From Molecular Snapshots to Longitudinal Prevention: The Personalized Optical Digital Twin

Imagine a future in which disease is not reconstructed from isolated clinical snapshots but followed as a changing biological trajectory. A patient arriving in an emergency department would not be assessed solely against population thresholds; the patient's current physiology and molecular state could also be interpreted against a personal baseline and the changes that preceded the crisis. Raman spectroscopy of a small blood sample offers one glimpse of that future by reading the biochemical phenotype of the patient's own leukocytes while microbiological results are still pending.

That future will not be built by treating a spectrum as a diagnosis. It begins when a reproducible molecular observation is connected to symptoms, history, laboratory medicine, imaging, microbiology, uncertainty, and outcomes. The published INTELLIGENCE-1 and INTELLIGENCE-2 trials provide an instructive benchmark: high-throughput Raman analysis of leukocytes proved operationally feasible across six centers in Greece and Germany, moving beyond a specialist research laboratory. Yet the Raman score did not add significant diagnostic value to the combined models in the evaluated ICU and emergency-department cohorts. This is not the end of the vision; it is a design lesson for the next generation. Technical feasibility must be coupled to a precisely defined clinical question, coherent biological phenotypes, longitudinal baselines, and evidence that the measurement improves a real decision.¹

Blood is only one window into the changing biology of a person. Fresh tissue can be transformed into label-free, histology-like images; compact nonlinear endomicroscopes can bring molecularly sensitive imaging into an interventional workflow; and future low-burden sensors may add further observations between clinical encounters. The opportunity is not to crown any one instrument as a digital twin, but to connect these molecular snapshots with longitudinal physiology, established clinical data, uncertainty-aware models, governed decisions, and outcomes. That connected learning loop is the Personalized Optical Digital Twin (PODT): a Photonics21 contribution to Europe's Virtual Human Twin (VHT) ecosystem that adds a distinctive longitudinal molecular capability without creating an optical-only or parallel platform.^2,3

Disease develops as a trajectory; clinical information is often episodic

Clinical examination, laboratory medicine, electrocardiography, blood-pressure measurement, imaging, and pathology are indispensable. Their limitation is often temporal: disease, recovery, and treatment response continue between encounters. The opportunity is therefore to add reliable context between visits at a burden and cadence appropriate to the clinical question - without devaluing established care.

Wearables can observe heart rate, oxygen saturation, respiratory rate, temperature, activity, and sleep over long periods. Their most useful contribution is not more data but an individual's baseline and persistent deviation from it. A deviation is not a diagnosis: exercise, stress, alcohol, heat, poor sleep, sensor error, and other factors can change these signals. Persistence, context, data quality, and false-alert burden must be checked before clinical review is recommended.

Molecular photonics can add a complementary biological layer where it improves a defined decision. PODT combines three information layers: longitudinal physiology; molecular observations from photonics, biosensors, biotechnology, and molecular assays; and established clinical and contextual data. Trustworthy AI and Virtual Human Twin models support quality control, personal baselines, multimodal fusion, uncertainty estimation, and interpretable decision support. Clinical need defines the question, the information required defines the modalities, and clinicians decide and act.

Figure 1. PODT as a longitudinal prevention contribution to Europe's Virtual Human Twin ecosystem. Longitudinal physiology, molecular observations, and established clinical and contextual data are fused by trustworthy AI and VHT models. The Observe-Understand-Escalate-Learn loop is clinically governed; observation, diagnosis, and treatment remain separate steps.

Photonics as a molecular observation layer - and where its limits begin

Photonics offers a rare combination: label-free access to multiple molecular constituents, spatial resolution, speed, and compatibility with living samples across several clinically relevant windows. This can make photonics a differentiating observation technology for PODT. It is not exclusive: electrochemical, biochemical, acoustic, imaging, and other modalities remain indispensable, and no technology should be privileged before the clinical information need is clear.

The physical limits are equally important. Light does not image the interior of a person from the outside at molecular resolution. Depending on wavelength and modality, useful penetration ranges from tens of micrometers in the mid-infrared, through millimeters for many optical imaging methods, to centimeters in favorable cases for diffuse near-infrared and photoacoustic approaches. Greater depth generally costs spatial resolution or molecular specificity.

