As an alternative to true isotropic 3D imaging, image super-resolution (SR) has been applied to reconstruct an isotropic 3D volume from multiple anisotropic scans. However, traditional SR methods struggle with inadequate performance, prolonged processing times, and the necessity for manual feature extraction. Motivated by the exceptional representational ability and automatic feature extraction of convolutional neural networks (CNNs), in this work, we present an end-to-end isotropic MRI reconstruction strategy based on deep learning. The proposed method is based on 3D convolutional neural networks (3D CNNs), which can effectively capture the 3D structural features of MRI volumes and accurately predict potential structure. In addition, the proposed method takes multiple orthogonal scans as input and thus enables the model to use more complementary information from different dimensions for precise inference. Experimental results show that the proposed algorithm achieves promising performance in terms of both quantitative and qualitative assessments. In addition, it can process a 3D volume with a size of 256 × 256 × 256 in less than 1 min with the support of an NVIDIA GeForce GTX 1080Ti GPU, which suggests that it is not only a quantitatively superior method but also a practical one.

In recent years, vision transformers have demonstrated remarkable superiority to CNNs in single image super-resolution (SISR), yet their heavy computational and storage costs hinder practical deployment. In this work, we revisit CNN-based residual learning for SISR and identify a specific yet overlooked problem: the residual signal is often biased by mapping errors during model training, and this bias can propagate and accumulate through layers. To this end, we revisit the commonly used residual learning and feature fusion in SISR and propose an adaptive residual correction network (AdaRCN) in this work. First, we introduce a residual correction mechanism that adaptively compensates for the bias in the residual, which is utilized to ease error accumulation and improve mapping accuracy. On the other hand, we generalize the standard identity shortcut to a weighted channel concatenation followed by a 1 × 1 convolution, which is a more versatile strategy for adaptive feature fusion. Our AdaRCN is built entirely upon a naive CNN without complex architecture design and training strategies, thus ensuring efficient inference and parallelization. Extensive experiments verify the benefits and effectiveness of residual correction and adaptive feature fusion in improving the representational capability of our model, enabling it to achieve impressive performance comparable to advanced SISR models with moderate overhead.