Spectroscopy Liquid Biopsy: FTIR and SERS Cancer Detection

A blood draw. A drop of serum on an infrared crystal. Twenty-five minutes later, a result: cancer detected or not detected, across multiple cancer types, from a single test. No sequencing. No PCR. No multi-week turnaround. Cost per test: a fraction of genomic alternatives.

This is not a theoretical future. Dxcover, a Glasgow-based diagnostics company, received CE-IVDR Class C certification for its FTIR-based brain cancer blood test in March 2026 and launched commercially in the UK the month before. Their published data shows 64% detection of Stage I cancers at 99% specificity across eight cancer types. They opened a US headquarters in Nashville in June 2025 and are targeting US market entry in 2027.

Meanwhile, SERS-based approaches are producing striking results from large clinical cohorts - a 2025 BMC Medicine study of 3,551 patients across six cancer types reported 95.4% sensitivity and 95.9% specificity using serum SERS combined with deep learning. Raman spectroscopy of blood plasma is achieving 90% sensitivity and 95% specificity for breast cancer subtype detection at Stage Ia.

Spectroscopic liquid biopsy is a fundamentally different approach from the genomic liquid biopsies (Grail Galleri, Guardant Shield) that dominate headlines. It does not sequence DNA. It reads the molecular fingerprint of the blood itself - proteins, lipids, carbohydrates, metabolites - capturing both tumor-derived and systemic host-response signals that genomic methods miss entirely. It is faster, cheaper, and requires dramatically less sample volume. And it is further along than most people realize.

How Spectroscopic Liquid Biopsy Works

The premise is straightforward. Cancer changes the molecular composition of blood. Tumor cells release proteins, lipids, exosomes, and metabolites into circulation. The immune system responds with its own molecular changes. Even at Stage I, before the tumor is large enough to produce detectable circulating tumor DNA, these molecular perturbations alter the overall biochemical profile of serum.

Spectroscopy detects these changes not by identifying specific biomarkers, but by measuring the aggregate molecular fingerprint. An FTIR spectrum of serum contains absorption peaks from every molecule present - albumin, immunoglobulins, lipoproteins, glucose, amino acids, hundreds of metabolites. A cancer patient's serum spectrum differs subtly but consistently from a healthy person's. Machine learning models trained on thousands of spectra learn to detect these patterns.

This is a fundamental difference from genomic liquid biopsy. Genomic approaches look for specific signals - circulating tumor DNA with cancer-associated mutations or methylation patterns. Spectroscopic approaches measure everything at once and let the classifier find discriminating features.

• The advantage: spectroscopy captures non-genomic signals (protein expression changes, lipid metabolism shifts, inflammatory markers) that circulating DNA tests cannot detect.
• The challenge: the spectral differences are subtle, and robust clinical validation requires large, well-controlled studies.

FTIR-Based Blood Analysis: Dxcover

Dxcover is the most clinically advanced spectroscopy-based liquid biopsy company. Founded in 2016 as a spin-out from the University of Strathclyde (originally ClinSpec Dx, rebranded in 2019), the company was founded by Professor Matt Baker, Dr. Holly Butler, Dr. Mark Hegarty, and Dr. David Palmer.

The PANAROMIC Platform

Dxcover's technology platform - PANAROMIC - uses attenuated total reflection Fourier transform infrared spectroscopy (ATR-FTIR) with proprietary machine learning algorithms. The workflow is designed for clinical simplicity:

1. Drop. Nine microliters of blood serum deposited onto a proprietary Sample Slide
2. Dry. Brief drying step to remove water interference
3. Detect. ATR-FTIR measurement followed by ML classification

Total time from sample preparation to result: 25 minutes. Single-day turnaround. No sequencing equipment, no reagent kits, no molecular biology expertise required.

The platform analyzes the complete biochemical fingerprint - proteins, lipids, carbohydrates, and other biomolecular components. This is a critical distinction from genomic liquid biopsies: PANAROMIC captures both tumor-derived signals and non-tumor-derived signals (host immune response, metabolic changes, inflammatory markers) that ctDNA-based tests miss entirely.

