Chinese researchers have developed a cutting-edge cervical cancer diagnostic model that combines spontaneous Raman spectroscopy, CARS imaging, and artificial intelligence to achieve 100% accuracy in distinguishing healthy and cancerous tissue.
A New Era in Cervical Cancer Detection
Cervical cancer, the fourth most common cancer among women worldwide, has long relied on histopathological analysis for diagnosis. However, this traditional method can be subjective, time-consuming, and invasive. Now, researchers from Jilin University and the Changchun Infectious Disease Hospital have created a non-invasive, highly accurate diagnostic model that could revolutionize cervical cancer screening and detection (1).
By integrating spontaneous Raman spectroscopy, Coherent anti-Stokes Raman spectroscopy (CARS), and a powerful ConvNeXt deep learning network, the team has introduced a rapid, objective, and label-free technique for distinguishing between healthy and cancerous cervical tissues (1).
Vibrant light waves: colorful spectrum visualization © StudioATC -chronicles-stock.adobe.com
Study Overview and Institutions Involved
This breakthrough research, recently published in Spectrochimica Acta Part A: Molecular and Biomolecular Spectroscopy, was conducted by Chenyang Liu, Caifeng Xiu, Yongfang Zou, Weina Wu, Yizhi Huang, Lili Wan, Shuping Xu, Bing Han, and Haipeng Zhang. The team represents a multidisciplinary collaboration across several departments at The First Hospital of Jilin University, Changchun Infectious Disease Hospital, and the State Key Laboratory of Supramolecular Structure and Materials at Jilin University (1).
Illuminating Biochemical Differences with Spectroscopy
The researchers first employed spontaneous Raman spectroscopy to analyze fresh-frozen cervical tissue samples from 10 patients diagnosed with squamous cell carcinoma. By targeting the high-wavenumber region (1800–3200 cm⁻¹) of the Raman spectrum, they identified biochemical markers such as elevated levels of lipids, fatty acids, and proteins in cancerous tissues.
The most significant spectral difference between cancerous and normal tissues was observed at 2928 cm⁻¹, a region associated with CH₂ symmetric stretching, common in lipid-rich environments. This served as the foundation for targeted imaging using CARS (1).
Visualizing Cancer with CARS Imaging
To overcome limitations of spontaneous Raman spectroscopy—such as low signal intensity and background noise—the team turned to CARS, a nonlinear optical imaging technique that amplifies Raman signals by up to 100,000 times (1–3). Using a Pico-Emerald laser system, the researchers captured high-resolution images of tissue sections at the 2928 cm⁻¹ vibrational frequency (1).
These CARS images revealed a stark contrast: normal cervical squamous cells displayed nearly zero signal intensity, while keratinized and non-keratinized cancerous cells showed strong Raman signals, allowing for naked-eye identification of cancerous lesions (1).
Moreover, the presence of keratin pearls—distinct structural features in keratinized squamous cell carcinoma—emerged as a potential marker to further subclassify cervical cancer types, providing clinicians with more nuanced diagnostic information (1).
AI-Driven Accuracy with ConvNeXt
To automate the diagnostic process, the team trained a ConvNeXt deep convolutional neural network using the acquired CARS images. This advanced artificial intelligence (AI) model, which excels at visual classification tasks, achieved a verification accuracy of 100% and a loss function of just 0.0927.
ConvNeXt eliminates the need for manual feature engineering, learning directly from raw image data to distinguish tissue types with exceptional precision. This represents a major leap forward in AI-powered pathology, particularly in the early and accurate detection of cervical cancer (1).
Clinical Impact and Future Potential
By combining biochemical specificity with visual clarity and machine learning (ML), the research team has demonstrated a highly sensitive, non-invasive, and scalable diagnostic tool for cervical cancer. The integration of spontaneous Raman and CARS spectroscopy with AI-based image classification opens the door for real-time, label-free cancer diagnostics in clinical settings.
Their findings not only reinforce the diagnostic potential of CARS imaging but also highlight the power of AI to streamline and enhance medical diagnostics (1).
References
(1) Liu, C.; Xiu, C.; Zou, Y.; Wu, W.; Huang, Y.; Wan, L.; Xu, S.; Han, B.; Zhang, H. Cervical Cancer Diagnosis Model Using Spontaneous Raman and Coherent Anti-Stokes Raman Spectroscopy with Artificial Intelligence. Spectrochim. Acta, Part A 2025, 327, 125353. DOI: 10.1016/j.saa.2024.125353
(2) Fujita, K. A Further Leap of Biomedical Raman Imaging. Spectroscopy 2020, 35 (7), 10. https://www.spectroscopyonline.com/view/further-leap-biomedical-raman-imaging (accessed 2025-07-08).
(3) Inoue, K.; Okuno, M. Coherent Anti-Stokes Hyper-Raman Spectroscopy. Nat. Commun. 2025, 16, 306. DOI: 10.1038/s41467-024-55507-0
Get essential updates on the latest spectroscopy technologies, regulatory standards, and best practices—subscribe today to Spectroscopy.
Integrating Spectroscopy with Machine Learning to Differentiate Seed Varieties
July 15th 2025Researchers at the University of Belgrade have demonstrated that combining Raman and FT-IR spectroscopy with machine learning algorithms offers a highly accurate, non-destructive method for identifying seed varieties in lettuce, paprika, and tomato.
sCMOS vs CCD: Advancing High-Speed CARS Spectroscopy
July 14th 2025A new comparative study shows that scientific CMOS (sCMOS) cameras could rival traditional CCD detectors in certain Raman CARS spectroscopy applications, offering faster readout and dynamic range despite slightly higher noise levels.
Nanometer-Scale Studies Using Tip Enhanced Raman Spectroscopy
February 8th 2013Volker Deckert, the winner of the 2013 Charles Mann Award, is advancing the use of tip enhanced Raman spectroscopy (TERS) to push the lateral resolution of vibrational spectroscopy well below the Abbe limit, to achieve single-molecule sensitivity. Because the tip can be moved with sub-nanometer precision, structural information with unmatched spatial resolution can be achieved without the need of specific labels.
How Analytical Chemists Are Navigating DOGE-Driven Funding Cuts
July 14th 2025DOGE-related federal funding cuts have sharply reduced salaries, lab budgets, and graduate support in academia. Researchers view the politically driven shifts in priorities as part of recurring systemic issues in U.S. science funding during administrative transitions. The impact on Federal laboratories has varied, with some seeing immediate effects and others experiencing more gradual effects. In general, there is rising uncertainty over future appropriations. Sustainable recovery may require structural reforms, leaner administration, and stronger industry-academia collaboration. New commentary underscores similar challenges, noting scaled-back graduate admissions, spending freezes, and a pervasive sense of overwhelming stress among faculty, students, and staff. This article addresses these issues for the analytical chemistry community.
Rapid Sweetener Detection Achieved Through Raman Spectroscopy and Machine Learning
July 10th 2025Researchers at Heilongjiang University have developed a rapid and accurate method for detecting sweeteners in food using Raman spectroscopy combined with a Random Forest machine learning algorithm, offering a powerful tool for improving food safety.
