We implemented a FC that uses pixel-wise T1-weighted signals and corresponding inversion time to estimate T1 values from a limited number of T1-weighted images. we studied how training the model using native, post-contrast T1 and a combination of both could impact performance of the MyoMapNet. We also explored two choices of number of T1 weighted images of four and five for native T1, selected to allow training of network using existing data from modified Look-Locker sequences (MOLLI). After a rigorous training using in-vivo T1 maps of 607 patients, undergoing clinical cardiac MR exams, collected by MOLLI, the performance of MyoMapNet was evaluated using in-vivo data of 61 patients by discarding the additional T1-weighted images from MOLLI. Subsequently, we implemented LL4 T1 mapping sequence and an inline implementation of MyoMapNet on a 3T Siemens scanner to imaging and inline reconstruction of T1 maps. The inline MyoMapNet was then used to collect LL4 T1 and MOLLI in 16 subjects to demonstrate feasibility of inline MyoMapNet.

We implemented and tested three different classes of deep learning architectures for MyoMapNet: (a) a fully connected neural network (FC), (b) convolutional neural networks (VGG19, ResNet50), and (c) encoder-decoder networks with skip connections (ResUNet, U-Net). Modified Look-Locker Inversion Recovery (MOLLI) images from 749 patients at 3T were used for training, validation, and testing. The first four T1 weighted images from MOLLI5(3)3 and/or MOLLI4(1)3(1)2 protocols data were extracted to create accelerated cardiac T1 mapping data. We also prospectively collected data from 28 subjects using MOLLI and LL4 to further evaluate model performance.

No description available