Surface Plasmon Resonance Imaging and Microscopy Modalities for Information-Rich, Label-Free Analysis of Biomolecular Interactions and Disease Biomarkers.

A variety of surface plasmon resonance imaging techniques have been developed, which differ fundamentally in the optical configurations used to generate images. These methods vary in how surface plasmons are excited, how signals are collected, and how spatial information is encoded. Scheme summarizes the operating principles of the major SPR imaging modalities and classifies them according to their underlying optical architectures. The most common type is prism-coupled SPR imaging, where excited surface plasmons propagate on a continuous metal film (Au/Ag/Al) in a Kretschmann (prism) geometry, with reflected intensity imaged over a wide field. SPR microscopy or objective-coupled SPR, on the other hand, uses a high-NA objective to couple light to surface plasmons (often via the substrate side), allowing higher spatial resolution than typical prism SPRi and more flexible imaging modes. Aside from SPRi and SPRM, there are a few other methods that are linked to plasmonic imaging. For example, plasmonic scattering microscopy (PSM) detects binding events by monitoring scattering from nanostructures or surface-bound particles enhanced by near fields. Interferometric plasmonic imaging (iSPR or plasmonic interferometry) combines a reference optical field with the SPR-modulated field so that phase-sensitive changes (interference fringes) can be measured, improving detectability of very small signals. Localized SPR (LSPR) imaging, a highly active and broadly pursued research direction, uses nanoparticles or nanostructured arrays (nanodiscs, nanoholes, nanotriangles) supporting localized plasmons rather than propagating ones. SERS (surface-enhanced Raman scattering) imaging utilizes plasmonic nanostructures to amplify Raman signals for chemical-fingerprint imaging. , Additionally, there have been reports for near-field plasmonic imaging (NSOM-plasmonic, TERS), which uses a nanoscale probe (tip) to access information beyond the diffraction limit of light, and waveguide-coupled plasmonic imaging, which excites plasmons using gratings or integrated photonics (planar chips) and images intensity/phase changes. Given the space constraints of this review, we focus our discussion on the SPR imaging modalities that are most commonly adopted and widely implemented. Scheme also illustrates the optical configuration of a typical reflectance-based SPRi imaging system in the Kretschmann geometry. In this system, monochromatic, p-polarized light is employed to excite propagating surface plasmons at the metal–dielectric interface, and imaging is performed at a fixed incident angle selected near the plasmon resonance condition. Binding-induced changes in the local refractive index are thus detected as variations in reflected intensity across the sensor surface. Beyond the conventional fixed-angle, intensity-based approach, several SPR imaging variants have been utilized that monitor alternative resonance parameters including wavelength shifts and phase changes associated with biomolecular interactions ( Table ). Angle-resolved SPR imaging enables the acquisition of reflectance images over a range of incident angles during a single measurement. By systematically scanning the excitation angle across the surface plasmon resonance condition, the full angular reflectance profile can be reconstructed for each pixel or region of interest. Since s-polarized light does not absorb, it is commonly used as an internal reference to normalize the p-polarized response, allowing reflectance changes to be expressed as a percentage of the maximum reflected intensity. Imaging across multiple incident angles increases the dimensionality of the data set and extends sensitivity to complex surface properties, including variations in feature size, thickness, and morphology. However, the resulting increase in data volume and image complexity often necessitates advanced postprocessing and modeling algorithms to accurately extract quantitative binding and structural information. Multiwavelength SPR imaging varies the excitation wavelength while maintaining a fixed incident angle. In contrast to single-wavelength, intensity-based SPRi, multiwavelength SPRi enables visualization of resonance shifts through full-color imaging, providing intuitive spectral contrast across the sensor surface. By sampling the SPR response over a range of wavelengths, this approach offers an increased dynamic range relative to fixed-wavelength reflectivity measurements. In practice, wavelength interrogation may be performed either by continuously scanning across a spectral range or by selectively monitoring multiple discrete wavelengths within the SPR reflectivity curve. Recent dual-wavelength systems exploit two approximately linear regions of the reflectivity spectrum, tracking their intensity changes simultaneously. The differential intensities are computationally tracked across regions of interest within an array, allowing faster sampling by limiting acquisition to predefined wavelength windows rather than the full spectrum. Multiwavelength SPRi can function either as a standalone analytical technique or as a complementary tool for identifying optimal operating conditions. Ongoing efforts continue to improve wavelength interrogation strategies in SPRi systems, and a recent study applied concepts derived from the Hadamard multiplexing configuration to achieve a sensitivity of 3407 nm/RIU and a refractive index resolution of 2.20 × 10 –6 RIU. In addition to intensity-based detection, SPR imaging in the Kretschmann configuration can be adapted to monitor phase changes between the incident and reflected light, with modification of the detection scheme. The phase shift of the reflected p-polarized light is measured relative to the incident light, and the resulting interference signal provides a sensitive readout of resonance perturbations, which can be directly related to shifts in the effective incident angle. Phase-based SPRi generally offers higher intrinsic resolution than conventional intensity-based SPRi, owing to the steep phase response near the plasmon resonance. However, its practical sensitivity can be limited by increased susceptibility to background noise and environmental fluctuations, as the measured phase variations often occur on very small scales. Incorporation of tunable optical filters has enabled phase detection over a broader wavelength range ( Figure ). An alternative strategy for phase interrogation involves indirectly extracting phase-like information from the SPR reflectivity curve itself. In this “quasi-phase” approach, wavelength-resolved SPRi data are analyzed using the derivative of the SPR curve, allowing phase-related information to be inferred without direct interferometric detection. This method differs fundamentally from traditional phase-resolved SPRi and offers several advantages, including a more linear relationship between the quasi-phase signal and refractive index unit (RIU) changes. Despite these advances, phase interrogation in SPRi remains challenged by a limited dynamic detection range and increased instrumental complexity relative to intensity-based methods. A recent effort with multichannel phase-sensitive SPR imaging platforms allowed a dynamic detection range of 3.7 × 10 –3 RIU and limit of detection (LOD) of 14.02 ng/mL for secondary antibody binding ( Figure ). A multiwavelength, phase-shift SPRi measurement of antibody binding. (A) Determination of optimal wavelengths for achieving highest phase-shift sensitivity. (B) Images of SPRi channels with colors representing the phase shift after sample introduction. The dashed boxes represent the wavelengths used for the enclosed channels with red, green, and blue corresponding to 696, 702, and 686 nm, respectively. (C) Overlayed real-time measurements (sensorgrams) from each channel. Reproduced with permission from ref . Copyright 2023 Elsevier. Surface plasmon resonance microscopy encompasses a broad class of plasmonic imaging techniques that employ a high-numerical aperture microscope objective, rather than a prism, to achieve total internal reflection and excite surface