Systems, devices and methods utilizing nanosensors for detection of tumor margins

Abstract

A method for characterizing a tissue sample is disclosed, which includes placing at least one nanostructured sensor in contact with said tissue sample, wherein said nanostructured sensor includes at least one response characteristic exhibiting dependence on a refractive index of said tissue sample when said nanostructured sensor is in contact with said tissue sample; directing interrogating optical radiation at a plurality of frequencies to said nanostructured sensor; detecting, for each of said optical radiation frequencies, radiation redirected by or transmitted through said nanostructured sensor in response to said interrogating optical radiation; analyzing said detected redirected or transmitted radiation to determine said at least one response characteristic of said nanostructured sensor at each of said plurality of interrogating optical radiation frequencies; and determining at least one attribute of said tissue sample based on said determined at least one response characteristic at said interrogating optical radiation frequencies.

Description

[0001] SYSTEMS, DEVICESAND METHODS UTILIZING NANOSENSORS FOR DETECTION OF TUMOR MARGINS

[0002] RELATED APPLICATION

[0003] The present application claims priority to Provisional Application No. 63 / 716,779, filed on November 6, 2024, which is herein incorporated by reference in its entirety.

[0004] TECHNICAL FIELD

[0005] The present disclosure relates generally to systems, devices, and methods utilizing nanostructured sensors for the characterization of various samples and environments.

[0006] BACKGROUND

[0007] In many applications, it is crucial to accurately determine the characteristics of a sample, whether for quality control in manufacturing, material science research, or biological analysis, such as disease detection. Distinguishing subtle variations in material properties, such as refractive index or optical characteristics, is often challenging with conventional methods.

[0008] By way of example, there is a need for the evaluation of the integrity and cleanliness of surfaces in semiconductor manufacturing. Further, there is a need for monitoring physical and chemical processes, such as adhesion of molecules to surfaces, and desorption.

[0009] Moreover, many medical applications also require characterization of a variety of samples. A particularly critical application where precise sample characterization is vital is in medical diagnostics and surgical procedures, such as cancer treatment. Cancer surgery often involves removing tumors along with some surrounding tissue to ensure that all cancerous cells are excised. However, surgeons frequently face difficulty in visually distinguishing between tumor tissue and healthy tissue in real-time. This can lead to the removal of either too much healthy tissue or, more critically, leaving cancerous cells behind, which may necessitate additional surgeries and cause unnecessary harm or physical trauma to patients.

[0010] Current methods for tumor margin detection, such as pathological examination, are complex, time-consuming (often taking hours to weeks), and provide limited real-time information during surgery. This uncertainty can result in incomplete resection of cancerous tissue, leading to patient anxiety and the need for follow-up interventions.

[0011] SUMMARY

[0012] The present disclosure generally provides systems, methods, and devices for characterizing samples, particularly biological tissue, using highly sensitive nanostructured sensors. In various embodiments, the present teachings overcome the limitations of existing methods by providing real-time, high-resolution feedback, for example, during surgical procedures, thereby enabling more precise and effective interventions. In some aspects, the method involves placing a sensor, such as a plasmonic sensor, in contact with a tissue sample and directing interrogating optical radiation at the sensor at multiple frequencies. The sensor's optical response characteristics, e.g., its spectral response over a range of interrogating optical frequencies, which can be highly dependent on at least one characteristic of the sample, e.g., on a tissue's refractive index, are measured by analyzing the redirected and / or transmitted radiation. As discussed in more detail below, in various embodiments, by interrogating different spatial locations of a sensing surface of the sensor with an optical beam, the response characteristics of different locations (pixels) of the sensor can be measured. An artificial intelligence (Al) engine can process these measured characteristics for each pixel to generate a pixel-by-pixel spatial map of at least one attribute of the sample under examination, e.g., a pixel-by-pixel probability of a tissue sample being cancerous, enabling rapid, high-resolution classification of the sample under examination. By way of example, in various embodiments, the methods and systems according to the present disclosure can provide real-time feedback during surgical procedures for assessing tumor margins, a significant advancement over slow conventional pathology. Furthermore, the disclosure includes a method for designing optimal sensors through an iterative, inverse design process that involves optimizing a figure-of-merit (FOM) function and Al-driven simulations and real measurements to engineer robust optical signatures for accurate classification.

[0013] In one aspect, a method for characterizing a tissue sample is disclosed, which includes placing a nanostructured sensor in contact with the tissue sample, where the nanostructured sensor includes at least one response characteristic exhibiting dependence on a refractive index of said tissue sample when the nanostructured sensor is in contact with the tissue sample, and directing interrogating optical radiation at a plurality of frequencies to the nanostructured sensor. For each of said optical radiation frequencies, radiation redirected by or transmitted through said nanostructured sensor in response to said interrogating optical radiation is detected and the detected redirected or transmitted radiation is analyzed to measure said at least one response characteristic of the nanostructured sensor at each of said plurality of interrogating optical radiation frequencies. At least one attribute of the tissue sample is identified based on said measured at least one response characteristic of the nanostructured sensor at said interrogating optical radiation frequencies. In various embodiments, this enables precise and non-invasive characterization of tissue properties by leveraging the optical response of a nanostructured sensor.

[0014] In various embodiments, a nanostructured sensor for use in various methods of the present teachings includes a plurality of nanostructured elements that are distributed over a sensing surface (herein also referred to as a metasurface) of the sensor. In various embodiments, the placement of the nanostructured sensor in contact with the sample includes placing the nanostructured elements (or at least a portion thereof) in touch with, or in very close proximity to (in the near field region of the nanostructured sensor, e.g., less than about 200nm for the visible regime), at least a portion of the tissue sample. In other words, at least a portion of the nanostructured elements will touch, at least a portion of the sample or will be in close proximity thereto.

