Medical image segmentation model based on local enhancement driven global optimization

In medical image segmentation, it is a challenging task to identify and locate the boundary of pathological tissue accurately. In response to this issue, this paper proposes a medical image segmentation model, named Local Enhancement Driven Global Optimization Network (LEGO-Net), and specially develops an Detail and Contour Recognition Module (DCRM) to accurately identify the boundaries of lesion tissue. Specifically, the DCRM has the following two main contributions: Firstly, it improves the network's capability to identify the boundaries of diseased tissue by examining the intricate spatial relationships between row and column elements on the feature map. Secondly, by integrating local modeling with global modeling, the network is able to not only capture the detailed local structural information of the lesion area but also take into account the tissue's overall structure, thereby enhancing the network's capability to delineate the boundaries of the lesion tissue more effectively. Furthermore, to further augment the network's capacity to discern lesion information, this paper introduces a Channel Feature Enhancement Module (CFEM). the CFEM can highlight the importance of elements that are beneficial to foreground feature discrimination. The outcomes demonstrate that the network architecture proposed in this paper is capable of effectively identifying and segmenting the boundaries of pathological tissues.