MF2-Net: A multipath feature fusion network for medical image segmentation

In this paper, we propose a multipath feature fusion convolutional neural network (MF2-Net) with noveland efficient spatial group convolution (SGC) modules with a multipath feature fusion network for theautomated segmentation of medical images. The proposed MF2-Net was designed with multiple encoder pathsto extract layer-specific multiscale information. Each encoder path employs SGC modules composed of stackedasymmetric kernels of different sizes (kx1 and 1xk). In the SGC modules, the context details of high-levelfeatures are encoded at varying scales, and neighbor feature information is incorporated with higher precision.In addition, the encoded features fused at the bottleneck layer capture abundant semantic features frominput images. Furthermore, the guided block mechanism is used to refine the segmentation boundaries at thedecoder stage by integrating the skip connection from the encoder stage. We verified that the SGC modulesin a multipath feature fusion network improve the segmentation accuracy with fewer learnable parameters.Experimental results demonstrated that the proposed model outperformed existing medical image segmentationmethods by an average score of 0.97 on publicly available datasets