[0015] The nanostructured sensors themselves can be realized in various forms. In some embodiments, the sensor is a plasmonic sensor comprising nano-sized electrically conductive elements, such as gold or silver elements, distributed in a periodic or a non-periodic pattern. In other embodiments, the sensor may be a photonic crystal sensor. In yet other embodiments, the sensor can be a hybrid structure that includes two-dimensional (2D) materials such as graphene or transition-metal di chalcogenides (TMDs).

[0016] The determination of the at least one attribute of said tissue sample can include comparing said measured at least one response characteristic with at least one respective reference simulation or measurement of said at least one response characteristic. In various embodiments, this provides a robust and quantifiable basis for tissue attribute identification, improving diagnostic accuracy. In various embodiments, the method further includes measuring, for at least one of said plurality of optical frequencies (and optionally for all of the optical frequencies), the redirected radiation at a plurality of angular orientations, e.g., measuring the intensity and / or polarization of the redirected radiation at said plurality of angular orientations. In some such embodiments, the method further includes determining an angular dependence of the redirected radiation as a function of the plurality of angular orientations. The measurement of the redirected radiation at multiple angular orientations can provide a deeper insight into the tissue's optical properties by analyzing the directional distribution of the redirected light and can allow for the capture of richer spatial information about the optical interaction, enhancing the sensitivity and specificity of tissue characterization. For example, in some embodiments, the analysis of the angular dependence of the redirected radiation can be used to compensate for effects of light polarization and / or noise, by way of example.

[0017] In various embodiments, such angular dependence of the redirected radiation can be considered as a response characteristic of the nano structured sensor, thereby integrating angular information regarding the redirected radiation into the characterization of the tissue sample and hence increasing its discriminative power.

[0018] In various embodiments, the method further includes directing, for at least one of said optical frequencies, the interrogating optical radiation to a plurality of different spatial locations of said sensor and detecting radiation redirected by or transmitted through said nanostructured sensor in response to illumination of each of said spatial locations by said interrogating optical radiation. In various embodiments, an intensity of the redirected or transmitted radiation associated with each illuminated spatial location can be determined to generate at least one response characteristic in the form of the intensity of the redirected or transmitted radiation as a function of the illuminated spatial locations of the sensor. Such a response characteristic can be analyzed, individually or more generally in combination with other response characteristics, to derive a map of at least one characteristic (attribute) of the tissue sample as a function of those spatial locations (herein also referred to as pixels). This can provide quantitative spatial information, crucial for identifying heterogeneities within the tissue, aiding in the visualization and identification of anomalous regions. In various embodiments, the spatial resolution of such a map can be equal to, or better than, about 1 micron, e.g., in a range of about 200 nm to about 1 micron.

[0019] While the optical illumination and collection paths remain subject to the conventional diffraction limit, the disclosed system does not rely on forming a direct optical image. Instead, each nanostructured element acts as a localized resonant transducer whose spectral response is modified in proportion to nanoscale variations in the surrounding refractive index. The resulting digital tissue map is reconstructed from these local sensing events rather than from far-field optical imaging.

[0020] Because the measurement relies on spectral modulation rather than optical image formation, the system is not constrained by geometrical -optics requirements regarding the sensor-tissue interface. Measurements can be obtained uniformly across irregular, curved, or scattering surfaces without the need for clear optical paths, thereby eliminating blind spots, diffraction artifacts, and contrast variations that typically limit imaging-based techniques. Optical access to the sensor surface for illumination and signal collection may be provided through free- space optics or optical fibers, facilitating operation even in confined or non-transparent environments. The contact-mode configuration allows quantitative, spatially resolved mapping that remains robust to sample geometry, scattering, and optical misalignment.

[0021] Accordingly, the platform achieves nanoscale sensitivity and quantitative accuracy that surpass the effective resolution of conventional imaging or microscopy techniques, which are constrained by diffraction and scattering artifacts. This resolution allows imaging at the cellular scale.

[0022] In various embodiments, this sensing modality provides distinct advantages for biological and clinical tissue evaluation, particularly for intraoperative cancer margin detection, where conventional optical imaging is degraded by scattering, opacity, and surface irregularities. Because the system collects spectral signals from the sensor surface rather than from the tissue, quantitative data can be obtained directly from moist, heterogeneous, or non-transparent specimens without staining, sectioning, or optical focusing. The platform thereby enables realtime, label-free, and surface-level assessment of tissue morphology and composition with nanoscale sensitivity. Furthermore, the resulting digital map is inherently quantitative and provides automatically classified outputs corresponding to tissue states, allowing immediate interpretation of margin status without subjective review or histological processing.

[0023] For example, the present methods and systems can be utilized to classify a tissue sample, e.g., a margin tissue sample, at a sub-cellular resolution (e.g., a resolution of better than 1 micron). As another example, the present methods and systems can be utilized to identify localized defects on a semiconductor surface at a sub-micron resolution. Further, in various embodiments, such spatial resolutions can be achieved in presence of various noise patterns. By way of example, in various embodiments, such spatial resolutions can be achieved at signal-to- noise ratios as low as 10 dB.

[0024] In various embodiments, an initial scan of the metasurface of the nanostructured sensor can be performed using the interrogating optical radiation beam at some initial large beam size and / or large scan step size to acquire a coarse dataset and identify one or more regions of interest associated with the sensor’s metasurface. A subsequent scan performed by using a smaller beam size and / or smaller scan step size of said one or more regions of interest can acquire a high- resolution dataset. In various embodiments, the method further includes analyzing the detected redirected or transmitted radiation associated with said plurality of spatial locations to generate a map indicating, for each of those spatial locations, the at least one response characteristic of the nanostructured sensor corresponding to that spatial location. Optionally, the analysis of the detected redirected or transmitted radiation can include analyzing an intensity of the detected redirected or transmitted radiation associated with each of the spatial locations. For example, the analysis can include analyzing an intensity spectrum (i.e., intensity as a function of the frequency of the interrogating optical radiation over a frequency band) of the redirected or transmitted radiation associated with each of those spatial locations. In addition, in various embodiments, for at least one of the spatial locations (and optionally all of the spatial locations), the intensity of the redirected or transmitted radiation can be detected and analyzed for a plurality of angular orientations, to supplement the dataset to be processed to determine at least one attribute of the sample at that location. In various embodiments, the at least one attribute of the tissue sample is determined at said spatial locations to generate a spatial map of that attribute of the tissue sample. For example, such a map can indicate the probability that each of a plurality of spatial locations of the tissue sample associated with the spatial locations of the sensing surface of the sensor illuminated with the interrogating optical radiation is cancerous.

