A1 / OverviewA2

Keep scrolling
to open How it works

NC STATE │ COS 100 │ GROUP 002-9A
// NAIWA:OBSERVATION BASE │ EXH 01FIELD RECORD观测记录

Making the
invisible
measurable.

Scientists make the invisible measurable by turning signals into trustworthy knowledge. This exhibit follows one chain of evidence through four disciplines: signal, measurement, reconstruction, validation.

所属 AFFIL.
NC STATE │ COS 100
编号 FILE
GROUP 002-9A
类型 CLASS
SCIENCE EXHIBIT
状态 STATUS
OBSERVING
开始观测START OBSERVATION
// NAIWA:OBSERVATION BASE │ A1OVERVIEW背景概览

Observation begins where direct perception ends.

No one has ever directly observed a lipid moving across a skin layer, or the two ends of a DNA strand moving a few nm away from each other. Instead, what is observed are the responses: ions detected by a detector or photons detected a few billionths of a second later than a laser flash.

Scientific imaging converts those responses into images and numbers researchers can test. This exhibit features two tools used by NC State researchers. Mass spectrometry imaging (MSI) records a full mass spectrum at every spot of a sample. It needs no labels, so hundreds to thousands of molecules can be mapped at once; NC State's IR-MALDESI extends it into three dimensions, removing tissue in layers under 7 µm deep (Paper #7). Fluorescence lifetime imaging with FRET (FLIM/PIE-FRET) attaches two fluorescent labels to a molecule. Energy transfer between them changes sharply over about 3–10 nm, so it works as a nanoscale ruler for how proteins and nucleic acids fold and move (Papers #6, #11).

The resulting image is not an image in the traditional sense. It is a claim, built from a sequence of choices: what to measure, how to measure it, how to reconstruct the structure from partial information, and whether the device can be trusted on that day. Each of the steps is from a different discipline, and each claim must be checked. That is why statistics and mathematics are not extra tools in this exhibit: they are the methods that decide whether these images can be trusted, and that rebuild images from incomplete data while reporting how uncertain the result is.

Approach highlighted
Quantitative scientific imaging as an evidence pipeline
Measurement examples
MSI (IR-MALDESI) │ FLIM / PIE-FRET
Why it matters
Images guide conclusions about tissue chemistry and molecular behaviour; an unreliable image misleads everything built on it.
Disciplines
Chemistry │ Physics │ Mathematics │ Statistics
// NAIWA:OBSERVATION BASE │ A2HOW IT WORKS观测原理

A signal is not yet knowledge.

Four steps separate a detector response from a claim someone else can rely on. Scroll to move the diagram through each step.

1 / 4Signal

Illustrative diagram of the evidence pipeline A continuous curve is sampled at thirteen positions, rebuilt as a connected estimate, then checked against a tolerance band. The shapes are invented for explanation and are not experimental data. ILLUSTRATIVE — NOT EXPERIMENTAL DATA RESPONSE 13 SAMPLES / FIXED INTERVAL RESIDUAL |SIGNAL − SAMPLE| TOLERANCE BAND SIGNAL │ MEASUREMENT │ RECONSTRUCTION │ VALIDATION

Chemistry │ Physics

Signal

Every measurement starts with a physical or chemical event. In mass spectrometry imaging, a laser samples one small spot of tissue and the released molecules are ionized. In fluorescence imaging, an excited donor fluorophore emits a photon or hands its energy to a nearby acceptor.

At this point there is no image, only a response that varies from place to place and moment to moment.

Chemistry │ Physics

Measurement

The instrument attaches numbers to the response. MSI records mass spectra together with the position each spectrum came from. FLIM / PIE-FRET uses alternating laser pulses and time-correlated single-photon counting to record donor and acceptor intensity and lifetime.

FRET efficiency changes sharply over roughly 3–10 nm, which is why physicists describe it as a nanoscale ruler (Silvernail et al., 2025; Erie et al., 2026).

Mathematics

Reconstruction

Many imaging systems never observe the object directly. Mathematicians treat the picture as an inverse problem: find the structure most consistent with indirect, noisy measurements.

Chung et al. (2024) developed hybrid projection methods that solve very large problems of this kind and split the answer into sparse and smooth parts.

Related example The paper demonstrates its methods on photoacoustic tomography and atmospheric inverse modeling. Our sources do not show it being applied to MSI or FLIM / FRET.

Statistics

Validation

Instruments keep producing data even when their performance has degraded. System suitability testing asks whether the instrument was working well enough for its measurements to be trusted.

Sohn et al. (2024) compared quality-control measurements from normal and deliberately compromised instrument conditions and combined them into a single suitability score. See the full case below.

