It’s a sunny day in Southern California and the developers of FreeSurfer—a suite of software tools for the analysis of neuroimaging data—are preparing for a training session to introduce scientists to the many benefits of the package. To help the scientists find the classroom they have hung “FreeSurfer Course” signs around the outside of the building, a stone’s throw from the beach and the restless waves of the Pacific. They switch on the computers as they await the attendees’ arrival.
The door opens and a gentleman drifts in. He’s young and tanned and dressed—if dressed is the right word—in flip flops and a tank top. It’s not your typical look for a neuroscientist but no matter. All are welcome here. He looks around the room in a quizzical sort of way and, after a moment, asks a single question.
“Is this the free surfer course?”
FreeSurfer—the software package—may not be a household name in the beach bum community but it has become an essential tool for researchers in a range of disciplines who work with neuroimaging data. Introduced and continuously developed and refined by investigators in the MGH Martinos Center in Boston, the suite has helped to provide ever-deeper insights into the structures of the brain, and thus has played an integral role in advancing our understandings of the brain in both health and disease.
But what does it do exactly?
Stated simply, FreeSurfer provides automated anatomical analysis of the brain. While it brings together a number of different tools for use with neuroimaging data, it is best known for—and named after—the first of these to be introduced, a tool designed for analysis of the surface of the cortex in the brain.
Martinos Center investigator Doug Greve offers a simple analogy to explain what this means for the typical neuroscientist; Greve joined the Center in the late 1990s and has been working with FreeSurfer since not long after. The analogy begins, as few in the neurosciences do, with a paper bag.
“The cortex is inherently a highly folded two-dimensional structure, like a paper bag that has been wadded up into a ball to fit inside a skull,” he says. With MRI, data is collected as a series of single images, in effect cutting the wadded-up bag into slices. If something were written on the bag—say, a topographic map, an image of the world as we perceive it projected onto the cortex—it would be nearly impossible to read from looking at the slices.
This is where the software package comes in. “FreeSurfer essentially stitches these sections together to reconstruct the folded bag, then unfolds it. The natural language of the cortex is written on the bag, so unfolding it makes it much easier to interpret.”
One might ask: Why is this important? In what ways can FreeSurfer actually help people? Even beyond the many benefits it confers for research applications, where it can help to make sense of functional neuroimaging studies by providing anatomical context, the software can bolster healthcare applications. For example, it has been used to track changes in disease due to pharmaceutical intervention, Greve says. “Prior to FreeSurfer, these structures would have to be manually designated—a tedious and error-prone process. Using FreeSurfer, we can quantify these changes automatically, which allows reliable studies with large sample sizes to be possible.”
A Brief History of Surfing the Brain
The origins of FreeSurfer can be traced to Ph.D. dissertation work by Anders Dale, done in the early 1990s under the supervision of Marty Sereno at the University of California, San Diego. Dale wrote the initial software code while trying to tackle the EEG/MEG inverse problem—essentially the problem of trying to glean information about the brain from electromagnetic signals recorded outside the skull—and in doing so enable reconstruction of the brain’s surface with EEG/MEG.
Others had sought ways to model the surface of the brain, to essentially flatten the organ for visualization and analysis purposes; Sereno himself had literally flattened a brain some years before, in the mid-1980s, seeking insight into how to achieve this. The breakthrough in Dale’s work came when he and Sereno realized they could infer the outlines of the pial surface—that is, the “top” of the gray matter, where adjacent banks of a sulcus are too close to one another to be resolved by MRI—by imaging its “bottom,” the boundary between the gray and white matter.
The next stage in the software’s history kicked off in 1996. After completing a fellowship at UCSD, Dale in 1996 joined what is now the MGH Martinos Center for Biomedical Imaging, where he continued a collaboration with Roger Tootell’s group, applying the code he had written in looking at the visual cortex with the nascent imaging technique functional MRI.
