Cardiac MR function analysis with DL-based super resolution reconstruction: application in the clinical setting.

Area of Science:

• Cardiovascular Imaging
• Medical Physics
• Artificial Intelligence in Medicine

Background:

• Cardiac magnetic resonance imaging (MRI) is crucial for diagnosing cardiomyopathies.
• Standardized sensitivity encoding (SENSE) and compressed SENSE (C-SENSE) are used for accelerated MRI acquisition.
• Deep learning offers potential for enhancing image quality and reducing scan times.

Purpose of the Study:

• To compare volumetry, image quality, and acquisition time of cardiac MRI using SENSE versus deep learning-based super-resolution (CS-SR) reconstruction from C-SENSE data.
• To evaluate the feasibility of CS-SR for routine cardiac imaging.

Main Methods:

• Retrospective analysis of 31 cardiac MRI exams (1.5T Philips Ingenia).
• Acquisition using SENSE (R=2) and C-SENSE (R=4) accelerated techniques.
• C-SENSE data reconstructed with a deep learning-based denoising and super-resolution algorithm (CS-SR).
• Manual left ventricular (LV) segmentation and volumetric analysis performed by blinded readers.
• Image quality assessed using Likert scales by three independent readers.

Main Results:

• High correlation (r=0.98-1.00) between SENSE and CS-SR for LV volumetry, with no significant differences in end-diastolic or end-systolic volumes.
• Comparable overall subjective image quality (p=0.061), with CS-SR showing improved sharpness but increased artifacts (p<0.001).
• Significantly reduced acquisition time with C-SENSE (165.6s) compared to SENSE (411.1s) (p<0.001).

Conclusions:

• Deep learning-based super-resolution reconstruction (CS-SR) effectively shortens cardiac MRI acquisition times.
• CS-SR preserves the accuracy of LV volumetric analysis, crucial for assessing cardiac function.
• CS-SR demonstrates potential for efficient and high-quality cardiac MRI, streamlining clinical workflows.