// NAIWA:OBSERVATION BASE │ B1DISCIPLINES学科交叉

Different questions. Connected evidence.

DisciplineRole in the chain What it uses and learnsCore paper

Chemistry

Fair measurement

A square top-hat laser spot and automatic height correction sample every tissue layer evenly, so label-free 3D MSI can compare lipids across the layers of mouse skin. Built with NC State Biological Sciences collaborators.

#7 │ Sohn et al., 2025

Physics

Self-consistency

Photon emission, excited-state decay and dipole–dipole energy transfer become quantitative measurements of nanoscale distance and motion in individual labeled molecules. Two independent analyses, checked against a theoretical prediction, test whether the answer holds.

#6 │ Silvernail et al., 2025

Mathematics

Fill gaps │ report uncertainty

Image reconstruction is an inverse problem: work backward from noisy or incomplete measurements to the object that produced them. PaLEnTIR recovers sharp material boundaries; hybrid projection methods separate sparse anomalies from a smooth background; Bayesian methods estimate how uncertain the result is. Related imaging example

#3 │ Ozsar et al., 2025
#10 │ Chung et al., 2024
#5 │ Antil & Saibaba, 2024

Statistics

Fit to use │ costliest error

PCA and suitability scores, then supervised machine learning, decide whether an MSI platform is fit to run, tuned against the costliest mistake: calling a faulty instrument healthy. The same logic guards against missed brain regions in presurgical fMRI.

#9 │ Sohn et al., 2024
#2 │ Kibbe et al., 2024
#1 │ Hu & Jeng, 2024

Four ways to check an image

Every scientific image is a claim. Chemistry makes the measurement fair, statistics decides whether the data are fit to use and which mistake matters most, physics checks a result against itself, and mathematics fills gaps while reporting uncertainty.

The two tools differ in kind. MSI is label-free and broad: many molecules, mapped across tissue at the micrometer scale. FRET is targeted and fine: two labeled sites, tracked at the nanometer scale over microseconds to milliseconds, though the labels can disturb the molecule (Paper #11).

The most direct connection in our sources is Chemistry and Statistics working on the same quality-control problem (Papers #9, #2), with authors from both NC State departments. Mathematics is connected more loosely, through related imaging research rather than a direct collaboration.

// NAIWA:OBSERVATION BASE │ B2RESEARCH研究实例

Four papers. Four roles in the evidence chain.

Statistics × Analytical Chemistry │ Paper #9 │ Original research

Can this instrument be trusted today?

Sohn, Kibbe, Dioli, Hector, Bai, Garrard & Muddiman (2024). A statistical approach to system suitability testing for mass spectrometry imaging. doi:10.1002/rcm.9725

0.93

Overall suitability threshold

Reported for the IR-MALDESI MSI setup the authors studied. Specific to that setup, not a universal threshold for MSI platforms.

Method

The researchers applied a quality-control standard mixture to glass slides and imaged it under normal conditions and under deliberately compromised instrument conditions. Principal component analysis in R separated the two conditions. Suitability metrics, including mass accuracy and spectral accuracy, were combined into a score for each compound and then into one overall score for the system.

What they learned

The statistical model could tell appropriate instrument operation from compromised operation. For the setup studied, the authors selected an overall suitability threshold of 0.93, and found that partial cleaning did not always restore satisfactory performance.

Limits

  • The authors present the work as a proof of concept, not a finished standard.
  • Other MSI platforms would need their own characterization before using a threshold.
  • A suitability test supports trust in the instrument; it cannot guarantee every result is correct.
  1. InputInstrument stateNormal vs. deliberately compromised
  2. MeasureQC metricsMass accuracy │ spectral accuracy
  3. ModelPCA + combined scorePer compound, then overall
  4. DecideSuitable / needs cleaningCompared with the study threshold
Illustrative process diagram drawn for this exhibit │ not a figure from the paper

Physics │ Paper #6

Quantitative FLIM / PIE-FRET

Silvernail et al. (2025) found that intensity-based and lifetime-based analyses gave consistent FRET efficiencies for benchmark DNA samples. For their RNA/DNA hybrid, the intensity-based efficiency was 0.61 and the lifetime-based value 0.58 for a single molecular trace, two independent routes to nearly the same answer, both matching the ~0.6 predicted from the molecule's geometry.

The smFRET review by Erie et al. (2026, Paper #11) describes a 15-thymine single-stranded DNA overhang that switched between three conformations on timescales from about 25 microseconds to 100 milliseconds, and direct observation of strand slippage in CAG-repeat hairpins linked to Huntington's disease.

