Multimodal microscopic imaging combining dual branch structure-component decoupled fusion for breast tumor image analysis and label-free intelligent diagnosis.

Authors

Chaoyi Ding, Sheng Ding, Jinjin Wu, Jianbo Zhu, Jing Pei, Chongyan Liu, Linwei Shang, Huijie Wang, Jianhua Yin

Published in

Analytica chimica acta. Volume 1425. Pages 346276. Dec 01, 2026. Epub Sep 16, 2026.

Abstract

Rapid and accurate analysis of breast tissue is essential for pathological studies and surgical decision support, whereas conventional frozen section pathology relies on time-consuming staining and subjective interpretation, limiting efficiency. This study introduced a multimodal fusion strategy that simultaneously integrates four complementary imaging modalities, including bright-field, phase-contrast, green fluorescence, and blue fluorescence imaging, to comprehensively capture tissue morphology, structural and molecular features. Further, a dual-branch structure-component decoupling network was designed to separately model structural and compositional information while allowing cross-branch interaction. An input-level spatial modality fusion module (SMF) was used to adaptively fuse paired modalities, while a contribution estimation module quantified structural, compositional, and coupling contributions to enable interpretable classification. The network achieved a three-class accuracy of 92.51% and an area under the curve (AUC) of 0.9852, outperforming dual-modality and tri-modality combinations. The four-modality fusion efficiently integrated structural and molecular features, enabling comprehensive tissue feature representation, and improved classification performance. This approach demonstrated that combining structural and compositional information via dual-branch decoupling and multimodal fusion allowed rapid, accurate, and interpretable tissue classification. It provided a reliable tool for tissue feature analysis and pathological assessment and shows strong potential for clinical applications.

PMID:
42951525
Bibliographic data and abstract were imported from PubMed on 29 Oct 2026.

Read full publication at:
Please sign in to see all details.

Stats

1-terrible, 9-excellent. How would you rate this publication? Sign in in to submit your rating.

Post a comment

You need to be signed in to post comments. You can sign in here.

Comments

There are no comments yet.