Approved Research

The application of deep learning in low-field MRI

Lay summary

MRI/mri: magnetic resonance imaging, SNR: signal-to-noise ratio

1 Research aim is to advance the mri in low field using deep learning, especially suppressing the noise and improving the scans visual quality.

2 Sientific rationale is that noise is severe in the MRI image in the low field for the MR signals are not as strong as those in high field. To deal with the case of such low SNR, we need to use depe learning to suppress the nosie and extract clean signals as good as the high-field scanners do.

Additionally, in the conventional high-field mri, the images have high SNR and cou;d be accelerated by some factors via some classical methods like compressed sensing, parallel imaging and low-rank reconstruction. However, for the mri in low field,the images have low SNR and the hardware has limited number of coils like one coil or two coil, this makes some traditioal acceleration methods infeasible. Now the deep learning bring promises to accelerate the scanning at low SNR case and this goes beyond the tradiotnal theory and reconstruction methdos.

3 The project duration will be 2-4 years. in the first two years, we will foues on the denoising and mri scanning acceleration to improve the image visual quality. in the resting time, we will focus on the various application of the MRI in low field like disease diagnosis, brain age prediction.

4 The public health impact is to make the mri  not expensive for  patients in developing countries and accelerate the MRI in low field to reudce the subjects scanned uncomfort. Apart from this, we hope the mri scanner becomes light-weighted and could move outside the hospital for some emergency cases.