A credible PODT therefore works through multiple accessible windows: skin and interstitial fluid, retina, mucosal surfaces, exhaled breath, biofluids brought to an instrument, and, when clinically justified, endoscopic or intraoperative surfaces. The system is assembled from complementary observations rather than one all-seeing sensor.

For readers of this journal, the toolbox is familiar. Raman and infrared spectroscopy provide label-free molecular information from cells, biofluids, and tissues. Surface-enhanced Raman scattering can improve sensitivity for selected trace targets. Coherent Raman techniques, including coherent anti-Stokes Raman scattering (CARS) and stimulated Raman scattering (SRS), enable fast, chemically specific imaging. Fluorescence-lifetime imaging microscopy (FLIM), second-harmonic generation (SHG), two-photon excited fluorescence (TPEF), optical coherence tomography (OCT), optical-photothermal infrared spectroscopy, and photoacoustics add complementary views.^4,5

Each accessible window contributes a different perspective. The scientific task is not to accumulate every available modality, but to determine which observations add actionable information at which time point, for which person, against which comparator, and at what burden.

What the field can already contribute: observation capabilities

The case for PODT starts with demonstrated measurement capability, but it cannot end there. Three families of work from Jena illustrate how molecular photonics can contribute at different points along a care pathway. They are examples from a much broader European community and occupy different levels of technological and clinical maturity.

Molecular blood analysis: toward longitudinal immune-state observation

Leukocytes are responsive reporters of infection and inflammation. Raman spectroscopy characterizes their composite biochemical phenotype without labels. It does not directly measure every cytokine or functional process, but it can detect reproducible spectral patterns associated with cell type and activation state. This may complement a conventional differential blood count, which reports cell proportions but only limited information about the cells' biochemical state.

In a prospective single-center study of 61 hospitalized patients, Raman analysis of peripheral blood leukocytes differentiated sterile inflammation, infection, and sepsis; combining Raman-derived scores with conventional biomarkers improved cross-validated classification performance in that cohort.⁶ High-throughput measurements in patients with COVID-19- and non-COVID-19-associated sepsis documented biochemical differences from acute disease to recovery at 6 and 12 months.⁷ The subsequent prospective multicenter INTELLIGENCE evaluation included 279 patients at six centers in two countries, measured approximately 1,500 leukocytes per patient within one hour, and showed that the Raman score did not add significant diagnostic value to combined clinical models in the assessed ICU and emergency-department cohorts.¹ Taken together, the studies demonstrate increasingly mature measurement capability, but not yet a validated clinical decision test. Independent prospective evaluation, careful cohort definition, and control of pre-analytical and demographic confounders remain essential.

The published multicenter result sharpens the translational question. The immediate opportunity is not to present leukocyte Raman spectroscopy as a validated sepsis classifier, but to determine whether it adds value for more coherent biological phenotypes, infection versus non-infection, serial immune-state changes, treatment-response monitoring, or recovery trajectories. Each use case requires predefined comparators and endpoints, patient-level validation, calibrated uncertainty, and evidence that the result improves a clinician-led decision.¹

Figure 2. Published multicenter evidence for Raman analysis of blood-derived leukocytes. INTELLIGENCE-1 and INTELLIGENCE-2 included 279 patients from six centers in two countries; approximately 1,500 leukocytes per patient were measured within one hour in clinical laboratory settings. The study demonstrated operational feasibility beyond the research laboratory, but the Raman score did not add significant diagnostic value to combined models in the assessed ICU and emergency-department cohorts. Technical feasibility and clinical utility must therefore be evaluated separately.¹ Schematic based on Giamarellos-Bourboulis et al.

From Raman2Go to RamanWatch: a staged development path

The route toward longitudinal molecular sensing is staged. The current Raman2Go research workflow still relies on external centrifugation, separate white-blood-cell preparation, manual loading onto a measurement chip, and an episodic ex vivo measurement. The immediate translational task is to make this end-to-end workflow robust, automated, quality-controlled, and clinically integrated - not simply to place the current chip in a smaller instrument.

A possible homecare stage would require a sealed sample-to-answer cartridge for finger-prick capillary blood and an untrained user. The cartridge would meter the sample, enrich leukocytes, position them in an optical window, and contain waste; a compact reader would perform quality control, imaging, and Raman spectroscopy. Its first output should be a quality-controlled molecular trend relative to a personal capillary-blood baseline, not an autonomous diagnosis. Capillary and venous blood must not be assumed interchangeable; paired validation and, if necessary, capillary-specific models will be required.