[0025] In various embodiments, such a spatial map of the tissue sample can be determined at a spatial resolution of better than one micron, e.g., in a range of about 200 nm to about 1 micron. In some embodiments, such resolutions allow generating a map of the desired attribute of the tissue sample at a sub-cellular resolution.

[0026] In some embodiments, such a spatial map of a tissue sample can designate each of the spatial locations as being cancerous when the likelihood of that tissue location being cancerous exceeds a threshold, e.g., it is greater than about 80%, or greater than about 85%, or greater than about 90%. In some cases, the threshold can be set by a user, e.g., a surgeon, while in other embodiments, the threshold can be preset by the system, e.g., based on available statistical data, e.g., for a particular cancer type and tissue type.

[0027] In various embodiments, the at least one response characteristic includes any of an intensity of said redirected or said transmitted radiation, at least one parameter associated with a spectral resonance feature, a phase shift between interrogating optical radiation incident on the nanostructured sensor and said redirected radiation, a change in polarization of the redirected radiation relative to the incident interrogating optical radiation, an integrated intensity or reflectivity over a spectral band, a temporal response of the redirected light when the interrogating optical radiation comprises a plurality of radiation pulses, a scattering angle distribution of said redirected radiation, a interferometer pattern generated via interference between the interrogating optical radiation incident on the nanostructured sensor and said redirected radiation. This broadens the types of optical signals that can be analyzed, increasing the versatility and diagnostic potential of the method.

[0028] In various embodiments, a response characteristic associated with said spectral resonance feature includes any of a spectral position, a phase angle, a quality factor (Q), a group delay, an integrated intensity and a spectral derivative of said spectral resonance feature. This allows for a detailed analysis of the spectral characteristics, providing fine-grained information about the tissue's interaction with light. In some embodiments, the response characteristic can include one or more parameters associated with a plurality of spectral resonance features measured in the redirected or transmitted radiation. For example, the parameter can include any of a spectral spacing between the plurality of spectral resonance features, or at least one ratio of intensity of peaks of said plurality of resonance features.

[0029] The interrogating optical radiation can provide a plurality of frequencies for illuminating the nanostructured sensor, e.g., as a broadband radiation having a plurality of optical frequencies or as multiple narrow-band radiation beams at different frequencies. For example, the interrogating optical radiation can be generated by tuning a narrow-band tunable optical source to said plurality of optical frequencies. This offers precise control over the interrogation frequencies, enabling targeted spectral analysis. This tunable optical source can include a tunable laser, or a supercontinuum laser.

[0030] The interrogating optical radiation can include any of a continuous and a pulsed radiation. This provides flexibility in the measurement technique, allowing for both steady-state and time- resolved analyses.

[0031] For example, the interrogating optical radiation can include a plurality of narrow-band optical beams each having one of said plurality of frequencies, where optionally said optical beams are directed to said nanostructured sensor in a plurality of different temporal intervals. In various embodiments, the plurality of narrow-band optical beams includes a plurality of laser beams. In various embodiments, the use of a plurality of narrow-band beams at a plurality of frequencies can enable sequential or multiplexed interrogation, optimizing signal acquisition and processing.

[0032] In some embodiments, the intensity or the frequency of the interrogating optical radiation can be modulated at a modulation frequency, and the detection of the redirected or transmitted radiation can be performed using lock-in detection at that modulation frequency.

[0033] In some embodiments, the interrogating optical radiation can include a plurality of narrow-band optical beams each having one of said plurality of frequencies, where optionally the optical beams are directed to the nanostructured sensor in contact with the tissue sample concurrently or in a plurality of different temporal intervals. In some such embodiments, at least two of the narrow-band optical beams have different polarizations. Further, in some such embodiments, at least two of the narrow-band optical beams are directed to the nanostructured sensor along different incidence angles. In some embodiments, the plurality of narrow-band optical beams includes a plurality of laser beams.

[0034] In various embodiments, the at least one response characteristic includes a plurality of response characteristics. In some embodiments, such response characteristics can collectively help characterize the tissue sample. While in some embodiments, the plurality of response characteristics can be independent of one another, in other embodiments, two or more of such response characteristics can be inter-related. For example, a change in one response characteristic when the nanostructured sensor is in contact with a tissue sample can be correlated with a respective change in another response characteristic. This allows for the exploitation of synergistic information from different optical responses, which can help reveal complex tissue properties. Further, in some cases, variation of two or more response characteristics of a nanostructured sensor in response to contact with a tissue sample can be independent of one another.

[0035] In various embodiments, an artificial intelligence (Al) engine can be used to process said measured at least one response characteristic of said nanostructured sensor to identify at least one attribute of said tissue portion. A variety of Al engines can be used in the practice of the present teachings. For example, classification algorithms, utilizing machine learning approaches together with Al technologies, can be utilized that are robust to data inaccuracies and systematic measurement bias. The Al engines can leverage different learning paradigms, including supervised, semi-supervised, and unsupervised algorithms, to analyze the complex response characteristics of the sensors. For example, algorithms such as Convolutional Neural Networks (e.g., ResNet architecture), Support Vector Machines (SVM), and Logistic Regression can be utilized for classification tasks allowing for the accurate categorization of various samples (e.g., classifying margin tissue as cancerous or non-cancerous) or other materials.

