A partnership between engineers at Villanova University and doctors at the Children's Hospital of Philadelphia underscores how data-driven approaches using physics, math, and computer algorithms can help doctors better diagnose and predict medical conditions.
The researchers are using these engineering techniques to help them diagnose pre ventricular leukomalacia, or PVL, a type of brain injury that most commonly affects premature newborns but it is also seen in babies with congenital heart issues.
Babies with the condition often develop lifelong problems, including movement issues and intellectual or learning disabilities. PVL has no treatment or cure; children are monitored and often undergo physical therapy, occupational therapy, and speech therapy. As with many medical conditions, failure to diagnose PVL or a delayed diagnosis, can contribute to someone becoming more ill or can result in death.
Researchers believe that the main factors that contribute to PVL are decreased blood or oxygen flow to certain parts of the brain, damaging the inner part of the brain, called white matter, that helps control motor functions. Doctors are working to better understand how to predict when and why PVL occurs, and researchers hope that including engineers in the process can answer some of their questions.
C. Nataraj director of the Dynamic Systems Laboratory at Villanova, says that the partnership with CHOP began because of a casual conversation he had with a friend who is a cardiologist at the hospital, who mentioned PVL was an unsolved problem and asked him to take a look at de-identified data. Over the course of roughly eight months, he had his research team analyze the data provided at CHOP and discovered that the collected data on OVL had some correlations. They then began officially studying the condition with a grant from the National Institutes of health, and collecting better data.
They have been able to develop algorithms to predict when PVL may occur, and hope this will lead to more effective treatments, save time, and reduce medical costs.
One of the reasons PVL is complicated to diagnose is because it can't be determined using only physical or behavioral characteristics.
"The infant exam is sort of a poor exam," says Daniel Licht, pediatric neurologist and director of Neurovascular Imaging Lab at CHOP. 'It doesn't really tell you very much. To understand the neurologic condition of the infants, you need static tools."
They use tools like ultrasounds or MRI and CT scans of the brain, among other measures such as blood pressure, heart rate and has concentration readings.
"It's really hard to make sense of these large data sets," Licht says. "You can compare two vital signs in a time series but when you start adding in different streams of data, it gets very complex to understand how these things are affecting cerebral blood flow."
Ali Jalali, a post-doctoral associate who works on the research team at Villanova, says the team was able to significantly reduce the amount of time it can take to diagnose PVL after an infant has had surgery because of heart issues.
"They do MRI just after surgery," he says of babies who with cardiac problems, undergo surgery. "They calculate volume of brain cells and do another MRI one week after surgery and calculate brain cells again. They can tell from the two tests whether children have PVL."
The team also has made strides in more accurately diagnosing the condition, thanks to engineers and doctors learning from each other. When engineers were looking at the data, they soon discovered they would need the input of physicians to have even more accurate outcomes than what they had already found. Noticing patterns in the data along had resulted in an accuracy rate, of correctly determining whether an infant had PVL, of more than half, but they had plateaued there, so they asked doctors about what else they should understand about PVL in analyzing outcomes.
Dieter Bender, a PhD student on Nataraj's research team, discussed his experience collaborating with physicians.
"Sometimes we gave them important information and they said, "Maybe you should look at it a bit closer,' or it was the other way around and we were able to do it a bit more with our algorithms to make better predictions," he says.
Nataraj said they told Licht they wanted his input and insight, particularly about how to watch different outcomes over time, including which times to watch most. Once they received it, they tweaked their algorithms and as a result saw an even greater increase in accuracy. Now, they have hit 92 percent.
"The doctors are a very important part of this, especially the researchers, who can come in with their expertise and understanding of why something happens," Natarj says.
"We are using physicians, we are using the doctor's expertise, and we are using data analytics," he continues. "We are using them all together in an optimal synthesis and I think that's what gets us much better answers."
The NIH grant, which totaled $1.9 million, ran for five years through 2015, and the team says it will apply for another year.