RamanWatch - a skin-contact module for continuous molecular trend sensing - remains a long-term research vision, not an available product. Transcutaneous specificity, photon budget, motion robustness, calibration, safety, and clinical benefit still have to be demonstrated. The development logic is therefore: close and validate the ex vivo workflow, enable repeated low-burden home measurements, and only then explore continuous non-invasive observation.

Molecular tissue imaging: rapid, label-free information at a clinical event

A second capability turns the microscope into a molecular imager. Multimodal nonlinear imaging combines contrast mechanisms that are individually informative and jointly powerful: CARS reports lipid- and protein-rich structures in the CH-stretch region; SHG reveals ordered collagen; TPEF and FLIM provide endogenous fluorescence and lifetime contrasts associated with tissue composition and cellular metabolism. Together, these channels can reconstruct tissue architecture and biochemical contrast without conventional dyes.

Two-color coherent Raman approaches acquire signals near the CH2 and CH3 stretching bands and transform them into a familiar virtual hematoxylin-and-eosin appearance. Stimulated Raman histology has demonstrated the clinical potential of this strategy in fresh, unprocessed neurosurgical specimens.⁸ In our own pilot work, CARS measurements on thick human biopsy material were mapped to an H&E-like display, as shown in Figure 3. The acquisition, preprocessing, and color transformation must be reported transparently because the familiar colors are computational output, not histochemical stains. This is valuable event-based information; a single virtual histology image is not itself a digital twin.

Figure 3. Pilot coherent Raman histology from a bulk biopsy. CARS signals near the CH2 and CH3 stretching bands are mapped to an H&E-like display that resolves nuclei, cell bodies, connective tissue, lipid-rich structures, and red blood cells without sectioning or staining. Scale bar 50 micrometers. Data from the authors' pilot analysis.

Image-guided intervention: a procedural feedback loop

Compact multimodal nonlinear endomicroscopes can bring high-resolution imaging to the tip of a rigid probe. A system with a 6-mm-diameter probe and a scan head only a few centimeters across has achieved approximately 1-micrometer lateral resolution while combining CARS, TPEF, SHG, and indocyanine-green fluorescence; in-vivo imaging and clinical deployment remain future steps.⁹

Coupled to femtosecond-laser ablation and automated image analysis, related instruments have demonstrated a 'See & Treat' workflow in experimental tissue settings.^10,11 In a 15-patient preclinical head-and-neck cohort, multimodal images were analyzed with deep-learning segmentation, and the same platform was used to demonstrate selective image-guided ablation as a proof of concept.¹⁰ These results support further clinical translation; they do not justify replacement of frozen-section pathology or autonomous treatment. Endomicroscopy can generate high-value event data for a longitudinal loop, but it is not a PODT on its own.

Figure 4. See & Treat proof of concept. Multimodal nonlinear images (CARS, TPEF, SHG) are acquired before and after selective femtosecond-laser ablation. Image processing produces an ablation mask for experimental, frame-by-frame tissue removal. Scale bars 100 micrometers. The workflow is preclinical and requires prospective validation.

None of these demonstrations provides continuous everyday molecular monitoring. Leukocyte analysis requires a blood sample; coherent Raman histology requires tissue; endomicroscopy requires an interventional procedure. A credible near-term PODT therefore uses a hybrid cadence: continuous or frequent low-burden physiological measurements where they add value, periodic molecular measurements, event-triggered tests, and existing clinical data. Genuinely continuous molecular sensing is a longer-term objective, not a prerequisite for beginning longitudinal care pathways.

From measurements to a longitudinal prevention loop

A collection of measurements becomes more than a dashboard only when it is organized around an individual and a decision. The model must maintain a personal baseline, identify persistent change with calibrated uncertainty, use clinical context to interpret that change, and update its expectations as new measurements and outcomes arrive. A useful operational loop is: Observe relevant change. Understand it against baseline and context. Escalate through a clinically governed pathway when warranted. Learn from outcomes to improve subsequent monitoring.

