Fig. 3: MRI image reconstruction with MIR.
From: Energy-efficient high-fidelity image reconstruction with memristor arrays for medical diagnosis
a Schematic of MRI image reconstruction with MIR followed by AI segmentation. b K-space raw data from MRI scanner. c Reconstructed sagittal plane images, computed with 2D IDFT by MIR. Combining all the sagittal plane images together, a 3D MRI image can be obtained. d, e Transverse plane image (d) and coronal plane image (e) of the reconstructed 3D MRI image. f, g Segmented left atriums from nnU-Net, with software-reconstructed images (f) and MIR-reconstructed images (g). Z, the coordinate on the z-axis which goes from inferior to superior. h The PSNR and SNR of each reconstructed MRI slice of No.16 dataset by MIR. i The reconstructed image quality for 20 MRI datasets. j DICE scores of nnU-Net segmentation results of software-reconstructed (S.) and MIR-reconstructed (M.) images for 20 MRI datasets. k The comparison of energy efficiency of GPU and MIR. l The comparison of the normalized image reconstruction speed of GPU and MIR. The cartoon pictures of human organs and medical equipment used in a was partly generated using Servier Medical Art, provided by Servier, licensed under a Creative Commons Attribution 3.0 unported license.