Multi-Cancer Detection: The BJC Study

The foundational clinical dataset was published in the British Journal of Cancer in 2023 - a study of 2,092 patients across eight cancer types. Performance by cancer type at tunable operating points:

At 99% specificity: 64% of Stage I cancers detected (overall sensitivity 57%). At a high-sensitivity operating point: 99% of Stage I cancers detected (specificity 59%). The platform allows tuning the sensitivity-specificity trade-off based on the clinical context - screening applications favor sensitivity, confirmatory applications favor specificity.

Brain Cancer: The EMBRACE Trial

Dxcover's most advanced clinical program is the EMBRACE trial (Early Multi-cancer Detection via Blood Raman Analysis for Clinical Efficiency) - a prospective, observational, multicenter study across seven sites in the UK, Belgium, Sweden, and Switzerland with over 2,200 participants.

Feasibility studies on 988 patients demonstrated:

• Sensitivity-tuned model: 96% sensitivity for brain tumors, 100% sensitivity for glioblastoma, 45% specificity, 99.3% NPV
• Specificity-tuned model: 47% sensitivity, 90% specificity, 28.4% PPV

For brain cancer specifically, the clinical utility is compelling. Brain tumors are diagnosed late because symptoms (headache, cognitive changes) are nonspecific and imaging (MRI) is expensive and access-limited. A blood test that rules out brain cancer with 99.3% NPV could dramatically reduce unnecessary MRI referrals while catching more tumors earlier.

Pancreatic and Other Cancers

Pancreatic cancer results are particularly notable: 92% sensitivity and 88% specificity discriminating cancer from asymptomatic healthy controls. Pancreatic cancer has one of the lowest five-year survival rates (12%) precisely because it is typically diagnosed at advanced stages. An effective early-detection blood test would be transformative.

At AACR 2026 in San Diego, Dxcover presented four abstracts reporting 90%+ sensitivity consistently across Stage I-IV for colorectal, pancreatic, and ovarian cancers.

Regulatory and Commercial Milestones

The CE-IVDR Class C certification is significant - it is one of the first AI/ML-integrated IVDR-compliant Class C medical device software certifications in any field, not just spectroscopy.

SERS-Based Biomarker Detection

While Dxcover pursues the "whole fingerprint" approach with FTIR, SERS-based liquid biopsy takes a different path. SERS amplifies Raman signals from specific molecules at nanostructured metal surfaces, enabling detection of individual biomarkers at concentrations far below what conventional Raman or FTIR can measure.

Large-Scale Multi-Cancer Screening

A landmark 2025 study published in BMC Medicine combined serum SERS with deep learning for pan-cancer screening:

• 3,551 participants: 1,655 early-stage cancer patients (breast n=569, lung n=513, thyroid n=220, colorectal n=215, gastric n=100, esophageal n=38) plus 1,896 healthy controls
• Deep learning architecture: Heatmap transformation and continuous wavelet transform for spectral dimensionality enhancement
• Results described as "highly effective" for multi-cancer early detection

A separate study of 1,582 patients across five cancer types achieved 95.81% overall accuracy, 95.40% sensitivity, and 95.87% specificity.

These are striking numbers, though important caveats apply. The studies used case-control designs (known cancer patients vs. healthy controls), which typically overestimate real-world screening performance. Prospective validation in asymptomatic screening populations - where cancer prevalence is low and the classifier must maintain specificity at scale - is the critical next step.

Exosome and Biomarker-Specific SERS

Beyond bulk serum SERS, targeted approaches are achieving impressive results:

• Breast cancer exosome SERS (2024–2025): ML models achieved greater than 94% accuracy for HER2-positive therapy outcome prediction. A dual-modal SERS-fluorescence approach reached 94% classification accuracy for breast cancer subtypes.
• Lung cancer ctDNA detection (2024): A pump-free microfluidic SERS biosensor detected EGFR E746-A750 mutations with 100 femtomolar sensitivity - pushing SERS into the domain traditionally occupied by PCR.
• Pancreatic cancer (2026): A digital SERS nanoprobe immunoassay for pancreatic cancer biomarker detection from clinical blood samples was published in Small Methods.
• Acute leukemia (2025): SERS and machine learning enabled liquid biopsy for early detection and recurrence prediction.