Two guardrails are non-negotiable. First, a longitudinal change is not a diagnosis; it may justify a repeated measurement, planned evaluation, or prompt clinical attention depending on the use case and uncertainty. Second, PODT supports healthcare professionals; it does not replace clinical responsibility. Observation, diagnosis, and treatment are separate steps. Treatment must never follow automatically from a sensor or algorithm.

Trustworthy AI begins with trustworthy measurement. Artificial intelligence can support quality control, estimate personal baselines, detect multimodal change, and model aspects of disease trajectory or treatment response. Its output must be interpretable at the decision point and accompanied by uncertainty. Training populations, missingness, dataset shift, false-positive burden, calibration, and performance over time must be governed and monitored.

For spectroscopy, a quieter problem comes first: turning a raw signal into a reliable molecular quantity. In optically complex samples, simple intensity-concentration assumptions can fail. Physics-informed and, where appropriate, complex-valued chemometrics can represent coupled absorption and dispersion, use available information more completely, and provide plausibility checks when a measurement leaves its validated domain.⁵ A longitudinal system also needs open ontologies, interoperable interfaces, provenance, auditable model updates, and a legally and technically governed privacy architecture. PODT should strengthen VHT, EDITH (the EU's Virtual Human Twin coordination and support action), and the European Health Data Space rather than duplicate them.³, ¹²

The honest part: PODT is a system innovation, not a sensor program

System innovations fail at their seams. Figure 5 summarizes four technical challenge families. Across all of them, the program also requires clinically led use cases, comparative evidence, workflow integration, regulation, reimbursement, patient acceptance, equitable access, and governed adoption. Success depends on integration and evidence - not on one isolated technology.

Figure 5. Core technology and implementation challenges for PODT: integrated advanced photonic sensing; reliable fusion with physiological data; trustworthy, explainable AI; and scalable, secure, usable systems. Clinical need, comparative evidence, governance, and equitable access cut across all four domains.

Integrating advanced photonic sensing

Many laboratory demonstrations remain large, expensive, and expert-operated. Translation requires compact, robust platforms; stable light-tissue interfaces; sufficient optical throughput; calibration transfer; and miniaturization without unacceptable loss of specificity. Photonic integrated circuits, CMOS-compatible processes, scalable packaging, and accessible pilot manufacturing will be important, but integration should be driven by a defined clinical decision point rather than by the availability of a device.

Fusing heterogeneous observations reliably

Multimodal fusion is a measurement-quality problem before it is a software problem. Unsupervised wearables and remote systems can generate plausible but erroneous data. Robust validation, redundancy where justified, metadata about measurement conditions, missing-data handling, and modality-specific quality flags are needed before an algorithm compares a person with their baseline. Preventive systems must explicitly control false alarms and the downstream burden they create.

Reliable, explainable, and governed AI

Because PODT outputs may influence prioritization and therapy monitoring, AI requires lifecycle governance, traceability, bias assessment across diverse populations, continuous performance monitoring, human oversight, and clear fallback states. The clinically relevant question is not whether a model can generate a score, but whether that score improves a defined decision compared with current care.

Scalable, secure, usable, and accessible systems

Devices and services must be energy-efficient, interoperable, maintainable, cybersecure, and usable in homes, clinics, and mobile settings. Intended purpose determines whether the Medical Device Regulation (MDR), the In Vitro Diagnostic Regulation (IVDR), and related AI requirements apply. Audit trails must distinguish model output from human decisions. Affordability, regional access, digital literacy, and alternatives for people who cannot or do not wish to use premium wearables are program design requirements, not afterthoughts.

Every entry-point use case needs a measurable decision point, an appropriate comparator, prospective multi-site validation, health-economic analysis, and predefined criteria for stopping or narrowing deployment. The value proposition is better clinical decisions - not more data. Benefits such as earlier recognition, better-timed diagnostics, therapy monitoring, and recovery support must be demonstrated rather than assumed.

Europe's longitudinal prevention mission

Europe already has strong assets in Virtual Human Twins, EDITH, health-data infrastructure, clinical networks, computing, medical technology, and trustworthy AI. PODT proposes to connect those assets and add what is still underdeveloped: longitudinal personal baselines, low-burden real-world observation, molecular photonics and biotechnology, uncertainty-aware multimodal fusion, and clinically governed prevention loops. It should not create a parallel proprietary platform.

