Attention U-Net for Segmentation of Pancreatic Neuroendocrine Neoplasms in Computer Tomograph

Traditional computer tomography (CT) segmentation techniques for pancreatic neuroendocrine neoplasms often encounter challenges such as indistinct tumor boundaries and variable contrast, which hinder accurate delineation. In response to these issues, our study proposes a novel deep learning framework based on attention U-Net architecture. By integrating specialized attention modules, the model selectively emphasizes critical tumor features, leading to enhanced boundary detection and overall segmentation accuracy. Evaluated on the Pancreas dataset, the framework employs advanced convolutional neural networks alongside customized loss functions designed specifically for medical imaging. In addition, the approach leverages multi-scale feature extraction and dynamic weighting strategies to address the complex and heterogeneous presentation of neuroendocrine tumors. Extensive quantitative analyses using metrics such as the Dice Similarity Coefficient and Intersection over Union demonstrate significant improvements over conventional segmentation methods. Overall, this AI-driven solution aims to improve diagnostic precision and support personalized treatment planning for patients with pancreatic neuroendocrine neoplasms.