, automatic registration with sub-millimeter accuracy in a minute or less. WE should see the availability of "brain" databases for comparison to normative values. In the question/answer period following this talk it was asked why there are no successful commercial methods available. Several impediments were mentioned: 1. The FDA regulations may be an impediment, since results depend upon the users, even in the best methods now available. FDA may be reluctant to approve such software and manufacturers may shy away form producing such software. 2. Bringing a user designed method to the marketplace is very difficult at most companies. It was mentioned that Focus Graphics is one of the few companies providing software for this application. 3. One of the significant impediments is just getting the images to the right workstation. Until DICOM is universal, little progress can be expected in commercial solutions, at least from "the big boys". Registration and Fusion of Nuclear Medicine Images Tom Miller Washington University, St. Louis Dr. Miller was given the job of describing the state of the art in body (or non-brain) registration. The methods available are not unlike those in use in the brain. They are more successful in the chest and less so in the abdomen. There is a lot of plasticity in the body below the neck. Miller indicated that manual methods alone had proved useful for several groups, but indicated it could only be considered successful if there was a high degree of interactivity between the user and the display/program. Rapid or real-time updating of the display is a must. Operator must be able to quickly change orientation and see immediate effect of shifts in registration. Fahey et al of Bowman Gray had success with registration of PET/CT images using the pleural surfaces much like the Pelizzari method. Hasegawa at UCSF has had success with a combined SPECT/CT imager where the patient is not moved between scans or the scans are performed simultaneously. Several groups (Wahl,( http://www.med.umich.edu/dipl/lab_biblio.html) Michigan, Noz (http://nucmed.nyu.edu/NucMed/noz-vita.html), New York) have had success in the abdomen and pelvic area registration by a combination of fiducials, skin marks and semiautomated methods of registration that involves substantial human input. Slomka's (http://www.irus.rri.uwo.ca/~pslomka/home.html) recent work in registering cardiac scans was mentioned as one of the few examples of cardiac registration across patient populations. In a survey of practitioners of the art Miller concluded that: 1. All registration methods in the body require significant human intervention. 2. Outside the brain there is little hope for fully automatic registration in near future 3. Fully manual registration method is not all bad 4. Most groups currently performing some form of registration thought registration very valuable but it was seldom used, perhaps because of the practical difficulties, i.e. getting the datasets together, controlling the fiducials, timing of scans, etc. 5. In the brain, many of Miller's respondents used registration all the time for such things as epilepsy where precise location of function and anatomy was required 6. His respondents saw great value in the chest related to small node localization 7. Enthusiasm in the abdomen was lower because of the problems with accuracy. Many thought a simply visual comparison using clinical judgment would suffice. Many thought warping in the pelvis and abdomen would be required. Perhaps our colleagues in radiation therapy could teach us about immobilization of the patient. Display Issues in Image Registration and Fusion Stephen M. Pizer University of North Carolina, Chapel Hill Dr. Pizer spoke on techniques that, for the most part, he and his colleagues have applied with some references to the methods of others. His breakdown of the techniques in use was a bit different from that of Alpert's: 1. manual 2. intensity matching 3. point matching 4. boundary matching 5. medial locus method 6. optimized, involving guided measure on locus introducing concepts of boundariness and medialness Manual registration essentially uses a fused display to allow the operator to interact with the registration process directly. This requires fast hardware and intelligently designed display algorithms. Pizer and his colleagues have found three dimensional reprojection and the ability to sample planes interactively important. He indicated this was a way to get started with registration without heavy development effort. Intensity matching algorithms attempt to maximize the mutual information in two datasets. Woods of UCLA, Viola and Wells of MIT (http://splweb.bwh.harvard.edu:8000/pages/papers/swells/regist.html)and Collignon of Belgium (http://www.esat.kuleuven.ac.be/~delaere/MIR/node15.html#SECTION00012400000 000000000) Point matching techniques minimize the interpolated distance between corresponding external or