Related Experiment Video
Updated: Aug 5, 2026
07:17
Exploring the Application of Surface-enhanced Raman Scattering-based Biosensing of Individual sEVs in Disease Diagnosis and Therapeutics
Published on: March 13, 2026
Cross-Device Risk Calibration for Surface-Enhanced Raman Spectroscopy-Based Urologic Disease Classification
Huafeng She1, Fuqiang Wang2, Xin Bai3
1School of Optoelectronic and Communication Engineering, Xiamen University of Technology, Xiamen, Fujian 361024, China.
Analytical Chemistry
|July 27, 2026
Summary
Risk-calibrated surface-enhanced Raman spectroscopy (SERS) with deep learning improves molecular screening for urologic diseases. This approach enhances accuracy and manages data variations, enabling reliable noninvasive disease detection.
Area of Science:
- Biomedical Engineering
- Spectroscopy
- Artificial Intelligence
Background:
- Surface-enhanced Raman spectroscopy (SERS) combined with deep learning shows promise for noninvasive molecular screening of urologic diseases.
- Current limitations include overconfident predictions near class boundaries and data distribution shifts across different devices and acquisition conditions.
- Addressing these challenges is crucial for broader analytical utility and clinical application.
Purpose of the Study:
- To evaluate risk-calibrated patient-level SERS analysis for reliable molecular screening under challenging data conditions.
- To test the system's ability to handle cross-device and cross-acquisition condition distribution shifts.
- To assess the performance in identifying reportable outputs and deferring unsupported inputs.
Main Methods:
- Developed a derivative-guided dual-stream classifier with soft-risk triage for internal data analysis, separating reportable spectra from high-risk samples.
- Implemented guarded patient-level calibration-path selection and support-dependent few-shot covariance alignment for external data analysis.
- Introduced a dual-evidence gate to manage unknown inputs, specifically testing with prostate cancer as an unknown class.
Main Results:
- Internally, the system maintained over 95% accuracy among retained spectra.
- Externally, patient accuracy recovered to 85.82% across four known classes, reaching 89.95% for retained patients.
- When prostate cancer was introduced as unknown, the deferral rate increased to 78.40% while maintaining 90.07% accuracy for known classes.
Conclusions:
- Risk-calibrated SERS analysis with deep learning effectively addresses challenges of overconfident predictions and distribution shifts.
- The developed workflow demonstrates robust performance in both internal validation and external data recovery.
- The system shows potential for reliable, noninvasive molecular screening of urologic diseases, including the deferral of unsupported or unknown inputs.