[0036] In various embodiments, an Al engine configured to employ a classification algorithm for determining at least one attribute of a tissue sample (e.g., whether a tissue portion is likely to be cancerous) can be trained on both general reference data (i.e., data corresponding to a group of individuals) as well as patient-specific data. For example, the Al engine can be trained on a general reference dataset including a diverse array of samples to create a pre-trained model. The pre-trained model can then be fine-tuned using a subject-specific calibration dataset. In various embodiments, collected spectral data associated with a sample under analysis can be analyzed under one or more of multiple computational paradigms: (1) supervised learning using pathology-labeled spectra; (2) semi-supervised and unsupervised clustering for unlabeled or partially labeled spectra; and (3) fine-tuning or calibration on patient-specific data. In some such embodiments, the training datasets can include mixed sources from human, animal, and in- vitro tissues to evaluate algorithm generalization. In various embodiments, an Al-based or a probabilistic modeling technique can be employed to assign uncertainty to the determined at least one response characteristic of the nanostructured sensor in contact with the tissue sample.

[0037] A variety of nanostructured sensors can be utilized in the practice of the present teachings. By way of example, the nanostructured sensor can be a plasmonic sensor. This can utilize the highly sensitive optical properties of plasmonic structures for detection of at least one target attribute of the tissue. In various embodiments, the plasmonic sensor includes a plurality of nano-sized electrically conductive elements, where optionally the nano-sized electrically conductive elements are distributed over an electrically insulating substrate.

[0038] In various embodiments, the nano-sized electrically conductive elements are arranged relative to one another according to a periodic pattern. By way of example, and without limitation, such a periodic pattern can be characterized by a unit cell having any of a square, or hexagonal geometry, among others. In various embodiments, the nano-sized electrically conductive elements include a metal, such as gold or silver. In various embodiments, the nanosized conductive elements can have a plurality of different shapes. By way of example, the nanosized conductive elements can be in the form of nano-sized rods (e.g., in the form of cylinders), or any other suitable shape. In various embodiments, the nano-sized conductive elements can have tapered shape. In various embodiments, the dimensions of the nano-sized conductive elements can be in a range of about X / 2 to about X / 20, where X refers to at least one wavelength (e g., the longest) of the interrogating optical radiation. The flexibility in selecting the different shapes and dimensions of the nanostructured elements can help optimize the optical response for the sensor.

[0039] In various embodiments, the nanostructured sensor can include a plurality of photonic crystal elements. The photonic crystal elements can be distributed relative to one another in a variety of different ways, e.g., as a periodic pattern or non-periodic pattern. This alternative sensor design leverages photonic bandgaps for sensitive optical detection of tissue properties.

[0040] In various embodiments, the nanostructured sensor can be a hybrid sensor that includes a plasmonic array and at least one two-dimensional (2D) material layer, where the 2D material can include any of graphene, hexagonal boron nitride (h-BN), or a transition-metal di chalcogenide (TMD).

[0041] In various embodiments, the tissue sample includes a margin tissue at least partially surrounding a tissue suspected of being cancerous. In various embodiments, in such applications, the methods and systems according to the present teachings allow determining the likelihood that the margin tissue is at least partially cancerous. In some such embodiments, an Al classification can be utilized to classify the margin tissue, e.g., different portions thereof, as being cancerous or non-cancerous. For example, a map of likelihood that various spatial locations of the margin tissue are cancerous can be determined and when the likelihood for a spatial location is greater than a predefined threshold, that spatial location can be identified as being cancerous. In various embodiments, the threshold can be set by a user (e.g., a surgeon), thereby providing the user with the choice to utilize the likelihood data in a way that the user deems appropriate, e.g., based on the user’s clinical judgement.

[0042] The methods according to various embodiments for characterizing a tissue sample can be used in-vivo or ex-vivo. For example, in some embodiments, the nanostructured sensor can be used to evaluate a biopsy tissue sample. In other embodiments, a nanostructured sensor can be incorporated into a surgical device to be placed in contact with a tissue sample, e.g., a margin tissue sample, in-vivo. Such in-vivo applications of various embodiments of the present teachings advantageously integrate the sensing methods into a practical surgical workflow, facilitating realtime intraoperative analysis.

[0043] By way of example, in various embodiments, a surgical device according to the present teachings can include a radiation source that generates radiation at a plurality of frequencies that can be directed via an interrogation optical path to a nanostructured sensor coupled to the device, or placed on a movable mount in case of an in-vitro device, to be in contact with a tissue sample under analysis. A mechanism for moving the interrogating optical radiation over the sensor’s metasurface is employed to interrogate different spatially localized regions of the sensor’s metasurface, e.g., different localized subsets of the nanostructured elements. By way of example, such a mechanism can be a movable stage on which the tissue sample is positioned, an optical radiation scanning mechanism, such as two mirrors configured to rotate about two orthogonal axes, among others. The radiation redirected by or transmitted through the sensor and the tissue sample is directed via a detection optical path to a data acquisition / analysis module that can detect the redirected or transmitted radiation to generate detection data that can be processed according to the present teachings to determine changes in one or more response characteristics of the sensor, to characterize the tissue sample. In various embodiments, the optical interrogation and detection paths can include various optical components, such as fiber optics, lenses, beam splitters, etc., for guiding the interrogating optical radiation to the sample and for directing the redirected or transmitted radiation to a data acquisition / analysis module.

[0044] By way of example, and without limitation, the surgical device can be a resectoscope, which includes a beam scanning and conditioning module that is configured to scan the interrogating optical radiation across an area of the nanostructured sensor’s metasurface. As another example, the surgical device can be an optical needle probe having a needle for penetrating a tissue sample with a nanostructured sensor coupled, removably or fixedly, to a tip of the needle. By way of example, and without limitation, the needle’s tip can be formed of sapphire.

