Pith. sign in

REVIEW 4 major objections 6 minor 63 references

Multi-spectral infrared imaging plus deep learning can turn non-ionizing limb scans into synthetic radiographs for radiation-free pediatric fracture triage.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-31 06:38 UTC pith:Q2BBTRCE

load-bearing objection Useful clinical framing of a multi-band IR+AI triage idea, but the load-bearing “fNIRS proves structural radiograph synthesis” leap overclaims what the cited physics actually shows. the 4 major comments →

arxiv 2607.24727 v1 pith:Q2BBTRCE submitted 2026-07-27 cs.CV

Infrared Imaging Empowered by Artificial Intelligence for Pediatric Skeletal Triage: A Narrative Review and Future Perspectives

Sajad Amiri , Pardis Afshar , Elham Anjomshoa This is my paper
classification cs.CV
keywords infrared imagingnear-infrared spectroscopyimage-to-image translationpediatric radiologyfracture diagnosisnon-ionizing imagingpoint-of-care triagedeep learning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

Pediatric emergency departments still lean on X-rays for common limb injuries, even though repeated low-dose radiation in childhood raises lifetime cancer risk. This narrative review argues that broad-spectrum infrared light—already known to pass through skin and bone in bedside brain monitoring—can be collected from thinner pediatric limbs in both transmission and reflection, then translated by modern image-to-image networks into radiograph-like pictures clinicians already know how to read. The authors ground feasibility in everyday fNIRS and photobiomodulation practice: if near-infrared light can round-trip the adult skull and still carry useful signal, illuminating a child’s forearm or distal leg is a milder optical problem. They sketch a full pipeline—multi-band capture, deep feature matching, cross-modal synthesis, and a confidence gate that escalates uncertain cases to real X-ray—and name the practical blockers: paired IR/X-ray datasets, device regulation, and fairness across body size and skin tone. If the argument holds, a portable, non-ionizing first-pass triage tool becomes a realistic complement to radiography and ultrasound rather than a speculative physics leap.

Core claim

The paper’s central claim is that coupling multi-spectral infrared acquisition across five windows (NIR-I through THz) with deep cross-modal translation is a credible path to radiation-free pediatric skeletal triage, because clinical near-infrared practice already recovers usable signal through thicker, denser paths than a pediatric limb, so the framework introduces no new light–tissue physics—only a redirection from hemodynamic sensing to structural reconstruction rendered as synthetic radiographs.

What carries the argument

Dual-geometry multi-spectral IR fusion plus confidence-gated image-to-image translation: transmission and reflection captures across NIR/SWIR/MIR/THz are aligned with deep matchers, then a generator (cGAN, Swin-Unet, or diffusion) trained on paired IR/X-ray data emits a synthetic radiograph and uncertainty map that routes only high-certainty cases to clinician triage.

Load-bearing premise

That infrared light passing through a child’s limb carries enough detailed bone-shape and fracture structure—not just bulk dimming or blood-flow signals—for AI to rebuild X-ray-faithful pictures clinicians can trust.

What would settle it

Build a paired multi-center IR and clinical X-ray dataset on pediatric distal limbs, train the proposed translation model, and test whether synthetic radiographs recover fracture presence and location at clinically useful sensitivity and specificity against real radiographs, with calibrated uncertainty correctly flagging failures.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Pediatric ED rule-out exams could start with a non-ionizing IR+AI pass, sending only positive or uncertain cases to X-ray or ultrasound.
  • Compact multi-LED/InGaAs hardware could deploy on ambulances, school clinics, and low-resource settings without lead shielding or dosimetry.
  • Ionizing exposure for dataset building would be limited to clinically indicated X-rays; deployment uses IR alone.
  • Safety qualification under IEC 60825-1 Class 1/1C becomes the main source-safety bar instead of radiation dose limits.
  • Progress hinges on federated paired IR/X-ray datasets that span age, habitus, and skin pigmentation before external clinical validation.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The information-content gap between fNIRS-style oxygenation sensing and fracture-line geometry is the make-or-break empirical question; success would likely need NIR-II/SWIR structural contrast far beyond what banana-path hemodynamics provide.
  • If confidence gating works, the system’s clinical value may rest as much on reliable escalation as on perfect synthesis—acting as a high-NPV rule-out filter rather than a full radiograph replacement.
  • The same pipeline could later target other thin pediatric sites (e.g., clavicle, digits) before any adult thick-limb attempt, matching the paper’s ‘pediatrics first’ logic.
  • Hallucinated cortical detail is a shared risk with other generative medical translators; mandatory uncertainty maps make this proposal a natural test bed for safe deployment patterns in SaMD.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 6 minor

Summary. This narrative review proposes a radiation-free pediatric skeletal-triage framework in which multi-band infrared (IR) data — acquired in transmission (NIR-I/NIR-II) and reflection (SWIR, MIR/LWIR, THz) geometries — are fused by feature-matching networks and translated by image-to-image models (cGANs, Swin-Unet, diffusion) into synthetic radiographs with uncertainty estimates and a confidence gate that escalates low-certainty cases to conventional X-ray. The clinical motivation (cumulative pediatric radiation risk) is well cited. The pivotal feasibility argument (§2.2, restated in §5.1 and the Conclusion) is that because fNIRS and transcranial photobiomodulation show NIR light traversing adult skin, skull, and cortex, transillumination of the thinner pediatric limb "introduces no new physics" and is "technologically modest by comparison." The review then covers wavelength-specific roles, the AI pipeline, dataset/regulatory roadmap, limitations (bias, hallucination, BMI effects), and a comparison with point-of-care ultrasound.

