基于去噪概率扩散模型的黑色素瘤图像分割算法  

作　　者：姚柳丞 陈小飞 王瑞[1] 

机构地区：[1]上海大学通信与信息工程学院,上海200444

出　　处：《工业控制计算机》2025年第2期93-94,共2页Industrial Control Computer

摘　　要：黑色素瘤是一种致命的皮肤癌,传统诊断挑战重重,病灶区域边界通常较为模糊,难以准确分离。开发了一种结合去噪概率扩散模型与改进的U-Net架构的图像分割算法,通过混合注意力机制精细捕捉病变区域的局部和全局特征。结果表明,所提出的模型能够精确地标定模糊边界,并在黑色素瘤图像分割任务中达到较高的准确率。Melanoma is a deadly form of skin cancer,with traditional diagnosis presenting significant challenges due to often blurred boundaries of lesion areas,making accurate separation difficult.This paper develops an image segmentation algorithm that combines a denoising probabilistic diffusion model with an improved U-Net architecture,capturing local and global features of the lesion area with a mixed attention mechanism.The results indicate that the proposed model can precisely delineate ambiguous boundaries and achieves high accuracy in the task of melanoma image segmentation.

关 键 词：深度学习 扩散模型 图像分割 黑色素瘤 U-Net 

分 类 号：R73[医药卫生—肿瘤]