Editorial for “Fully Automated Plane Prescription for Cardiac <scp>MRI</scp> : A Prospective Cohort Study”

Cardiac MRI is known to be superior when compared with other imaging modalities and has the potential to become the gold standard for assessing heart structure and function. It can be useful for a wide range of applications, including ischemic heart diseases and congenital heart defects, among others [1]. However, despite these advantages, clinical adoption of cardiac MRI is low, and it remains underutilized globally. One of the reasons for this underutilization is the complexity of planning in cardiac MRI. The conventional imaging planes including sagittal, coronal, and transverse planes which are adequate for other body regions are inadequate for cardiac imaging. The unique anatomical position of the heart requires cardiac-specific imaging planes established by professional societies [2]. Multiple orientations are imaged in a specific sequence of scans to achieve these cardiac-specific views. This complexity necessitates highly trained and experienced technologists. This also means there is notable intra- and inter-operator variability hampering clinical adoption. Further, even experienced technologists require significant time for planning, affecting the overall patient throughput. As a result, widespread adoption of cardiac MRI has been limited to large hospitals or specialized diagnostic centers. To enhance workflow efficiency, consistency, and precision, automatic plane prescription would be beneficial. As is the case in many domains, recent advances in artificial intelligence and hardware capabilities are starting to bear fruit; automatic cardiac MRI planning tools are in different stages of development from multiple teams and vendors [3-6]. One such effort has resulted in the AIRxHeart prototype from GE Healthcare, designed to automate the prescription of standard, widely used cardiac imaging planes. In the study titled “Fully Automated Plane Prescription in Cardiac MRI: A Prospective Cohort Study,” [7] the authors have evaluated the functionality of this prototype. This study is also one of the first to prospectively evaluate a fully integrated AI-based tool for cardiac planning. In this work, the authors enrolled a cohort of 57 healthy volunteers and 20 patients indicated for cardiac MRI and acquired four standard cardiac imaging planes—short axis, two-chamber, three-chamber, and four-chamber views—using both manual and automated planning workflows. They hypothesized that there would be no preference for one method over the other and that automated planning could perform as well as manual planning. They measured this equivalence both subjectively and quantitatively. Qualitatively, they compared the two planning methods with the help of two experienced radiologists, and quantitatively, they compared the angles for these different planes. The authors found that automatic planning was on par with manual planning, with angulation differences within acceptable limits. Additionally, the authors conducted volumetric analysis, such as the calculation of left ventricle end-diastolic volume and systolic volume, and confirmed there was minimal bias, indicating the automatic and manual planning methods were similar. They also reported that in 91% of the patient scans, the automatic prescriptions needed no manual correction from the technologists. Thus, they established that the tool is diagnostically equivalent to manual planning. While these results are promising, and the study takes the right step toward standardized cardiac MRI, there are certain limitations to this study. Observer bias in manual planning was present, since automated planning was done first and the technologists were not blinded to the automatic planning views. Also, the cardiovascular radiologist reviewing the volumetric contours was not blinded to the planning mode. Further, the heterogeneity and abnormalities seen in regular clinical practice are difficult to fully capture in the training of automated solutions and can lead to planning failure in extended clinical usage. However, these issues can be addressed in further studies, which the cardiac MR community would consider, including multi-center trials, quantification of time, cost and efficiency savings, and evaluation of downstream impact in other sequences, like LGE T1 mapping, among others. Overall, this study is an important step in bringing the goal of zero-click, completely automatic planning of cardiac MRI to reality, aiming to improve diagnostic consistency, reduce technologists' burden, and ultimately lead to wider clinical adoption of cardiac MRI.