Mid-Infrared Photothermal Microscopy

• Mid-infrared photothermal microscopy is a vibrational imaging technique that converts IR absorption into localized heating, enabling detection via visible or near‑IR probes with subwavelength resolution.
• It employs diverse detection architectures—such as confocal, wide‑field, and interferometric designs—to enhance sensitivity and throughput for applications ranging from single viruses to live-cell imaging.
• Recent advances in hyperspectral, tomographic, and flow cytometry modalities broaden its applications in virology, cellular phenotyping, and materials analysis.

Searching arXiv for recent and foundational papers on mid-infrared photothermal microscopy to ground the article in current literature. Mid-infrared photothermal microscopy is a vibrational imaging modality in which mid‑IR absorption is converted into a visible or near‑infrared optical signal. In its basic form, a pulsed mid‑IR pump excites molecular vibrations, the absorbed energy is converted to heat through non‑radiative relaxation, and a co‑focused visible or near‑IR probe senses the resulting refractive‑index, phase, scattering, or fluorescence change. Because the readout is performed at visible wavelengths, the spatial resolution is set by the probe optics rather than by the mid‑IR wavelength, while the chemical specificity remains that of mid‑IR absorption. Across recent implementations, the method spans confocal, wide‑field, interferometric, holographic, fluorescence‑detected, dual‑comb, tomographic, and flow‑cytometric architectures, with applications ranging from single nanoparticles and single viruses to live‑cell volumetric imaging and high‑throughput single‑cell phenotyping (Zhang et al., 2020, Zhang et al., 2021, Fukushima et al., 4 Mar 2026).

1. Fundamental photothermal mechanism

In mid‑infrared photothermal microscopy, a mid‑IR pump pulse excites a vibrational resonance, and the resulting local heating perturbs optical properties sampled by a shorter‑wavelength probe. The basic thermal–optical relation is commonly written as

$$\Delta n = \left(\frac{\partial n}{\partial T}\right)\Delta T,$$

with additional contributions from thermal expansion. For phase-sensitive readout, the induced optical phase shift follows the visible-probe optical path; for scattering-based readout, the heating modifies the particle polarizability and hence the scattered field; for fluorescence-based readout, the temperature dependence of fluorescence quantum yield provides the signal. In the confocal interferometric formulation, the detected intensity can be written as

$$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$

so the photothermal change is encoded in the change of the scattered field relative to a strong reference field. This linearization is central for subwavelength objects, because pure scattering scales as $r^6$ whereas interferometric detection scales effectively as $r^3$ through the particle polarizability (Zhang et al., 2020).

The chemical contrast derives from the vibrational spectrum. In the fingerprint region, commonly targeted bands include amide I near $1650\ \text{cm}^{-1}$, amide II near $1550\ \text{cm}^{-1}$, phosphate near $1080\ \text{cm}^{-1}$, and ester carbonyl near $1750\ \text{cm}^{-1}$. In the CH stretching region, lipid-rich structures are emphasized near $2860$ and $2920\ \text{cm}^{-1}$, whereas protein-rich environments contribute strongly near $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$0. Spectral acquisition is obtained by tuning the pump wavenumber and recording the photothermal response, so that the measured signal reproduces the local IR absorption spectrum with visible-light spatial resolution (Zhang et al., 2021, Fukushima et al., 4 Mar 2026).

A recurrent point in the literature is that the readout observable is not unique. Some systems detect probe phase, some detect intensity or backscatter, and others detect fluorescence modulation. This suggests that “mid‑infrared photothermal microscopy” is better regarded as a family of pump–probe transduction schemes unified by the same absorption–heating step rather than by a single detection geometry.

2. Instrumental architectures and detection geometries

Confocal implementations use tightly overlapped pump and probe foci and often rely on lock‑in detection. In the single‑virus confocal interferometric microscope, a tunable pulsed quantum cascade laser covering $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$1–$$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$2 pumps the sample from below through CaF$$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$3, while a continuous‑wave 532 nm probe is focused from above with a $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$4, NA 1.2 water‑immersion objective. The reflected substrate field provides the interferometric reference, a pinhole enforces confocal detection, and a lock‑in amplifier demodulates the signal at the QCL repetition rate. This architecture reached single‑virus sensitivity by combining interferometric enhancement, axial phase optimization, and power normalization with an MCT detector (Zhang et al., 2020).

