1351986 - Joint cardiac T1 mapping and cardiac cine using a deep manifold framework

In this work, we introduce a deep manifold framework for the recovery of inversion recovery prepared free-breathing and ungated cardiac MRI. We model the image frames in the time series as a non-linear function of three variables: cardiac and respiratory phases and inversion time. The non-linear function is realized using a convolutional neural network (CNN) generator. We use a dense conditional auto-encoder (VAE) [1] to estimate the cardiac and respiratory phases from the k-space samples. Once the phases are estimated, we pose the image recovery as the learning of the parameters of the CNN generator from the measured k-t space data [2]. The learned CNN generator is used to generate synthetic data on demand, by feeding it with appropriate latent vectors. The framework enables the generation of synthetic breath-held CINE with different contrasts and the estimation of the T1 maps with specific phases.

Methods:

The free-breathing and ungated data acquired for joint cardiac T1 mapping and cardiac cine are based on an inversion recovery sequence with the details shown in Fig. 1.

After the data was acquired, we use a two-step strategy for data processing to jointly obtain the cardiac T1 mapping and cardiac cine. In the first step, we try to estimate the cardiac and respiratory motions from the central k-t space data using a conditional variational VAE. The known inversion timing signal is fed as a conditional vector to the network. Once the VAE optimization is complete, the outputted motion vectors will capture the cardiac and respiratory motion. Together with the known inversion timing signal, the three latent vectors are used in the second step, where the reconstruction of the cardiac MR images happens. For the reconstruction, we model the image frames in the time series as the output of a CNN generator. We propose to estimate the parameters of the CNN generator by fitting the model to the undersampled k-t space data. The idea of the method is illustrated in Fig. 2.

Results:

Once the generator parameters are learned, we can generate images with arbitrary cardiac/respiratory phases and inversion time. These images can be used to evaluate the cardiac function, and to quantify the T1 parameters based on MR fingerprints.

We first fix the inversion signal and respiratory phase and vary only the cardiac phase to generate synthetic breath-held cine images. The results for cine image generation can be found in Fig. 3. The generator can also be excited with the constant latent signals and varying inversion signals to generate images with different contrast. The T1 maps are derived by comparing the fingerprints to the pre-computed dictionary. The results for T1 mapping estimation are illustrated in Fig. 3.

Conclusion:

In this study, we proposed a manifold-based recovery scheme for the recovery of inversion recovery prepared free-breathing and ungated cardiac MRI. The framework enables the generation of CINE images with different contrast and the estimation of the T1 maps with specific phases.