Fourier transform infrared (FTIR) microscopy provides label-free molecular contrast but is diffraction-limited to roughly 3–6 µm resolution in the fingerprint region. Mankar et al. (Analyst, 2021) showed that an unsupervised curvelet-based fusion (pansharpening) of FTIR hyperspectral images with high-resolution dark-field visible images can sharpen IR images of breast and ovarian cancer tissue, improving tissue-class separation, with only qualitative comparison against optical photothermal infrared (O-PTIR) super-resolution images. The method's own authors left its validity envelope open: it depends on the sample showing dark-field contrast; compatibility with alternative contrast modalities (auto-fluorescence, Raman, second-harmonic generation) was conjectured but never tested; validation covered only two gynecologic tissue types at a single 5.5 µm pixel size; and no quantitative spectral benchmark against a true super-resolved chemical reference exists, since the O-PTIR comparison was by visual inspection only. As of 2026 these gaps remain open: a different group demonstrated fusion of FTIR with hematoxylin-and-eosin (H&E) stained optical images in oral cancer tissue with a different (regression) algorithm, and the original group adapted the curvelet machinery within O-PTIR, but no work validates the multimodal FTIR-fusion approach with alternative label-free contrast, across tissue types and coarser pixel sizes, or against quantitative super-resolved standards.

Background

Fourier transform infrared (FTIR) microscopy images tissue by measuring mid-infrared absorbance (wavenumbers roughly 750–4000 cm⁻¹), giving label-free molecular contrast: proteins, DNA, lipids and collagen have characteristic absorbance bands, so no staining is needed. Because mid-IR wavelengths are long (2.5–12.5 µm), the spatial resolution is diffraction-limited and wavelength-dependent — at best about 3–6 µm in the chemically informative fingerprint region (900–1800 cm⁻¹) — which is too coarse for sub-cellular digital histopathology. Commercial FTIR microscopes use a focal plane array (FPA) detector; 'high-definition' modes reach about 1.1 µm pixels but are extremely slow (order 2 hours per 1 mm tissue core versus minutes in standard 5.5 µm mode). Optical photothermal infrared (O-PTIR) spectroscopy sidesteps the diffraction limit by using a visible probe beam to detect photothermal expansion induced by a pulsed IR beam, achieving sub-micrometer spatial resolution, but full hyperspectral O-PTIR cubes remain too slow for routine clinical work. A complementary computational route is pansharpening — the fusion of a low-resolution hyperspectral image with a high-resolution image from another modality, originally developed for satellite data. Curvelets are multiscale basis functions sparsely representing curved edges, making curvelet-based fusion well suited to biomedical images with curved discontinuities: high-frequency curvelet coefficients are taken from the sharp image, low-frequency (spectral) content from the FTIR image. Mankar et al. (2021) fused 5.5 µm FTIR data of breast and ovarian cancer tissue microarrays (TMAs; many small tissue cores on one slide) with co-registered dark-field visible images (dark-field microscopy collects only scattered light, highlighting scattering structures without staining) and showed improved class separability of epithelium, stroma, adipocytes and lymphocytes. Their unsupervised method deliberately avoided the supervised alternative of Falahkheirkhah et al. (2021), a deep-learning generative adversarial network (GAN) that uses hematoxylin-and-eosin (H&E) stained images as a substitute super-resolved ground truth — a choice Mankar et al. argue can inject wavenumber-dependent spatial artifacts because H&E carries no spectroscopic information. Independently, Al Jedani et al. (2021, 2023) developed a regression-based FTIR + H&E fusion for oral squamous cell carcinoma metastases in lymph nodes. Within this landscape, the open question is the validity envelope of multimodal FTIR sharpening: when does computational fusion actually improve spatial resolution, with which contrast modalities, for which tissues and acquisition settings, and how can the improvement be benchmarked quantitatively rather than by eye.

Problem Statement

Establish the validity envelope of multimodal optical-contrast-assisted computational sharpening of FTIR microscopy images, as posed by the open questions left by Mankar et al. (2021): (1) Modality compatibility — validate whether the curvelet-based fusion actually improves spatial resolution when the high-resolution channel is an alternative contrast modality such as auto-fluorescence, Raman or second-harmonic generation microscopy, including for samples that show no dark-field contrast; and characterize when fusion fails, i.e. for samples whose high-resolution channel carries no relevant contrast. (2) Generalization — determine whether the fusion improves resolution and preserves spectral fidelity across tissue types and disease states beyond gynecologic cancer microarrays, and at FTIR pixel sizes coarser than 5.5 µm (which is the practical lever for accelerating acquisition, since coarser pixels shorten collection time). (3) Quantitative benchmarking — replace the current qualitative, visual-inspection comparison with quantitative spectral and spatial benchmarks of fused FTIR data against a true super-resolved chemical reference (e.g. O-PTIR or an established super-resolved chemical standard), including how well fused spectra recover the reference's spectroscopic content. The three demands are jointly posed as parts of one validity envelope, and the contrast modalities named in (1) are illustrative rather than a required set. None of the acquisitions the three demands require are available for this task — no alternative-contrast or coarser-pixel FTIR data, no out-of-family tissue and no super-resolved chemical reference. The question asks when multimodal FTIR sharpening genuinely improves spatial resolution without sacrificing spectral fidelity, and how the magnitude of that improvement is to be quantified under specified conditions.