No single laboratory, company, clinic, or country can establish the required measurement standards, longitudinal reference datasets, open interfaces, manufacturing capacity, multi-site evidence, reimbursement pathways, and public trust. The mission must be co-designed by patients, clinicians, health systems, researchers, regulators, payers, technology providers, and manufacturers. Photonics contributes a distinctive molecular capability, but clinical need remains the organizing principle.

The route to FP10 (the EU's 10th Framework Programme for Research and Innovation) should be staged in the same spirit: validate the mission and its boundaries, then select two or three anchor use cases with defined comparators, endpoints, and interfaces to VHT, EDITH, and the European Health Data Space. The funding model should follow the evidence requirements, not precede them.

Now is the time to connect the observations

Return to the patient in the emergency department. The clinically meaningful outcome is not a spectrometer declaring a diagnosis. It is a better-timed, better-evidenced decision: a reliable observation is compared with the patient's baseline and context, uncertainty is visible, and the result enters a governed pathway. Later outcomes improve the next observation and the next decision.

For the spectroscopy community, the ambition is both narrower and more consequential than building the entire twin. We must deliver molecular observations with the quality, burden, cadence, provenance, and interoperability that longitudinal prevention requires. PODT will exist when those observations are connected to personal baselines, clinical context, escalation, and learning - not when another instrument produces another isolated signal. The task is to build on what Europe already has, add the missing longitudinal capability, and validate what actually matters.

References

(1) Giamarellos-Bourboulis, E. J.; Thomas-Rüddel, D.; Pistiki, A.; et al. Blood Biomarker and Raman Analysis of Leukocytes in the Multicenter INTELLIGENCE Trials: At the Intensive Care Unit (INTELLIGENCE-1) and the Emergency Department (INTELLIGENCE-2). Shock 2026, 65 (1), 15–21. DOI: 10.1097/SHK.0000000000002699.

(2) Photonics21. Investing in Light: The Vision to Secure Europe’s Competitiveness in the Global Tech Race. Position Paper, Photonics21, 2025. Annex: Personalised Optical Digital Twin: Photonic Health Monitoring for Europe’s Preventive and Precision Medicine. https://photonics21.org/download/ppp-services/photonics-downloads/Photonics21_Position_Paper_Investing_in_Light_The_Vision_to_Secure_Europes_Competitiveness_in_the_Global_Tech_Race.pdf (accessed September 1, 2026).

(3) European Commission. European Virtual Human Twins Initiative. Shaping Europe’s Digital Future, 2025. https://digital-strategy.ec.europa.eu/en/policies/virtual-human-twins (accessed September 1, 2026).

(4) Ban, S.; Yi, H.; Park, J.; Huang, Y.; Yu, K. J.; Yeo, W.-H. Advances in Photonic Materials and Integrated Devices for Smart and Digital Healthcare: Bridging the Gap between Materials and Systems. Adv. Mater. 2025, 37 (49), 2416899. DOI: 10.1002/adma.202416899.

(5) Taccheo, S.; Mayerhöfer, T. G.; Doron, M.; et al. Pathway to Versatile, Point of Care, and Wearable Photonics in the Mid-Infrared and Fingerprint Region Based on Quantum-Cascade Lasers and Analytical and Computational Advances. Opt. Mater. Express 2026, 16 (3), 662–691. DOI: 10.1364/OME.580918.

(6) Ramoji, A.; Thomas-Rüddel, D.; Ryabchykov, O.; et al. Leukocyte Activation Profile Assessed by Raman Spectroscopy Helps Diagnosing Infection and Sepsis. Crit. Care Explor. 2021, 3 (5), e0394. DOI: 10.1097/CCE.0000000000000394.

(7) Ramoji, A.; Baumbach, P.; Ryabchykov, O.; et al. Raman Spectroscopy Can Identify Acute and Persistent Biochemical Changes in Leukocytes from Patients with COVID-19 and Non-COVID-19-Associated Sepsis. Biotechnol. J. 2025, 20 (9), e70105. DOI: 10.1002/biot.70105.