Significance. If the central framing is corrected, the manuscript is a useful, well-organized synthesis of three literatures (IR–tissue biophysics, cross-modal I2I translation, multi-spectral feature matching) around a genuinely important problem. To its credit, the manuscript is not naive about failure modes: it builds in calibrated uncertainty and a mandatory escalation pathway (Graphical Abstract, Fig. 3, §5.2), names skin-pigmentation and BMI bias explicitly (§4.2, §5.2), positions the method as complementary to POCUS and photoacoustics rather than a replacement (§5.3), and ends with falsifiable next steps — paired multi-center IR/X-ray datasets, external validation, and prospective comparison against radiography and POCUS. The regulatory roadmap (IEC 60825-1/60601-1, FDA adaptive-AI guidance, CLAIM) is unusually concrete for a perspective piece. These features make the review potentially valuable as a research agenda — but only if the load-bearing feasibility claim is re-stated honestly, because as written it invites readers to overestimate what the cited evidence shows.

major comments (4)
  1. [§2.2 / §5.1 / Conclusion] §2.2 (restated in §5.1, ¶'the optical feasibility...' and in the Conclusion): the core argument conflates photon penetration with spatially resolved structural information. fNIRS demonstrates that NIR photons survive a diffuse round trip and return a *bulk hemodynamic* signal integrated over a banana-shaped volume at ~2–3 cm resolution — a figure the manuscript itself concedes in §2.2. Fracture triage requires recovering millimeter-scale cortical discontinuities and growth-plate geometry through several cm of scattering soft tissue. In the diffuse regime the paper invokes via the μ_eff equation in §2.1, ballistic photons at limb depths are negligible and transillumination resolution scales as a substantial fraction of the slab thickness; for a ~4–5 cm pediatric forearm this implies centimeter-scale blurring, not cortical edges. 'Penetration envelope' is therefore the wrong metric: the sk
  2. [§2.1–2.3, Table 1] The citations offered as evidence that IR encodes recoverable *structural* bone information do not demonstrate that claim. (a) §2.1 and §2.3 attribute 'cortical-bone shadowing and gross trabecular geometry' in transmission and 'sharper images of cortical bone' in NIR-II to refs [12, 19]; ref [12] (Mi et al., Nat. Commun. 2023) is NIR-II *fluorescence* imaging of bone disease in small animals using an exogenous probe — label-free, through-limb structural imaging of human cortical bone is not shown, and ref [19] is in vivo calcium imaging with a fluorescent indicator. (b) §2.3 invokes ref [63] (NIR spectroscopy of intact bovine teeth) as 'an encouraging analog for learning skeletal features'; that is surface compositional spectroscopy of hydroxyapatite chemistry, not spatial reconstruction of buried structure, and the hydroxyapatite-chemistry argument does not bear on spatial information c
  3. [§3.1, §3.4, §4.1, refs [30, 38, 39]] Several citations in the AI-methods section do not support the claims they are attached to, which matters because one of them anchors the paper's central safety mechanism. (a) §3.4 and §5.2 cite ref [38] for uncertainty estimation and anatomical-consistency post-processing; ref [38] is Chang et al., 'Multispectral visible and infrared imaging for face recognition' (CVPRW 2008), which contains neither. (b) §3.1 attributes 'bone-mask losses and perceptual losses anchored on skeletal atlases' to ref [30], a short WIECON-ECE conference paper titled simply 'Image-to-image translation with conditional adversarial networks' that does not appear to contain those loss terms. (c) §4.1 cites ref [39] (Parri et al., ultrasound detection of pediatric skull fractures) for the claim about 'rule-out fracture' examinations not requiring ionizing imaging — a mismatch. Since the confidence-gate/uncertainty
  4. [§4.1] §4.1 states that growth plates 'produce characteristic radiographic features that may be more recoverable from optical data.' No citation or mechanism is given for why cartilaginous growth plates — whose X-ray appearance is a lucency between ossified structures — would carry a distinctive *optical* signature, and this cuts against the manuscript's own observation that the synthetic radiograph must ultimately reproduce radiographic, not optical, contrast. This sentence should either be supported or removed; as written it is an ungrounded assertion in the section arguing 'why pediatrics first.'
minor comments (6)
  1. [§1.1] §1.1: the search window 'January 2000 – April 2026' includes a future end date; please confirm and, given the 'narrative review' designation, state explicitly how many sources were screened and whether any inclusion/exclusion criteria beyond 'prioritized by relevance' were applied.
  2. [§2.3, Table 1] §2.3 and Table 1: the THz window is given as '15 µm – 1 mm'; 15 µm is conventionally still LWIR, and THz is usually taken as ~30 µm–1 mm (0.1–10 THz). Please check the band boundaries for internal consistency with the MIR/LWIR row ('3–15 µm').
  3. [§3.1] §3.1: diffusion models are introduced as 'a third architecture family' with calibrated uncertainty but no citation; given the weight placed on uncertainty elsewhere, a reference (e.g., to diffusion-based medical image synthesis) should be added.
  4. [§5.2] §5.2: the transferred fNIRS mitigation strategies (short-separation channels, multi-distance layouts) assume a reflection geometry; for the proposed transmission geometry the analogy is incomplete — a sentence clarifying which strategies apply to which geometry would help.
  5. [§2.3, ref [13]] Ref [13] (Jeffery et al., systemic effects of longer-wavelength sunlight) is cited in §2.3 for NIR-I 'centimeter-scale soft-tissue transillumination... volumetric mapping of muscle, vasculature'; the source concerns photobiomodulation-type systemic effects, not imaging — please verify the attribution.
  6. [Figures] The Graphical Abstract and Figure 3 captions describe the same confidence-gated pipeline; consider merging or differentiating them to avoid redundancy.

