The Surgical Planning Laboratory is a computer science-oriented laboratory that is housed in a clinical environment. The SPL is part of the MRI Division, Department of Radiology, Brigham and Women's Hospital, a teaching affiliate of Harvard Medical School.
The fields of computer science, digital imaging, visualization, and information technologies that the SPL specializes in have been part of some fields of medicine for more than twenty-five years. However, doctors have not always been able to access the totality of the data about their patients in a way that can lead to a best diagnosis or treatment. For example, the exact shape of a brain tumor might be contained in a patient's CT (computed tomography) or MRI (magnetic resonance imaging), but might elude a surgeon looking at a series of slice images on pieces of film.
Today's advanced high-speed imaging devices in some ways only compound the problem of information understanding by hiding potentially life-saving measurements in a torrent of other data. While performing extraction of this data may be difficult, the challenge is only multiplied for researchers attempting to learn about diseases such as schizophenia, multiple sclerosis, or cancer by examining anatomical changes of perhaps thousands of patients over a period of years.
It seems clear that without proper computational tools, medicine cannot understand and treat with greatest effectiveness the diseases that humanity continues to face.
The SPL's Research
The Surgical Planning Lab was created to bring the power of computers and the information available in modern imaging technologies to the benefit a wider range of medical fields than traditionally used them. The SPL provides a computer hardware infrastructure (powerful graphics workstations connected to a high-speed network of high performance computer clusters and large amounts of data storage), an extensive software environment (image processing and analysis tools, graphics editing and display programs), an expert staff (computer scientists, physicians, applications developers, imaging physicists, and other researchers), and an environment that encourages diverse collaboration. Colleagues from within the hospital as well as from hospitals and universities from throughout the world come to the SPL to advance the future of medicine.
The main research of the SPL is to develop post-processing methods for
digital medical imaging data and to use these methods for real-life applications.
These applications include, but are not limited to the following examples:
the goals of the computational work conducted in the SPL is to have fully
automated segmentation (i.e., identification of relevant structures in
the images), high-speed registration of data sets acquired using different
methods and at different times, interactive editing of and navigation
through complex and dynamic multi-dimensional data, and interactive and
intuitive display of the graphical models that result. The ability to
perform these difficult tasks will allow us to address the computational
needs of a wide range of current and future image-based medical applications.
While we have still not reached our goal, the current processing time
of between a few minutes and a few hours for many tasks allows us to process
several hundred studies per year.
Examples: the SPL at work
Currently, MR and CT scanners produce hundreds of megabytes of data per day and unit. Generally, this information is analyzed by a visual evaluation of cross-sectional slices, extracted from the 3D rastergrid of the examination, and then incorporated as grey-scale images on radiological film. The filming is done by technicians, and the reading of the films is done by radiologists. The films are also used to communicate with the referring physicians and for reference during patient procedures.
However, it is important to keep in mind that using the above procedures, anatomical structures appear on cross-sectional grey-scale images in a way that is very different from their real appearance. This difference in appearance requires the physician to perform a major mental translation of the information. This translation requires highly specialized training and is very difficult.
While radiologists undergo this specialized training, their clinical partners usually have more problems with the translation. The goal of the work in the SPL is to try to make this job easier by taking the raw imaging data, segmenting out relevant structures, and then generating three-dimensional reconstructions. This work has several potential applications, such as:
Surgery today relies conceptually on the same principles as it did three thousand years ago: the surgeons use their hands to directly control instruments and they use their eyes to provide them with feedback about the effect of their manipulations. Accordingly, a surgeon needs access to the site of an operation for both reasons: visualization and mechanical access.
The modern trend in surgery is an evolution towards minimally invasive approaches, where the damage set for accessing the surgical site is reduced by using rigid or flexible long-necked instruments introduced through natural openings or small incisions into the target areas. These instruments typically carry some form of visualization equipment and some way to introduce instruments for procedures.
The big problems arising with this type of approach are:
An additional area of noninvasive treatment methods uses some form of energy deposition in focal hotspots to prevent access damage at all (e.g. focussed ultrasound). In this scenario, the lack of feedback information for monitoring of the treatment success is even more urgent.
Since 1991 we have been providing three-dimensional reconstructions for intraoperative display (lower monitor) and, after alignment, overlay the renderings with live video of the patient(upper monitor). Over 100 surgeries have been supported in this way over the years. This research has been performed on Sun workstations.
