Key Takeaways
- A recent review highlights how machine learning (ML) is enhancing the capabilities of SERS in clinical diagnostics by improving data interpretation and disease detection accuracy.
- The article details how ML techniques, such as PCA, SVM, CNN, and t-SNE, are applied throughout the SERS workflow, from signal acquisition to deep learning-based disease prediction.
- Practical applications discussed include rapid bodily fluid diagnostics and real-time surgical guidance, pointing to a future of portable, AI-powered SERS devices in point-of-care settings.
- The integration of AI with SERS faces challenges such as data standardization, reproducibility, and the need for large annotated data sets
Recently, a review article published in TrAC Trends in Analytical Chemistry highlights the increased use of artificial intelligence (AI), especially machine learning (ML), in clinical diagnostic applications (1). This study, which was led by Alfred Chin Yen Tay from the University of Western Australia and Liang Wang of Southern Medical University and how it is improving the application of surface-enhanced Raman spectroscopy (SERS) in clinical diagnostics.
AI is a growing area in not only spectroscopy, but in many different research areas, including medical and clinical applications. For example, AI algorithms have been used to improve the accuracy of diagnosing diseases by analyzing complex medical data more effectively than traditional methods (2). AI has also been used to evaluate medical images to detect abnormalities, process diverse data types, and accelerate the diagnosis process to improve patient outcomes (2).
Because SERS can detect trace levels of biological molecules, this technique has been increasingly used in medical diagnostics as well. SERS has been used in areas such as pathogen detection, biomarker discovery, and intraoperative guidance (1). However, the increasingly complex data sets produced by modern SERS instruments have outpaced traditional linear data processing methods, necessitating the need for AI integration.
In this review article, the authors outline how the different ML techniques are being integrated with SERS. Covering various parts of the SERS workflow including signal acquisition and preprocessing to feature extraction, unsupervised clustering, and deep learning (DL), the authors provide an extensive look into how researchers can use AI to unlock the full potential of SERS (1).
According to the review article, the ability of ML algorithms to learn from and interpret vast and complex data sets makes them ideal for handling SERS signals, which often exhibit high dimensionality and variability (1). Some of the ML algorithms that are most popular include principal component analysis (PCA), support vector machines (SVM), convolutional neural networks (CNN), and t-distributed stochastic neighbor embedding (t-SNE). These techniques have all been successfully and recently used to classify spectral data, identify biomarkers, and even predict disease states with high accuracy and speed (1).
There are numerous important takeaways researchers should digest from this article. First, the authors break down the computational workflows and talk in length about how SERS and ML fit into them. Unlike many earlier reviews that focused heavily on nanomaterial design or broad SERS applications, this article emphasizes the computational strategies required to analyze Raman spectral data effectively (1). Therefore, the authors offer insight on the theoretical underpinnings of ML and DL, offering practical insights into model training, evaluation, and interpretation (1).
Another main takeaway of this article is that the authors provide examples of ML-assisted SERS applications in action, offering specifics on what this integration looks like. Notable examples highlighted in this review include rapid fluid diagnostics, where ML models help interpret complex bodily fluid spectra in minutes, and intraoperative cancer margin detection, where AI-integrated SERS systems provide real-time feedback to surgeons (1). These innovations point to a future in which portable SERS devices, powered by ML algorithms, could be deployed in point-of-care settings to deliver fast and reliable diagnostics (1).
Although AI is becoming a routine part of scientific research, there are still some challenges and limitations that emerge when using AI in clinical diagnostics. The authors acknowledge in their article that there is a lack of standardized SERS data format, difficulties in reproducing results across platforms, and the need for larger annotated data sets to train robust ML models (1).
To address these issues, the authors suggest that interpretable ML models need to be developed further. As the authors emphasize in their article, the continued evolution of both SERS instrumentation and AI methodologies could lead to many positive outcomes for healthcare diagnostics (1).
References
- Tang, J.-W.; Yuan, Q.; Zhang, L.; Marshall, B. J.; et al. Application of Machine Learning-assisted Surface-enhanced Raman Spectroscopy in Medical Laboratories: Principles, Opportunities, and Challenges. TrAC Trends Anal. Chem. 2025, 184, 118135. DOI: 10.1016/j.trac.2025.118135
- Al-Antari, M. A. Artificial Intelligence for Medical Diagnostics—Existing and Future AI Technology! Diagnostics (Basel). 2023, 13 (4), 688. DOI: 10.3390/diagnostics13040688