Circularity Check

0 steps flagged

No circular derivation: narrative synthesis proposes a future IR+AI triage framework from external literatures without fitting parameters or self-justifying premises.

full rationale

This is a narrative review and future-perspectives piece, not a derivation paper. It does not fit free parameters, does not redefine its conclusion as an input equation, and does not rest load-bearing uniqueness or feasibility claims on self-citation. The central feasibility argument (NIR already traverses skin/skull/cortex in fNIRS and photobiomodulation, so a thinner pediatric limb is within the demonstrated penetration envelope) cites independent external literatures [55–62] and standard I2I/feature-matching architectures (Pix2Pix, CycleGAN, Swin-Unet, SuperPoint, LightGlue) from unrelated authors. The proposed pipeline explicitly uses X-ray only to build paired training labels and IR alone at deployment—an ordinary supervised-learning setup, not a circular construction. Author list (Amiri, Afshar, Anjomshoa) does not overlap the cited foundational works. Any weakness is an information-content leap (bulk hemodynamic sensing vs. millimeter skeletal geometry), which is a correctness/evidence-gap concern, not circularity under the stated rules. Score 0; steps empty.

Axiom & Free-Parameter Ledger

0 free parameters · 5 axioms · 2 invented entities

The central feasibility claim rests on domain optical assumptions imported from fNIRS/biomedical optics, on the untested premise that IR contrast is information-rich enough for radiograph synthesis, and on standard deep I2I inductive biases—not on fitted constants. No free parameters are estimated because there is no quantitative model fit. The main invented construct is the end-to-end clinical system (dual-geometry multi-band IR + matcher + translator + confidence gate) as a deployable triage device.

axioms (5)
  • domain assumption Effective attenuation μ_eff = sqrt[3 μ_a (μ_a + μ_s')] governs diffuse penetration depth in tissue; the 650–1350 nm window permits centimeter-scale paths in pediatric limbs.
    Invoked in §2.1 with standard biomedical-optics citation; used to justify NIR/SWIR transmission through pediatric extremities.
  • ad hoc to paper Successful fNIRS/transcranial NIR through adult scalp+skull+cortex implies that structural IR imaging of thinner pediatric cortical bone is within the same penetration envelope.
    Core feasibility bridge in §2.2 and Conclusion; equates path length/attenuation success with recoverability of skeletal structure for X-ray-like synthesis.
  • domain assumption Paired supervised or cycle-consistent I2I networks (cGAN, Swin-Unet, diffusion) can map multi-spectral IR stacks to radiograph-equivalent images without clinically dangerous hallucination if uncertainty gating is used.
    §3.1–3.4; standard ML transfer assumption from other modality-translation tasks, not validated here for IR→X-ray bone.
  • domain assumption Deep local features (SuperPoint, SuperGlue, ALIKED, LightGlue) can align multi-band IR frames under pediatric motion and radiometric band differences at subpixel accuracy.
    §3.3; imported from general CV matching literature.
  • domain assumption Class 1 / 1C operation under IEC 60825-1 yields acceptable ocular/dermal risk for routine pediatric non-contact use.
    §2.4 and regulatory roadmap §4.4; standard laser-safety premise.
invented entities (2)
  • Confidence-gated multi-spectral IR→synthetic-radiograph pediatric triage system no independent evidence
    purpose: End-to-end clinical device concept that fuses dual-geometry multi-band IR, deep matching, cross-modal translation, uncertainty maps, and escalation to real X-ray.
    Introduced in Graphical Abstract and Figure 3 as the paper's proposed framework; no working implementation or external clinical evidence exists in the manuscript.
  • Dual-geometry (transmission + reflection) multi-band IR acquisition scheme for limbs no independent evidence
    purpose: Separate deep cortical/structural cues (NIR-I/II transmission) from superficial biochemical maps (SWIR/MIR reflection) as priors for synthesis.
    Proposed in §2.1 and Figure 2; geometries are known in optics, but the specific fused skeletal-triage protocol is paper-specific and untested.

pith-pipeline@v1.2.0-grok45-kimik3 · 24233 in / 3687 out tokens · 78535 ms · 2026-07-31T06:38:06.502395+00:00 · methodology