Not long after, in January of 1997, the Center welcomed researcher Bruce Fischl, a one-time software developer working in industry and a recent PhD graduate in Cognitive and Neural Systems. In a series of conversations in the following months, Fischl and Dale huddled together to discuss the potential for Dale’s code; especially with greater numbers of researchers adopting functional MRI, they knew, there would be a strong need for accurate anatomical guidance. They decided to join forces to develop the code for broader use in the neurosciences.
The researchers set to work, collaborating with Sereno back in San Diego in seeking ways to improve upon and extend the applicability of the software. Over the next couple of years, endless hours of discussions and of clacking away in writing new code paved the way for the introduction of FreeSurfer. This finally came in 1999 with a pair of papers in the journal NeuroImage and an official launch at that year’s Human Brain Mapping meeting in Dusseldorf, Germany—the latter made possible by the indomitable efforts of the Center’s Doug Greve and Thomas Witzel.
The road to this point wasn’t always a smooth one, of course. Along the way the investigators encountered the occasional obstacle, the intermittent snag, that would put at risk the continued vitality of the project.
Fischl points to one of these, in particular. In what may have been the greatest existential threat to the software, he says, the researchers couldn’t agree on a name for it. Dale suggested “B-Vis,” short for “Brain Visualization” but also a nod to the crude yet often hilarious MTV cartoon “Beavis and Butthead.” This idea was “universally panned,” as was Sereno’s characteristically esoteric contribution: “D,” a roundabout reference to the little-known “B,” a 1980s programming language developed by Bell Labs. Fischl himself, presumably joking, suggested they charge a premium for the software and call it ExorbitantSurfer.
Finally, after considerable debate, they settled on the more pithy—and indeed more accurate—FreeSurfer.
The software package was an immediate hit with the neuroscience community, and its popularity has grown exponentially in the years since. Today, FreeSurfer boasts nearly 33,000 active licenses, with users around the world applying it to a wide range of basic science and clinical problems. It has driven important advances in psychiatry and genetics and seemingly countless other areas, and helped to launch new techniques: among them diffusion imaging, in which the diffusion of water molecules in the brain provides the MR contrast.
Led by Fischl, the FreeSurfer group itself—more formally known as the Laboratory for Computational Neuroimaging—has also continued to make a mark. Over the past 20 years, the group has endlessly advanced and extended the software, and worked tirelessly to serve the ever-expanding FreeSurfer community.
What Tomorrow May Bring
On a brisk day in Boston in November 2017, the Martinos Center hosted a daylong symposium honoring two decades of FreeSurfer. The speakers in the morning looked back on the early days of the technique—a time, said researcher David Salat, maybe slightly wistfully, when Tiger Woods was a fresh young face on the professional golf circuit and Hanson and The Spice Girls ruled the pop music landscape—and on the many successes it has seen since then. Fischl, Sereno and Greve stepped up to the podium, as did several other developers and prominent users of the software: Nancy Kanwisher and Arthur Liu as well as Salat, who emphasized the importance of FreeSurfer in his work with structural imaging of the brain during the aging process.
In the afternoon, the speakers turned their gaze to the horizon, to what might be achieved with FreeSurfer in the coming years. Here, they focused on existing applications of the software as well as ones just now emerging. Jon Polimeni spoke about its use for anatomically informed analysis of high-resolution fMRI data while Anastasia Yendiki discussed the many ways it could improve diffusion imaging and Polina Golland looked toward its ongoing use with data from stroke patients and other clinical populations. Turning to a bold new application, Mark Michalski and Jayashree Kalpathy-Cramer outlined the ways in which the software could bolster artificial intelligence and deep learning in medical imaging.
In his introductory remarks, Fischl noted that, while coding can be wildly enjoyable in and of itself, people who get into this business do so because they want to make an impact, to somehow change the world for the better. With FreeSurfer, he and the many others who have helped to expand upon and improve the software over the years have clearly already done this. And the impact they have, on biomedical research and in the world generally, will surely only continue to grow.