%0 Journal Article
%T 基于3D CMFF深度网络模型的肺结节分类<br>Classification of Pulmonary Nodules Based on 3D CMFF Deep Network Model
%A 刘一展
%J Modeling and Simulation
%P 588-596
%@ 2324-870X
%D 2024
%I Hans Publishing
%R 10.12677/MOS.2024.131057
%X 当今基于深度学习的肺小结节分类成为了研究热点。然而，现有的网络模型大都精度过低，且只能处理2D肺结节切片，无法在肺部CT中学习到3D特征。为了应对这些挑战，提出了一个具有高精度的多尺度特征融合网络模型3DCMFF。3DCMFF的主干基于3DECABlock，可以在提高网络的特征表达能力。模型中的3D PSA模块，能够提取特征信息更丰富的3D精细化特征图。多尺度特征融合将阶段2~4层输出的特征图进行融合，在几乎不增加计算量的同时提升精度。在LUNA16数据集上进行综合消融实验和对比实验中，3DCMFF取得了93.58的精度。实验结果表明，我们的方法对分类精度有着可观的提升，并且和现有的先进模型相比较，3DCMFF具有精度高，鲁棒性优秀的特点，并能够给医生提供辅助诊断。<br />
Nowadays, deep learning based classification of pulmonary nodules has become a research hotspot. However, most existing network models have low accuracy, as they can only handle 2D lung nodule slices and cannot learn 3D features in lung CT. To address these challenges, a high-precision mul-ti-scale feature fusion network model 3D CMFF has been proposed. The backbone of 3D CMFF is based on 3D ECA Block, which can improve the feature expression ability of the network. The 3D PSA module in the model can extract 3D refined feature maps with richer feature information. Multi scale feature fusion integrates feature maps output from stages 2~4, improving accuracy with al-most no increase in computational complexity. In the comprehensive ablation and comparative ex-periments conducted on the LUNA16 dataset, 3D CMFF achieved an accuracy of 93.58. The experi-mental results show that our method has significantly improved classification accuracy, and com-pared with existing advanced models, 3D CMFF has the characteristics of high accuracy and excel-lent robustness, which can provide auxiliary diagnosis for doctors.
%K 肺癌，肺结节，深度学习，多尺度特征融合<br>Lung Cancer
%K Lung Nodules
%K Deep Learning
%K Multi-Scale Feature Fusion
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=79760