0 comments
read the original abstract

Background. Pediatric musculoskeletal trauma represents up to 18% of pediatric ED visits, yet diagnosis still depends on ionizing radiography. Cumulative low-dose radiation in early life raises lifetime leukemia and brain malignancy risk, motivating radiation-free triage alternatives. Objective. To synthesize evidence for a hybrid framework coupling broad-spectrum infrared (IR) imaging with deep-learning cross-modal translation to generate clinically interpretable synthetic-radiograph reconstructions from non-ionizing data. Approach. We review five IR spectral windows spanning 650 nm to 1 mm - NIR-I, NIR-II, SWIR, MIR/LWIR, and THz - and how dual-geometry (transmission/reflection) acquisition exploits wavelength-specific tissue depth and biochemical sensitivity. We summarize image-to-image translation networks (Pix2Pix, CycleGAN, Swin-Unet) and feature-matching algorithms (SuperPoint, SuperGlue, ALIKED, LightGlue) used to align and fuse IR data into radiograph-equivalent reconstructions. Implications. Pediatric anatomy - smaller cross-sections, thinner cortical bone - favors IR penetration, enabling compact, portable, non-ionizing triage hardware. Feasibility is grounded in fNIRS and transcranial photobiomodulation evidence: near-infrared light passes through skin, skull, and cortex with sufficient signal for hemodynamic monitoring - a longer, more attenuating path than through a pediatric forearm or distal leg. Key barriers: paired IR/X-ray dataset construction, AI-as-medical-device regulatory pathways, generalization across body habitus and skin pigmentation, and acquisition-protocol standardization. Conclusions. Integrated multi-spectral IR+AI imaging is a promising radiation-free complement to pediatric skeletal radiography. Progress requires multi-center paired datasets, externally validated models, and IR source safety qualification under IEC 60825-1.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

63 extracted references · 34 canonical work pages

  1. [1]

    Epidemiology of musculoskeletal pain in a pediatric emergency department,

    M. P. Mintegi-Raso et al., “Epidemiology of musculoskeletal pain in a pediatric emergency department,” Rheumatology International, vol. 35, no. 11, pp. 1791–1797, 2015, doi: 10.1007/s00296-015-3335-9

    work page doi:10.1007/s00296-015-3335-9 2015

  2. [2]

    Rational usage of fracture imaging in children and adolescents,

    R. Kraus and K. Dresing, “Rational usage of fracture imaging in children and adolescents,” Diagnostics (Basel), vol. 13, no. 3, p. 538, 2023, doi: 10.3390/diagnostics13030538

    work page doi:10.3390/diagnostics13030538 2023

  3. [3]

    P. Langevin et al., “Direct-access physiotherapy to improve access to quality care for children and adolescents presenting to the pediatric emergency department with musculoskeletal problems: the PEDPTMSK pilot randomized control trial,” Journal of Orthopaedic and Sports Physical Therapy, vol. 55, pp. 440–450, 2025, doi: 10.2519/jospt.2025.13321

    arXiv 2025

  4. [4]

    Epidemiology of fractures in children and adolescents,

    R. M. Hedström et al., “Epidemiology of fractures in children and adolescents,” Acta Orthopaedica, vol. 81, no. 1, pp. 148–153, 2010, doi: 10.3109/17453671003628780

    work page doi:10.3109/17453671003628780 2010

  5. [5]

    Cancer risk in 680,000 people exposed to computed tomography scans in childhood or adolescence: data linkage study of 11 million Australians,

    J. D. Mathews et al., “Cancer risk in 680,000 people exposed to computed tomography scans in childhood or adolescence: data linkage study of 11 million Australians,” BMJ, vol. 346, p. f2360, 2013, doi: 10.1136/bmj.f2360

    work page doi:10.1136/bmj.f2360 2013

  6. [6]

    Computed tomography — an increasing source of radiation exposure,

    D. J. Brenner and E. J. Hall, “Computed tomography — an increasing source of radiation exposure,” New England Journal of Medicine, vol. 357, no. 22, pp. 2277–2284, 2007, doi: 10.1056/NEJMra072149

    doi:10.1056/nejmra072149 2007

  7. [7]

    Magnetic resonance imaging protocols for paediatric neuroradiology,

    D. E. Saunders, C. Thompson, R. Gunny, R. Jones, T. Cox, and W. K. Chong, “Magnetic resonance imaging protocols for paediatric neuroradiology,” Pediatric Radiology, vol. 37, no. 8, pp. 789–797, 2007, doi: 10.1007/s00247-007-0462-9

    work page doi:10.1007/s00247-007-0462-9 2007

  8. [8]

    Tracked 3D ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry,

    M. Krönke et al., “Tracked 3D ultrasound and deep neural network-based thyroid segmentation reduce interobserver variability in thyroid volumetry,” PLoS ONE, vol. 17, no. 7, e0268550, 2022, doi: 10.1371/journal.pone.0268550

    work page doi:10.1371/journal.pone.0268550 2022

  9. [9]

    Evidence base for point-of-care ultrasound (POCUS) for the diagnosis of skull fractures in children: a systematic review and meta-analysis,

    G. Alexandridis, E. W. Verschuuren, A. V. Rosendaal, and D. A. Kanhai, “Evidence base for point-of-care ultrasound (POCUS) for the diagnosis of skull fractures in children: a systematic review and meta-analysis,” Emergency Medicine Journal, vol. 39, no. 1, pp. 30–36, 2022, doi: 10.1136/emermed-2020-209893

    work page doi:10.1136/emermed-2020-209893 2022

  10. [10]

    Accuracy of point-of-care ultrasound in detecting fractures in children: a validation study,

    C. Caroselli et al., “Accuracy of point-of-care ultrasound in detecting fractures in children: a validation study,” Ultrasound in Medicine and Biology, vol. 47, no. 1, pp. 68–75, 2021, doi: 10.1016/j.ultrasmedbio.2020.09.019

    work page doi:10.1016/j.ultrasmedbio.2020.09.019 2021

  11. [11]

    Ultrasound in forearm fractures: a pragmatic study assessing the utility of point-of-care ultrasound in identifying and managing distal radius fractures,