plasmons. , Depending on the optical configuration and detection scheme, SPRM systems can achieve spatial resolution and sensitivity that are generally superior to conventional SPR imaging, particularly for nanoparticle-based and single-entity analyses. , − Recent advances in SPRM have focused on improvements in optical design and signal processing. , , , Parallel, postacquisition image analysis approaches, assisted by machine learning processing and Fourier-domain transformations, have been introduced to improve contrast, suppress noise, and extract quantitative information from complex SPRM data sets. , Several emerging instrumental innovations further highlight the expanding scope of SPRM. , Total internal reflection-based SPR microscopy (TIR-SPRM) has recently been demonstrated as a means to improve spatial resolution by combining TIR excitation with Fourier-domain reconstruction. Terahertz (THz) microscopy has also been integrated with SPRM imaging principals to form a THz-SPRM platform ( Figure ). In this system, plasmon excitation is achieved in the Otto configuration ( Figure A), rather than the Kretschmann geometry. This configuration permits precise control over the coupling gap between the prism/objective and the sample surface, offering improved image quality and partially mitigating the spatial resolution limitations imposed by long-wavelength THz excitation. It is interesting to note that although this platform is described as an iteration of SPR microscopy, it retains features characteristic of SPRi, such as prism-based excitation, while omitting elements typically associated with SPRM, i.e., the use of a high-numerical-aperture objective. The presence of attributes from both approaches underscores that THz sensing principles can be adapted to configurations resembling both SPRi and SPRM. The THz-SPRM system was validated by characterizing the semiconducting and plasmonic properties of indium antimonide (InSb) partially covered with polypropylene or graphene ( Figure B,D), with reflectance measured as a function of the objective-substrate gap distance ( Figure C). Optical configuration and imaging data of an InSb terahertz SPRM sensor. (A) Instrumental design for terahertz SPR microscopy in the Otto configuration utilizing an InSb sensor chip. (B) Images of InSb chip partially obscured by a 35 nm thin film of graphene at variable distances of objective from substrate. (C) A graph of reflectance vs distance from objective to substrate for 35 nm graphene thin film. (D) Images of InSb chip partially obscured by a 20 μm film of polypropylene at variable distances of objective from substrate. Reproduced with permission from ref . Copyright 2023 MDPI. Beyond the common plasmonic imaging methods like SPRi and SPRM, an emerging suite of lesser-explored techniques has been growing, including methods such as interferometric plasmonic microscopy, plasmonic scattering microscopy, and waveguide-coupled SPR. These iterative techniques build on the core principles of SPRi and SPRM, offering improvements in sensitivity, resolution, and the ability to extract complementary information that adds greater dimensionality to existing plasmonic imaging approaches. These methods highlight promising innovations with the potential to expand sensing capabilities into new regimes. It is known that the reliance on reflected light in SPRM limits achievable image contrast, and by extension, sensing resolution. , Further, analyte scattering patterns resulting from interference between the analyte and the surface plasmon wave produces a visual tailing effect, limiting resolution and analysis of heterogeneous samples, especially in the absence of corrective models to compensate. , While the plasmonic scattering has typically been considered an undesirable byproduct, it can serve as measurable optical signal that forms the basis of more sensitive techniques such as interferometric plasmonic imaging, plasmonic scattering microscopy, and near-field plasmonic imaging. Interferometric plasmonic microscopy (iPM) utilizes common-path interferometry, producing images from the interference between reflected light and the near-field scattering interaction of surface plasmons with surface nanostructures. Although the scattering produces relatively small intensity variations, which are particularly difficult to elucidate among the parabolic patterning in unprocessed images, recent advances in holographical image reconstruction and convolutional neural network (CNN) processing have aided in extracting high quality amplitude and phase information from iPM images. , The ability of iPM to image individual particles has been utilized to characterize gold nanostructure surfaces as well as the surface absorption, liposome fusion, and antibody capture of exosomes. Image reconstruction drastically increased the working resolution of iPM images, but the technique is still bottlenecked by the need for standalone postprocessing procedure. Plasmonic scattering microscopy (PSM) utilizes the scattering interactions between analytes and the surface plasmon wave by selectively monitoring the out-of-plane interference patterns, which are detectable in the far-field. By not relying on reflected light, contrast is significantly improved, enabling higher brightness and improved spatiotemporal resolution necessary for analyzing individual surface structures and their binding events in heterogeneous samples. Since PSM sensing can be achieved with an objective above the sensing surface, it is technically compatible with both prism-coupled SPRi and SPRM while retaining their respective sensing capabilities. , Recent work of PSM includes protein-antibody kinetic analysis, antibodies and small molecules binding to cellular membranes, as well as exosome immobilization and surface protein characterization for biomarker analysis. Further, the incorporation of an interferometric phase parameter in PSM data processing yields material-specific parameters, such as nanoparticles’ refractive index, which can enable nanoparticle differentiation in complex samples. Therefore, interference-coupled PSM could be a potential platform for drug delivery studies using nanoparticles as recognition tags. It is worth noting that by relying on scattered light instead of reflected light, PSM could be considered an application of interferometric scattering microscopy (iSCAT) instead of SPRi or SPRM. While scattering techniques such as iSCAT possess enhanced imaging resolution relative to SPRi and SPRM due to not being diffraction-limited, , they usually lack the sensitivity, kinetic resolution, and multiplex capabilities of plasmonic imagining, − while PSM possesses both enhanced sensitivity and resolution relative to standard SPRM again. , Plasmonic scattering also opens the doors for further iterations of scattering-based analysis, including methods like evanescent scattering microscopy (ESM), which shows promise in alternative surface materials and greater compatibility with fluorescent emission. While surface optimizations in film thickness tuning, materials, and 3D functionalization have improved the sensitivity in SPR methods, novel approaches to the fundamental physical design of the sensor can also offer sensitivity enhancements. Waveguide-coupled SPR (WCSPR) utilizes a metal–dielectric-metal surface configuration rather than the metal–dielectric configuration in conventional SPR. This allows for the excitation of SPs at both the lower metal–dielectric interface and the upper metal-waveguide layer, resulting in greater sensitivity and sharper reflectivity curves compared to normal SPR. These principles also apply to waveguide-coupled SPR imaging, which retains the strong sensitivity improvements and high-throughput microarray compatibility while requiring minimal modifications to existing SPRi hardware. While few studies have implemented waveguides in SPR imaging, recent works have demonstrated its biosensing capabilities by analyzing protein A-antibody interactions, as well as single-cell tracking of wheat-germ agglutinin ligand binding events on HeLa cells. The potential ease in incorporating waveguide-coupled surfaces into existing SPRi hardware warrants further exploration as a viable sensitivity enhancement method.

