Related Experiment Video
Updated: Jun 17, 2026
Computed Tomography-guided Time-domain Diffuse Fluorescence Tomography in Small Animals for Localization of Cancer Biomarkers
Published on: July 17, 2012
Domain-adaptive Raman spectral calibration transfer for cross-instrument glioma detection
Qingbo Li1, Yefan Zhang1, Xupeng Shao1
1School of Instrumentation and Optoelectronic Engineering, Precision Opto-Mechatronics Technology Key Laboratory of Education Ministry, Beihang University, Beijing 100191, China. qbleebuaa@buaa.edu.cn.
This study introduces a new Subdomain Feature Alignment Network (SFAN) for accurate glioma detection using Raman spectroscopy. SFAN enhances model transferability across different instruments, even with limited target data, improving diagnostic reliability.
Area of Science:
- Biomedical Engineering
- Spectroscopy
- Machine Learning
Background:
- Glioma detection is critical for patient prognosis, demanding rapid and precise tissue discrimination.
- Raman spectroscopy offers real-time detection but faces challenges in cross-instrument model transfer due to spectral variations.
- Existing transfer learning methods require impractical paired measurements and struggle with limited target domain data, leading to overfitting.
Purpose of the Study:
- To develop a robust method for cross-domain model transfer in Raman spectroscopy for glioma detection.
- To overcome limitations of existing methods, particularly the need for transfer sets and susceptibility to overfitting with small sample sizes.
- To enhance the generalizability and stability of Raman spectroscopy models across different instruments and measurement conditions.
Main Methods:
- Proposed a Subdomain Feature Alignment Network (SFAN) that aligns features in the feature space using Local Maximum Mean Discrepancy (LMMD).
- Implemented a collaborative soft-hard label weighting mechanism to improve transfer stability under small sample conditions.
- Introduced a two-stage network migration strategy to decouple shared feature learning from target domain adaptation.
Main Results:
- The SFAN method demonstrated superior performance compared to conventional approaches on a human glioma dataset.
- Achieved domain-invariant and discriminative feature representations, enhancing model transferability.
- The proposed approach effectively addressed overfitting and improved transfer stability with limited target domain samples.
Conclusions:
- SFAN provides a more general and effective framework for cross-domain Raman spectroscopy analysis, especially in small sample scenarios.
- Shifting from spectral mapping to feature alignment fundamentally improves model transferability across instruments.
- The method offers a promising solution for reliable real-time glioma detection in clinical settings.