    D. Wood et al., “Ultrasound in forearm fractures: a pragmatic study assessing the utility of point-of-care ultrasound in identifying and managing distal radius fractures,” Emergency Radiology, vol. 28, no. 6, pp. 1107–1112, 2021, doi: 10.1007/s10140-021-01957-8

    work page doi:10.1007/s10140-021-01957-8 2021

  12. [12]

    Bone disease imaging through the near-infrared-II window,

    C. Mi et al., “Bone disease imaging through the near-infrared-II window,” Nature Communications, vol. 14, p. 6018, 2023, doi: 10.1038/s41467-023-42001-2

    work page doi:10.1038/s41467-023-42001-2 2023

  13. [13]

    Longer wavelengths in sunlight pass through the human body and have a systemic impact which improves vision,

    G. Jeffery, R. Fosbury, E. Barrett, C. Hogg, M. R. Carmona, and M. B. Powner, “Longer wavelengths in sunlight pass through the human body and have a systemic impact which improves vision,” Scientific Reports, vol. 15, 09785, 2025, doi: 10.1038/s41598-025-09785-3

    work page doi:10.1038/s41598-025-09785-3 2025

  14. [14]

    Shortwave-infrared meso-patterned imaging enables label-free mapping of tissue water and lipid content,

    Y. Zhao et al., “Shortwave-infrared meso-patterned imaging enables label-free mapping of tissue water and lipid content,” Nature Communications, vol. 11, p. 5355, 2020, doi: 10.1038/s41467-020-19128-7

    work page doi:10.1038/s41467-020-19128-7 2020

  15. [15]

    L. V. Wang and H. Wu, Biomedical Optics: Principles and Imaging. Hoboken, NJ, USA: Wiley, 2007

    2007

  16. [16]

    Penetration depth of photons in biological tissues from hyperspectral imaging in shortwave infrared in transmission and reflection geometries,

    H. Zhang, D. Salo, D. M. Kim, S. Komarov, Y.-C. Tai, and M. Y. Berezin, “Penetration depth of photons in biological tissues from hyperspectral imaging in shortwave infrared in transmission and reflection geometries,” Journal of Biomedical Optics, vol. 21, no. 12, p. 126006, 2016, doi: 10.1117/1.JBO.21.12.126006

    work page doi:10.1117/1.jbo.21.12.126006 2016

  17. [17]

    Time-resolved diffuse reflectance using small source–detector separation and fast single-photon gating,

    A. Pifferi et al., “Time-resolved diffuse reflectance using small source–detector separation and fast single-photon gating,” Physical Review Letters, vol. 100, no. 13, p. 138101, 2008, doi: 10.1103/PhysRevLett.100.138101

    work page doi:10.1103/physrevlett.100.138101 2008

  18. [18]

    Vo-Dinh, Ed., Biomedical Photonics Handbook, 2nd ed

    T. Vo-Dinh, Ed., Biomedical Photonics Handbook, 2nd ed. Boca Raton, FL, USA: CRC Press, 2014

    2014

  19. [19]

    In vivo calcium imaging in the near-infrared II window,

    D. Xu et al., “In vivo calcium imaging in the near-infrared II window,” Journal of the American Chemical Society, vol. 148, pp. 349–359, 2026, doi: 10.1021/jacs.5c13641

    work page doi:10.1021/jacs.5c13641 2026

  20. [20]

    Short-wave infrared light imaging measures tissue moisture and distinguishes superficial from deep burns,

    S. Mironov et al., “Short-wave infrared light imaging measures tissue moisture and distinguishes superficial from deep burns,” Wound Repair and Regeneration, vol. 28, no. 2, pp. 185–193, 2020, doi: 10.1111/wrr.12779

    work page doi:10.1111/wrr.12779 2020

  21. [21]

    A reappraisal of the use of infrared thermal image analysis in medicine,

    B. F. Jones, “A reappraisal of the use of infrared thermal image analysis in medicine,” IEEE Transactions on Medical Imaging, vol. 17, no. 6, pp. 1019–1027, 1998, doi: 10.1109/42.746635

    work page doi:10.1109/42.746635 1998

  22. [22]

    The analysis of human cortical bone by terahertz time-domain spectroscopy,

    M. R. Stringer et al., “The analysis of human cortical bone by terahertz time-domain spectroscopy,” Physics in Medicine and Biology, vol. 50, no. 14, pp. 3211–3219, 2005, doi: 10.1088/0031-9155/50/14/001

    work page doi:10.1088/0031-9155/50/14/001 2005

  23. [23]

    Handheld near-infrared spectroscopy: state-of-the-art instrumentation and applications in material identification, food authentication, and environmental investigations,

    H. Yan et al., “Handheld near-infrared spectroscopy: state-of-the-art instrumentation and applications in material identification, food authentication, and environmental investigations,” Chemosensors, vol. 11, no. 5, p. 272, 2023, doi: 10.3390/chemosensors11050272

    work page doi:10.3390/chemosensors11050272 2023

  24. [24]

    Advances in spectral imaging: a review of techniques and technologies,

    S. Mukhtar, A. Arbabi, and J. Viegas, “Advances in spectral imaging: a review of techniques and technologies,” IEEE Access, vol. 13, pp. 31678–31703, 2025, doi: 10.1109/ACCESS.2025.3544476

    arXiv 2025

  25. [25]

    Geneva, Switzerland: IEC, 2014

    International Electrotechnical Commission, IEC 60825-1: Safety of Laser Products — Part 1: Equipment Classification and Requirements, Edition 3.0. Geneva, Switzerland: IEC, 2014