[0045] In a related aspect, a method of designing a nanostructured sensor is disclosed, which includes generating one or more initial theoretical designs of a nanostructured sensor by defining an initial configuration of a plurality of nanostructured elements via a plurality of sets of design parameters, initializing each of said sets of plurality of design parameters via assignment of a respective set of initial parameter values thereto, theoretically determining at least one response characteristic of said one or more initial theoretical designs of the sensor when said sensor is in contact with a theoretically-defined tissue sample and illuminated with interrogating optical radiation at a plurality of frequencies, evaluating a figure-of-merit function for said one or more initial theoretical designs based on said determined at least one response characteristic for said one or more initial theoretical designs, accepting at least one of said sets of said initial design parameters associated with at least one of said one or more initial theoretical designs as at least one initial optimal sensor design when said figure-of-merit function evaluated based on said at least one of said sets of said initial design parameters is optimal, iteratively adjusting one or more of said sets of initial design parameters to optimize said figure-of-merit function and thereby generating at least one initial optimal sensor design when said evaluated figure-of-merit function for said sets of one or more initial designs is sub-optimal, simulating or fabricating at least one test nanostructured sensor based on said at least one initial optimal sensor design, utilizing said simulated or fabricated test nanostructured sensor to determine at least one response characteristic thereof when said test nanostructured sensor is placed in contact with a reference simulated or real tissue sample and illuminated via simulated or real interrogating optical radiation at a plurality of optical frequencies, and analyzing said determined at least one response characteristic to perform at least one of (1) accepting said at least one initial optimal sensor design as a final sensor design, (2) adjusting at least one of said figure-of-merit function and one or more of sets of said design parameters to generate a revised initial optimal design of said nanostructured sensor.

[0046] In various embodiments, the step of analyzing the determined at least one response characteristic includes determining at least one attribute of said reference tissue sample based on said at least one response characteristic.

[0047] In various embodiments, the method can further include adding a simulated or previously-observed noise pattern to said at least one response characteristic and determining said at least one attribute in presence of the noise pattern.

[0048] In various embodiments, the figure-of merit function can be adjusted based on the determined at least one attribute.

[0049] In various embodiments including the above method of designing a sensor, a variety of figure-of-merit functions can be used. By way of example, the figure-of-merit function can include at least one of a number of spectral resonances exhibited by the nanostructured sensor over a target optical frequency band, a sharpness of at least one spectral resonance exhibited by the nanostructured sensor, a spectral separation between at least two spectral resonances exhibited by the nanostructured sensor, a classification accuracy of said tissue sample based on an Al algorithm trained based on previously-obtained simulated or real tissue classification data, and impact of simulated noise on classification accuracy of said tissue sample based on an Al classification algorithm. In various embodiments, the figure-of-merit function is defined as a weighted function of a number of spectral resonances exhibited by said nanostructured sensor over a target optical frequency band, a sharpness of at least one spectral resonance exhibited by said nanostructured sensor, a spectral separation between at least two spectral resonances exhibited by said nanostructured sensor, a classification accuracy of said tissue sample based on an Al algorithm trained based on previously -obtained simulated or real tissue classification data, impact of simulated noise on classification accuracy of said tissue sample based on an Al classification algorithm.

[0050] In various embodiments, the theoretically-defined tissue sample is defined based on a refractive index thereof. For example, the theoretically-defined tissue sample can be defined based on a distribution of the refractive index at a plurality of spatial locations of said tissue sample.

[0051] In various embodiments, the step of theoretically determining at least one response characteristic includes determining, for an array of initially designed nanostructured elements of at least one initially designed sensor in contact with the tissue sample, a map of response characteristics each corresponding to a localized region of the tissue sample, e.g., by simulating interrogation of different subsets of the nanostructured elements associated with different localized regions of the tissue sample with an interrogating optical radiation beam. Such a map of response characteristics can then be utilized to generate a corresponding spatial map of at least one attribute of the tissue sample. The figure-of-merit function can be evaluated via a comparison of the theoretically -determined spatial map of the at least one attribute with a previously-defined attributes of said tissue sample and the figure-of-merit function can be adjusted based on that comparison.

[0052] Various methods and systems according to the present teachings are not limited to characterization of tissue sample, but rather can be employed for characterization of a variety of different samples, such as semiconductor samples.

[0053] In a related aspect, a method for characterizing a sample is disclosed, which includes placing at least one nano structured sensor in contact with said sample, wherein said nanostructured sensor includes at least one response characteristic exhibiting dependence on a refractive index of said sample when said nanostructured sensor is in contact with said sample, directing interrogating optical radiation at a plurality of frequencies to said nanostructured sensor, detecting, for each of said optical radiation frequencies, radiation redirected by or transmitted through said nanostructured sensor in response to said interrogating optical radiation, analyzing said detected redirected or transmitted radiation to determine said least one response characteristic of said nanostructured sensor at each of said plurality of interrogating optical radiation frequencies, and identifying at least one attribute of said sample based on said measurements of said at least one feature at said interrogating optical radiation frequencies.

[0054] In some embodiments, the sample under investigation includes a tissue sample, a semiconductor sample, among others.

[0055] In another aspect, a method for correlating one or more optical properties of fresh and fixed tissue states is provided, which can enhance model accuracy and versatility. In this method, a tissue specimen can first be characterized in its fresh state using the present teachings. Subsequently, the same specimen can undergo a standard fixation or preservation procedure, for example, by immersion in saline, a short-chain alcohol, or a formaldehyde solution. During this process, the same spatial region of the specimen can be repeatedly interrogated by the sensor at predefined time intervals. This sequential measurement protocol generates a time-resolved dataset that quantifies the spectral evolution of the tissue as it transitions from a fresh state to partially fixed and fully fixed states.

[0056] Further understanding of various aspects of the present teachings can be obtained from the following detailed description in conjunction with the associated drawings, which are described briefly below.

[0057] BRIEF DESCRIPTION OF THE DRAWINGS

[0058] FIG. 1 is a flowchart depicting various steps of an embodiment of a method for characterizing a tissue sample.