Wide‑field systems replace raster scanning with parallel camera acquisition. A QPI-based wide‑field molecular-vibrational microscope used diffraction phase microscopy to record MIR‑OFF and MIR‑ON phase maps over a $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$5 field of view, achieving 430 nm spatial resolution and wide‑field molecular imaging at 1 frame per second (Tamamitsu et al., 2019). A later wide‑field quantitative phase implementation using a homemade nanosecond MIR optical parametric oscillator and a high full‑well‑capacity image sensor demonstrated single-live-cell imaging beyond video rate, with 440 nm spatial resolution and 50 fps single‑frame imaging (Ishigane et al., 2022). Another wide‑field branch, WIDE‑MIP, introduced interferometric defocus enhancement: the photothermal contrast was maximized not at the axial position where interferometric contrast is strongest, but at a defocus offset where the interferometric response is most sensitive to small thermal perturbations. That system used a 520 nm nanosecond probe, wide‑field interferometric detection on silicon, and achieved single‑virus fingerprinting with approximately $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$6 higher throughput than an earlier scanning implementation (Xia et al., 2022).

Detection geometry has become a topic in its own right. Forward‑scattering geometries provide quantitative, shape‑independent phase contrast but require two‑sided optical access. Backward‑scattering geometries offer convenient single‑sided access, but their signals are strongly distorted by depth‑dependent interference for micron‑scale objects. “Bidirectional phase sensitivity in holographic phototransient microscopy” introduced internal forward scattering, in which the back‑reflection generated at the top surface of an MIR‑transparent substrate acts as an internally generated forward-scattering illumination wave. Using temporal coherence gating, the internally generated forward field was isolated from direct backscatter, and the resulting IFS signals reproduced the magnitudes and temporal dynamics of true forward‑scattering measurements while retaining single‑sided accessibility (Robertson et al., 3 Jul 2026).

Background suppression has also been engineered at the pupil plane. In a wide‑field epi configuration on silicon, a central titanium blocker at a pupil conjugate plane selectively attenuated low‑angle reflected light, yielding over three orders of magnitude background suppression and a 6‑fold signal‑to‑background noise ratio improvement without sacrificing lateral resolution. This quasi‑darkfield strategy enabled background-suppressed chemical fingerprinting at a single nanoparticle level across a $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$7 field of view (Zong et al., 2021).

3. Spectroscopy, speed, sensitivity, and resolution

The spectroscopic fidelity of mid‑IR photothermal microscopy has been repeatedly benchmarked against FTIR or other vibrational references. A single 200 nm PMMA bead measured by confocal interferometric MIP showed a prominent C=O stretching band at $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$8 and spectral agreement with an FTIR spectrum of a PMMA film. In the QPI implementation, the mid‑IR photothermal spectrum of index‑matching oil in the $$I_{\text{det}} = |E_r + E_s|^2 = |E_r|^2 + |E_s|^2 + 2E_rE_s\cos\phi,$$9–$r^6$0 range matched ATR‑FTIR with 3 cm$r^6$1 resolution (Zhang et al., 2020, Tamamitsu et al., 2019). Wide‑field interferometric defocus‑enhanced MIP similarly recovered virus spectra in the $r^6$2–$r^6$3 region, including protein and nucleic‑acid bands (Xia et al., 2022).

Reported modulation depths are typically small. For a 100 nm PMMA bead at $r^6$4, the confocal interferometric system measured a modulation depth of $r^6$5 with an estimated temperature rise of $r^6$6–$r^6$7 K, and its probe-power dependence had a log–log slope of $r^6$8, close to the theoretical 0.5 expected for shot-noise-dominated detection (Zhang et al., 2020). In wide‑field QPI, the improved system achieved approximately two orders of magnitude higher signal-to-noise ratio than previous MIP imaging techniques and reached 50 fps single-live-cell imaging (Ishigane et al., 2022). WIDE‑MIP reported a lateral MIP resolution of approximately 417 nm and a $r^6$9 throughput increase over an earlier single-virus scanning MIP scheme (Xia et al., 2022). For background-suppressed dark‑field wide‑field MIP, 300 nm PMMA beads were detected with signal-to-background noise ratio $r^3$0, and 500 nm beads reached SBNR values around 100 under suitable averaging (Zong et al., 2021).

Speed has been a central driver of recent work. Dual‑comb photothermal microscopy parallelized excitation across hundreds of wavelengths by using two GHz MIR combs spanning $r^3$1–$r^3$2, so that each modulation tone $r^3$3 corresponded to a distinct MIR channel. The method acquired hyperspectral data across roughly $r^3$4 with apodized spectral resolutions between $r^3$5 and $r^3$6, and demonstrated super‑resolved chemical imaging relative to the MIR diffraction limit (Chang et al., 2024). Mid‑infrared energy deposition spectroscopy pushed the temporal encoding further by using a beam‑combined monolithic DFB QCL array with 32 channels from 940–1056 cm$r^3$7, each delivering a 100 ns pulse within a single 3.2 µs burst. By numerically differentiating the transient heating curve, MIRED read out $r^3$8, which corresponds to instantaneous energy deposition and hence instantaneous infrared absorption. This work argued that the upper limit for spectrum encoding shifts from thermal diffusion to vibrational relaxation (Yin et al., 2024).