In 1989, the MR division of the Department of Radiology of Brigham and Women's Hospital and Harvard Medical School initiated a project to develop MR-guided interventional procedures.
The components of this project are:
General Electric Medical Systems has participated in the project and has built an open magnet for surgical applications. This is a whole environment including MR compatible instruments, MR compatible anesthesia and monitoring equipment and so on. The first machine was installed at Brigham and Women's Hospital in December, 1993.
For an extended information on this topic check out our paper of Image Guided Procedures and the Operating Room of the Future.
The Schizophrenia Research Project was first discussed in 1987 when Drs. Jolesz, McCarley, and Shenton met to talk about the use of MRI scans to evaluate brain structures in schizophrenic patients. The project began in earnest in 1988, when Dr. Shenton received funds from the National Institute of Mental Health (NIMH) to conduct MRI studies in schizophrenia, and when Dr. Ron Kikinis joined the MRI division with his expertise in neuroradiology and image processing. These four investigators have been collaborating on MR studies of schizophrenia ever since, and the project has grown to include a large number of post-doctoral fellows, junior faculty, visiting faculty, and research assistants. This project is also well funded through grant support from private foundations, NIMH, and the Veterans Affairs Medical Center in Brockton, MA.
For more information about this project check out our Schizophrenia Project pages.
It is widely known that the aging process brings changes to the structure of the human brain. What causes these changes, however, is not well understood. The aging project uses MR imaging and computer image analysis to learn more about the mechanisms and result of aging on the brain.
The Multiple Sclerosis (MS) Project at Brigham and Women's Hospital is being carried out in the Surgical Planning Lab (SPL). The MS project is headed by Dr. Charles Guttmann .The purpose of this project is to monitor the progression of MS tumors or lesions over a time period of several years. The study required fifty patients to be scanned in the Brigham and Women's Magnetic Resonance Imaging (MRI) scanner twenty four times each. The scans were done over time intervals of from one week to two months in between scans.
MS lesions usually appear as bright spots in MRI scans. Our goals were to quantify and categorize these lesions in order to better understand this disease. Our approach to understanding our data was an elaborate one that evolved over time and was primarily developed by Dr. Kikinis and Dr. Guttmann, with many novel image processing algorithms supporting the project developed by numerous computer scientists and programmers at MIT Artificial Intelligence Laboratory, and GE Corporate R&D as well as here in the SPL. An image processing pipeline was applied to the datasets generated at each patient visit. The major components of the pipeline are registration and segmentation algorithms.
We see an increase of the data production of diagnostic imagers. Just the implementation of methods that are already used in laboratory settings will result in up to two orders of magnitude increase of data production per imager. It can be expected that the unit price of scanners will decrease, while at the same time the reimbursement per scan will decrease as well. This means more data from more scanners and less money to analyze them. This means that the specialists will have to become more efficient in the analysis of diagnostic data and its interpretation by nonradiologists will have to be facilitated because more non-specialists are likely to do some of the reading.
We predict that this will require computerized programs to prescreen the data and point out suspicious areas to the human diagnostician.
Minimally invasive procedures will require an increasing amount of visualization in the operating room. While different implementations are thinkable, there is a large common ground.
Preoperatively, a high spatial resolution, high contrast definition diagnostic data set is acquired. The data is segmented and prepared for 3D reconstruction. During the procedure the models are updated using available imaging information. This might be MR images (in the case of the open magnet), plain X-ray images, ultrasound images or video images. The updated models are then used to generate 3D renderings for the surgeon. This can be done using different technologies: monitors in the room, projection into the operation area, projection into a scope (operating room microscope, endoscope), headmounted display.
Significant work remains to be done both in the algorithmic domain as well as in the computational domain. Some of the classes of segmentation algorithms:
Our ability to visualize is directly related to our ability to understand; to understand the difficult problems computationally assisted medicine we must have the tools that permit anatomic and abstract data visualization in a way that frees, rather than limits, the physician.
With the development of new and more powerful computational tools, the SPL will strive to make it possible for doctors to better heal their patients, for scientists to detect and cure disease, and for researchers to better understand the complexity of the human anatomy. In this way, we hope to play a part in the development of the future of medicine.
Surgical Planning Laboratory, Department of Radiology, Brigham and Women's Hospital