When Signals Cross
The human body and the systems that maintain it are, at their most basic, bundles of crackling electricity. Impulses, currents and waves can be found in every part of our world, and they offer much in the way of information if they can be properly read and interpreted. At the abstract level, Professor John Guttag and his research team are engaged in applied signal processing. But the marriage they have made between computer systems and medical research is vigorous and thriving. While it has already spawned impressive accomplishments, the most exciting opportunities to positively impact the practice of medicine lie in the team’s future.
When Guttag began working in the area of medical systems, he had no background in medicine. He was approached by a group of doctors who had grown frustrated with the limitations of current technology for diagnosis and treatment of cardiac patients. Cardiovascular disease is, at present, the leading cause of death worldwide. In the United States alone, nearly two million people a year will suffer acute coronary syndrome (ACS) of one sort or another. And after these events, there is no accurate way of predicting future outcomes. It is unknown who in that population is at high risk for recurrence and should be treated aggressively, and who would benefit most from something like a change in lifestyle and diet. The stakes for that uncertainty are high; some patients will go on to be perfectly healthy, and others will die – five percent of them in the first ninety days following the event. In a partnership with cardiologists Ben Scirica (Brigham and Women’s) and Collin Stultz (VA Hospital, MIT), the researchers have analyzed EKG data from over six thousand patients who have suffered ACS, in search of micro-instabilities – invisible to the eye of even the most well-trained cardiologist. Their thesis is that these nearly unnoticeable irregularities, much like small seismic tremors, can predict future instabilities with greater adverse significance. This re-stratification of patients has had remarkable results; the population that was classified as high risk was between six and eight times more likely to die from a recurrent cardiac event. When restricted to the first thirty days after the initial event, the gap was even larger. This finding was independent of nearly every other traditional bio-marker, including left ventricular ejection fraction – currently the most accurate predictive tool available to physicians. Recent studies have shown that the effect of implantable defibrillators to prolong life show measurable success in only a small number of patients; it is the group’s hope that their findings will enable the devices to be implanted in a more targeted segment of the population that would benefit the most. Pairing signal processing with medical research in this way allows students to engage with computer engineering problems while testing them against real world application. When Jenna Wiens, a current graduate student and project lead on the cardiac effort, was applying to schools, she was uncertain whether she wanted to pursue medical school or electrical engineering. In her work at CSAIL, she is able to have the best of both worlds. “Some researchers that work in machine learning work only in machine learning, and they stay very theoretical. But working in medical systems also helps you gain a new perspective of how the theory can apply.” The next project takes the processing of signals one step closer to patients, removing data not from post-event EKGs but from ongoing brain waves. Guttag and his team learned of a group at Children’s Hospital that wanted to automate diagnostics to determine what part of an epileptic’s brain gave rise to seizures. Part of that automation involved building a seizure detector, which is where the team from CSAIL got started. Via a system of electrodes on the scalp, researchers were able to record multiple seizures from a single individual. One interesting aspect of seizures is that each individual who suffers from them tends to have a predictable pattern of brainwaves for each event – it is almost as if the brain has a signature seizure. By building a device that used machine learning to leverage this fact, the researchers have constructed one of the world’s most accurate seizure detectors; it can recognize the onset of a particular patient’s seizures just as we would recognize our own signature. Epilepsy affects slightly less than one percent of the world’s population, comprising about fifty million people. In the United States, two thirds of those individuals’ conditions can be regulated fairly well with medication, but roughly thirty percent of patients present drug-resistant symptoms. The next step for the detector, once it was working well, was to pair it with a device called a vagus nerve stimulator, or VNS. The vagus is a long cranial nerve that moves from the skull down through the torso to the abdomen, relaying messages about the state of the internal organs back to the brain. Named for the Latin word “to wander,” it is the only nerve that originates from the brainstem (within the medulla oblongata). The left vagus nerve has the added distinction of being the main conduit for information between the brain and the heart: it is here that the VNS is implanted. Recently completed clinical trials on the ability of the device to control the VNS in adults showed encouraging results. Now one of the lead researchers, newly minted PhD Ali Shoeb, is working as a postdoctoral assistant at MGH to more clearly understand the potential therapeutic benefits of the system. Having started the research with no previous training in biology or medicine, the project has fundamentally altered his course as a researcher. “I was always interested in signal processing, extracting information from signals of any kind,” he says. “But this work has gotten me truly interested in biomedical signal processing and systems that can be used to enhance therapies and improve care. It was through my research experience that I developed a passion for this.” It is his hope that his work will eventually be used not only to address the symptoms of epilepsy, but to provide quality of life improvements for patients and their caregivers one day. The third main project of the group does not involve signals from within the body at all. A group of students headed by Asfandyar Qureshi is working to improve person to person communication in telemedicine, allowing physicians and medical personnel to provide a sustained, informed level of care. Playing on the omnipresence of pre-existing cell phone towers, the scientists have been able to construct a mobile streaming application that provides continuous, high quality video from a device comprised of a laptop with some strategic additions. Via a collaboration with Children’s Hospital, its test application has been in the area of neonatal monitoring during transit. Because of the specialization necessary for treating premature infants, this high-risk population frequently needs to be transported across long distances to personnel trained in their care. The babies’ very fragility makes them impossible to airlift in helicopters due to issues of air pressure and other variables, so these trips must be undertaken on the ground. But while the infant is in transit, the waiting physician has no way of seeing the patient or communicating with the emergency medical technicians about treatments that may become necessary to stabilize conditions or keep them from deteriorating further. The opportunistic networking protocol that was built to handle this problem relies on streaming through a net of cell phone reception areas. If one carrier drops a call, the bits are automatically shifted to the others so that quality is maintained and the signal never dies. This seamlessness is of paramount importance when a physician is making a diagnosis.
Adwoa Gyimah-Brempong, CSAIL