AI-based Platform Accurately Analyzes MRIs of Children with Heart Defects

Jan. 22, 2021 - A UC Irvine engineering-led team has developed an artificial intelligence method to analyze the MRI scans of pediatric patients with congenital heart defects, and in a new study, they’ve demonstrated that the platform performs as accurately as physicians in analyzing the scans. Their research was recently published in the Journal of Cardiovascular Magnetic Resonance.

Cardiac magnetic resonance imaging, or MRI, is the method of choice for assessing heart function and anatomy in children with complex congenital heart diseases (CHD). Segmenting and analyzing individual heart chambers in these children are essential steps toward understanding their conditions. But hearts in children with CHD differ from those in healthy children and adults, and analyzing their MRI data is highly challenging, time-consuming and prone to variability in interpretation by physicians.

The research team’s artificial intelligence platform is based on deep-learning algorithms that can automatically and efficiently analyze cardiac MRIs for this growing group of patients.

“Our learning-based framework provides an automated, fast and accurate model for left ventricle and right ventricle segmentation, and its outstanding performance in children with complex CHDs implies its potential to be used in clinics across the pediatric age group,” said Dr. Arash Kheradvar, professor of biomedical engineering.

Compared to the existing automated approaches, UCI’s platform does not make any assumption about the image or structure of the heart, but instead performs segmentation while learning features of the image on its own, fully automating the process without requiring any predefined input. This makes its results more reliable than those from commercially available platforms. The researchers trained and validated their algorithm on a dataset of 64 pediatric patients with complex cases of CHD from Children’s Hospital of Los Angeles. They found no significant statistical difference in accuracy between the advanced AI method and manual segmentation.

“A major challenge in AI-based segmentation and analysis of cardiac MRI of children with congenital heart disease is lack of a diverse dataset. To mitigate that, for the first time, we employed a novel deep-learning method that synthetically generates new segmented MRI data from noise,” said Saeed Karimi-Bidhendi, the article’s first author and a UCI doctoral candidate in electrical engineering and computer science.

“This means that machines can segment and analyze the cardiac MRI data of these patients as well as a pediatric cardiologist,” said Kheradvar. “Eventually machines will be able to perform the analyses, replacing physicians. This paper indicates one more step toward fully automating the diagnostic imaging.”

Additional team members include UCI Chancellor’s Professor Hamid Jafarkhani; Dr. Andrew Cheng, a pediatric cardiologist at Children’s Hospital of Los Angeles and assistant professor at USC Keck School of Medicine; Arghavan Arafati, UCI mechanical and aerospace engineering alumna currently at Harvard Medical School; and Yilei Wu, who recently finished his undergraduate studies in computer science and is working on a doctorate at National University of Singapore. The research was supported by the American Heart Association.

– Lori Brandt