Informatica

Focusing on the problems of failing to make full use of spatial context information and limited local receptive field when U-Net is utilized to solve MRI brain tumour segmentation, a novel 3D multi-scale attention U-Net method, i.e. MAU-Net, is proposed in this paper. Firstly, a Mixed Depth-wise Convolution (MDConv) module is introduced in the encoder and decoder, which leverages various convolution kernels to extract the multi-scale features of brain tumour images, and effectively strengthens the feature expression of the brain tumour lesion region in the up and down sampling. Secondly, a Context Pyramid Module (CPM) combining multi-scale and attention is embedded in the skip connection position to achieve the combination of local feature enhancement at multi-scale with global feature correlation. Finally, MAU-Net adopts Self-ensemble in the decoding process to achieve complementary detailed features of sampled brain tumour images at different scales, thereby further improving segmentation performance. Ablation and comparison experiment results on the publicly available BraTS 2019/2020 datasets well validate its effectiveness. It respectively achieves the Dice Similarity Coefficients (DSC) of 90.6%/90.2%, 82.7%/82.8%, and 77.9%/78.5% on the whole tumour (WT), tumour core (TC) and enhanced tumour (ET) segmentation. Additionally, on the BraTS 2021 training set, the DSC for WT, TC, and ET reached 93.7%, 93.2%, and 88.9%, respectively.

In recent years, Magnetic Resonance Imaging (MRI) has emerged as a prevalent medical imaging technique, offering comprehensive anatomical and functional information. However, the MRI data acquisition process presents several challenges, including time-consuming procedures, prone motion artifacts, and hardware constraints. To address these limitations, this study proposes a novel method that leverages the power of generative adversarial networks (GANs) to generate multi-domain MRI images from a single input MRI image. Within this framework, two primary generator architectures, namely ResUnet and StarGANs generators, were incorporated. Furthermore, the networks were trained on multiple datasets, thereby augmenting the available data, and enabling the generation of images with diverse contrasts obtained from different datasets, given an input image from another dataset. Experimental evaluations conducted on the IXI and BraTS2020 datasets substantiate the efficacy of the proposed method compared to an existing method, as assessed through metrics such as Structural Similarity Index (SSIM), Peak Signal-to-Noise Ratio (PSNR) and Normalized Mean Absolute Error (NMAE). The synthesized images resulting from this method hold substantial potential as invaluable resources for medical professionals engaged in research, education, and clinical applications. Future research gears towards expanding experiments to larger datasets and encompassing the proposed approach to 3D images, enhancing medical diagnostics within practical applications.

This paper presents a non-iterative deep learning approach to compressive sensing (CS) image reconstruction using a convolutional autoencoder and a residual learning network. An efficient measurement design is proposed in order to enable training of the compressive sensing models on normalized and mean-centred measurements, along with a practical network initialization method based on principal component analysis (PCA). Finally, perceptual residual learning is proposed in order to obtain semantically informative image reconstructions along with high pixel-wise reconstruction accuracy at low measurement rates.