    2014

  26. [26]

    Image-to-image translation with conditional adversarial networks,

    P. Isola, J.-Y. Zhu, T. Zhou, and A. A. Efros, “Image-to-image translation with conditional adversarial networks,” in Proc. IEEE Conf. Computer Vision and Pattern Recognition (CVPR), 2017, pp. 5967–5976, doi: 10.1109/CVPR.2017.632

    doi:10.1109/cvpr.2017.632 2017

  27. [27]

    Unpaired image-to-image translation using cycle-consistent adversarial networks,

    J.-Y. Zhu, T. Park, P. Isola, and A. A. Efros, “Unpaired image-to-image translation using cycle-consistent adversarial networks,” in Proc. IEEE Int. Conf. Computer Vision (ICCV), 2017, pp. 2242–2251, doi: 10.1109/ICCV.2017.244

    doi:10.1109/iccv.2017.244 2017

  28. [28]

    Narrative review of generative adversarial networks in medical and molecular imaging,

    K. Koshino et al., “Narrative review of generative adversarial networks in medical and molecular imaging,” Annals of Translational Medicine, vol. 9, no. 9, p. 821, 2021, doi: 10.21037/atm-20-6325

    work page doi:10.21037/atm-20-6325 2021

  29. [29]

    All-in-one medical image-to-image translation,

    L. Han et al., “All-in-one medical image-to-image translation,” Cell Reports Methods, vol. 5, p. 101138, 2025, doi: 10.1016/j.crmeth.2025.101138

    arXiv 2025

  30. [30]

    Image-to-image translation with conditional adversarial networks,

    M. Tahmid, M. S. Alam, N. Rao, and K. M. A. Ashrafi, “Image-to-image translation with conditional adversarial networks,” in Proc. IEEE WIECON-ECE, 2023, doi: 10.1109/WIECON-ECE60392.2023.10456447

    arXiv 2023

  31. [31]

    Swin-Unet: Unet-like pure transformer for medical image segmentation,

    H. Cao et al., “Swin-Unet: Unet-like pure transformer for medical image segmentation,” in Proc. ECCV Workshops, 2022, pp. 205–218

    2022

  32. [32]

    Swin Transformer: hierarchical vision transformer using shifted windows,

    Z. Liu et al., “Swin Transformer: hierarchical vision transformer using shifted windows,” in Proc. IEEE Int. Conf. Computer Vision (ICCV), 2021, pp. 10012–10022, doi: 10.1109/ICCV48922.2021.00986

    arXiv 2021

  33. [33]

    SuperPoint: self-supervised interest point detection and description,

    D. DeTone, T. Malisiewicz, and A. Rabinovich, “SuperPoint: self-supervised interest point detection and description,” in Proc. IEEE CVPR Workshops, 2018, pp. 224–236, doi: 10.1109/CVPRW.2018.00060

    arXiv 2018

  34. [34]

    SuperGlue: learning feature matching with graph neural networks,

    P.-E. Sarlin, D. DeTone, T. Malisiewicz, and A. Rabinovich, “SuperGlue: learning feature matching with graph neural networks,” in Proc. IEEE CVPR, 2020, pp. 4938–4947, doi: 10.1109/CVPR42600.2020.00499

    arXiv 2020

  35. [35]

    ALIKED: a lighter keypoint and descriptor extraction network via deformable transformation,

    X. Zhao, X. Wu, W. Chen, L. Chen, Q. Li, and S. Wang, “ALIKED: a lighter keypoint and descriptor extraction network via deformable transformation,” IEEE Transactions on Instrumentation and Measurement, vol. 72, pp. 1–16, 2023, doi: 10.1109/TIM.2023.3271000

    arXiv 2023

  36. [36]

    LightGlue: local feature matching at light speed,

    P. Lindenberger, P.-E. Sarlin, and M. Pollefeys, “LightGlue: local feature matching at light speed,” in Proc. IEEE ICCV, 2023, pp. 17627–17638, doi: 10.1109/ICCV51070.2023.01616

    arXiv 2023

  37. [37]

    Deep-image-matching: a toolbox for multiview image matching of complex scenarios,

    L. Morelli, F. Ioli, F. Maiwald, G. Mazzacca, F. Menna, and F. Remondino, “Deep-image-matching: a toolbox for multiview image matching of complex scenarios,” Int. Arch. Photogramm. Remote Sens. Spat. Inf. Sci., vol. XLVIII-2/W4-2024, pp. 309–316, 2024, doi: 10.5194/isprs-archives-XLVIII-2-W4-2024-309-2024

    work page doi:10.5194/isprs-archives-xlviii-2-w4-2024-309-2024 2024

  38. [38]

    Multispectral visible and infrared imaging for face recognition,

    H. Chang, A. Koschan, M. Abidi, S. G. Kong, and C. H. Won, “Multispectral visible and infrared imaging for face recognition,” in Proc. IEEE CVPR Workshops, 2008, pp. 1–6, doi: 10.1109/CVPRW.2008.4563054

    arXiv 2008

  39. [39]

    Ability of emergency ultrasonography to detect pediatric skull fractures: a prospective, observational study,

    N. Parri, B. J. Crosby, C. Glass, et al., “Ability of emergency ultrasonography to detect pediatric skull fractures: a prospective, observational study,” Journal of Emergency Medicine, vol. 44, no. 1, pp. 135–141, 2013, doi: 10.1016/j.jemermed.2012.02.038

    work page doi:10.1016/j.jemermed.2012.02.038 2013

  40. [40]

