Photon Absorption Remote Sensing (PARS)

• PARS is a suite of advanced optical imaging techniques that capture both radiative and non-radiative relaxation events following photon absorption.
• It employs dual-wavelength, interleaved excitation and synchronized detection to generate high-contrast, label-free images for clinical and histopathological applications.
• The method utilizes quantifiable metrics such as Quantum Efficiency Ratio and Total Absorption to achieve sub-micron resolution and precise biomolecular mapping.

Photon Absorption Remote Sensing (PARS) refers to a suite of advanced, all-optical imaging modalities that exploit the simultaneous detection of both radiative and non-radiative relaxation processes following photon absorption in biological, clinical, and material samples. Initiated in the context of label-free microscopy, PARS now encompasses multi-modal biomedical imaging, time-domain molecular fingerprinting, advanced image analysis in histopathology, functionally-resolved in vivo imaging, and even algorithmic schemes in machine learning and privacy theory. This article focuses on the physical, technical, and computational principles of PARS in the context of label-free absorption microscopy and digital pathology, as well as related extensions in contemporary research.

1. Physical Principles and Detection Mechanisms

The foundational concept of PARS is that absorption of a pulsed or modulated photon by an endogenous chromophore results in partitioned relaxation through two primary channels:

• Radiative processes: Emission of photons, e.g., autofluorescence or other radiative decays.
• Non-radiative processes: Conversion of absorbed energy to heat and pressure, giving rise to photothermal and photoacoustic effects, typically sensed as fast index or reflectivity changes.

For an excitation wavelength $\lambda$, the absorption cross-section $\sigma(\lambda)$ and photon flux $\Phi$ determine the total absorbed energy per pulse: $$E_{\mathrm{abs}}(\lambda) \propto \sigma(\lambda) \cdot \Phi \cdot \Delta t$$. The signal channels are:

• Non-radiative: $$\Delta I_{\mathrm{scatter}} = I_{\mathrm{scatter,baseline}} - I_{\mathrm{scatter,modulated}}$$
• Radiative: $$I_{\mathrm{fluor}} = \max[I_{\mathrm{total}}(\lambda)] - I_{\mathrm{scatter,baseline}}$$

By sampling both channels from a single absorption event, PARS encodes a “quantum efficiency ratio” (QER), which quantifies the ratio of radiative to non-radiative pathways, yielding high specificity to biomolecular composition. For example, UV excitation at 266 nm strongly excites nucleic acids, yielding pronounced non-radiative “nuclear” contrast, while visible to UVA excitation (e.g., 355 nm or 532 nm) probes extracellular (e.g., collagen, elastin) and pigmentary features through both channels (Tweel et al., 5 Sep 2025, Ecclestone et al., 25 Jun 2025, Ecclestone et al., 2024).

2. System Architectures: Illumination, Acquisition, and Contrast Extension

State-of-the-art PARS microscopes realize their multi-channel detection via synchronized dual-wavelength pulsed excitation (e.g., interleaved 266 nm and 355 nm pulsed lasers, each at 50 kHz), co-focused with a probe beam (e.g., 405 nm CW) onto the sample. The detection head comprises:

• High-speed photodetectors capturing both radiative emissions and non-radiative fast intensity modulations.
• On-the-fly digitization and gridding into multichannel pixel arrays, enabling whole-slide or depth-resolved imaging (Tweel et al., 5 Sep 2025, Ecclestone et al., 2021).

The interlaced acquisition scheme enables both wavelengths to be sampled in alternation with submicron pixel separation, maintaining maximal throughput (e.g., 1 cm²/min is achievable at 12.5 mm/s linear stage velocity). The inclusion of complementary excitation bands, such as 355 nm for stromal and pigmentary contrast, vastly extends the molecular coverage obtainable in a single scan (Tweel et al., 5 Sep 2025).

In some implementations, spectral unmixing methods (Gaussian Mixture Modeling, NNLS) further resolve individual biomolecular contributions directly from raw radiative and non-radiative channel profiles, achieving quantitative, label-free tissue composition maps (Ecclestone et al., 25 Jun 2025).

3. Quantification Metrics: Energy Partitioning and Novel Contrast

The simultaneous detection of radiative ($E_R$) and non-radiative ($E_{NR}$) energies enables definition and extraction of uniquely informative, physically grounded contrast metrics:

• Total Absorption (TA): $TA = E_{NR} + E_R$
• Quantum Efficiency Ratio (QER): $QER = \frac{E_R - E_{NR}}{E_R + E_{NR}}$

TA serves as the most complete direct proxy for chromophore-specific extinction, invariant to branching through radiative or non-radiative routes. QER encodes local quantum yield plus vibrational relaxation effects, introducing chemical specificity fundamentally inaccessible to single-channel modalities (Ecclestone et al., 2024, Ecclestone et al., 2021, Ecclestone et al., 25 Jun 2025).

Performance metrics demonstrate sub-micron lateral resolution (as fine as 250–350 nm in thin-sample PARS, ∼1 µm in vivo), detection of sub-percent changes in signal, and contrast-to-noise ratios (>10:1 for key features). Image and segmentation fidelity have been verified by virtual labeling, deep-learning transforms, and direct histological correlation (Ecclestone et al., 25 Apr 2025, Ecclestone et al., 25 Jun 2025, Pellegrino et al., 2022).