While an ideal SPR imaging or microscopic sensorgram exhibits well-defined phases of association, equilibrium, and dissociation, experimental noise frequently obscures these features and complicates reliable identification of binding events. Sources of noise, including bulk refractive index shifts, nonspecific interactions, cross-reactivity, and weak optical signals, are often intrinsic to the instrumentation and become increasingly pronounced as sample matrices grow more complex. Measurements performed in biologically relevant media such as patient serum or urine are particularly challenging, as additional matrix components introduce background fluctuations and confounding signals that mask true surface-binding responses. Though incremental improvements can be achieved through optimization of optical alignment, illumination stability, and flow control, such hardware-based approaches alone are often insufficient to fully resolve these issues. Consequently, a growing array of computational and machine learning-assisted analysis strategies has been introduced into SPR imaging and microscopy workflows to enhance signal extraction, suppress noise, and enable more robust interpretation of complex plasmonic data sets. Common approaches for SPR imaging data cleanup rely on mathematical models that facilitate identification of the resonance angle or reduce noise in sensorgram traces. Widely used strategies include fitting reflectance curves with analytical functions such as sigmoid or asymmetric models, which can be combined to more accurately locate the SPR minimum in noisy data sets. As an alternative, centroid-based algorithms estimate the resonance condition by comparing pixel-wise reflectance values to a defined baseline, which may be fixed or dynamically updated depending on the implementation. Polynomial fitting represents another commonly employed strategy, in which averaged subsets of reflectance data are used to generate simplified polynomial representations of the sensorgram. , This approach was demonstrated using Horner’s polynomial fitting method to extract average reflectance values on a row-by-row basis across an SPRi microarray, enabling reconstruction of smoothed reflectance curves from subdivided spatial data. Such methods can accommodate the inherent asymmetry of SPR dips, an important consideration given that conventional polynomial fitting often performs poorly in regions immediately surrounding the reflectance minimum. While centroid algorithms and polynomial fitting substantially simplify SPR image analysis and enable rapid extraction of resonance parameters, their capacity to resolve highly complex or low-signal data sets is limited. More advanced modeling approaches therefore offer improved fitting accuracy and enhanced discrimination of weak or convoluted binding signals. In addition, postprocessing strategies can be applied following data acquisition, spanning from relatively simple mathematical treatments for kinetic analysis to sophisticated machine-learning-based frameworks designed to handle large, multidimensional SPR imaging data sets. SPRi data sets are often challenging to analyze due to weak analyte responses superimposed on background fluctuations arising from bulk refractive index shifts. Fitting such complex data sets to analytical curves is further complicated by the rapid growth in data volume, as the number of collected points increases inversely with experiment runtime and proportionally with the computational resources required for polynomial fitting. This challenge is worsened in SPRi configurations that acquire multidimensional data such as angle- or wavelength-resolved SPRi, where a full sensorgram may be generated at each iteration of the scanned parameter, leading to rapid data accumulation. One strategy to reduce data complexity involves sampling a limited subset of wavelengths in multiwavelength SPRi experiments, thereby simplifying analysis and reducing acquisition time. However, this approach is only effective when the optimal wavelength range can be accurately predicted a priori. Alternatively, it has been demonstrated that an initial, coarse polynomial fit to spectral reflectance curves can be refined through postprocessing to improve resonance peak identification and suppress background noise. This refinement employs independent longitudinal and transverse scaling and translation parameters to iteratively correct the initial polynomial approximation. By reducing the total number of data points required for accurate curve reconstruction, the approach introduced by Wang et al. enables shorter acquisition times and increased experimental throughput. Notably, this method was shown to enhance resolution in the vicinity of the SPR dip, where conventional fitting routines typically exhibit the greatest uncertainty. Common statistical modeling approaches have been successfully applied to SPR imaging data to elucidate binding events and enable analyte identification in complex media. Principal component analysis (PCA), for example, has been used to differentiate SPRi responses arising from antibody interactions with fluid-specific biomarkers in blood, semen, saliva, urine, and sweat. Because each biological fluid produced a distinct SPRi response pattern, PCA clustering enabled correct identification of 62.5% of unknown samples containing one of the five fluids. The unidentified samples were appropriately labeled as inconclusive rather than being misidentified, demonstrating the robustness of this approach. Partial least-squares-discriminant analysis (PLS-DA) represents a complementary statistical approach that can be targeted to specific regions of sensorgrams corresponding to known binding events, enabling predictive identification of molecular interactions. In studies of biomarker cross-reactivity associated with multiple sclerosis (MS), PLS-DA analysis of sensorgrams collected from three MS-related antibodies achieved improved differentiation of ganglioside-antibody interactions compared to PCA alone. While effective at resolving subtle binding differences, both PCA and PLS-DA remain limited in their ability to compare across large or heterogeneous data sets, as they are typically applied to individual experiments or narrowly defined data subsets. More broadly, the performance of statistical models in SPR imaging is constrained by the dimensionality of the input data. Excessive data points can lead to overparameterization and increased computational burden, whereas overly sparse data sets risk oversimplification and loss of critical binding information. These trade-offs motivate the growing integration of advanced computational and machine-learning strategies to more effectively manage high-dimensional SPR imaging data sets. The integration of machine learning (ML) into SPR imaging has emerged as a powerful strategy for improving method development, system simulation, and analyte discrimination in complex media. ML-based approaches have been shown to enhance the differentiation of closely related targets and biomarkers in data sets where conventional fitting or statistical models struggle to resolve subtle signal features. Because surface plasmon polaritons are well described by Fresnel-based electromagnetic models, SPR system responses can be accurately simulated in silico. , These physics-based