References

  1. Mankar, Rupali; Gajjela, Chalapathi Charan; Shahraki, Farideh Foroozandeh; Prasad, Saurabh; Mayerich, David; Reddy, Rohith. 'Multi-modal image sharpening in fourier transform infrared (FTIR) microscopy.' The Analyst 146(15):4822–4834 (2021). DOI 10.1039/d1an00103e, https://doi.org/10.1039/d1an00103e
  2. Al Jedani, Safaa; Smith, Caroline I.; Ingham, James; Whitley, Conor A.; Ellis, Barnaby G.; Triantafyllou, Asterios; Gunning, Philip J.; Gardner, Peter; Risk, Janet M.; Shaw, Richard J.; Weightman, Peter; Barrett, Steve D. 'Tissue discrimination in head and neck cancer using image fusion of IR and optical microscopy.' The Analyst 148(17):4189–4194 (2023). DOI 10.1039/d3an00692a, https://doi.org/10.1039/d3an00692a
  3. Al Jedani, Safaa; Whitley, Conor A.; Ellis, Barnaby G.; Triantafyllou, Asterios; Smith, Caroline I.; Gunning, Philip J.; Gardner, Peter; Risk, Janet M.; Weightman, Peter; Barrett, Steve D. 'Image fusion of IR and optical microscopy for mapping of biomolecules in tissue.' The Analyst 146(19):5848–5854 (2021). DOI 10.1039/D1AN01161H, https://doi.org/10.1039/D1AN01161H
  4. Reihanisaransari, Reza; Gajjela, Chalapathi Charan; Wu, Xinyu; Ishrak, Ragib; Corvigno, Sara; Zhong, Yanping; Liu, Jinsong; Sood, Anil K.; Mayerich, David; Berisha, Sebastian; Reddy, Rohith. 'Rapid Hyperspectral Photothermal Mid-Infrared Spectroscopic Imaging from Sparse Data for Gynecologic Cancer Tissue Subtyping.' Analytical Chemistry 96(40):15880–15887 (2024). DOI 10.1021/acs.analchem.4c01093, https://doi.org/10.1021/acs.analchem.4c01093 (preprint: arXiv:2402.17960, https://arxiv.org/abs/2402.17960)
  5. Gajjela, Chalapathi Charan; Brun, Matthew; Mankar, Rupali; Corvigno, Sara; Kennedy, Noah; Zhong, Yanping; Liu, Jinsong; Sood, Anil K.; Mayerich, David; Berisha, Sebastian; Reddy, Rohith. 'Leveraging mid-infrared spectroscopic imaging and deep learning for tissue subtype classification in ovarian cancer.' The Analyst 148(12):2699–2708 (2023). DOI 10.1039/d2an01035f, https://doi.org/10.1039/d2an01035f
  6. Falahkheirkhah, Kianoush; Yeh, Kevin; Mittal, Shachi; Pfister, Luke; Bhargava, Rohit. 'Deep learning-based protocols to enhance infrared imaging systems.' Chemometrics and Intelligent Laboratory Systems 217:104390 (2021). DOI 10.1016/j.chemolab.2021.104390, https://doi.org/10.1016/j.chemolab.2021.104390
  7. Sihvonen, T.; Duma, Zina-Sabrina; Reinikainen, S. 'AB-PLS-DA: Pansharpening tailored for scanning electron microscopy and energy-dispersive X-ray spectrometry multimodal fusion.' Micron 177:103578 (2024). DOI 10.1016/j.micron.2023.103578, https://doi.org/10.1016/j.micron.2023.103578
  8. Yin, Tao; Yang, Tong; Dong, Xiaoyu; Liu, Bin; Qiao, Zhixing. 'Mid-infrared spectroscopy and its applications in biomedical analysis and imaging.' Biochemical and Biophysical Research Communications 790:152890 (2025). DOI 10.1016/j.bbrc.2025.152890, https://doi.org/10.1016/j.bbrc.2025.152890