Accelerated cardiac T₁ mapping in four heartbeats with inline MyoMapNet: a deep learning-based T₁ estimation approach

Area of Science:

• Cardiovascular MRI
• Quantitative Imaging
• Artificial Intelligence in Medicine

Background:

• Myocardial T1 mapping is crucial for assessing diffuse myocardial disease.
• Traditional T1 mapping methods like MOLLI can be time-consuming.
• There is a need for faster, accurate T1 mapping techniques.

Purpose of the Study:

• To develop and evaluate MyoMapNet, a novel, rapid myocardial T1 mapping approach.
• To utilize fully connected neural networks (FCNN) for T1 value estimation.
• To enable T1 mapping using only four T1-weighted images acquired in four heartbeats (LL4).

Main Methods:

• Implemented an FCNN (MyoMapNet) to estimate T1 values from reduced T1-weighted images.
• Trained and tested MyoMapNet using in-vivo MOLLI T1 mapping data from a large patient cohort.
• Explored training with native, post-contrast, or combined T1 data and evaluated using four vs. five images.
• Implemented a prototype LL4 sequence with inline MyoMapNet reconstruction on a 3T scanner.

Main Results:

• MyoMapNet trained on combined native and post-contrast data demonstrated excellent T1 accuracy compared to MOLLI.
• Using four T1-weighted images yielded performance comparable to using five images.
• Inline LL4 and MyoMapNet enabled successful T1 map acquisition and reconstruction.
• MyoMapNet showed high agreement with MOLLI for native and post-contrast myocardial and blood T1 values.

Conclusions:

• A FCNN trained on MOLLI data can accurately estimate T1 values from only four T1-weighted images.
• MyoMapNet facilitates rapid myocardial T1 mapping in four heartbeats.
• The technique achieves accuracy comparable to MOLLI with the advantage of inline map reconstruction.