Deep learning‐based 3D MRI contrast‐enhanced synthesis from a 2D noncontrast T2Flair sequence

Abstract

Purpose Gadolinium‐based contrast agents (GBCAs) have been successfully applied in magnetic resonance (MR) imaging to facilitate better lesion visualization. However, gadolinium deposition in the human brain raised widespread concerns recently. On the other hand, although high‐resolution three‐dimensional (3D) MR images are more desired for most existing medical image processing algorithms, their long scan duration and high acquiring costs make 2D MR images still much more common clinically. Therefore, developing alternative solutions for 3D contrast‐enhanced MR image synthesis to replace GBCAs injection becomes an urgent requirement. Methods This study proposed a deep learning framework that produces 3D isotropic full‐contrast T2Flair images from 2D anisotropic noncontrast T2Flair image stacks. The super‐resolution (SR) and contrast‐enhanced (CE) synthesis tasks are completed in sequence by using an identical generative adversarial network (GAN) with the same techniques. To solve the problem that intramodality datasets from different scanners have specific combinations of orientations, contrasts, and resolutions, we conducted a region‐based data augmentation technique on the fly during training to simulate various imaging protocols in the clinic. We further improved our network by introducing atrous spatial pyramid pooling, enhanced residual blocks, and deep supervision for better quantitative and qualitative results. Results Our proposed method achieved superior CE‐synthesized performance in quantitative metrics and perceptual evaluation. In detail, the PSNR, structural‐similarity‐index, and AUC are 32.25 dB, 0.932, and 0.991 in the whole brain and 24.93 dB, 0.851, and 0.929 in tumor regions. The radiologists’ evaluations confirmed that our proposed method has high confidence in the diagnosis. Analysis of the generalization ability showed that benefiting from the proposed data augmentation technique, our network can be applied to “unseen” datasets with slight drops in quantitative and qualitative results. Conclusion Our work demonstrates the clinical potential of synthesizing diagnostic 3D isotropic CE brain MR images from a single 2D anisotropic noncontrast sequence.

Artificial Intelligence to Reduce or Eliminate the Need for Gadolinium-Based Contrast Agents in Brain and Cardiac MRI

“…The whole-brain peak signal-tonoise ratio was 32.25 dB (in tumor regions, it was 24.93 dB), structural similarity index measure was 0.932 (tumor regions, 0.851), and AUC was 0.991 (tumor regions, 0.929). 38 As a result, this method was found to be potentially useful in synthesizing 3D isotropic contrast-enhanced brain MR images from a single 2D anisotropic noncontrast sequence. Because this study was intended to evaluate the use of precontrast 2D T2 FLAIR images to predict 3D postcontrast T2 FLAIR images, the research focused on superresolution imaging rather than contrast synthesis, which is typically evaluated using T1-weighted images.…”

“…34 Finally, another research group was motivated to find a solution to synthesizing and replacing 3D contrast-enhanced MR images, given the long scan duration and relatively high acquisition costs of these 3D sequences. 38 A DL-based model was used to obtain 3D isotropic full-dose postcontrast T2-weighted FLAIR images from 2D anisotropic noncontrast T2-weighted FLAIR images. An identical generative adversarial network was used to perform superresolution and synthesize contrast-enhanced characteristics.…”

“…An identical generative adversarial network was used to perform superresolution and synthesize contrast-enhanced characteristics. The whole-brain peak signal-to-noise ratio was 32.25 dB (in tumor regions, it was 24.93 dB), structural similarity index measure was 0.932 (tumor regions, 0.851), and AUC was 0.991 (tumor regions, 0.929) 38 . As a result, this method was found to be potentially useful in synthesizing 3D isotropic contrast-enhanced brain MR images from a single 2D anisotropic noncontrast sequence.…”

Real time image processing and operation management of multicultural communication platforms based on deep learning

“…Faster R-CNN uses another convolutional net-work (RPN) to extract candidate frames, applies a convolutional network to the entire image to obtain feature maps, and passes feature maps into RPN networks to extract candidate boxes [10][11][12]. This chapter considers the use of encryption algorithms of different strengths and different convolutional network structures for datasets, and analyzes whether the encrypted image leaks the information of the original image [13][14][15][16]. By using the encrypted image as the training set of the convolutional network, in which the label of each image is unchanged, if the judgment accuracy of the convolutional network in the test set is higher than the random prediction effect, it indicates that the convolutional network has learned some commonalities in the encryption process and analyzed the encryption mode.…”

Contrast‐enhanced MRI synthesis using dense‐dilated residual convolutions based 3D network toward elimination of gadolinium in neuro‐oncology

“…MRI synthesis approaches based on deep learning currently serve as an emerging field of research in neuro-oncology. 8,9 In particular, various deep learning-based approaches have been investigated for contrast-enhanced MRI synthesis toward reduction 7,[10][11][12][13] or even elimination 8,[14][15][16][17][18][19][20][21][22] of gadolinium contrast agents in glioma patients. The former involves methods that propose the synthesis of full-dose contrast-enhanced images from their low-dose counterparts (e.g., 10% low-dose).…”