Fluorescence and reflectance-based dual-modal hyperspectral image fusion for caries diagnosis
Lists
Only you see this
Abstract
• We propose a dual-modal hyperspectral imaging method fusing fluorescence and reflectance to improve caries diagnostic performance. • We design a two-branch global and local fusion network model, GLFusionNet. • GLFusionNet use a hybrid fusion approach to decompose, extract, and fuse shared and specific features of fluorescence and reflectance hyperspectral images. • The proposed method achieved the best performance and generalization ability with 99.10% accuracy, 99.15% sensitivity, and 99.77% specificity. • The proposed method can potentially develop an auxiliary diagnostic tool for precisely classifying clinical caries. Caries is a common disease with high morbidity in the field of the oral cavity, and its accurate diagnosis is beneficial to the prevention and treatment of the disease. However, the existing methods based on single-modality optical imaging have problems with low accuracy and limited scope of application. Here, we present an approach to improve diagnostic performance by combining fluorescence and reflectance hyperspectral imaging with deep learning. We design a two-branch global and local fusion network model, GLFusionNet, using a hybrid fusion approach to decompose, extract, and fuse shared and specific features of fluorescence and reflectance hyperspectral imaging. GLFusionNet can excel caries grading diagnostic performance in a dataset containing 4,435 spectral blocks with an accuracy of 99.10 %, sensitivity of 99.15%, and specificity of 99.77%. The proposed method can potentially develop an auxiliary diagnostic tool for precisely classifying clinical caries.
BibTeX
@inproceedings{wang2025fluorescence,
author = {Cheng Wang and Huaxing Xu and Hongyu Tang and Xin Lin and Xueying Huang and Luoqi Wang and Xinming Zhao and Xiaoling Wei and Rongjun Zhang},
title = {{Fluorescence and reflectance-based dual-modal hyperspectral image fusion for caries diagnosis}},
booktitle = {Measurement},
year = {2025},
doi = {10.1016/j.measurement.2025.116701},
}Loading related content...