simulations provide high-quality training data for deep learning models, which can subsequently be applied to experimental SPR data sets to improve resonance identification and signal interpretation. In this framework, simulated reflectance curves effectively serve as labeled ground truth, enabling supervised learning without the need for extensive experimental calibration data sets. It has been demonstrated that incorporating realistic noise profiles into training simulations, by randomly introducing variations that mimic experimental bulk shifts, baseline drift, and optical instability, can significantly improve model robustness. In addition to physics-based synthetic data sets, experimentally acquired SPR imaging data can serve as valuable training inputs for both supervised and unsupervised machine-learning algorithms. Machine learning algorithms such as neural networks, k-nearest neighbor (k-NN), support vector machines (SVM), and principal component analysis are effective tools for analyzing the large data sets generated by arrayed and multiplexed SPR imaging platforms. , In addition to downstream image analysis, advanced models can be applied during image preprocessing to improve data quality, reproducibility, and robustness. Convolutional neural networks, including architectures such as U-Net image transformation models and generative adversarial networks (GAN) for signal recovery from noisy or suboptimal data, offer computational approaches that in certain cases can reduce or replace the need for dedicated optical or electronic hardware. By incorporating features derived from association, steady-state, and dissociation regimes of analyte–surface interactions, ML models can be trained to predict and differentiate binding behaviors with high specificity ( Figure ). Beyond temporal sensorgram features, spatial analysis of signal homogeneity and relative feature positioning across SPRi arrays provides additional insight into analyte binding dynamics. This approach has been demonstrated using both convolutional neural networks (CNNs) and k-NN classifiers, which showed marked improvements in distinguishing closely related analytes compared to conventional statistical methods. Additionally, matrix components can be explicitly incorporated into the training process, enabling ML models to differentiate true binding events from bulk refractive index shifts and nonspecific background responses ( Figure ). In neural-network implementations, raw sensorgram data can be introduced at the input layer, while weighted statistical descriptors are incorporated into hidden layers to establish relative relationships among data points. These architectures enable computation of performance metrics, including accuracy, precision, sensitivity, and specificity, associated with identified resonance features. In addition, some models generate self-evaluating error metrics (e.g., false-negative indicators, FN), allowing bidirectional or iterative architectures to flag uncertain classifications and trigger additional processing steps when needed. A notable advantage of neural-network-based approaches is their adaptability: models can be incrementally retrained as new experimental data sets become available, facilitating continuous performance improvement and enabling seamless redeployment in future SPRi experiments without complete model reconstruction. Schematic representation and resulting data of machine learning-assisted multiple sclerosis detection using fixed-angle SPR imaging. (A) Graphical representation of autoimmune attack in MS, involving antibodies targeting gangliosides located in the myelin sheath. (B) Schematic of the utilized SPRi methodology for this work, with different gangliosides captured on microarray spots followed by interaction with their respective antibodies. (C) 3D bar plot showing the response of 25 ng/mL of each antibody in serum over each ganglioside-functionalized surface, illustrating a lack of strong specificity for the interactions. (D) PCA analysis is able to reliably determine antibody identity based on response across the entire ganglioside array. (E) Including concentrations below 25 ng/mL negatively affects identification and more advanced statistical methods (machine learning) is needed. (F) Nnet of 10 ng/mL anti-GT 1b over the three ganglioside surfaces (top) showing model predicted ganglioside based on response at different time points. A confusion matrix (bottom) shows the ability of the model to accurately classify the interactions occurring. (G) kNN generated using the same data as the neural net of 10 ng/mL anti-GT 1b . This model worked well during steady state and after dissociation, achieving 96% accuracy at classification. Reproduced with permission from ref . Copyright 2022 Elsevier. An example of applying neural network machine learning to biosensor development can be best seen in an inflammation sensor developed during the 2019 SARS-CoV2-Pandemic. A 6 × 8 SPRi array with 6 repeats each was used to produce 288 sensorgrams in 100 min of antibody capture for six cytokines in blood serum (see Table ). To increase data processing throughput and accuracy, a neural network was utilized to distinguish binding events from background noise. The data was preprocessed using dark-field image processing to reduce nose and enhance contrast, then a CNN model using 6 layers of downsampling and 5 layers of upsampling was utilized to quantify binding signals. Standards of each cytokine were individually spiked into serum and calibration curves were obtained and utilized to determine levels of each cytokine in 40 patients with COVID-19 to determine which patients were at risk of dangerous immune responses. Besides disease applications, Thadson et al. demonstrated the capabilities of CNN ML to build a phase retrieval algorithm capable of extracting phase and reflectivity information directly from single-shot SPRM images, aiming to replace dedicated SPRM hardware interferometers ( Figure ). Their model utilized a context aggregation network (CAN) trained on Fresnel simulations, which incorporated reflection and transmission coefficients, gold thickness, as well as sample and substrate refractive index to extract phase and reflectivity profiles from SPRM back focal plane (BFP) images. Additionally, the model could simulate and account for simpler sources of shot noise to reduce artifacts. This deep learning model showed marked agreement of both the extracted phase and reflectivity profiles between simulated and experimental images ( Figure ), yielding a 3.5× lower LOD compared to simpler polynomial fitting methods. Deep learning phase extraction from single-shot SPRM images. (A) Schematic representation of the CAN network, which used BFP images as inputs to extract phase and intensity profile images. A single-shot BFP SPRM image (B) can then be processed in the network, producing an extracted phase profile image (C). The model was trained on simulations derived from Fresnel equations of a 46 nm gold surface, which produced both simulated BFP intensity (D) and phase training images (E). To show agreement between the simulated and extracted intensities, an extracted BFP intensity line scan is overlaid with the Fresnel-calculated intensity profile (F). Similarly, the extracted phase profiles can be overlaid with the Fresnel simulation phase