Only three percent of radiologists are experts in reading pediatric MRI scans, often leaving children outside of major metropolitan areas at greater risk of misdiagnoses. Enter the algorithm.
When a young child shows signs and symptoms of developmental delay, among the many tests they receive is a brain scan called an MRI.
In the U.S., one in six kids has some sort of neurological delay. Of those, 650,000 of them may undergo an MRI, yet only about three percent of radiologists specialize in pediatric neuroradiology. Within that three percent, most of these experts tend to work in major metropolitan hospitals. For children who live in rural communities and need an MRI, diagnostic decisions are left to general practitioners and radiologists who may be unfamiliar with detecting signs and symptoms of abnormalities. That’s because abnormalities appear much differently on a brain scan of a child than they do on an adult.
It’s estimated that pediatric neuroradiologists in high-volume settings, such as the world-renowned Boston Children’s Hospital, collectively read studies that comprise of 30,000 to 50,000 images a day, with each scan containing anywhere from 1,000 to 4,000 images. These pediatric neuroradiologists belong to the three percent – the specialists, the experts. But their knowledge and insights required to read such scans are limited to the confines of walls of hospitals like Boston Children’s.
Because a child’s brain develops at such a rapid pace, normal development can often be misconstrued as disease on MRI scans, leading to unnecessary fear, referrals, appointments and follow-on imaging. In many cases, children who have no pathology at all may be falsely diagnosed by a general practitioner, only to be cleared by an expert.
“If you’re looking at adult scans all the time, it’s incredibly difficult to transition to pediatric scans and understand what is considered ‘normal’ and ‘abnormal’,” said Dr. Sanjay Prabhu, Pediatric Neuroradiologist, Boston Children’s Hospital.
During the first four years of life, a child’s brain undergoes a tremendous amount of change. Compared to an adult’s brain, it is incredibly watery and myelin has yet to develop around the fibers. The amount of change varies by child, making the process of diagnosis even more complicated.
If a radiologist had access to a normative tool at the point of care on their desktops, appearing alongside the patient’s information and corresponding to the age of the patient they are assessing, they would be able to confidently diagnose and point to next steps. From there, the clinician can determine if additional scanning is required – saving the child and their family’s time and unnecessary stress.
-Dr. Sanjay Prabhu, Pediatric Neuroradiologist, Boston Children’s Hospital
GE Healthcare and Boston Children’s are working to co-create that tool which would be an algorithm that could make it easier for non-specialists to determine what’s normal or not in a child’s brain. At the core of the collaboration between GE Healthcare and Boston Children’s is the creation of a pediatric MRI image database and in the future the team envisions a machine-learning algorithm that helps to read and analyze the images within it. The database is intended to be populated with various brain scans that indicate signs of normal developmental stages and those that show disease states, and the algorithm would be targeted to help non-specialists to distinguish between the two.
The first set of algorithms to be co-developed by GE Healthcare and Boston Children’s will look for myelination developmental patterns – a process in which a fatty white substance surrounds nerve cells in the brain to create a healthy central nervous system. Diseases such as multiple sclerosis and Pelizaeus-Merzbacher affect the pace and shape of nerve connections in the brain which can dramatically impact the development of the child. Any delay in the diagnosis – or a misdiagnosis – may lead to delayed or wrong therapy and additional diagnostics which impacts the quality of life.
A database of normal pediatric myelination would be focused on facilitating machine reading of these images. Over time, the machine learning algorithms could warn radiologists in the event of a potentially abnormal scan.
By gathering the necessary information into one database and comparing it with thousands of patients in the back end, clinicians and radiologists may be able to leverage the power of population health for precision medicine, tailoring a patient’s therapy based on a host of images and specialist diagnoses.
“In a decade or so, I could envision a time when patients could go in to a hospital and immediately get a diagnosis and prognosis based on previous patients’ scans and clinical data,” Prabhu said. “These algorithms will not only enhance the work of clinicians around the world, but will also dramatically reduce a child and family’s level of stress and uncertainty.”
“This collaboration exemplifies our long-term commitment to working with leading academic institutions and health systems to build a library of deep learning algorithms that target key disease states,” said David Hale, Vice President and General Manager, Digital Solutions, GE Healthcare.
By 2020, together with our academic collaborators, we’ll have hundreds of algorithms, analytics and apps in the GE Health Cloud, enabling insights that will truly transform healthcare in multiple disease areas.
-David Hale, Vice President and General Manager, Digital Solutions, GE Healthcare
Technology in development that represents ongoing research and development efforts. These technologies are not products and may never become products. Not for sale. Not cleared or approved by the U.S. FDA or any other global regulator for commercial availability.