A recent unified sensitivity analysis compared stimulated Raman scattering, stimulated Raman photothermal microscopy, and MIP under shot‑noise‑limited assumptions. It concluded that MIR absorption coefficients for representative polar vibrational modes are typically about two orders of magnitude larger than the effective absorption coefficients of SRS, but that in MIP this intrinsic advantage is partly offset by a small photothermal factor, limiting the net sensitivity gain over SRS to several-fold and, at most, about one order of magnitude under experimentally realistic conditions (Toda et al., 24 Jun 2026). This suggests that large MIR absorption cross‑sections do not, by themselves, determine practical detectability; the readout architecture, detector well depth, and thermal operating regime remain decisive.

4. Biological, virological, and materials applications

Single-particle and single-virus applications were among the earliest demonstrations of the method’s sensitivity. The confocal interferometric system produced label‑free spectra of individual vesicular stomatitis virus and poxvirus particles, with dominant amide I and amide II peaks and reproducible spectra within each virus type. The amide I/amide II peak ratio differed between the two classes, with $r^3$9 for poxvirus and $1650\ \text{cm}^{-1}$0 for VSV, supporting label‑free differentiation at the single‑virion level (Zhang et al., 2020). WIDE‑MIP extended this to higher throughput and a broader virological fingerprinting scope. In single vaccinia virions it resolved thymine‑associated bands near $1650\ \text{cm}^{-1}$1, whereas in vesicular stomatitis virus it resolved uracil‑associated bands near $1650\ \text{cm}^{-1}$2, allowing DNA–RNA discrimination at the single‑virus level. In varicella‑zoster virus it additionally revealed enriched $1650\ \text{cm}^{-1}$3-sheet components in proteins through the amide I region (Xia et al., 2022).

Bacterial and cellular imaging followed closely. Background-suppressed wide‑field MIP produced hyperspectral dark‑field fingerprints of single S. aureus and E. coli bacteria in the amide I–amide II region, with amide-I-dominated contrast at $1650\ \text{cm}^{-1}$4 and distinct spectral variability across individual cells (Zong et al., 2021). Fluorescence‑enhanced MIP, while not label‑free, showed that thermo‑sensitive fluorescent probes can increase the modulation response to approximately $1650\ \text{cm}^{-1}$5 per Kelvin, about 100 times larger than scattering-based thermo‑response, and enable bond-selective imaging of single bacteria, mitochondria, and membrane-associated proteins. In wide‑field mode it reached 20 FE‑MIP frames/s and directly observed $1650\ \text{cm}^{-1}$6 modulation in GFP-expressing Shigella flexneri (Zhang et al., 2021). Fluorescence-detected PTIR was also applied outside cell biology, including polyethylene glycol/silica mixtures and ritonavir/PVPVA amorphous solid dispersions, where it characterized phase‑separated drug‑rich domains upon water absorption (Li et al., 2021).

Live-cell thermometry has revealed a distinct application domain. Label-free MIP‑ODT was used to measure intracellular thermal diffusivity from transient thermal decay and found values of 93–94% of water in the cytoplasm and nucleus. In the same study, MIP thermometry and fluorescent nanothermometry were compared during heat-induced temperature variations in living cells. The label‑free MIP measurement showed the fast temperature change expected from heat diffusion, while fluorescent nanothermometry showed an additional slowly varying signal. The reported conclusion was that the 10$1650\ \text{cm}^{-1}$7 gap issue arises from comparing two fundamentally different physical quantities: the equilibrium temperature field measured via refractive-index change and an additional quantity to which fluorescent nanothermometers are sensitive (Toda et al., 2024).

High‑throughput single‑cell phenotyping under flow has now become feasible. Mid‑infrared photothermal imaging flow cytometry used single-shot nanosecond-dual-pulse MIP microscopy to encode MIR‑ON and MIR‑OFF states into separate holographic channels within a single camera exposure, reducing their temporal separation to 20 ns. This suppressed motion-induced subtraction artifacts and increased the allowable sample velocity for artifact‑free MIP imaging by five orders of magnitude compared with frame‑sequential MIP. The system operated at 500 frames per second and achieved cellular event rates up to $1650\ \text{cm}^{-1}$8, with applications to oleic‑acid‑induced lipid accumulation, adipocyte differentiation, and confluence‑dependent heterogeneity (Sugawara et al., 18 Jun 2026).

