Automatic calculation of myocardial perfusion reserve using deep learning with uncertainty quantification.

Area of Science:

• Cardiovascular Imaging
• Artificial Intelligence in Medicine
• Medical Image Analysis

Background:

• Myocardial perfusion reserve index (MPRI) measured via MRI is crucial for detecting ischemia.
• Current MPRI measurement techniques are often manual and time-consuming.
• Developing automated methods is essential for efficient clinical application.

Purpose of the Study:

• To develop a fully automatic method for estimating MPRI using deep learning and machine learning.
• To evaluate the performance and accuracy of the proposed automated MPRI measurement technique.
• To compare the automated method against a manual reference standard.

Main Methods:

• Utilized Monte Carlo dropout U-Net for myocardium segmentation in DCE-MRI.
• Employed machine learning for landmark localization to divide the myocardium into segments.
• Calculated MPRI based on left ventricular and myocardial contrast upslopes.
• Compared automated results with manual adjustments for contours and upslope ranges.

Main Results:

• Segmental MPRI measurements from the automatic method showed strong correlation with the manual reference (ICC = 0.75).
• The mean difference between automatic and manual MPRI values was -0.02.
• The 95% limits of agreement were (-0.86, 0.82), indicating good consistency.

Conclusions:

• The developed automatic method leverages deep learning segmentation and machine learning landmark detection for MPRI measurement.
• This approach enables efficient and quantitative assessment of myocardial ischemia in DCE perfusion MRI.
• The fully automated process eliminates the need for user interaction, enhancing clinical utility.