How Raman Spectroscopy Sees Through Cancer Cells: Label‑Free Cancer Diagnosis

Have you ever wondered where those red‑and‑blue slides that pathologists examine under a microscope every day come from? Tissue removed from the operating room must undergo fixation, dehydration, paraffin embedding, sectioning, and H&E staining — the entire process takes at least 12 to 24 hours. Then a doctor examines slide after slide under a microscope to look for atypical cells and determine whether the tissue is benign or malignant. This workflow has been used for more than 100 years. It is the “gold standard” of pathological diagnosis, but it is slow, relies on staining reagents, depends on physician experience, and cannot provide real‑time results on the operating table.

What if there were a technology that, without staining or labelling, could read a cell’s biochemical “fingerprint” with just a beam of light — directly telling you whether it is benign or malignant?

This is exactly what Raman spectroscopy does.

Chapter 1

Basic Principles of Raman Spectroscopy and Molecular Fingerprinting

The principle of Raman spectroscopy can be explained in one sentence: when light hits a molecule, the molecule “returns” a beam of light with a changed frequency — and this frequency change is the molecule’s fingerprint.

To expand slightly: when a photon collides with a molecule, the vast majority are bounced back unchanged — this is Rayleigh scattering, with no frequency change. However, about one in a million photons will “exchange a little more energy” with the molecule. If the molecule happens to be vibrating, the photon may transfer part of its energy to the molecule, lowering its own frequency (Stokes line); or it may take energy from the vibrating molecule, raising its frequency (anti‑Stokes line).
This frequency shift is called the Raman shift, measured in cm−1. It depends on the vibrational mode of the chemical bond — a C–H bond has one vibrational frequency, a C=O bond another, and a P–O bond yet another. The vibrational modes of every bond in a molecule combine to form a unique “molecular fingerprint spectrum.” Just as you cannot forge your own fingerprint, a molecule cannot “disguise” its Raman spectrum. A qualified Raman spectrometer can detect peak information in the range of approximately 200∼4000cm⁻¹ — covering almost all characteristic vibrations of biomolecules.
 

Figure 1 | Molecular-level schematic of Rayleigh scattering (elastic, wavelength unchanged) and Raman scattering (inelastic, wavelength shifted).
 

Figure 2 | C. V. Raman (1930 Nobel Prize in Physics) and schematic diagrams of Raman scattering energy-level transitions and spectra: Rayleigh scattering has an unchanged frequency, Stokes lines have a lower frequency, and anti-Stokes lines have a higher frequency.
 

So the question is — since every molecule has a fingerprint, where exactly do the “fingerprints” of cancer cells and normal cells differ?.

Chapter 2

Raman Characteristics of Cancer Cells — Key Peaks and Diagnostic Criteria

Cancer cells and normal cells look very similar — both have a cell membrane surrounding cytoplasm, and both contain a nucleus. But at the biochemical level, they are different species.

Cancer cells undergo frenzied acceleration of DNA synthesis, and nucleic acid content skyrockets; protein conformation changes, with shifts in the ratio of α‑helix to β‑sheet; lipid metabolism becomes abnormal, with an increased proportion of unsaturated fatty acids. All of these biochemical differences are faithfully written into the Raman spectrum.

The table below shows several of the most critical characteristic peaks in Raman cancer diagnosis:

Table 1 | Key characteristic peak positions in Raman cancer diagnosis and their biochemical significance

By combining the intensity ratios, peak shifts, and full width at half maximum of these peaks, cancer cells can be distinguished from normal cells — without any staining or antibody labelling.

This sounds wonderful. But in actual research, how solid is the data?

In recent years, the accuracy of Raman spectroscopy in diagnosing various cancers has become quite impressive:

Cervical cancer — Single‑cell Raman combined with a Stacking ensemble learning model achieved an AUC of 0.987, accuracy of 99.2%, sensitivity of 98.9%, and specificity of 99.3%. Another study using CARS (Coherent Anti‑Stokes Raman Scattering) combined with ConvNeXt deep learning achieved 100% tissue classification accuracy.

Gastric cancer — Raman spectra of normal gastric mucosa and cancerous tissue show significant differences at 1003, 1250, 1445, and 1655 cm⁻¹, with machine learning classification accuracy exceeding 92%.

Lung cancer — Raman spectroscopy based on human plasma samples combined with machine learning achieved diagnostic accuracy of 0.77∼0.85; a multicenter study of lung adenocarcinoma tissue Raman imaging combined with PCA‑LDA achieved sensitivity of about 90%, comparable to the pathological gold standard.

SERS multi‑cancer screening — In 2025, an AI‑guided SERS chip could distinguish 10 common cancers from a single serum sample, with 97.4% accuracy in distinguishing cancer from healthy controls.

Figure 3 | Cancer cells show increased intensity at the nucleic acid phosphate peak (1095 cm⁻¹) and lipid peak (1445 cm⁻¹), with a shift in the amide I peak (1660 cm⁻¹).
 

Chapter 3
Technical Challenges and AI Empowerment — From Weak Signals to Strong Discrimination

The principle of Raman spectroscopy is elegant, but engineering implementation is difficult.

The first problem: the signal is too weak. The Raman scattering cross‑section is extremely small — roughly one millionth of the incident photon count or even less. The Raman signals emitted by proteins and nucleic acids in biological samples are inherently weak, and coupled with strong autofluorescence “noise” from cells, the signal‑to‑noise ratio challenge is enormous.