Values belong to the reported sample and analysis │ #11 is a review, not an original experiment.

Chemistry │ Paper #7

Three-dimensional MSI

Sohn et al. (2025) built a 3D MSI workflow using top-hat IR-MALDESI and automatic z-axis correction, with collaborators in NC State Biological Sciences, and evaluated it on mouse skin with two biological replicates.

With a laser energy of 1.3 mJ per burst through the top-hat optics and a 120 µm step in x and y, each layer removed was under 7 µm deep, a depth resolution comparable to earlier studies. In the skin, lipids such as cholesterol were most abundant in the first layer, the epidermis.

Parameters belong to this IR-MALDESI setup and tissue.

Mathematics │ Papers #3, #10

Reconstruction from indirect data

In a simulated three-dimensional X-ray CT test with only 7 projection angles, PaLEnTIR (Ozsar et al., 2025) reached a peak signal-to-noise ratio of 54.7, against 33.8 for total-variation regularization and 18.4 for a radial-basis-function level-set method, while using about ten times fewer unknowns. Chung et al. (2024) showed hybrid projection methods separating sparse anomalies from smooth components in photoacoustic tomography and atmospheric inverse modeling.

Simulated data │ one test, not a claim that the method always performs best.

Related example │ not used in MSI or FLIM/FRET

// NAIWA:OBSERVATION BASE │ CIMPACT影响与传播

Trust must be communicated, not assumed.

How this research reaches the public

National MagLab │ 23 Jan 2020

Molecular Maps of Animal Tissue With Unprecedented Detail

Strength
Vivid and accessible: compares mass spectrometers to "fancy molecular scales" and shows a colour-coded lipid map of a rat brain.
Could improve
It credits "special software" for turning data into images without explaining it, and never mentions quality control, replicates or instrument checks.

NC State Chemistry │ 3 Aug 2022

The Power of Mass Spectrometry Imaging

Strength
Defines MSI as a way to map molecules across a sample, gives accessible applications, and ties NC State's work to health research.
Could improve
A simple diagram of how quality-control measurements and statistical tests catch unreliable results would make the data-analysis contribution visible.

Morgridge Institute for Research │ 4 Nov 2025

Fluorescence Imaging Explained: Microscopic Life in Many Colors

Strength
Explains fluorophores, microscopy and image processing in plain language with vivid biomedical examples, and admits limits such as photobleaching and autofluorescence.
Could improve
A comparison chart of widefield, confocal, FRET and FLIM: purpose, advantages, limitations.

MathWorks

What Is Image Reconstruction?

Strength
Connects an abstract mathematical process to a familiar medical technology, using CT examples and a reconstruction illustration.
Could improve
A side-by-side of reconstructions from complete versus limited or noisy data, showing why algorithms and regularization change the final image.

Pattern across all four: they show the final image and skip the checks that make it trustworthy.

What NC State could do better

  1. Show the check, not just the picture

    Pair imaging news with a simple pipeline diagram, one before/after example such as a clean vs. compromised instrument, and one sentence on which error the researchers guarded against. Where a story says "special software," say what that software checks.

  2. Compare the methods side by side

    One chart: what each method measures, at what scale, whether it needs labels, and its main limitation.

  3. Open the instruments and the evidence

    Student tours of NC State's shared facilities (METRIC, the Analytical Instrumentation Facility), a cross-department series building on "The Mathematics behind Imaging," and news stories linked to the open code and data researchers already share.

// NAIWA:OBSERVATION BASE │ SSOURCES引用与分工

Every claim leaves a trail.

Works cited

  1. 01

    Hu, Y., & Jeng, X. J. (2024). Spatially adaptive variable screening in presurgical functional magnetic resonance imaging data analysis. Biometrics, 80(4), ujae157. https://doi.org/10.1093/biomtc/ujae157

  2. 02

    Kibbe, R. R., Sohn, A. L., & Muddiman, D. C. (2024). Leveraging supervised machine learning algorithms for system suitability testing of mass spectrometry imaging platforms. Journal of Proteome Research, 23(10), 4384–4391. https://doi.org/10.1021/acs.jproteome.4c00360

  3. 03

    Ozsar, E., Kilmer, M., de Sturler, E., Saibaba, A. K., & Miller, E. (2025). Parametric level-sets enhanced to improve reconstruction (PaLEnTIR). Inverse Problems, 41(2), 025004. https://doi.org/10.1088/1361-6420/ada798

  4. 04

    Ahmad, B. I. Z., Keasler, K. T., Stacy, E. E., Meng, S., Hicks, T. J., & Milner, P. J. (2023). MOFganic chemistry: Challenges and opportunities for metal–organic frameworks in synthetic organic chemistry. Chemistry of Materials, 35(13), 4883–4896. https://doi.org/10.1021/acs.chemmater.3c00741