[0059] FIG. 2A is a schematic depiction of a plasmonic nanostructured sensor with periodically distributed nanostructured elements.

[0060] FIG. 2B is a schematic side view of the plasmonic nanostructured sensor illustrated in

[0061] FIG. 2A, FTG. 2C is a schematic depiction of the plasmonic nanostructured sensor of FTG. 2A in contact with a tissue sample.

[0062] FIG. 3 is a schematic depiction of a plasmonic nanostructured sensor with non- periodically distributed nanostructured elements.

[0063] FIG. 4A is a schematic depiction of a plasmonic nanostructured sensor having nanosized holes in a free-standing metal sheet.

[0064] FIG. 4B is a schematic depiction of a side view of the plasmonic nanostructured sensor of FIG. 4A,

[0065] FIG. 5 is a schematic depiction of a hybrid nanostructured sensor.

[0066] FIG. 6 is a schematic depiction of a dielectric metasurface / 2D photonic crystal-based nanostructured sensor.

[0067] FIG. 7A is a schematic diagram of a system for characterizing a sample, such as a tissue sample.

[0068] FIG. 7B is a schematic diagram of a processing module of the system of FIG. 7A, including an Al engine.

[0069] FIG. 8A is a schematic depiction of a plasmonic sensor including nanostructured elements distributed as co-located periodic lattices.

[0070] FIG. 8B is a schematic depiction of a sensor including irregularly shaped nanostructured elements.

[0071] FIG. 8C is a schematic depiction of a sensor including two sets of nanostructured elements distributed as different arrays on two different parts of the sensor.

[0072] FIG. 9A is a schematic diagram illustrating partially overlapping optical beam spots on a sensor surface.

[0073] FIG. 9B is a schematic diagram illustrating disjointed optical beam spots on a sensor surface. FTG. 10 is a flowchart of an embodiment of a sensor design process utilizing an inverse design approach.

[0074] FIG. 11 is a schematic diagram of a system for performing ex-vivo characterization of a tissue sample.

[0075] FIG. 12 is a schematic depiction of an endoscopic surgical device (such as a resectoscope) for in-vivo tissue characterization.

[0076] FIG. 13A is a schematic depiction of a surgical device (motorized needle probe) according to an embodiment for volumetric tissue interrogation.

[0077] FIG. 13B is a schematic depiction of an inner portion of the device of FIG. 13A.

[0078] FIG. 13C is a schematic depiction of a surgical robotic system in which the surgical device depicted in FIGS. 13 A, 13B, integrating a motorized needle probe in which a nanostructured sensor according to various embodiments is incorporated.

[0079] FIG. 14A is a schematic depiction of a plasmonic nanostructured sensor with gold nanoantennas on a chromium layer.

[0080] FIG. 14B is a schematic depiction of a plasmonic nanostructured sensor similar to FIG. 14A, with an additional SiCh spacer.

[0081] FIG. 14C is a schematic illustration of the sensor of FIG. 14A in contact with a first sample (A).

[0082] FIG. 14D is a schematic illustration of the sensor of FIG. 14B in contact with a first sample (A).

[0083] FIG. 14E is a schematic illustration of the sensor of FIG. 14A in contact with a second sample (B).

[0084] FIG. 14F is a schematic illustration of the sensor of FIG. 14B in contact with a second sample (B).

[0085] FIG. 14G is a schematic diagram illustrating geometric parameters of nanostructured elements in a plan view. FTG. 14H is a schematic diagram illustrating geometric parameters of nanostructured elements in a side view.

[0086] FIG. 141 is a Scanning Electron Microscope image of a prototype sensor according to an embodiment.

[0087] FIG. 15A is a graph showing reflection data from a sensor in contact with glycerol-water mixtures.

[0088] FIG. 15B is a graph showing normalized reflection data from a sensor in contact with muscle and fat tissue.

[0089] FIG. 15C is an image of a labeled tissue sample showing fat and muscle portions.

[0090] FIG. 15D presents a confusion matrix associated with the data provided in FIG. 15A,

[0091] FIG. 16A is a graph showing measured spectra of labeled fat samples used for classifier training.

[0092] FIG. 16B is a graph showing measured spectra of labeled muscle samples used for classifier training.

[0093] FIG. 17A is an image representation of prediction functions for a background class.

[0094] FIG. 17B is an image representation of prediction functions for a fat class.

[0095] FIG. 17C is an image representation of prediction functions for a muscle class.

[0096] FIG. 17D is a predicted map of an unknown sample with confidence scores.

[0097] FIG. 17E is a graph illustrating the importance of spectral features for classification.

[0098] FIG. 18A is a schematic illustration of a sensor in contact with a healthy tissue.

[0099] FIG. 18B is a schematic illustration of a sensor in contact with a cancerous tissue.

[0100] FIG. 18C is simulated spectra of the sensors shown in FIGS. 18A and 18B in contact with healthy and cancerous tissue, respectively.

[0101] FIG. 19A shows the spectra from FIG. 18C with added noise.

[0102] FIG. 19B is the confusion matrix for a classification of the noisy spectra from Fig. 19A. FTG. 19C is the Receiver Operating Characteristic (ROC) curve for the classification.

[0103] FIG. 19D is the precision-recall curve for the classification.

[0104] FIG. 20A is an image of a histological tissue sample.

[0105] FIG. 20B is the "Ground Truth" classification for the tissue sample shown in FIG. 20A.

[0106] FIG. 20C is a predicted tissue map using a pretrained model.

[0107] FIG. 20D is a predicted tissue map using another pretrained model.

[0108] FIG. 20E is a predicted tissue map using a self-calibrated model only.

[0109] FIG. 20F is a predicted tissue map using both the pretrained and self-calibrated model.

[0110] FIGS. 21A-FIG.21D depict simulated spectra of two different sensors with noise following two different noise distributions.

