Machine Learning-Enhanced SERS Technology Advances Cancer Detection

Rendition of two cancerous tumor cells on healthy tissue © NikahGeh - stock.adobe.com

The early detection of cancer is crucial for effective treatment and improved patient outcomes. One promising method for early cancer diagnosis is the identification of circulating or wandering tumor cells (CTCs) in the bloodstream. CTCs, which are cancer cells that shed from tumors and enter the bloodstream, carry vital information about tumor progression and metastasis. However, detecting these cells is challenging due to their low concentration, often only 1–10 cells per milliliter of blood. A team of researchers from the Chinese Academy of Sciences, led by Chenguang Zhang, has developed an innovative strategy that combines surface-enhanced Raman spectroscopy (SERS) with machine learning (ML) to overcome these challenges, offering a high-precision method for CTC detection. This work is published in the journal Biosensors and Bioelectronics (1).

Dual-Modal SERS Bioprobes for CTC Detection
SERS is a sensitive spectroscopy technique that enhances Raman signals using nanomaterial surfaces, enabling highly sensitive detection of molecules, even in trace amounts (1,2). This makes it ideal for applications like cancer diagnostics, where detecting rare cells such as CTCs is essential. The research approach utilizes dual-modal SERS bioprobes designed to both enhance the Raman signal and magnetically separate CTCs from other cells in the blood (1,2). The bioprobes are composed of gold (Au)-based and iron oxide (IO)-based nanoparticles, each equipped with Raman reporter molecules, such as 4-mercaptobenzoic acid (4-MBA) and alizarin red (AR), and folic acid (FA) for targeting CTCs (1).

The gold-based bioprobes generate strong Raman signals, while the iron oxide bioprobes provide magnetic separation, facilitating the removal of unwanted white blood cells (WBCs) that could interfere with the detection process. When co-incubated with blood samples, these bioprobes can identify tumor cells at concentrations as low as two cells per milliliter, a significant achievement given the rarity of CTCs (1).

Integration with Machine Learning
The researchers further enhanced the CTC detection process by integrating ML algorithms. Principal component analysis (PCA) and a Random Forest (RF) classification model were employed to analyze the complex Raman spectra generated by the SERS bioprobes. These ML tools enabled the team to distinguish between CTCs and other blood cells with remarkable accuracy. The model achieved a 98% detection rate for tumor cells and a 90% accuracy rate in classifying different types of tumor cells, such as HeLa, MDA-MB-231, and A549 cells, alongside the WBCs (1).

By analyzing large datasets of Raman spectra, the ML model could identify subtle differences in the chemical composition of various tumor cells, overcoming the limitations of traditional methods that often struggle with high background noise or interference from non-target cells. This makes the dual-modal SERS system not only more efficient but also more reliable than previous methods (1).

Spectroscopic Analysis and Bioprobe Performance
The team characterized the performance of their dual-modal SERS bioprobes. The Raman spectra of the gold-based probes revealed characteristic peaks at 1076 cm⁻¹ and 1580 cm⁻¹, indicative of 4-MBA, while the iron oxide probes exhibited peaks at 1441 cm⁻¹ and 1459 cm⁻¹, corresponding to AR. The SERS intensity of the probes was optimized to achieve the best signal-to-noise ratio (S/N) for accurate detection (1).

The bioprobes demonstrated excellent stability, with minimal variation in signal intensity, a key factor for reproducible and reliable detection. The targeting effect of the bioprobes was also confirmed through confocal microscopy, showing strong binding to CTCs in blood samples, further validating their specificity (1).

Practical Application and Future Potential
The new detection strategy was tested using rabbit blood samples, where tumor cells were successfully separated and identified using the ML model. The validation set, which included three types of tumor cells and WBCs, achieved a 95% prediction accuracy, underscoring the robustness of the system in real-world applications (1).

The study's success in combining dual-modal SERS bioprobes with ML algorithms potentially opens the door for more accurate, rapid, and non-invasive cancer diagnostics. This approach could significantly improve early cancer detection, enabling timely interventions and personalized treatment plans. The integration of various Raman reporter molecules and targeting agents with SERS technology, paired with ML, could further refine CTC detection, providing an analytical tool for precision oncology (1).

This research presents an innovative approach to cancer diagnosis by merging advanced spectroscopy with ML. The dual-modal SERS bioprobes not only enhance the detection of rare CTCs but also eliminate interference from non-target cells, achieving a highly sensitive detection rate and classification accuracy. This work holds immense potential for advancing early cancer detection and improving patient outcomes (1).

References

(1) Zhang, C.; Xu, L.; Miao, X.; Zhang, D.; Xie, Y.; Hu, Y.; Zhang, Z.; Wang, X.; Wu, X.; Liu, Z.; Zang, W. Machine Learning Assisted Dual-Modal SERS Detection for Circulating Tumor Cells. Biosens. Bioelectron. 2025, 268, 116897. DOI: 10.1016/j.bios.2024.116897

(2) Ge, H.; Gao, X.; Zhang, H.; Wang, F.; Gong, X.; Lin, J. A Novel Image-Based Machine Learning Method Assists SERS for Multiple Types of Urological Cancers Detection. Measurement 2025, 242, 115831. DOI: 10.1016/j.measurement.2024.115831