The Lingyan Shi group at UC San Diego - winner of the 2025 Emerging Leader in Molecular Spectroscopy Award - is developing exosome-based Raman spectroscopy with machine learning for noninvasive liquid biopsy diagnostics, a promising approach that combines the specificity of exosome isolation with the molecular fingerprinting capability of Raman.

Raman Spectroscopy for Blood and Serum Cancer Screening

Conventional (non-SERS) Raman spectroscopy of blood and serum is producing clinical-grade results for cancer detection, though with smaller sample sizes than the SERS studies.

The lung cancer serum Raman study (BMC Cancer, 2024) is particularly practical - it used an SVM classifier on serum Raman spectra and also distinguished benign pulmonary lesions from healthy controls (92.22% sensitivity, 95.56% specificity), demonstrating the ability to triage between cancer, benign disease, and normal.

A 2025 multimodal approach combined FTIR, Raman, and excitation-emission matrix (EEM) fluorescence spectroscopy with XGBoost classifiers for breast and colorectal cancer detection, exploiting complementary biochemical information from each spectroscopic technique.

Comparison with Genomic Liquid Biopsy

The field of multi-cancer early detection (MCED) tests is dominated by genomic approaches. Spectroscopy competes - and in several dimensions, wins - against these established platforms.

Three advantages stand out for spectroscopy:

• Speed. Twenty-five minutes versus two weeks. In a clinical workflow where a patient is anxious and a physician needs to make treatment decisions, same-day results change the care pathway. No sample shipping to a central lab. No two-week wait for sequencing.
• Cost. ATR-FTIR spectrometers cost $25K to $80K - a one-time capital expenditure that performs unlimited tests. The per-test consumable cost is negligible (a disposable sample slide). Genomic liquid biopsies require sequencing infrastructure ($500K+) or central lab processing at $500 to $1,000 per test. At scale, spectroscopy is an order of magnitude cheaper.
• Sample volume. Nine microliters of serum. That is a single drop. Genomic tests require several milliliters of blood for adequate cfDNA recovery. The minimal sample requirement makes spectroscopic liquid biopsy viable in settings where blood volume is limited - pediatric patients, neonates, resource-limited clinics.

The disadvantage is coverage breadth: Galleri screens for 50+ cancer types, Dxcover has validated eight. And the genomic approaches have larger clinical datasets and more advanced regulatory submissions.

But Dxcover's 64% Stage I sensitivity at 99% specificity compares favorably to Galleri's approximately 40% overall sensitivity at similar specificity - suggesting that the spectroscopic approach may be more sensitive at early stages, likely because it captures non-genomic biological signals that are present before tumors shed sufficient ctDNA.

The Software Pipeline

Deploying spectroscopy-based liquid biopsy at clinical scale requires a software pipeline that is deceptively complex. Each stage introduces sources of variability that must be controlled.

Sample Preparation to Spectral Acquisition

The "drop, dry, detect" workflow sounds simple, but the drying step introduces physics: as serum dries on the ATR crystal, the "coffee ring effect" creates spatial heterogeneity in the dried film. The spectral measurement location on the dried drop matters. Acquisition protocols must specify exact positioning, and the software must enforce it - either through automated stage positioning or by acquiring spectra at multiple positions and averaging.

Spectral quality control is critical. The software must automatically flag and reject spectra with anomalous features: inadequate signal-to-noise, atmospheric CO₂ interference (a common FTIR artifact), incomplete drying, or crystal contamination. A single bad spectrum entering the classifier can produce a false result.

Preprocessing

This is where spectroscopic liquid biopsy pipelines succeed or fail. A 2025 assessment published in Diagnostics demonstrated that differences in preprocessing substantially affect diagnostic performance - the same raw spectra can yield different clinical conclusions depending on how they are processed. Standardization of preprocessing is the key barrier to reproducibility across laboratories.