5. Extensions to hyperspectral, tomographic, and volumetric imaging

Tomographic variants extend MIP from 2D contrast to quantitative 3D chemical imaging. Mid‑infrared photothermal optical diffraction tomography combined wide‑field MIR excitation with multi‑angle interferometric visible probing and reconstructs the 3D distribution of MIR‑induced refractive‑index change rather than only static refractive index. In the video‑rate implementation, 11 illumination angles and alternating MIR‑ON/MIR‑OFF frames yielded 22 frames per volume, giving 19.2 volumes per second. Under these conditions, a representative peak photothermal RI change of $1650\ \text{cm}^{-1}$9 against $1550\ \text{cm}^{-1}$0 produced SNR $1550\ \text{cm}^{-1}$1, enabling video-rate three-dimensional tracking of lipid droplets in living cells and anomalous-diffusion analysis from full volumetric trajectories (Fukushima et al., 4 Mar 2026).

A related intensity-diffraction-tomography branch, photothermal relaxation intensity diffraction tomography, introduced a dual-delay photothermal relaxation scheme to suppress water background and reconstruct 3D photothermal volumes from intensity-only data. PRIDT achieved video-rate volumetric chemical imaging with up to 15 Hz per wavelength, lateral and axial resolutions of 264 nm and 1.12 $1550\ \text{cm}^{-1}$2m, and a volumetric field of view of $1550\ \text{cm}^{-1}$3. It was used for volumetric mapping of proteins, lipids, and azide-tagged fatty acid uptake, and for live-cell lipid-droplet dynamics (Jia et al., 30 Dec 2025).

Hyperspectral volumetric imaging has also been demonstrated. The video-rate MIP‑ODT system swept the MIR idler wavelength and performed high-speed hyperspectral volumetric chemical imaging across a $1550\ \text{cm}^{-1}$4 spectral window within 1 s, enabling compartment-specific spectra of lipid droplets, nucleoli, and perinuclear structures (Fukushima et al., 4 Mar 2026). Dual‑comb photothermal microscopy approached the same challenge from the spectral side by replacing sequential wavelength stepping with simultaneous measurements at hundreds of wavelengths (Chang et al., 2024). These developments collectively suggest that the long‑standing trade‑off between spectral richness and imaging throughput is being relaxed by parallel excitation, fast angular control, and time-domain multiplexing.

6. Limitations, controversies, and outlook

Several limitations are persistent across the field. Water absorption in the mid‑IR remains both an opportunity and a constraint: it enables strong photothermal conversion but complicates live-cell spectroscopy, especially outside reflection geometries and thin sample chambers. Many of the highest-sensitivity implementations still operate on dried or fixed samples on CaF$1550\ \text{cm}^{-1}$5 or silicon substrates, and several wide‑field or interferometric approaches rely on reflective substrates or carefully controlled interface phases (Zhang et al., 2020, Xia et al., 2022, Robertson et al., 3 Jul 2026). Label‑free operation is not universal; fluorescence‑enhanced MIP and fluorescence-detected PTIR trade label freedom for stronger thermo‑response, lower speckle, and organelle targeting (Zhang et al., 2021, Li et al., 2021).

A second issue is interpretability of the readout. Scattering-based systems may exhibit sign inversions, defocus dependencies, and geometry‑dependent interference, especially in epi and backscattering modes. The size-dependent sign reversal observed for 300 and 500 nm PMMA beads in dark‑field wide‑field MIP and the depth-dependent interference emphasized in bidirectional holographic phototransient microscopy both illustrate that the photothermal observable is not always a simple scalar proxy for local absorption (Zong et al., 2021, Robertson et al., 3 Jul 2026). This suggests that quantitative MIP microscopy increasingly depends on explicit forward models of scattering, heat transport, and optical detection rather than on qualitative image interpretation alone.

The field nonetheless shows a clear trajectory. Recent work has delivered single‑virus vibrational spectroscopy, background-suppressed single‑nanoparticle imaging, wide‑field live-cell imaging beyond video rate, dual‑comb hyperspectral acquisition, video‑rate volumetric tomography, and artifact‑free MIP imaging flow cytometry (Xia et al., 2022, Ishigane et al., 2022, Chang et al., 2024, Fukushima et al., 4 Mar 2026, Sugawara et al., 18 Jun 2026). A plausible implication is that mid‑infrared photothermal microscopy is evolving from a set of specialized pump–probe demonstrations into a broader quantitative imaging framework that links vibrational spectroscopy, phase imaging, and high‑throughput biophysical measurement. The remaining challenges are less about proof of principle than about standardizing quantitative reconstruction, extending performance in aqueous and thick specimens, and clarifying which physical quantity each readout actually measures.