2022/01 - Accelerated Cardiac T1 Mapping in Four Heartbeats with Inline MyoMapNet: A Deep Learning-Based T1 Estimation Approach [Journal Club]-Journal Club Webinar

Video Transcription

Video Summary

The video is a JCMR Journal Club session introducing and discussing a recent paper on <strong>MyoMapNet</strong>, a deep-learning approach to <strong>cardiac T1 mapping</strong>. The host explains that T1 mapping is widely used for myocardial tissue characterization, but standard MOLLI-style acquisitions often require long breath-holds and multiple heartbeats, which can be difficult for patients.

Dr. Reza Nasirfat presents the study, which aims to <strong>replace traditional curve-fitting with a neural network</strong> and <strong>reduce the number of T1-weighted images</strong> needed for mapping. Instead of using the full MOLLI acquisition, the model estimates T1 values from only <strong>4–5 images after a single inversion pulse</strong>, shortening scan time to about <strong>4 heartbeats</strong>. The team trained and tested the model using a large dataset of patients with cardiomyopathies, then validated it in phantoms, retrospective data, prospective scans, and even compared different neural network architectures.

Key findings: MyoMapNet produced <strong>T1 maps and ECV values very similar to standard MOLLI</strong>, with small average errors and comparable accuracy. A simple fully connected neural network performed about as well as more complex architectures like U-Net, so the simpler model was favored for clinical use. The method was also integrated directly into the scanner workflow using Siemens prototype software, allowing near real-time map reconstruction.

In discussion, the presenters emphasized that the method <strong>does not solve all site-to-site variability issues</strong>, and it still depends on the training data and scanner platform. However, it may make T1 mapping more practical and patient-friendly. The team also made the code and data publicly available.

Keywords

MyoMapNet

cardiac T1 mapping

deep learning

MOLLI

myocardial tissue characterization

neural network

ECV

T1 mapping

cardiomyopathy

scanner workflow