profile (G). Reproduced with permissions from ref . Copyright 2021 Springer Nature. Deep learning processing has also been employed to fill the role of dedicated SPRM autofocusing equipment. Focusing tools are beneficial for SPRM reproducibility because image-to-images variations in focusing and long-term drift can produce shifts between single-tailed and double-tailed scattering artifacts. While these can be supplemented with dedicated focusing hardware or processing of individual images, , adaptable virtual refocusing tools provide an alternative that lowers cost while adding versatility for focus across longer experiments. Xu at al. utilized convolutional neutral networks and U-net processing to computationally reconstruct unfocused SPRM images into more consistent, focused images. Built upon the principles of common path holography, their model was trained on nanoparticle images at a variety of focal lengths and was split into two submodels. The model’s generator component extracted feature maps from input images and produced focused output images, while the discriminator component’s convolutional networks scored processed images to determine if further focusing was necessary. The approach points to a potential future where deep learning models can serve as powerful replacements for SPRM accessory hardware. Beyond enhancing data analysis of biological interactions, computational modeling plays a critical role in the rational design and optimization of SPR imaging experimental platforms. As noted previously, such computational-first design strategies have been shown to accurately predict angular shifts associated with common sensing media and reagents, including water, ethanol, polydimethylsiloxane (PDMS), and phosphate-buffered saline (PBS). Fresnel equation simulations have also facilitated the exploration of alternative plasmonic materials, most notably aluminum, which offers distinct optical properties compared to traditional gold or silver substrates. In addition, these simulations enable systematic evaluation of surface-layer thicknesses, oxide compositions, and multilayer architectures, supporting both the development of novel sensor chips and the fine-tuning of established surface designs. By extending computational modeling to the design of SPRi arrays incorporating nontraditional metals or engineered surface architectures, it becomes possible to introduce new surface functionalities and expand the accessible parameter space of plasmonic imaging. This capability opens avenues for investigating biological and chemical systems that are difficult or impossible to probe using standard gold- or silver-based SPRi substrates. Computational acceleration of post-data-acquisition analysis can further enhance the development of novel SPR imaging platforms by guiding selection of optimal experimental parameters, such as illumination wavelength or incident angle, for future measurements. This capability has been demonstrated using machine-learning approaches that track reflectance minima across multiple wavelengths, enabling efficient identification of incident wavelengths that are most informative for subsequent fixed-wavelength experiments. Similarly, local similarity algorithms have been employed to determine optimal operating angles for specific SPRi platforms, allowing future experiments to be conducted under constant-angle conditions. Additional computational strategies have been developed to rapidly process angle-scanning SPRi images and identify optimal resonance angles for specialized surfaces. For example, Lee et al. reported an angle-scanning SPRi approach enhanced through contrast-limited adaptive histogram equalization (CLAHE) to improve sensitivity toward low-molecular-weight analytes. In that study, the method was validated by detecting gaseous acetone in nitrogen at concentrations ranging from 1.0 to 10.2 ppm using SiO 2 /Au thin films functionalized with chloride, methyl, amine, or hydroxyl groups. While presented as a proof of concept, this approach demonstrates promise for multiplexed SPRi analysis of small-molecule analytes with accelerated angle sampling. Across the full SPRi workflow, from computationally guided experimental design and surface engineering to high-throughput data acquisition and postprocessing of complex imaging data sets, computational tools play an increasingly central role. The integration of machine-learning and advanced data-processing algorithms has proven particularly effective in managing large data sets, improving experimental efficiency, and elucidating binding events obscured by noise, especially in complex biological or chemical matrices.

Surface plasmon resonance methodologies enable quantification of analytes at nanomolar to picomolar concentrations, owing to their exceptional sensitivity to changes in surface refractive index. When implemented in imaging formats, multiwell or array-based plasmonic chips further enable multiplexed and high-throughput analysis of multiple targets and samples in parallel. These capabilities naturally position plasmonic imaging techniques as powerful platforms for biosensor development. Hence, a growing body of literature has reported the application of SPR imaging to disease biomarker detection, with representative examples summarized in Table . Recent studies have focused particularly on biomarkers associated with oncological conditions, autoimmune disorders, inflammatory responses, and infectious diseases, highlighting the broad clinical relevance and analytical versatility of plasmonic imaging-based biosensing approaches. Cancer diagnostics and therapeutics have remained a major focus of biomedical research through 2025. Analytes investigated using SPR imaging techniques span a broad range of oncological biomarkers, including growth factors, apoptosis-related protein, and cancer-derived extracellular vesicles. One notable study examined the detection of cells overexpressing human epidermal growth factor receptor 2 (HER2) using an SPRi-based biosensor. In this work, nanobodies were employed as capture ligands on an SPRi array chip, and a suspension of cancer cells was flowed directly over the sensor surface. Cells exhibiting HER2 overexpression were detected with a limit of detection of ∼1 × 10 5 cells/mL. This approach is particularly promising, as it enables rapid identification of HER2-positive cells directly from dilute cell suspensions, requiring smaller sample volumes and offering faster analysis compared to conventional techniques such as flow cytometry. Neuropilin-1 detection using SPRi has also been demonstrated through immobilization of mouse monoclonal antibodies on a sputtered gold sensor chip. This platform achieved high analytical sensitivity, with a limit of quantification of 0.038 ng/mL and a linear response over the concentration range of 0.01–25 ng/mL. Other related work has focused on quantifying soluble programmed death-ligand 1 (sPD-L1) in human serum using SPRi. In this approach, a DNA aptamer was hybridized to the gold sensor surface to serve as the capture ligand, rather than a nanobody or antibody. Signal amplification was achieved by introducing anti-sPD-L1 antibodies conjugated to gold nanoparticles, resulting in a significantly enhanced plasmonic response. Using this strategy, picomolar sensitivity was obtained, with a linear detection