基于深度残差神经网络的肝脏局部病灶分类  

Local Liver Lesion Classification Based on Deep Residual Neural Network

作　　者：黄林 孙小淇 

机构地区：[1]青岛大学数学与统计学院,山东 青岛

出　　处：《应用数学进展》2025年第1期352-359,共8页Advances in Applied Mathematics

基　　金：山东省自然科学基金省、补助/拨款编号:ZR2024MA097。

摘　　要：该研究提出了一种基于特征提取的深度神经网络(ResNet),用于对局灶性肝脏病变的超声波图像进行分类。这些优势使其能够有效地从超声图像中提取肝脏损伤的迹象,并实现准确的分类。基于包含血管瘤、脂肪肝、肝转移瘤、肝囊肿和正常肝脏的超声图像数据集,该模型在实验验证中提供了出色的分类效果。在测试集上,该模型的准确率达到了93.99%。本文将该模型与AlexNet和VGGNet模型进行了比较,结果表明作者获得的模型分类效果更好,在准确率、参数数量和学习效率方面都有显著优势,并且具有很强的泛化能力。这项研究在病灶性肝脏病变的超声波图像分类任务中具有潜在的应用价值,可为临床医生提供准确、快速的辅助诊断工具。This study proposes a feature extraction-based deep neural network (ResNet) for classifying ultrasound images of focal liver lesions. These advantages enable it to effectively extract signs of liver damage from ultrasound images and achieve accurate classification. Based on an ultrasound image dataset containing hemangioma, fatty liver, liver metastases, liver cysts, and normal liver, the model provided excellent classification results in experimental validation. On the test set, the model achieved an accuracy of 93.99%. This article compares this model with AlexNet and VGGNet models. The results show that the model obtained by the author has better classification results, has significant advantages in accuracy, number of parameters, and learning efficiency, and has strong generalization ability. This study has potential application value in the task of ultrasonic image classification of focal liver lesions and can provide clinicians with an accurate and rapid auxiliary diagnostic tool.

关 键 词：ResNet AlexNet VGGNet CNN 特征提取 超声图像 肝脏病变 图像分类 

分 类 号：R73[医药卫生—肿瘤]