[0111] FIG. 21E is a graph showing the classification accuracy of the two sensors on the simulated spectra from FIG. 21A-FIG.21D.

[0112] FIG. 22A is a schematic diagram illustrating an example of a sensing method according to the present teachings for detecting the presence of a nanoparticle.

[0113] FIG. 22B presents the total intensity obtained from the multi -spectral measurements of the sensor in close contact with the nanoparticle, as depicted in FIG. 22A.

[0114] FIG. 22C is the decision function for each pixel in FIG. 22B indicating its likelihood of being an outlier.

[0115] FIG. 22D shows the estimated outlier positions, indicating the presence of a nanoparticle.

[0116] FIG. 22E shows the spectral response from each pixel of the nanosensor in FIGS 22A, classified as an inlier or an outlier.

[0117] DETAILED DESCRIPTION

[0118] In various aspects, the present disclosure generally relates to utilizing nanostructured sensors to characterize various samples, such as biological tissues, by analyzing the optical response of the sensor to interrogating optical radiation when the sensor, or at least a portion thereof, is in contact with a sample. These nanostructured sensors, which can be, for example, plasmonic or dielectric metasurfaces, or photonic crystal-based nanostructured sensors, such as nanostructured sensors based on hybrid structures, 2D materials, among others, are designed to exhibit response characteristics that depend on the properties of the sample, such as its refractive index, including a spatial distribution of the refractive index.

[0119] In various embodiments, a sensor according to the present teachings can be used in a versatile manner to enhance the information that can be derived about a sample under investigation. For example, the interrogating optical radiation can be directed at the nanostructured sensor in a variety of ways, including adjusting the size of the interrogated area. This adjustable interrogation size allows for encompassing different numbers of nanostructures within the interrogating beam, thereby enabling a tunable resolution for determining the characteristics of the tissue sample. For example, the pixel size, determined by the spot size of the beam on the sensor can be adjusted to adapt it to specific needs such as different cancer types or cell sizes, as discussed in more detail below.

[0120] Furthermore, the physical arrangement of the nanostructures (herein also referred to as nanostructured elements) within the sensor itself can be highly customized to optimize performance for specific applications. For example, the nanostructures forming a sensor can be distributed in either periodic or non-periodic patterns, with the spacing between individual nanostructures being uniform or varied. For instance, the design can feature nanostructures that are more closely packed in certain portions of the sensor, such as a central area, compared to more peripheral portions, or vice versa, to achieve specific sensing characteristics. Further, in some embodiments, a portion of the nanostructures of a sensor can be arranged according to a periodic pattern and another portion can be arranged according to a non-periodic pattern. This design flexibility, which includes tunable parameters like periodic spacing of the nanostructures, e g., along two dimensions, and their dimensions, allows for the engineering of complex spectral signatures, which can in turn facilitate the determination of attributes of a sample under investigation.

[0121] This customizable architecture of sensors according to various embodiments can contribute to the sensor's ability to generate distinct spectral features that act as independent response characteristics channels, enhancing detection robustness and adaptability to diverse sensing challenges. It is noted that in various embodiments, the response characteristics with the spectral channels can be inter-related. For example, the shift in the center frequency of a resonance feature in response to contact of the sensor with a particular tissue type may be correlated with a respective shift of another resonance feature of the sensor.

[0122] In various embodiments, by combining nanosensor designs with advanced computational support, including artificial intelligence (Al) and machine learning algorithms, the present disclosure offers a robust and user-friendly platform for precise material classification and mapping. This unique combination addresses the limitations of existing methods by providing real-time, accurate, and efficient characterization capabilities across a wide range of applications, from industrial quality control to medical diagnostics. For example, in various embodiments, the methods and systems according to the present teachings can be utilized for accurate classification of tissue types, e.g., for distinguishing healthy from cancerous tissue. In some embodiments, methods, systems and devices according to the present teachings can be used to detect multiple attributes in a sample simultaneously using the same physical sensor, for example, distinguishing between more than two tissue types in the same sample.

[0123] In various embodiments, the methods and systems according to the present teachings overcome the limitations of existing characterization methods, e.g., by combining high resolution and user-friendly operation with advanced computational support. For example, unlike existing methods, in various embodiments, the methods according to the present teachings can provide intraoperative feedback to surgeons regarding tissue characteristics (e.g., whether a margin tissue sample is cancerous or not) with a high resolution, tissue preservation as well as easy, and direct readout, without any need for a special interpretation by an expert. For example, a false-color map of a target tissue sample can be provided to assist the surgeon during surgery to distinguish tissue locations that are likely to be cancerous from those that are likely non-cancerous. In some such embodiments, the surgeon can select the likelihood threshold that is suitable for identifying a tissue portion as being cancerous. Moreover, as discussed above, the use of Al engines for analysis allows flexibility in using different levels of Al analysis for characterizing a tissue sample. For example, a surgeon can be provided with the choice of whether to a) turn on a prediction layer trained on a database of certain lesion types, or certain patient group, or b) display the raw pixel-by-pixel classification result only. Tn various embodiments, the AT engine provided with a system according to various embodiments can be updated, e.g., periodically, via a remote connection.

[0124] As noted above and discussed in more detail below, in various embodiments, systems and methods according to the present teachings utilize label-free nanophotonics designs, such as nanostructures (e.g., nano-holes or nano-antenna arrays) that support surface waves and localized field enhancements and light confinement. The resonance frequency of these nanostructures is highly sensitive to their cavity shape, geometry, and the surrounding tissue or material. The design process involves optimizing the nanophotonic structure to generate an optical signature that can enable an accurate tissue type classification, under conditions of biological variability (caused by, e g., a different skin composition, or a difference in individual tissue density) and even from noisy measurements. In various embodiments, this is achieved by constructing a figure-of-merit function that favors structures with multiple high-Q spectral resonances in the optical range, which is then used in an inverse design algorithm to obtain sensor designs that allow accurate characterization of a sample of interest, e.g., a tissue sample. Additional information about constructing a figure-of-merit function for use in various embodiments of the present teachings is provided below.

