摘要
针对肺结节分类方法仍存在缺乏推理过程的可解释性和判别性特征表示等问题,提出了一个基于移动窗口注意力机制和编码解码器肺结节分类方法(SWAC)来对图像进行特征提取。该模型结合了卷积神经网络(CNN)和移动窗口注意力机制的优势,通过关注结节分类所必需的区域进行结节分类,有效地提取了结节的浅层特征和深层特征。该卷积神经网络引入了Focal损失函数,对网络主干进行特征约束来关注难分类样本,以此提升网络的判别表征能力。在LIDC-IDRI数据集上通过消融实验分析了该方法中各部分的贡献和影响,结果表明,SWAC分类方法具有优异的性能。
School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China Existing methods for pulmonary nodule classification still exist problems such as lack of interpretability in reasoning process and discriminative feature representation.To address these issues,pulmonary nodule classification network based on shifted window attention and codec(SWAC)was proposed.The SWAC model combines the advantages of convolutional neural networks(CNN)and the shifted window attention mechanism,effectively extracted shallow and deep features of nodules by focusing on the necessary regions for classification.The CNN introduces the Focal loss function to constrain the main network’s features and focus on difficult samples,thus improving the discriminative representation ability of the network.The contribution and impact of each part of the method was analyzed through ablation experiments on the LIDC-IDRI dataset.The results showed that the proposed method has excellent performance in pulmonary nodule classification.
作者
张琮昊
迟子秋
王占全
王喆
ZHANG Conghao;CHI Ziqiu;WANG Zhanquan;WANG Zhe(School of Information Science and Engineering,East China University of Science and Technology,Shanghai 200237,China)
出处
《大连工业大学学报》
CAS
2024年第1期73-78,共6页
Journal of Dalian Polytechnic University
基金
国家自然科学基金项目(62076094)
上海市科技计划项目(21511100800,20511100600)。
关键词
肺结节分类
深度学习
注意力机制
classification of pulmonary nodules
deep learning
attention mechanism
