CMR Innovations
Adam B. Christopher, MD
Assistant professor
UPMC Children's Hospital of Pittsburgh
Adam B. Christopher, MD
Assistant professor
UPMC Children's Hospital of Pittsburgh
Kinsey Brassaw, RT(MR)
Cardiovascular MRI Technologist
Boston Children's Hospital
Nadine Choueiter, MD
Director of Pediatric cardiac non invasive imaging
Kravis Children's Hospital at Mount Sinai
Conner Earl, BSc
MD/PhD Candidate
Purdue University
Jason N. Johnson, MD
Chief, Pediatric Cardiology
Le Bonheur Children's Hospital
Giulia Pasqualin, MD
Cardiologist
IRCCS Policlinico San Donato, Italy
Marian Pop, PhD
Associate Professor of radiology and medical imaging
UMFST, Romania
Sruti Rao, MD
Clinical Instructor
Johns Hopkins Medicine
soha romeih, MD, PhD
Director of advanced cardiovascular imaging
Aswan Heart Center, Egypt
Tobias Rutz, MD
Medical doctor
University Hospital and University of Lausanne, Switzerland
Animesh Tandon, MD, MS
Director of Cardiovascular Innovation
Cleveland Clinic Children's
There has been growing interest in CMR efficiency to promote wider adoption, greater accessibility, and sustainability of CMR.1,2 Efforts have focused on decreasing acquisition time yet post processing solutions remain a bottleneck for CMR throughput. High-resolution 3D cine and 4D flow magnitude images provide reliable anatomic and volumetric data with limited planning and are of particular interest in the pediatric and congenital CMR space. These datasets also hold promise for non-congenital patients that have limited breath hold compliance or claustrophobia. The adoption of these datasets has been hindered by the lack of robust post-processing solutions. Cumbersome manual segmentation remains a major limitation for the use of these datasets especially in congenital cardiac imaging.
Measurable Goal:
Automated AI segmentation of a dynamic 3D volumetric dataset
Current vs. Goal Capabilities:
Current clinical software requires manual processing of 3D datasets to orient to the long axis plane and create a short axis cine stack. The short axis stack must then be segmented in both systole and diastole. The ideal software would segment the volumes of each chamber (atria and ventricles) in 3D throughout each phase of the cardiac cycle, eliminating many mathematical assumptions associated with the summation of disks.
Must-Have Features:
Ability to reliably identify the borders of all cardiac chambers and specifically the endocardial and epicardial borders of both ventricles.
Ability to easily modify automated segmentation borders.
Ability to easily modify end diastolic and end systolic phases for each ventricle.
Ability to store segmentation either in the software or PACS system for future comparison.
Accurate for both simple and complex forms of congenital heart disease.
Nice-to-Have Features:
Ability to copy volumetric data to the clipboard to insert into the reporting tool of choice.
Ability to easily visualize, report, and summarize regional differences for each chamber.
Ability to compare volumetric data to a normative reference of choice.
Ability to incorporate new AI frameworks for segmentation and risk prediction in specific populations