internal reference points or fiducials. Chamfer matching fall in this category, according to Dr. Pizer. Difficulty with the point matching approach is the difficulty of selecting matching points especially outside the head and across different modalities. The number of points are necessarily limited and this can lead to less robust registrations. Boundary point correspondence or boundary matching methods are represented by Pelizzari's method and Best and Macay. The major difficulty is in defining the correct boundary. Medial locus matching is a technique developed mostly by the group at UNC with Dr. Pizer. Features are described by properties such as their spatial locus and width, resulting in structures called cores. Cores have the properties of medial locus and width. Corresponding cores in two datasets are the basis for performing registration. Cores have been found to be stable in both 2D and 3D sense. Examples in angiography were shown. Optimizing these graded measure on locus is the current area of interest for the UNC group. It applies the properties of medialness and boundariness to more difficult registration problems. Much of Dr. Pizer talk was concerned with properties of displays to be used in fusion work. This seems to be an area ripe for exploitation and in need of a few good (better) ideas. Present techniques such as "magic window" - partial overlay of one plane within the display of a reference plane is extremely useful but difficult to apply to a whole dataset. Superimposed bicolor displays, while appearing to be useful can be very misleading about registration accuracy. Dr. Pizer noted the usefulness of simulated images to check the quality of registrations as used by Perry, Rosenman and Tsui with SPECT-PET datasets. It was noted that a display for showing the results of fusion should be equally useful for verification of accuracy. There was a distinct need for more work in this area to extract measures of match and display them to the viewer. Quantitative Parametric Imaging in Nuclear Medicine Robert A. Koeppe U Michigan Parameters that can be displayed are perfusion, rate of transport or metabolism, receptor density, neurotransmitter density and volume of distribution. It was emphasized that this does not have to be an exact representation, an index or related quantity is often satisfactory. PET And SPECT are capable of unique measurements using tracer kinetic techniques. Displays of dynamic processes, rates, compartments and concentration can be made. It is important in all of this work to know the sources of error and bias. Validation is often required. Correction for such things as partial volume maybe necessary. The group at Johns Hopkins and others have addressed this with some success. The Web-Based Nuclear Medicine Department J Anthony Parker Beth Israel-Deaconess Medical Center, Boston Dr. Parker (http://www.med.harvard.edu/JPNM/Publications/ePublications.html) described the work his group has done in making a current PACS system compliant with WWW browsers as a means of easily making images available within and without the hospital for staff and referring physicians. WWW browsers are ubiquitous and will be around for a long time. It makes great sense to base a PACS system in this universal technology. Dr. Parker deems the project a complete success but points out that security issues may bother some, but present security is probably good enough in that passwords and username are required to view the mulitmodality report. Easily implemented security measures also include a secure server, use of a non-standard port and allowing only limited search capability on patient information. Nuclear Medicine Perspectives on PACS Edward M. Smith U Rochester, Rochester, NY Dr. Smith provided a very detailed process of specification, selection and installation of a nuclear medicine PACS system. His group decided to be the general contractor and specify what was needed, get bids, evaluate them and select a vendor(s) and proceed. At the time of this meeting installation was proceeding. At the outset they were determined NOT to fall into any of the pits that the earlier workers in PACS had discovered. Theirs had to work AND be cost effective. By limiting themselves primarily to nuclear medicine they felt it was a feasible goal. The major objectives were: 1. all images systems in nuclear medicine should be on the PACS 2. integration of hard copy printing for all stations 3. interchange files between all vendors for optimal processing 4. soft reading of all images 5. digital archive of all data with reversibility 6. modular approach 7. less than $400k 8. had to interface to IDXrad, the RIS for the department 9. adhere to standards such as Interfile 3.3, DICOM 3.0 and HL7. Having decided on the objectives the spec were written. It was decided to not go with a single vendor as such systems have been too closed and are too expensive. Multiple vendors were utilized and they