range of 50 pM-10 nM without signal enhancement and 100 fM-50 pM with nanoparticle enhancement, corresponding to a minimum detection limit of 19 fM. Recent improvements utilizing nanoparticle enhanced nucleic acid detection have produced limits of detection as low as 0.5 attomolar. Additional oncological and metastasis-associated biomarkers have likewise been quantified in complex biological matrices using SPRi. Fibroblast growth factor 23 (FGF23), a biomarker associated with bone carcinogenesis and metastasis, was measured directly in blood plasma, yielding a linear working range of 1–75 pg/mL, a lower limit of detection of 0.033 pg/mL, and a limit of quantification of 0.107 pg/mL. In related work, SPRi was employed to evaluate candidate imaging peptides for multiple myeloma by characterizing their binding interactions with CD38, identifying two promising probes with K D values of 7.4 × 10 –7 M and 8.9 × 10 –7 M. Beyond cancer-specific biomarkers, a recent study investigated poly­(ADP-ribose) polymerase-1 (PARP-1), a potential marker for pathologically altered endometriosis, using SPRi in comparison with an enzyme-linked immunosorbent assay (ELISA) for analysis of blood serum samples. The SPRi sensor exhibited a linear detection range of 10–1000 pg/mL and recoveries between 95% and 105%. Taken together, these studies underscore the continued maturation of SPRi methodologies, particularly their expanding capability to operate reliably in complex media such as blood plasma and serum, an area that has historically posed significant challenges for SPR imaging. Additional efforts have focused on oncological biomarkers for which effective capture reagents are not yet commercially available. One such example is ETV1, a transcription factor commonly associated with prostate cancer when overexpressed, for which no suitable monoclonal antibody currently exists. To address this limitation, researchers generated and cultured polyclonal rabbit antibodies and evaluated their binding performance using both ELISA and SPRi. Antibody development was largely successful, yielding high-affinity interactions with a measured K D of 479 pM, demonstrating the feasibility of SPRi as a platform for validating newly developed capture agents for previously inaccessible oncological targets. Similar oncological developments have been seen in prostate cancer diagnostics where multiplexed SPRi has been investigated to improve throughput and selectivity of screening. In a recent example, traditional testing for prostate cancer using prostate specific antigen (PSA) was augmented with multiplex SPRi detection by simultaneously detecting PSA with three other antigens to create an assay resistant to false positives from benign prostatic hyperplasia and prostatitis ( Figure ). The biosensor utilized a home-built SPRi system with a 3D-printed microfluidic flowcell to simultaneously detect the four nanoparticle-conjugated analytes, PSA, insulin-like growth factor 1 (IFG-1), vascular endothelial growth factor D (VEGF-D), and cluster of differentiation 14 (CD14), in a single sample injection with pg/mL sensitivity for each analyte ( Figure , also Table ). Magnetic nanoparticles utilization allowed for sensitivity 100-fold greater than traditional ELISA and enabled removal of analytes from native media to prevent interference from complex matrix conditions. While patient samples were not tested, previous work by the authors utilizing a similar immunoassay showed 84.7% accuracy in discriminating cancer samples from benign controls with the SPRi samples showing improved sensitivity and throughput. SPRi assay for sensitive and high-throughput cancer diagnostics. (left) Graphical depiction of the SPRi setup and assay. Antigens (Ag) are preincubated with biotinylated antibodies (AB 2 ) bound to streptavidin coated 1 μm diameter superparamagnetic particles (MBs). The MG-AB 2 -Ag samples are then captured using antibodies for the antigen (AB 1 ) covalently immobilized on the sensor chip. (right) SPRi signals for three analytes, CD14, IGF-1, and VEGFD, contrasted to a BSA control. Each spot represents a different concentration of the injected protein. Reproduced with permission from ref . Copyright 2025 Elsevier. Another oncologically relevant analyte is vascular endothelial growth factor receptor 2 (VEGF-R2), which plays a central role in angiogenesis and metastasis. One research group developed two SPRi biosensor platforms for VEGF-R2 detection in plasma by immobilizing monoclonal rabbit antibodies via cysteamine linkers onto either a pure gold sensor surface or a silver/gold bimetallic chip. The gold sensor exhibited superior analytical performance, achieving a precision of 1.4% and a broader linear detection range (0.03–2 ng/mL), compared to the silver/gold bimetallic chip, which showed a precision of 2.2% and a linear range of 0.03–1 ng/mL. The binding parameters were further characterized using both SPRi and quartz crystal microbalance (QCM), yielding K D values of 2.05 × 10 –12 M and 2.10 × 10 –12 M, respectively, highlighting strong agreement between the two techniques. Building on this work, the same group applied similar SPRi strategies in plasma and peritoneal fluid to evaluate proteasome and immunoproteasome levels as potential diagnostic markers for endometriosis and to assess anti-CDH12 monoclonal antibodies for Cadherin-12 biosensing. While proteasome-based markers did not show a statistically significant correlation with endometriosis, the Cadherin-12 sensor exhibited exceptional sensitivity and diagnostic promise, with a reported K D of 7.52 × 10 –9 M, a limit of detection of 1.49 pg/mL, and a linear working range of 1–80 pg/mL. The inherent multiplexing capability of SPR imaging makes it a powerful platform for categorizing antibody–antigen interactions in autoimmune disease research. Multiple sclerosis (MS), a neurodegenerative autoimmune disorder, arises from aberrant immune responses against components of the central nervous system, most notably the myelin sheath surrounding neurons. Of particular interest are antibodies secreted by B cells that erroneously recognize neuronal membrane constituents, especially gangliosides. These antiganglioside antibodies have been correlated with specific MS symptoms, positioning them as promising disease biomarkers. However, their clinical utility has been limited by their low abundance in blood and the analytical challenges associated with detecting such targets in complex biological matrices. Malinick et al. introduced a fixed-angle SPR imaging methodology designed to quantitatively probe interactions between gangliosides immobilized on a nearly superhydrophobic sensor surface and MS-associated antibodies spiked into human serum. This approach enabled sensitive, multiplexed analysis of ganglioside-antibody binding and represented a significant advance in both biosensor development and mechanistic understanding of neurodegenerative immune processes ( Figure ). The ganglioside self-assembled monolayer (SAM) employed in this work exhibited strong antifouling behavior, attributable to its near-superhydrophobic character, which facilitated reliable detection of MS-specific antibodies in 10% human serum. To further address the complexity of serum-based measurements, the same research team subsequently integrated SPRi with machine-learning-assisted