AIUNet: A Medical Image Segmentation Network Composed of Attention and Inception Blocks

Abstract

Background: Glioma is the most common brain tumor disease. Magnetic resonance can help the clinical diagnosis according to the location of the glioma and the degree of malignancy, in which the segmentation of glioma site plays an important role for clinicians. The work of manual segmentation is very time-consuming and cumbersome, therefore automatic and efficient segmentation methods are very necessary. Methods: This paper proposed an AIUNet to give a more efficient segmentation of glioma, where a new block---Attention based Inception Block (AI Block), combining convolution and Self-Attention, is introduced. This module combines the smaller receptive field of convolution with the larger receptive field of Self-Attention, so as to extract more diverse feature maps to meet the needs of refined segmentation, and some deformations of AI block will be introduced and applied to the segmentation of glioma lesion. The AIUNet image segmentation network is constructed by merging AI block with the U-shaped network, this network has excellent segmentation performance, as well as low amount of parameters and calculation. Moreover, a loss function combined with GHM loss and Dice loss is plugged into the network, thereby improving the robustness of the network. Results: Experiments show that the proposed network can improve segmentation effects comparing with state-of-the-art methods.

Interface engineered Sb2O3/W18O49 heterostructure for enhanced visible-light-driven photocatalytic N2 reduction

“…4d, the transient photocurrent responses of the samples with several on-off cycles of intermittent irradiation were found to be very stable and reversible. Especially, the Sb2O3/W18O49 gave a remarkably higher photocurrent density as compared with W18O49 and oxidized 2D Sb in accord with the PL spectra, suggesting enhanced photogenerated charge migration in the composite [44,45]. Fig.…”