were nuclear medicine was system integrator. Dr. Smith was very encouraging that this process will be successful. He distributed an outline of his major points that would be useful for anyone contemplating a similar path. PACS in Radiology /Nuclear Medicine Paul Christian University of Utah, Salt Lake City Mr. Christian described the PACS system that is in operation within the nuclear medicine division at their facility. It involved the main hospital and several outlying facilities that must be able to readily communicate images and results between the main hospital. The heart of the system is built around the UNIX operating system of a Picker Odyssey nuclear medicine platform. Interfile is relied on to obtain images from other systems. Interfile translation for General Electric computers are handled by PC system using Numa software which converts GE to Picker Interfile. Storage is done using hierarchical storage system involving nearline disks storage, and tape jukebox. Some of this part of the system was provided buy APUNIX. Access from home for emergency reading is supported. This system has proved to be very reliable and robust. Archival Options for Nuclear Medicine Datasets Andrew J. Van Abs NETegrity, Waltham MA formerly with Dicomex, Inc. Mr. Abs gave a very enlightening talk about current storage options giving capacity and cost figures. It was emphasized that nuclear medicine does not stand alone and we need to be aware of the bigger digital world out there. A lesson Radiology should also take to heart. Notable comments of Mr. Abs: Proprietary formats - shoot them; DICOM is of limited general utility since it has no database possibilities within it. These sentiments were generally shared by the audience. Internet in the Future of Nuclear Medicine Jim Halama Loyola U, Chicago Dr. Halama (http://www.meddean.luc.edu/lumen/Staff/Halama.html) discussed a WWW activities within the nuclear medicine community. A few NM vendors are taking note of the Web, but slowly. ADAC (http://www.adaclabs.com/) for example is talking about WebView a potential viewing product. There is a possibility a DICOM Web Server may be developed as means to distribute DICOM images to Web browsers. Some pilot projects exist, both commercial and user based. Dr. Halama went on to describe the project at Loyola to integrate multiple vendors onto a WWW system based on NT Server with an SQL database using ActiveX and VBscript as the programming environment. This system would bring together Siemens, Toshiba and ADAC systems into a common environment for reading, archiving and printing. It is the way of the future. Instead of basing the NM PACS on radiology/hospital standards base it on general computer industry standards. It should e very cost effective. Tuesday Panel Discussion on Clinical Utility of Quantitation, PACS, and Data Management in Nuclear Medicine Tom R. Miller, J. Anthony Parker, Michael Graham A wide ranging panel discussion emceed by Gary Hutchins allowed for much audience participation. Important points that were made here: 1. Image registration is useful but not clinically usable and has the potential for widespread use. Most of the time the clinical focus is known and it is not hard to cross localize by eyeball. If tools are easy to use it will be done. 2. Image registration of high quality is essential for epilepsy. Nuclear cardiology today uses registration tools and reorientation daily. 3. It is time to stop having fun with registration and get commercial products out there. We need papers to demonstrate the need and utility of image registration and fusion. 4. Registration will not be justified until we have easy availability of images across all modalities. This will require a DICOM archive. Presently film is all too common and it becomes a big impediment to registration 5. Parametric Imaging is used every day with bullseyes, phase amplitude imaging, parathyroid scans, three phase bone scans, etc. 6. Biggest benefit of PACS is no lost films, CRT readouts are better than film, fusion become easier, and rapid access to images is a reality. Funding comes down to whose budget is going to pay for it: hospital administration or radiology? 7. Leasing can be used to cover the obsolescence factor and web based tools are free and robust 8. In the years post 2000 it was predicted that a computer whiz will be needed as standard type of person in nuclear medicine and distance nuclear medicine will be the norm Manufacturer's Reports I had to leave for my plane before the AGFA representative finished with his video and pretty slides. It was probably just as well. This could have potentially been a valuable part of the meeting but this time it did not work. Commercial speakers need to be given a short leash and have to be screened carefully. Robert E. Zimmerman Joint Program in Nuclear Medicine Harvard Medical School Department of Radiology 25 Shattuck St. Boston MA 02115 zimmer@bwh.harvard.edu