Related Experiment Video
Updated: Jan 17, 2026
Author Spotlight: Advancing 3D Modeling for Enhanced Diagnosis and Treatment of Pulmonary Nodules in Early-Stage Lung Cancer
Published on: October 13, 2023
Multimodal deep learning method based on multiple spectra for lung cancer early diagnosis
Haolin Zhang1, Yafeng Qi1, Han Xu1
1State Key Laboratory of Electromechanical Integrated Manufacturing of High-performance Electronic Equipment, School of Electro-Mechanical Engineering, School of Life Science and Technology, Xidian University, Xi'an, Shaanxi 710071, China.
This study introduces a novel multimodal deep learning approach for early lung cancer detection using four spectral types. The method achieves high accuracy, offering a promising alternative to traditional diagnostic techniques.
Area of Science:
- Biomedical Engineering
- Artificial Intelligence
- Cancer Diagnostics
Background:
- Lung cancer is a leading cause of mortality globally, necessitating accurate and early diagnostic methods.
- Conventional diagnostic tools like imaging and histopathology have limitations, including cost, radiation, and subjectivity.
- There is a critical need for advanced, reliable, and accessible lung cancer detection technologies.
Purpose of the Study:
- To develop and validate a multimodal deep learning framework for enhanced lung cancer detection.
- To integrate diverse spectral data (FTIR, UV-Vis, fluorescence, Raman) into a unified diagnostic model.
- To improve the accuracy and efficiency of early lung cancer diagnosis through spectral analysis.
Main Methods:
- A deep learning model was designed integrating four spectral types: Fourier transform infrared (FTIR), UV-Vis absorbance, fluorescence, and Raman spectra.
- Spectral data were represented as 1D sequences and 2D Gramian Angular Summation Field (GASF) images.
- A dual-branch architecture with a MambaVision-based fusion module was employed for feature extraction and cross-modal interaction.
Main Results:
- The proposed multimodal deep learning method achieved high diagnostic performance: 97.65% accuracy, 98.14% precision, 97.52% recall, 97.82% F1-score, and 99.76% AUC.
- Ablation studies and comparative experiments validated the effectiveness of the training strategy and the proposed model.
- Interpretability analysis confirmed the model's focus on diagnostically relevant spectral regions.
Conclusions:
- The multimodal spectral deep learning approach demonstrates significant potential for accurate and early lung cancer diagnosis.
- This method offers a promising, intelligent, and potentially more accessible paradigm for spectral-based cancer diagnostics.
- The findings highlight the power of integrating diverse spectral information for improving diagnostic capabilities in oncology.
