Super-resolution and denoising of MRI data with machine learning

Qualitative results of different SR methods. The top row shows the whole-brain SR images of different methods; the bottom row shows zoom-in view of the region in the green boxes with corresponding residual error map next to them (note: the dark blue color indicates smaller error and dark red for the larger).

Qualitative results of different SR methods. The top row shows the whole-brain SR images of different methods; the bottom row shows zoom-in view of the region in the green boxes with corresponding residual error map next to them (note: the dark blue color indicates smaller error and dark red for the larger).

MRI super-resolution (SR) and de-noising tasks are fundamental challenges in the field of deep learning, which have traditionally been treated as distinct tasks with separately paired training data. We propose an innovative method that addresses both tasks simultaneously using a single deep learning model, eliminating the need for explicitly paired noisy and clean images during training.

Our proposed model is primarily trained for SR, but also exhibits remarkable noise-cleaning capabilities in the super-resolved images. Instead of conventional approaches that introduce frequency-related operations into the generative process, our novel approach involves the use of a Generative Adversarial Network (GAN) model guided by a frequency-informed discriminator network. To achieve this, we harness the power of the 3D Discrete Wavelet Transform (DWT) operation as a frequency constraint within the GAN framework for the SR task on magnetic resonance imaging (MRI) data. Specifically, our contributions include: 1) a 3D generator based on residual-in-residual connected blocks; 2) the proposed module of the 3D DWT followed by a convolution layer in the discriminator; 3) the use of the trained model for high-quality image SR, accompanied by an intrinsic de-noising process. We dub the model ”De-noising Induced Super-resolution GAN (DISGAN)” due to its dual effects of SR image generation and simultaneous de-noising. Different from the traditional approach of training SR and de-noising tasks as separate models, our proposed DISGAN is trained only on the SR task, but also achieves exceptional performance in de-noising. The model is trained on 3D MRI data from dozens of subjects from the Human Connectome Project (HCP) and further evaluated on previously unseen MRI data from subjects with brain tumors and epilepsy to assess its de-noising and SR performance.

Wang, Mahler, Steiglechner, Birk, Scheffler, Lohmann
DISGAN: Wavelet-informed Discriminator Guides GAN to MRI Super-resolution with Noise Cleaning.
ICCV 2023 CVAMD workshop. (2023)