0090

View Abstract / View Presentation

Deep learning based acceleration of multi-contrast MRI examinations by acquiring contrast and sharing inter-contrast structure information

Sudhanya Chatterjee¹, Suresh Emmanuel Joel¹, Sajith Rajamani¹, Shaik Ahmed¹, Uday Patil¹, Ramesh Venkatesan¹, and Dattesh Dayanand Shanbhag¹
¹GE Healthcare, Bangalore, India

We used deep learning to accelerate MRI examinations where multiple contrasts (T2FSE, FLAIRT1, FLAIRT2) are acquired for the same subject using a contrast preserving and structure sharing approach.

Figure-3 Accelerated FLAIR T1 and FLAIR T2 images (at 2x and 3x) are shown for a test case. For this MR examination, all three contrasts were acquired on a 1.5T GE Signa Creator MR system. The FLAIR (accelerated scans) and T2FSE (reference scan) were obtained at same in-plane resolution but different slice thickness and slice spacing (FLAIR T2 being thicker). The proposed method is designed to deal with such acquisitions. All the accelerated reconstructions have high SSIM values showing robust and high quality reconstruction

Figure-2 The training workflow is shown here. The network is trained on both reference image and image obtained from the grafted k-space. It predicts the grafting artifact which is then removed from the accelerated image to obtain a clean image