4. Image Processing and Virtual Staining: Algorithms and Validation

The robust multi-spectral and multi-channel outputs of PARS are leveraged in advanced deep-learning pipelines for virtual (digital) staining and multi-marker emulation. Paradigmatic pipelines include:

• RegGAN (Registration-aware GAN): Consumes 4-channel (266_NR, 266_R, 355_NR, 355_R) images for virtual colorization of diverse special stains (H&E, PAS, Masson’s, Silver), unifying image-translation with tolerance to paired data misalignment. The generator is a deep ResNet, the registration network a ResUNet-based deformable flow predictor, and the discriminator a PatchGAN. Losses are composed of adversarial, registration-corrected ℓ₁, and flow smoothness terms. RegGAN is quantitatively superior to Pix2Pix and CycleGAN in MS-SSIM and DISTS metrics (Tweel et al., 5 Sep 2025).

Extensive masked evaluation by board-certified pathologists (virtual vs. traditional stained images) consistently demonstrates that virtual stains from PARS data are diagnostically equivalent to gold-standard chemical stains. Pathologists are unable to reliably distinguish real from virtual images (chemical 2.60±0.31 vs. virtual 2.67±0.31 diagnostic quality; masked identification close to random) (Tweel et al., 5 Sep 2025, Ecclestone et al., 25 Apr 2025).

Spectral unmixing via GMM and NNLS allows direct mapping from PARS channels to quantitative biomolecule abundance, with subcellular correspondence to histological benchmarks ($\sigma(\lambda)$0; Dice coefficient $\sigma(\lambda)$1 for tissue subtypes) (Ecclestone et al., 25 Jun 2025).

5. Applications: Histopathology, Intraoperative Assessment, and In Vivo Imaging

PARS has demonstrated capabilities in a wide array of label-free imaging tasks:

• Whole-slide virtual histology: Large-area datasets from FFPE or fresh-frozen tissues, with virtual multi-stain outputs suitable for digital pathology workflows. Multi-channel data preserves biomolecular information lost in conventional chemical processing, and all imaging is non-contact and non-destructive (Tweel et al., 5 Sep 2025, Ecclestone et al., 25 Apr 2025).
• Intraoperative guidance: High-throughput, reflection-mode imaging directly on thick or irregular tissue surfaces, providing near-real-time margin assessment for tumor resection and margin evaluation, with virtual H&E-like output aiding surgical decision-making (Ecclestone et al., 2020, Ecclestone et al., 2021).
• In vivo and functional imaging: When combined with multi-wavelength excitation and time-resolved detection, PARS can extract vascular structure, blood flow metrics, and even local oxygenation status through spectral absorption analysis, enabling functional vasculature and retinal imaging without physical contact (Simmons et al., 2023, Hosseinaee et al., 2021, Hosseinaee et al., 2021).
• Label-free molecular mapping: Direct quantitation of nuclei, cytoplasmic components, stroma, hemoglobin, melanin, and more is possible in tissue and cellular contexts, either via statistical unmixing or through QER/TA-based segmentation (Ecclestone et al., 25 Jun 2025, Ecclestone et al., 2024).

6. Limitations and Technical Challenges

Despite its versatility, several limitations remain:

• Alignment: Although registration-aware algorithms (e.g., RegGAN) can tolerate residual misalignment, large-scale tissue deformations or distortions pose challenges for precise image translation.
• Spectral coverage: While 355 nm and 266 nm excitation extend contrast substantially, certain histochemical features (e.g., lipids, mucins) may remain inaccessible without further wavelength expansion or nonlinear techniques.
• Throughput: Current point-scanning implementations (e.g., 50 kHz PRR) limit areal imaging speed (∼1 cm²/min). Approaches such as high-repetition-rate lasers, parallelization, or line-scanning modalities are being explored to mitigate speed limitations.
• Penetration depth: Optical sectioning is typically limited to ∼100 µm due to strong scattering and absorption at UV/visible wavelengths, restricting 3D imaging to superficial tissue layers (Tweel et al., 5 Sep 2025, Ecclestone et al., 2024).

7. Future Directions

Research in PARS is rapidly extending in multiple directions:

• Spectral and time-domain multiplexing: Integration of more excitation bands (405 nm, 532 nm, NIR) and time-domain waveform analysis for even richer molecular and functional inference.
• Higher-dimensional and multimodal approaches: Fusion with scattering, phase contrast, or other optical modalities to provide holistic specimen characterization and to address biomolecular specificity gaps.
• Real-time, high-throughput systems: Ongoing development targets MHz-range lasers, multiplexed detection arrays, and advanced GPU/FPGA data processing to achieve whole-slide and volumetric throughput suitable for clinical deployment.
• AI and model-based analytics: Coupling physical signal modeling, time-domain clustering, and deep learning will enable automated, explainable diagnosis, real-time virtual staining, and new diagnostic endpoints derived from the unique information content of PARS data (Tweel et al., 5 Sep 2025, Ecclestone et al., 25 Jun 2025, Ecclestone et al., 2024).

PARS thus represents a paradigm shift in optical microscopy and digital pathology by uniting physical specificity, multi-modal signal acquisition, and advanced computational pipelines into a