COMPARATIVE ANALYSIS OF LANET AND FE-NET IN IMAGE SEGMENTATION

Image segmentation in infocommunications helps analyze infrastructure, enhance video quality, manage networks to improve efficiency and service quality. Biomedical image segmentation is very important for the medical imaging. It requires precise delineation of anatomical structures and pathological regions for various diagnostic tasks (for example, tumor detection and treatment planning) in improving clinical outcomes. In recent decades, the automatic medical segmentation methods which are based on deep learning (DL) models, such as convolutional neural network (CNN) architectures, have facilitated the automatic delineation of organ and lesion boundaries. This changed traditional manual segmentation approaches and improved efficiency and accuracy in clinical practice. This paper presents a comparative analysis of two DL-based models: LANet and FE-Net. LANet employs an Efficient Fusion Attention (EFA) module and an Adaptive Feature Fusion Decoder (AFF) module to improve segmentation efficiency and precision. In contrast, FE-Net integrates a Feature Awareness Module (FAM) for enhances its features to capture multi-scale and process segmentation details. The advantages of both methods in handling different scales, details, and edges were investigated and experimentally evaluated on various public datasets, and their performance in specific scenarios was assessed. The experiment showed that LANet is superior in computational efficiency and feature refinement. FE-Net shows superior performance in handling complex variations and edge details. The source code of LANet and FE-Net can be found on GitHub at [https://github.com/tyjcbzd/LANet] and [https://github.com/tyjcbzd/FE-Net].