A typical preprocessing pipeline includes:

1. Replicate averaging - multiple spectra per sample, averaged to reduce noise
2. Spectral region selection - fingerprint region (900-1800 cm⁻¹), CH-stretch region (2800-3100 cm⁻¹), or both, depending on the clinical application
3. Baseline correction - rubber band, polynomial, or asymmetric least squares
4. Scatter correction - multiplicative scatter correction (MSC) or standard normal variate (SNV)
5. Smoothing - Savitzky-Golay filter
6. Derivative computation - first or second derivative to resolve overlapping peaks
7. Normalization - vector, area, or min-max normalization

Each step has parameters that must be optimized for the specific clinical application and locked down for clinical deployment. The preprocessing pipeline is part of the validated medical device - it cannot be modified without revalidation.

Classification and Clinical Output

Classification algorithms range from traditional chemometrics (PCA-LDA, PCA-SVM, Mahalanobis distance) to deep learning (CNNs on transformed spectral data, autoencoders for feature extraction). Dxcover uses proprietary PANAROMIC algorithms. Open-source toolkits - RamanSPy, PyFasma, SSNet - provide building blocks for research groups developing their own pipelines.

The clinical output must include not just a classification (detected/not detected) but:

• A confidence score
• The cancer type if detected
• The operating point at which the classification was made (the sensitivity-specificity trade-off)

Clinicians need to understand what a positive result means in terms of positive predictive value, which depends on the pre-test probability in their patient population. The software must support tunable thresholds so institutions can set operating points appropriate to their clinical context - screening programs favor sensitivity, symptomatic workup favors specificity.

For the integration layer that connects this pipeline to EHR systems, see our clinical workflow architecture and HL7v2 guide.

Where This Is Heading

Spectroscopic liquid biopsy is at an inflection point. Dxcover has commercial product in the UK, CE-IVDR certification, and a clear path to US market entry. SERS-based platforms have produced the large-cohort clinical data needed to attract clinical trial funding. The question is no longer whether spectroscopy can detect cancer from blood - the published data says it can - but whether it can do so reliably enough, across diverse enough populations, to earn regulatory clearance and clinical adoption.

Near-term expectations (2026–2028):

Dxcover will be the bellwether. Their US launch in 2027 - if it happens on schedule - will be the first spectroscopy-based liquid biopsy available in the world's largest diagnostics market. The CREATE2 colorectal cancer trial and the lung cancer trial, both announced in January 2024, will generate the prospective validation data needed for regulatory submissions beyond brain cancer.

SERS-based multi-cancer detection will move from case-control studies to prospective screening trials. The BMC Medicine dataset of 3,551 patients is large enough to attract the institutional investment needed for multicenter prospective validation.

The standardization bottleneck. The single biggest risk to the field is preprocessing variability. Two labs using the same spectrometer on the same serum sample can get different clinical results if their preprocessing pipelines differ. The 2025 assessment in Diagnostics made this explicit. The field needs consensus preprocessing protocols - the spectroscopy equivalent of standardized PCR cycling conditions - before multi-site clinical deployment is feasible. This is a software problem, not a spectroscopy problem, and it is solvable.

Complementary, not competitive. The most likely near-term deployment model is not spectroscopy replacing genomic liquid biopsy but complementing it. A spectroscopic test as a rapid, low-cost first-line screen - same-day result for a few dollars - followed by genomic confirmation (Galleri, Shield) for positive results. This tiered approach uses spectroscopy's speed and cost advantages while relying on genomic methods' broader cancer-type coverage for confirmatory diagnosis.

The infrastructure needed to deploy spectroscopic liquid biopsy at clinical scale - quality-controlled acquisition, standardized preprocessing, validated classification, HL7/FHIR output - is exactly the kind of clinical workflow software that determines whether a scientifically validated test reaches patients or stays in the journal.

The spectroscopy works. The clinical evidence is accumulating. The remaining challenge is engineering.

Further Reading

• FTIR vs. Raman vs. NIR for Diagnostics - modality comparison for clinical applications
• Building AI Pipelines for Spectral Classification - the ML pipeline behind spectral diagnostics
• Building Clinical Workflow Software for Spectroscopy-Based Diagnostics - system architecture for clinical spectroscopy
• SaMD Classification for Spectroscopy Software - regulatory implications for software as a medical device
• Raman Spectroscopy for Rapid Bacterial ID and AST - another clinical application of spectral diagnostics
• Spectroscopy in the Emergency Department: Rapid Drug Identification - Raman for substance identification