Heldt says the goal is to achieve accuracy within four to five mmHg, which will enable physicians to distinguish between a safe pressure—a healthy person’s intracranial pressure ranges from about seven to 15 mmHg—and one that requires intervention. When pressure rises to between 20 to 25 mmHg, physicians try to bring it down to a safer range, either through steps as simple as making the patient sit up, or as severe as taking away a piece of the skull to relieve pressure.
Researchers are about to begin a new test of the technology with collaborators at Beth Israel Deaconess Medical Center in Boston using data collected in real time from intensive care unit (ICU) patients. They hope that better-quality data will improve the accuracy of the measure. (The previous data set was collected more than a decade ago, with older equipment.) They also hope to show that a noninvasive method of collecting arterial pressure will work as well as intra-arterial monitoring.
While the researchers are initially focused on validating the technology in ICU patients, where they can compare the measure to intracranial catheters, they say the biggest potential for the tool is in examining patients with mild traumatic brain injury, recurrent migraine, and certain vestibular disorders.
The cumulative effect of mild brain injury is of great concern to both athletes and the military, given growing evidence that repetitive damage can have serious long-term effects. “For mild traumatic brain injury, we don’t know what intracranial pressure does,” says Heldt. Recent research in rats has shown that exposure to a blast, which generates a pressure wave, triggers an increase in intracranial pressure; the bigger the blast, the bigger the increase in pressure. Eventually, the researchers plan to develop miniaturized devices that could be deployed on the battlefield or the sports field.
Heldt adds that his team isn’t the first to try to assess intracranial pressure based on arterial and cerebral blood flow. But previous efforts used data mining or machine learning approaches to create the algorithm. Such approaches require a database of previous measures. If a new patient is substantially different from those in the database, the algorithm fails. By incorporating simple physiological knowledge of the brain, his team could create a model that doesn’t require any previous knowledge of the patient or anyone else.