    Multi-modality approaches for medical support systems: a systematic review of the last decade,

    M. Salvi et al., “Multi-modality approaches for medical support systems: a systematic review of the last decade,” Information Fusion, vol. 103, p. 102134, 2024, doi: 10.1016/j.inffus.2023.102134

    arXiv 2024

  41. [41]

    Assessing the generalizability of artificial intelligence in radiology: a systematic review of performance across different clinical settings,

    M. U. Suleman et al., “Assessing the generalizability of artificial intelligence in radiology: a systematic review of performance across different clinical settings,” Annals of Medicine and Surgery, vol. 87, pp. 8803–8811, 2025, doi: 10.1097/MS9.0000000000004166

    doi:10.1097/ms9.0000000000004166 2025

  42. [42]

    Racial bias in pulse oximetry measurement,

    M. W. Sjoding, R. P. Dickson, T. J. Iwashyna, S. E. Gay, and T. S. Valley, “Racial bias in pulse oximetry measurement,” New England Journal of Medicine, vol. 383, no. 25, pp. 2477–2478, 2020, doi: 10.1056/NEJMc2029240

    work page doi:10.1056/nejmc2029240 2020

  43. [43]

    Multispectral extended depth-of-field fluorescence microscopy with co-designed meta-optics and neural reconstruction,

    I. A. Atalay Appak et al., “Multispectral extended depth-of-field fluorescence microscopy with co-designed meta-optics and neural reconstruction,” Light: Science and Applications, vol. 15, p. 242, 2026, doi: 10.1038/s41377-026-02337-y

    work page doi:10.1038/s41377-026-02337-y 2026

  44. [44]

    Geneva, Switzerland: IEC, 2020

    International Electrotechnical Commission, IEC 60601-1: Medical Electrical Equipment — Part 1: General Requirements for Basic Safety and Essential Performance, Edition 3.2. Geneva, Switzerland: IEC, 2020

    2020

  45. [45]

    U.S. Food and Drug Administration, Marketing Submission Recommendations for a Predetermined Change Control Plan for Artificial Intelligence/Machine Learning-Enabled Device Software Functions: Guidance for Industry and Food and Drug Administration Staff. Silver Spring, MD, USA: FDA, 2024

    2024

  46. [46]

    Underdiagnosis bias of artificial intelligence algorithms applied to chest radiographs in under-served patient populations,

    L. Seyyed-Kalantari, H. Zhang, M. B. A. McDermott, I. Y. Chen, and M. Ghassemi, “Underdiagnosis bias of artificial intelligence algorithms applied to chest radiographs in under-served patient populations,” Nature Medicine, vol. 27, no. 12, pp. 2176–2182, 2021, doi: 10.1038/s41591-021-01595-0

    doi:10.1038/s41591-021-01595-0 2021

  47. [47]

    Artificial intelligence in acute and critical care: current challenges and strategic solutions,

    T. Gao, T. Fan, W. Leng, S. Yu, and Q. Lyu, “Artificial intelligence in acute and critical care: current challenges and strategic solutions,” Frontiers in Public Health, vol. 14, p. 1818726, 2026, doi: 10.3389/fpubh.2026.1818726

    arXiv 2026

  48. [48]

    Declaration of Helsinki — ethical principles for medical research involving human subjects,

    World Medical Association, “Declaration of Helsinki — ethical principles for medical research involving human subjects,” JAMA, vol. 310, no. 20, pp. 2191–2194, 2013, doi: 10.1001/jama.2013.281053

    arXiv 2013

  49. [49]

    Optical imaging in vivo with a focus on paediatric disease: technical progress, current preclinical and clinical applications and future perspectives,

    J. Napp, J. E. Mathejczyk, and F. Alves, “Optical imaging in vivo with a focus on paediatric disease: technical progress, current preclinical and clinical applications and future perspectives,” Pediatric Radiology, vol. 41, no. 2, pp. 161–175, 2011, doi: 10.1007/s00247-010-1907-0

    work page doi:10.1007/s00247-010-1907-0 2011

  50. [50]

    Upconverting nanoparticles for the near infrared photoactivation of transition metal complexes: new opportunities and challenges in medicinal inorganic photochemistry,

    E. Ruggiero, S. Alonso-de Castro, A. Habtemariam, and L. Salassa, “Upconverting nanoparticles for the near infrared photoactivation of transition metal complexes: new opportunities and challenges in medicinal inorganic photochemistry,” Dalton Transactions, vol. 45, pp. 13012–13020, 2016, doi: 10.1039/c6dt01428c

    work page doi:10.1039/c6dt01428c 2016

  51. [51]

    Tutorial on methods for estimation of optical absorption and scattering properties of tissue,

    R. Tao, J. Gröhl, L. Hacker, A. Pifferi, D. Roblyer, and S. E. Bohndiek, “Tutorial on methods for estimation of optical absorption and scattering properties of tissue,” Journal of Biomedical Optics, vol. 29, no. 6, p. 060801, 2024, doi: 10.1117/1.JBO.29.6.060801

    work page doi:10.1117/1.jbo.29.6.060801 2024

  52. [52]

