Regional Classification Method of Pulmonary Fibrosis Lesions Based on Dual Pathway Multi-level Hybrid Network
SU Shuzhi
YIN Xinle
ZHENG Xuejia
DAI Yong
Abstract：To address the diagnostic challenges posed by the structural complexity and morphological heterogeneity of pulmonary fibrosis lesions in computed tomography(CT)images,a region classification method of pulmonary fibrosis lesions based on dual pathway multi-layer hybrid network(DMH-Net)was proposed.Firstly,a texture feature enhancement module was designed to quantify the density gradient distribution of lesion micro-textures,and a calibration mechanism was used to optimize the modeling efficiency of the vision transformer(ViT)for morphological features.Secondly,a dynamic mamba transformer(DMTransformer)encoder based on mamba-like linear attention(MLLA)and dynamic hyperbolic tangent activation was constructed to achieve pixel-level lesion boundary localization.Finally,a method of global-local tandem dual-path information interaction was built to realize the coupled expression of macroscopic morphological features and microscopic texture features.The results showed that the accuracy of the DMH-Net model was improved by 16.87%compared with the ViT model.This study provided a new technical paradigm for the intelligent diagnosis of pulmonary fibrosis.
Keywords：pulmonary fibrosisconvolutional neural networkTransformerCT imagetexture featuremultiscaleinterstitial
Publication Date：2025-06-20
Online Publishing Date：2025-08-15(First online date of this platform, not the publication date of the document)
Pages：6( 196-201 )