According to a study published by In, deep learning, a type artificial intelligence, can increase the power of MRI when it comes to predicting attention deficit hyperactivity disorder, (ADHD). Radiology: Artificial Intelligence. Researchers suggested that the approach could also be applied to other neurological conditions.
The human brain is complex. It has many networks. Functional MRI, a type if imaging that measures brain activity using changes in blood flow to detect brain activity, has allowed for the mapping of brain connections. This comprehensive brain map is known as the connectome.
The connectome is being increasingly recognized as a key to understanding brain disorders such as ADHD. This condition affects people who are unable to pay attention or control their restlessness.
According to the National Survey of Children’s Health (6.1 million), approximately 9.4% of U.S. kids aged 2-17 years have been diagnosed with ADHD in 2016. A single test or medical imaging exam cannot diagnose ADHD in a child. ADHD diagnosis is based upon a combination of symptoms and behavior-based tests.
Brain MRI could be used to diagnose ADHD. Research suggests that ADHD may be caused by a breakdown or disruption of the connectome. The MR image contains a number of spatial regions known as parcellations that make up the connectome. Brain parcellations can either be based on functional criteria or anatomical criteria. Different brain parcellations can be used to study the brain at different scales.
Prior studies focused on the single-scale approach. This is where the connectome is built based on one parcellation. Researchers from the University of Cincinnati College of Medicine as well as Cincinnati Children’s Hospital Medical Center took an even more comprehensive view for the new study. They devised a multi-scale method that used multiple connectome maps, each based on multiple parcellations.
The NeuroBureau ADHD 200 dataset was used by the researchers to build the deep learning model. The model was built using the multi-scale brain connectome data of the project’s 973 participants, as well as relevant personal characteristics such gender and IQ.
Multi-scale ADHD detection performed significantly better than a single-scale approach.
“Our results highlight the predictive power the brain connectome,” said Lili, Ph.D., senior author of the study, at the Cincinnati Children’s Hospital Medical Center. “The brain functional connectome constructed on multiple scales provides additional information that can be used to depict the networks throughout the entire brain.”
Deep-learning-aided MRI based diagnosis could improve diagnostic accuracy. This could make it possible to implement early interventions in ADHD patients. ADHD has been diagnosed in approximately 5% of American school-aged and pre-school children. These children and adolescents are at high risk of failing academic studies and building social relationships. This can lead to financial hardship for families and a huge burden on society.
She said that the model could be used to predict other neurological disorders. “We use it already to predict cognitive deficiencies in preterm infants. To predict neurodevelopmental outcomes at age two years, we scan them shortly after birth.”
The researchers anticipate that the deep learning model will improve as it is exposed more neuroimaging data. They also hope to better understand ADHD-related disruptions and breakdowns in the connectome that the model has identified.