    Near-infrared spectroscopy,

    A. Y. Denault, M. Shaaban-Ali, A. Cournoyer, A. Benkreira, and T. Mailhot, “Near-infrared spectroscopy,” in Neuromonitoring Techniques. Amsterdam, The Netherlands: Elsevier, 2018, ch. 7, pp. 179–233, doi: 10.1016/B978-0-12-809915-5.00007-3

    doi:10.1016/b978-0-12-809915-5.00007-3 2018

  53. [53]

    Checklist for Artificial Intelligence in Medical Imaging (CLAIM): a guide for authors and reviewers,

    J. Mongan, L. Moy, and C. E. Kahn, Jr., “Checklist for Artificial Intelligence in Medical Imaging (CLAIM): a guide for authors and reviewers,” Radiology: Artificial Intelligence, vol. 2, no. 2, p. e200029, 2020, doi: 10.1148/ryai.2020200029

    work page doi:10.1148/ryai.2020200029 2020

  54. [54]

    Multicontrast photoacoustic in vivo imaging using near-infrared fluorescent proteins,

    A. Krumholz, D. M. Shcherbakova, J. Xia, L. V. Wang, and V. V. Verkhusha, “Multicontrast photoacoustic in vivo imaging using near-infrared fluorescent proteins,” Scientific Reports, vol. 4, p. 3939, 2014, doi: 10.1038/srep03939

    work page doi:10.1038/srep03939 2014

  55. [55]

    Noninvasive, infrared monitoring of cerebral and myocardial oxygen sufficiency and circulatory parameters,

    F. F. Jöbsis, “Noninvasive, infrared monitoring of cerebral and myocardial oxygen sufficiency and circulatory parameters,” Science, vol. 198, no. 4323, pp. 1264–1267, 1977, doi: 10.1126/science.929199

    work page doi:10.1126/science.929199 1977

  56. [56]

    A brief review on the history of human functional near-infrared spectroscopy (fNIRS) development and fields of application,

    M. Ferrari and V. Quaresima, “A brief review on the history of human functional near-infrared spectroscopy (fNIRS) development and fields of application,” NeuroImage, vol. 63, no. 2, pp. 921–935, 2012, doi: 10.1016/j.neuroimage.2012.03.049

    doi:10.1016/j.neuroimage.2012.03.049 2012

  57. [57]

    A systematic review of functional near-infrared spectroscopy-based task paradigms in stroke rehabilitation,

    Y. Huang, X. Zhan, H. Zeng, S. Li, J. Shi, Z. Cui, Q. Fan, B. Li, Y. Sui, F. Liang, and Z. Song, “A systematic review of functional near-infrared spectroscopy-based task paradigms in stroke rehabilitation,” Frontiers in Human Neuroscience, vol. 19, p. 1633142, 2026, doi: 10.3389/fnhum.2025.1633142

    arXiv 2026

  58. [58]

    Functional near-infrared spectroscopy and its clinical application in the field of neuroscience: advances and future directions,

    W. L. Chen, J. Wagner, N. Heugel, J. Sugar, Y. W. Lee, L. Conant, et al., “Functional near-infrared spectroscopy and its clinical application in the field of neuroscience: advances and future directions,” Frontiers in Neuroscience, vol. 14, p. 724, 2020, doi: 10.3389/fnins.2020.00724

    arXiv 2020

  59. [59]

    Progress of near-infrared spectroscopy and topography for brain and muscle clinical applications,

    M. Wolf, M. Ferrari, and V. Quaresima, “Progress of near-infrared spectroscopy and topography for brain and muscle clinical applications,” Journal of Biomedical Optics, vol. 12, no. 6, p. 062104, 2007, doi: 10.1117/1.2804899

    work page doi:10.1117/1.2804899 2007

  60. [60]

    Transcranial near-infrared light in treatment of neurodegenerative diseases,

    D. Nizamutdinov, C. Ezeudu, E. Wu, J. H. Huang, and S. S. Yi, “Transcranial near-infrared light in treatment of neurodegenerative diseases,” Frontiers in Pharmacology, vol. 13, p. 965788, 2022, doi: 10.3389/fphar.2022.965788

    arXiv 2022

  61. [61]

    Three-photon imaging of mouse brain structure and function through the intact skull,

    T. Wang, D. G. Ouzounov, C. Wu, N. G. Horton, B. Zhang, C.-H. Wu, et al., “Three-photon imaging of mouse brain structure and function through the intact skull,” Nature Methods, vol. 15, no. 10, pp. 789–792, 2018, doi: 10.1038/s41592-018-0115-y

    work page doi:10.1038/s41592-018-0115-y 2018

  62. [62]

    Red and NIR light dosimetry in the human deep brain,

    A. Pitzschke, B. Lovisa, O. Seydoux, M. Zellweger, M. Pfleiderer, Y. Tardy, et al., “Red and NIR light dosimetry in the human deep brain,” Physics in Medicine and Biology, vol. 60, no. 7, pp. 2921–2937, 2015, doi: 10.1088/0031-9155/60/7/2921

    work page doi:10.1088/0031-9155/60/7/2921 2015

  63. [63]

    Near infrared spectroscopy for the characterisation of bovine teeth in terms of sex, diet, tooth type and place of origin,

    N. E. Pretorius, N. Anderson, A. S. Forrest, and K. B. Walsh, “Near infrared spectroscopy for the characterisation of bovine teeth in terms of sex, diet, tooth type and place of origin,” Journal of Near Infrared Spectroscopy, vol. 32, no. 6, pp. 207–217, 2024, doi: 10.1177/09670335241300641

    work page doi:10.1177/09670335241300641 2024