(8) Orringer, D. A.; Pandian, B.; Niknafs, Y. S.; Hollon, T. C.; Boyle, J.; Lewis, S.; Garrard, M.; Hervey-Jumper, S. L.; Garton, H. J. L.; Maher, C. O.; Heth, J. A.; Sagher, O.; Wilkinson, D. A.; Snuderl, M.; Venneti, S.; Ramkissoon, S. H.; McFadden, K. A.; Fisher-Hubbard, A.; Lieberman, A. P.; Johnson, T. D.; Xie, X. S.; Trautman, J. K.; Freudiger, C. W.; Camelo-Piragua, S. Rapid Intraoperative Histology of Unprocessed Surgical Specimens via Fibre-Laser-Based Stimulated Raman Scattering Microscopy. Nat. Biomed. Eng. 2017, 1, 0027. DOI: 10.1038/s41551-016-0027.

(9) Lai, C.; Calvarese, M.; Reichwald, K.; Bae, H.; Vafaeinezhad, M.; Meyer-Zedler, T.; Hoffmann, F.; Mühlig, A.; Eidam, T.; Stutzki, F.; Messerschmidt, B.; Gross, H.; Schmitt, M.; Guntinas-Lichius, O.; Popp, J. Design and Test of a Rigid Endomicroscopic System for Multimodal Imaging and Femtosecond Laser Ablation. J. Biomed. Opt. 2023, 28 (6), 066004. DOI: 10.1117/1.JBO.28.6.066004.

(10) Calvarese, M.; Corbetta, E.; Contreras, J.; Bae, H.; Lai, C.; Reichwald, K.; Meyer-Zedler, T.; Pertzborn, D.; Mühlig, A.; Hoffmann, F.; Messerschmidt, B.; Guntinas-Lichius, O.; Schmitt, M.; Bocklitz, T.; Popp, J. Endomicroscopic AI-Driven Morphochemical Imaging and fs-Laser Ablation for Selective Tumor Identification and Selective Tissue Removal. Sci. Adv. 2024, 10 (50), eado9721. DOI: 10.1126/sciadv.ado9721.

(11) Meyer, T.; Ackermann, R.; Kammel, R.; Schmitt, M.; Nolte, S.; Tünnermann, A.; Popp, J. CARS-Imaging Guidance for fs-Laser Ablation Precision Surgery. Analyst 2019, 144 (24), 7310–7317. DOI: 10.1039/C9AN01545K.

(12) European Parliament; Council of the European Union. Regulation (EU) 2025/327 of the European Parliament and of the Council of 11 February 2025 on the European Health Data Space and Amending Directive 2011/24/EU and Regulation (EU) 2024/2847. Off. J. Eur. Union 2025, L 2025/327. https://eur-lex.europa.eu/eli/2025/327/oj (accessed September 1, 2026).

About the Authors

Jürgen Popp and his interdisciplinary team work at the interface of spectroscopy, biomedical research, and clinical translation. Popp is Scientific Director of the Leibniz Institute of Photonic Technology (Leibniz IPHT) and Professor of Physical Chemistry at Friedrich Schiller University Jena, Germany. His research focuses on translating Raman spectroscopy and multimodal optical imaging into biomedical tools for infection diagnostics, oncology, and image-guided intervention.

A central theme of his work is the development of complete diagnostic workflows – from molecular contrast mechanisms and photonic instrumentation to automated data analysis, clinical validation, and transfer into practical systems. He has initiated major international collaborations linking spectroscopy, artificial intelligence, clinical research, and technology transfer. His work has also contributed to patents, spin-offs, and compact Raman-based diagnostic platforms.

Popp is one of the scientific driving forces behind the Personalized Optical Digital Twin (PODT), a Photonics21 initiative currently under development. PODT aims to bring together molecular photonics, physiological sensing, clinical information, and artificial intelligence to enable longitudinal, personalized models of human health. Its long-term ambition is to develop into a European health flagship that translates Europe’s deep-tech capabilities into clinically validated solutions for prevention, earlier detection, therapy monitoring, and personalized intervention.

His contributions have been recognized with the Charles Mann Award for Applied Raman Spectroscopy, the SPIE Biophotonics Technology Innovator Award, and the Ellis R. Lippincott Award.

The co-authors represent the wider interdisciplinary team behind the article. Thomas Mayerhöfer and Michael Schmitt are scientists affiliated with Leibniz IPHT and Friedrich Schiller University Jena. Andrea Borowsky, Lavinia Meier-Ewert, and Gabriele Hamm contribute expertise in international relations and science communication at Leibniz IPHT.