The second problem: fluorescence background. Biological samples excited by short‑wavelength lasers (such as 532 nm) produce strong autofluorescence that drowns out the Raman signal. One solution is to choose longer‑wavelength excitation — 785 nm is the most commonly used excitation wavelength for biological samples, greatly reducing fluorescence background; 1064 nm further suppresses fluorescence, but detection efficiency decreases.

The third problem: data interpretation. A Raman spectrum has hundreds to thousands of data points, and comparing peak positions by eye is no longer realistic. Researchers have introduced PCA (Principal Component Analysis), LDA (Linear Discriminant Analysis), Random Forest, Stacking ensemble learning, and even deep learning models such as ConvNeXt for classification. These models are not showing off — they can reduce hundreds of dimensions of spectral data to a few principal components, then clearly separate cancer cells from normal cells in two‑dimensional space.

SERS (Surface‑Enhanced Raman Scattering) is another important breakthrough direction. Through the plasmonic resonance enhancement effect of gold and silver nanoparticles, Raman signals can be amplified by 10⁶∼10¹⁰  times, achieving single‑molecule‑level detection sensitivity. In 2025, AI‑guided SERS chips can already distinguish 10 cancers simultaneously from a single tube of serum — unimaginable five years ago.

But challenges remain. Different instruments, excitation wavelengths, and sample preparation workflows affect spectral comparability, and standardization and clinical translation remain key bottlenecks. Most studies are still at the ex vivo sample stage; in vivo原位 detection and large‑sample multicenter validation are the next mountains to conquer.

Chapter 4

Application of ATR8800 in Biomedical Raman Detection

From the previous chapter, it is clear that Raman detection of biological samples places three hard requirements on instruments: multiple wavelengths to adapt to different samples, extremely high signal‑to‑noise ratio to capture weak signals, and confocal imaging capability for spatial resolution.

Optosky’s ATR8800 confocal Raman imaging system is a research‑grade platform designed precisely for these needs.

Figure 4 | ATR8800 confocal Raman spectrometer main unit and analysis software interface.
 

The table below shows the core parameters of the ATR8800 and their significance in biomedical detection:

Table 2 | Key technical parameters of ATR8800 and their application significance in cancer detection

A few more words here:

Why emphasize multiple wavelengths? There is no “universal wavelength” for Raman detection of biological samples. 532 nm has a large scattering cross‑section for proteins and nucleic acids, giving strong signals, but biological tissues have strong autofluorescence in this band.785 nm is recognized as the “sweet spot” for biological samples, greatly reducing fluorescence background.

1064 nm further suppresses fluorescence, but InGaAs detectors have lower efficiency. The ATR8800 supports up to four wavelengths integrated into one instrument, allowing researchers to flexibly switch for different sample types, and even perform multi‑wavelength cross‑validation on the same sample — impossible with single‑wavelength instruments.

Why emphasize confocal and 0.35 μm resolution? On a tumor tissue section, cancer cells and normal cells may be tightly adjacent. If spatial resolution is insufficient, what you measure is a mixed spectrum of two cell types, and discrimination loses meaning. The confocal system’s lateral resolution of 0.35 μm means point‑by‑point scanning can be performed at the single‑cell level (typical cell diameter 10–30 μm) — each pixel is an independent Raman spectrum, thereby constructing a “chemical composition distribution map” that directly visualizes tumor boundaries.

The significance of EMCCD is this — when the signal is so weak that a normal CCD requires hundreds of seconds of integration to see a peak, EMCCD pushes effective quantum efficiency more than 1000‑fold higher through on‑chip multiplication, completing a single‑point measurement in milliseconds. For observing dynamic processes in living cells and detecting weakly fluorescent samples, this is a key link.

Chapter 5

Conclusion and Outlook

What Raman spectroscopy brings to cancer diagnosis is not a "replacement for pathological sections," but information from an entirely new dimension—molecular-level chemical fingerprints, requiring no staining, no labeling, and no destruction.

At present, the three most mature application scenarios are: intraoperative real-time margin assessment—when surgically removing a tumor, Raman spectroscopy can quickly scan the margin to determine whether cancer cells remain, reducing the probability of secondary surgery; single-cell-level malignancy discrimination—biochemical screening of early lesions and precancerous lesions; and liquid biopsy—SERS screening of multiple cancer types from serum, aimed at health checkup scenarios.

But to reach routine clinical use, several mountains must still be crossed:

Standardization—how can spectral differences produced by different instruments, different wavelengths, and different sample preparation workflows be aligned? Reference materials and standard protocols need to be established.

Large-sample validation—most current studies have sample sizes ranging from dozens to hundreds of cases. To reach NMPA/FDA certification level, multicenter, large-sample clinical trials are needed.

In vivo in situ—the vast majority of current studies use ex vivo samples. In vivo detection must solve engineering problems such as miniaturization of fiber-optic probes, tissue depth penetration, and in vivo fluorescence interference.

These challenges are precisely the directions in which research-grade platforms such as the ATR8800 are focusing—multi-wavelength, EMCCD enhancement, confocal mapping, and AI image processing. Every upgrade in these parameters shortens the distance from "laboratory to operating table" a little further.

Finally, a boundary statement: Raman spectroscopy is currently a research tool and an auxiliary diagnostic method, not a basis for clinical confirmed diagnosis. If you or a family member need cancer screening or diagnosis, please rely on the pathology report from your attending physician. The mission of Raman spectroscopy is to make that report arrive faster, more accurately, and with stronger evidence.

For more information, please contact:

Email: optoskyphotonics@gmail.com

Web: www.optosky.net