analysis, combining k-nearest neighbor classification and neural networks to differentiate binding interactions between multiple ganglioside species and individual antibodies. This SPRi-ML framework enabled quantitative discrimination across a working concentration range of 1–100 ng/L and achieved low limits of detection of 6.6 ng/mL for anti-GA1, 5.6 ng/mL for anti-GM1, and 4.5 ng/mL for anti-GT1b in whole human serum. Notably, these detection limits fall well within the clinically relevant concentration range for MS-associated antibodies (3–25 ng/mL), underscoring the potential of SPRi-based, data-driven platforms to interrogate the molecular underpinnings of autoimmune neurodegeneration with diagnostic relevance. Additional SPRi-based studies have focused on biomarker panels for the diagnosis of relapsing-remitting multiple sclerosis (RRMS). In one such investigation, SPRi was employed to quantify ubiquitin carboxyl-terminal hydrolase L1 (UCHL1), leptin, and fibronectin directly in serum collected from 100 RRMS patients and 46 healthy controls. Among the three analytes, UCHL1 demonstrated the strongest diagnostic performance, achieving a negative predictive value of 100%, compared to 43% for leptin and 92% for fibronectin. Despite its superior diagnostic specificity, UCHL1 exhibited a relatively narrow linear dynamic range, with linearity observed only between 0.5 and 3.0 ng/mL. In contrast, leptin and fibronectin displayed broader linear ranges of 0.1–10 ng/mL and 5–400 ng/mL, respectively. These results highlight a common trade-off between diagnostic specificity and analytical dynamic range in biomarker development. Importantly, the multiplexed and high-throughput nature of SPRi biosensor platforms enables the simultaneous monitoring of multiple biomarkers, suggesting that combined analysis of UCHL1, leptin, and fibronectin may provide a more robust and clinically informative strategy for RRMS diagnosis than any single marker alone. Fibronectin has also been investigated as a serum biomarker for bladder cancer using SPRi by an independent research group. In this study, patients with transitional cell carcinoma (TCC) of the bladder exhibited elevated serum levels of laminin-5, fibronectin, and collagen IV relative to healthy control populations. While all three extracellular matrix proteins demonstrated strong diagnostic sensitivity (98–100%), their specificity varied considerably, ranging from 92% down to 62%. These findings highlight both the promise and the limitations of early SPRi-based biosensors operating in complex media, where nonspecific interactions and biological variability can significantly impact assay selectivity. Similarly, a lectin-modified SPRi detector surface was developed to probe aberrant glycosylation patterns on circulating glycoproteins. This platform enabled correlation of disease progression from myelodysplastic syndromes (MDS) to acute myeloid leukemia (AML) by monitoring changes in glycan structures, specifically increased sialylation and truncation of O-glycans. Fetal and neonatal alloimmune thrombocytopenia (FNAIT) represents another clinically significant condition amenable to SPRi-based biosensing. In FNAIT, maternal IgG antibodies recognize and bind platelet antigens expressed by the neonate, which can lead to severe thrombocytopenia and life-threatening hemorrhage. To address limitations of current diagnostic approaches, researchers proposed a cellular SPRi assay that quantifies FcγRIII binding to antibody-coated platelets as an alternative to established IgG glycoanalytics based on mass spectrometry. This SPRi-based method was validated using 143 plasma samples and demonstrated strong discriminatory power between clinically relevant sample groups, with results showing good agreement with mass spectrometry-based analyses. Although less prevalent in the literature than oncological and autoimmune applications, surface plasmon resonance imaging techniques have increasingly been applied to characterize inflammatory and anti-inflammatory interactions. Pfizer researchers employed SPRM to complement their investigation of lipid ligands that bind chemokine receptor-like 2 (CCRL2). SPRM was used to quantify binding kinetics between CCRL2-expressing cells and candidate ligands, including chemerin, revealing that chemerin binds HEK-huCCRL2 cells with a K D of 5.49 nM. SPRi has been similarly utilized to elucidate the interaction of TLR4 with cycloastragenol. SPRi has also been utilized to evaluate anti-inflammatory compounds with potential cardiovascular relevance. One recent study investigated ethyl ferulate, the active component of the traditional Chinese medicine Ligusticum chuanxiong, as a therapeutic agent for preventing myocardial infarction in at-risk patients. SPRi measurements demonstrated that ethyl ferulate binds transforming growth factor-β receptor 1 (TGF-βR1) with a K D of 4.01 nM, providing a molecular basis for the compound’s reported cardioprotective and anti-inflammatory effects. Similarly, SPRi was applied to study anti-inflammatory constituents of another traditional Chinese medicine, Glycyrrhizae . In this work, investigators examined whether glycyrrhizin-derived compounds (GL) interact with toll-like receptor 4 (TLR4), a key mediator of inflammatory signaling. Binding studies revealed that licochalcone A (LicoA) exhibited high affinity toward TLR4, with a measured K D of 387 nM. Complementary Western blot analyses further showed that GL extracts reduced TLR4 protein expression in lung tissue. Collectively, these results supported the molecular binding models and suggested that the anti-inflammatory activity of Glycyrrhizae arises from both inhibition and downregulation of TLR4 signaling pathways. Compared to oncological and autoimmune applications, SPR imaging-based biosensors remain relatively underexplored for direct infectious disease diagnostics. To date, most SPRi studies in this area have focused on characterizing molecular binding interactions and probing immunological mechanisms rather than clinical detection. For example, SPRi has been used to quantify the interaction between the HIV-neutralizing monoclonal antibody CH31 and the HIV envelope glycoprotein AE.A244 gp120, yielding a K D of 1.61 × 10 –8 M. SPRi has also been applied to elucidate protein function in bacterial pathogens, including Staphylococcus aureus . Previous work has shown that functional amyloid motifs in S. aureus exhibit diverse biological roles, yet the function of one important protein class, phenol-soluble modulins (PSMs), has remained poorly understood. One proposed mechanism suggests that PSMs may act as bioadhesives or contribute to biofilm stabilization. To test these hypotheses, a recent study employed a multifaceted membrane-mimetic SPRi platform to examine PSM aggregation on surfaces with varying hydrophobicity and surface charge, including positively charged, negatively charged, and neutral hydrophilic oxidized polystyrene, as well as hydrophobic polystyrene substrates Kinetic analysis revealed that PSMs exhibit adaptive binding behavior, with a pronounced preference for surfaces that are traditionally considered low-binding. These findings support a functional role for PSMs as versatile bioadhesives and suggest their importance in biofilm adhesion and structural integrity within bacterial communities.