  5. 05

    Antil, H., & Saibaba, A. K. (2024). Efficient algorithms for Bayesian inverse problems with Whittle–Matérn priors. SIAM Journal on Scientific Computing, 46(2), S176–S198. https://doi.org/10.1137/22M1494397

  6. 06

    Silvernail, I., Morgan, A. N., Gordon, K., Kerr, A. N., Borgognoni, K., Atisa, A. M., Clark, B. S., Castaneda, J. F., Stanley, R. E., & LeBlanc, S. J. (2025). Quantitative single-molecule FLIM and PIE-FRET imaging of biomolecular systems. BMC Methods, 2(1), 27. https://doi.org/10.1186/s44330-025-00048-1

  7. 07

    Sohn, A. L., Witherspoon, J. G., Smart, R. C., & Muddiman, D. C. (2025). Three-dimensional mass spectrometry imaging (3D MSI): Incorporating top-hat IR-MALDESI and automatic z-axis correction. Analytical and Bioanalytical Chemistry, 417(8), 1649–1661. https://doi.org/10.1007/s00216-025-05755-w

  8. 08

    Jin, A.-H., Muttenthaler, M., Dutertre, S., Himaya, S. W. A., Kaas, Q., Craik, D. J., Lewis, R. J., & Alewood, P. F. (2019). Conotoxins: Chemistry and biology. Chemical Reviews, 119(21), 11510–11549. https://doi.org/10.1021/acs.chemrev.9b00207

  9. 09

    Sohn, A. L., Kibbe, R. R., Dioli, O. E., Hector, E. C., Bai, H., Garrard, K. P., & Muddiman, D. C. (2024). A statistical approach to system suitability testing for mass spectrometry imaging. Rapid Communications in Mass Spectrometry, 38(9), e9725. https://doi.org/10.1002/rcm.9725

  10. 10

    Chung, J., Jiang, J., Miller, S. M., & Saibaba, A. K. (2024). Hybrid projection methods for solution decomposition in large-scale Bayesian inverse problems. SIAM Journal on Scientific Computing, 46(2), S97–S119. https://doi.org/10.1137/22M1502197

  11. 11

    Erie, D. A., LeBlanc, S. J., & Weninger, K. R. (2026). Resolving the dynamics of atypical DNA structures using smFRET. Current Opinion in Structural Biology, 101, Article 103351. https://doi.org/10.1016/j.sbi.2026.103351

  12. 12

    Mo, X., Rao, D. P., Kaur, K., Hassan, R., Abdel-Samea, A. S., Farhan, S. M., Bräse, S., & Hashem, H. (2024). Indole derivatives: A versatile scaffold in modern drug discovery—An updated review on their multifaceted therapeutic applications (2020–2024). Molecules, 29(19), Article 4770. https://doi.org/10.3390/molecules29194770

Core case in this exhibit │ other entries form the research archive from Benchmark #2

Public communication sources

  1. P1

    Coyne, K. (2020, January 23). MagLab scientists capture molecular maps of animal tissue with unprecedented detail. National High Magnetic Field Laboratory. https://nationalmaglab.org/news-events/news/molecular-maps-animal-tissue-maldi/

  2. P2

    Nahas, N. (2022, August 3). The power of mass spectrometry imaging. Chemistry at NC State. https://chemistry.sciences.ncsu.edu/2022/08/03/the-power-of-mass-spectrometry-imaging/

  3. P3

    Williams, A. (2025, November 4). Fluorescence imaging explained: Microscopic life in many colors. Morgridge Institute for Research. https://morgridge.org/feature/fluorescence-imaging-primer/

  4. P4

    MathWorks. (n.d.). What is image reconstruction? https://www.mathworks.com/discovery/image-reconstruction.html

Group contributions

  • StatisticsTerence ZhengMeasurement reliability │ website design & build#9#2
  • ChemistryJessica De Lauder3D MSI & instrument quality control#7#2#9
  • PhysicsBilly HedlundFLIM / PIE-FRET & smFRET#6#11
  • MathematicsChuanqiu DuImage reconstruction#3#10

Each member wrote one discipline section and its public-communication review. The Group Synthesis was written together.

Media credits

All diagrams and the 奶蛙观测基地 wordmark are original work made for this exhibit. Diagrams marked Illustrative are explanatory drawings and do not show experimental data. No journal figures, author photographs or news images are reproduced.

Typefaces: Saira, Geist and Geist Mono; Montserrat (subtitle outlines). All under the SIL Open Font License 1.1.