DeepPlanner4Cardio: An automatic multi-view planning tool for Cardiac MRI

Cardiac magnetic resonance imaging (Cardiac MRI) is an ionizing radiation-free medical imaging technology used to monitor the function and structure of the cardiovascular system. Its disadvantage is that it’s a lengthy exam, demanding high levels of patient cooperation, and can take up to an hour, depending on the operator’s skill. A time-consuming element is the planning of the four standard cardiac views (2-chamber, 3-chamber, 4-chamber, short axis) needed for obtaining cardiac volumes and other measurements essential for a precise diagnosis.

As cardiovascular diseases (CVDs) are the leading cause of death globally, simplifying and automating parts of the Cardiac MR exam is vital for accessibility. Manual cardiac view planning is a multi-step process customized for the patient’s anatomy and subject to operator variability. Deep learning (DL) is rapidly advancing in the medical field and has been proposed to address this; however, previous DL-based methods rely on extensive manual annotations and involve multiple models.

This thesis builds on DeepCardioPlanner, a single-view planning DL tool. We proposed DeepPlanner4Cardio, a multi-view DL model predicting four cardiac views simultaneously from low-resolution 3D Cardiac MRI by leveraging inter-view relationships, reducing computational demands, and improving scalability.

DeepPlanner4Cardio trains in under 3 minutes, a significant improvement over models requiring up to 19 hours. Two Cardiac MRI experts evaluated the planning accuracy of DeepPlanner4Cardio, confirming its clinical relevance with 67.5% accurate predictions and 26.5% acceptable ones. This fully automated tool offers a faster, more accessible solution for Cardiac MRI planning.