This review highlights the growing capabilities of surface plasmon resonance imaging and surface plasmon resonance microscopy for characterizing a broad spectrum of biological and inorganic entities. Continued advances in instrumental design have significantly improved spatial resolution, sensitivity, and analytical specificity, thereby expanding the applicability of these techniques across biosensing, nanoscience, and environmental research. In parallel, innovations in surface chemistry and signal-amplification strategies have enabled more reliable detection of low-molecular-weight and low-abundance analytes while mitigating nonspecific adsorption and cross-reactivitychallenges that have historically limited SPR-based analyses in complex media. The integration of computational methods, particularly machine learning, has further accelerated progress in SPR imaging. Data-driven approaches now support both postacquisition signal interpretation and the rational design of next-generation SPRi platforms, enabling improved noise reduction, enhanced sensitivity, and increased adaptability to diverse experimental conditions. When combined with advanced surface chemistries and amplification reagents, these computational tools substantially broaden the versatility and analytical power of SPRi and SPRM methodologies. Application-wise, SPR imaging has been most extensively explored in oncological and autoimmune disease research, where multiplexing and label-free detection offer clear advantages. In contrast, emerging areas such as virology, drug-target characterization, extracellular vesicle analysis, and environmental sensing remain comparatively underdeveloped and present significant opportunities for future investigation. In particular, the continued adoption of machine learning-assisted analysis is expected to play a critical role in enhancing specificity for small analytes in complex biological matrices. Despite substantial advances in hardware and software, several factors continue to limit the broader adoption of SPRi and SPRM, particularly in clinical settings. Conventional SPR instruments have evolved into standardized, largely “black box” platforms through the integration of automated fluid delivery, degassing, and simplified software. By contrast, SPRi and SPRM systems remain less mature commercially. Many reported implementations rely on custom-built optical assemblies constructed from general-purpose components. While this modularity is suitable for academic studies due to simple modification, it increases operational complexity and restricts routine use to laboratories with specialized expertise. The use of SPRi and SPRM places greater technical demands on users than traditional SPR. In addition to understanding SPR fundamentals, imaging- and scattering-related effects must be considered during data interpretation and assessment of reproducibility. These requirements increase training demands and limit robustness in environments where consistent operation by nonspecialist users is required. Data processing and standardization represent additional challenges for clinical translation. SPRi and SPRM experiments typically generate large, image-based data sets that require extensive postprocessing before being interpreted. At present, analysis workflows are often implemented using custom scripts or general image analysis software rather than standardized pipelines. Variability in data formats, processing methods, and reporting practices complicates interlaboratory comparison and presents obstacles for standardization, verification, as well as regulatory acceptance. Recent developments nonetheless suggest pathways toward broader adoption. Continued progress in instrument design and increased commercial development are expected to improve robustness while reducing SPRi/SPRM system complexity. In parallel, computational approaches, including machine learning and AI-assisted methods, are increasingly being applied to address challenges in calibration, image interpretation, and large-scale data processing. These advances may reduce user-dependent variability and improve reproducibility, lowering barriers to wider use. It remains to be seen if recent advances in AI can be leveraged to assist in the processing and interpretation of SPRi and SPRM data. As these efforts advance, SPRi and SPRM are likely to complement established SPR techniques by enabling spatially resolved measurements in applications where the added information justifies the increased complexity.

When investigating biomolecular interactions, a wide range of analytical tools are available to probe, elucidate, and quantify binding processes. The choice of method is often governed by factors such as spatial and temporal resolution, sensitivity, throughput, and ease of implementation, all of which influence experimental efficiency and data reliability. In recent years, label-free techniques, most notably surface plasmon resonance (SPR), have emerged as robust alternatives for real-time analysis of biomolecular interactions while maintaining high sensitivity. Beyond conventional spectroscopic SPR, surface plasmon resonance imaging (SPRi) extends these capabilities by enabling parallel, high-throughput, and multiplexed detection across spatially resolved sensing surfaces. − The advances in SPRi and related imaging-based plasmonic techniques in the past decade have substantially broadened their applicability in bioanalysis through the development of new optical configurations, computationally assisted data analysis for complex data sets, improved surface and nanostructure characterization, and targeted strategies for biomarker screening. The excitation of surface plasmons using a prism-based optical coupler was first reported by Kretschmann, a configuration that remains widely employed in modern SPR instrumentation. In conventional spectroscopic SPR measurements, a sensorgram is generated by monitoring shifts in the resonance condition, typically angular or wavelength changes, which reflect association, dissociation, steady-state binding, and surface regeneration processes. , Integration with microfluidic systems further enhances performance by enabling controlled and efficient analyte delivery across plasmonic sensor surfaces. In SPRi, the entire sensing surface is interrogated simultaneously using a charge-coupled device (CCD), allowing high-throughput and multiplexed analysis that is not readily achievable with point-detection SPR spectroscopy ( Scheme ). While SPRi generally employs optical configurations similar to those used in spectroscopic SPR, detection is typically performed at a fixed angle. Alternative interrogation modes include wavelength-scanning SPRi and phase-interrogation SPRi, each offering distinct advantages in sensitivity and contrast. Comprehensive discussions of these variations have been reported elsewhere. , a Developments in SPRi platforms have included the incorporation of plasmonic waveguides and expansion into terahertz sensing while developments in SPRM have added far-field scattering and interferometric imaging analysis. Developments in SPRi platforms have included the incorporation of plasmonic waveguides and expansion into terahertz sensing while developments in SPRM have added far-field scattering and interferometric imaging analysis. Surface plasmon resonance microscopy (SPRM) shares conceptual similarities with SPRi but differs fundamentally in optical implementation. In SPRM, the prism coupler is replaced by a high-numerical-aperture (NA) objective lens, and the signal arising from the evanescent plasmonic field is directly imaged − ( Scheme ). In many operations, the reflected light associated with surface plasmon excitation is also recorded. While SPRM can offer enhanced spatial resolution and access to single-particle or single-event measurements, the detected signal is generally weak. Thus, signal enhancement is often utilized through plasmonically active nanoparticles on the sensor surface. , SPRi and SPRM represent two complementary extensions of classical SPR that transform refractive-index sensing into spatially and temporally resolved, information-rich measurements. Both approaches retain the core advantages of SPR, capable of real-time, label-free detection and quantitative kinetic analysis. This review focuses on recent progress in SPRi and SPRM with representative literature published prior to 2025 to provide a comprehensive overview of the field. We first discuss advances in instrumental design and optical configuration, followed by emerging applications of machine learning and computational approaches for SPR-based data analysis. We then present key applications of SPRi and SPRM for information-rich characterization of molecular interactions on bioinspired and nanoscale interfaces. A substantial portion of the review is devoted to recent developments in disease biomarker assessment by imaging-based approaches. Finally, we present perspectives on the role of SPRi and SPRM in disease research, with emphasis on multiplexed analysis and underexplored areas that may offer significant opportunities for future investigation.