[0125] To further enhance accuracy and efficiency, in various embodiments, the present teachings integrate state-of-the-art generative and physics-based AT simulation tools during the sensor hardware design phase, allowing for rapid exploration of design possibilities and convergence to optimal multi-resonant structures.

[0126] As noted above, and explained in more detail below, various embodiments of the methods and systems according to the present teachings employ an array of nanostructured elements that are designed to be placed in contact with the sampled environment, e.g., a tissue sample. The nanostructured elements can be optically interrogated to derive information about the sampled environment. For example, at least one optical characteristic of the nanostructured elements, e.g., the frequency or the shape of one or more optical resonances exhibited by the nanostructured elements, can be sensitive to the sampled environment, thus allowing obtaining information about the sampled environment via optical interrogation of the nanostructured elements and measuring a change is said at least one characteristic of the nanostructured elements. Tn various embodiments, the high sensitivity of one or more optical characteristics of a nanostructure sensor according to the present teachings allows determining a map of a target attribute of a sample under investigation, e.g., a tissue sample.

[0127] In various embodiments, the contact between a sensor according to various embodiments and a sample, i.e., at least a portion of the sensing surface of the sensor touching at least a portion of a sample, such as a tissue sample, enables high sensitivity to nanoscale environmental changes, since each metasurface pixel locally senses its surroundings independently, eliminating cross-interference between regions and providing a clear, quantitative signal unlike far-field imaging methods that reconstruct the image from distorted scattered waves. In some embodiments, the very close proximity of the sensor’s sensing surface to the sample, such as a semiconductor sample, provides high sensitivity to the presence of nanoscale defects on the sample surface, such as nanosized metal residues, micro-scratches, or other contaminants, and enables robust detection of these defects by the metasurface pixels.

[0128] As discussed in more detail below, in various embodiments, Al and / or machine learning classification algorithms can be utilized to process the collected optical data acquired via optical interrogation of a nano structured sensor to classify a sample, such as a tissue sample, under investigation, e.g., to distinguish a healthy tissue sample from a cancerous tissue sample. Tn particular, in various embodiments, the Al and / or machine learning classification algorithms can process measured redirected or transmitted radiation associated with each pixel of the sensing surface of the sensor (i.e., associated with each spatially localized portion of the sensing surface interrogated by a radiation beam) to generate a pixel-by-pixel classification of the sample, e.g., a margin tissue sample, under investigation. In various embodiments, the classification of each pixel can be performed independent of the classification of the other pixels.

[0129] Further, in various embodiments, advanced Al-based and probabilistic modeling techniques can be employed for uncertainty quantification and image enhancement techniques can be used to improve data quality and provide clearer representations of the analyzed environment. In various embodiments, the Al engine may further integrate contextual information — such as prior results or clinical data — to guide next steps and enhance classification. Such a comprehensive approach can ensure a highly accurate, reliable, and adaptable characterization of various samples.

[0130] In some embodiments, the Al system further provides contextual recommendations for next procedural steps — such as extending excision, how many lymph nodes to remove, repeating measurement, or marking resection areas — based on statistical analysis and prior outcomes. This can be presented as a decision-support layer for the surgeon.

[0131] Without any loss of generality and simply for ease of description, in the following detailed description, various features of the present teachings will be described in connection with the characterization of tissue margins surrounding a tissue portion (e.g., a tumor) that is suspected of being cancerous.

[0132] Various terms are used herein in accordance with their ordinary meanings in the art. By way of further elucidation, several terms are defined below.

[0133] As used herein, a “nanostructured sensor” refers to a small sensor, e.g., having at least one dimension that is equal to or less than about 1 micrometer, that is designed to exhibit a change in one or more of its characteristics when in contact with a surrounding material and / or environment, e.g., when in contact with a semiconductor or a tissue sample. Such a change in one or more of the sensor’s characteristics can be detected using optical interrogating radiation. Some examples of nanostructured sensors suitable for use in the practice of the present disclosure include, without limitation, metasurface sensors, plasmonic sensors, sensors including photonic crystals, and hybrid sensors (e.g., a plasmonic sensor formed of a plurality of nanostructured metal antennas disposed on an underlying substrate and a 2-D sheet of material, such as a graphene layer, deposited over the nanostructured metal antennas). A “sensing surface” of a sensor refers to a surface of a nanostructured surface on which or in which the nanostructured elements are formed.

[0134] For the purposes of this disclosure, a nanostructured sensor being in contact with a sample, such as a tissue sample, broadly refers to a configuration where there is either a direct physical contact between at least a portion of the nanostructured sensor and the sample being analyzed, or the sample is in the very close proximity (e.g., at a distance less than about 100 nm nanometers, as in the near field region of the nanostructured sensor) to at least a portion of the nanostructure sensor. In cases in which the nanostructured sensor is in close proximity to the sample, the gap between the nanostructured sensor and the sample may be an air gap, or may be filled with a dielectric material. This contact (either direct or with a small separation between the nanostructured sensor and the sample) allows for an effective interaction of the sensor's active elements with the sample's surface or immediate subsurface, enabling the detection of changes in the sample's properties, such as its refractive index, which can directly influence the sensor's response characteristics. Typically, the opposed surface of the sensor (i.e., the sensor’s surface opposite to the surface that is in contact with the tissue sample) is illuminated by an optical interrogating beam. This definition encompasses scenarios ranging from a sensor’s surface, or at least a portion thereof, gently touching a sample to the sensor’s surface being slightly pressed onto a sample to ensure proper contact or being in close proximity to the sample. In other embodiments, the interrogating beam may pass through the sample first (such as a transparent liquid layer or a thin solid layer to be analyzed) and then reach the surface of the nanostructured sensor; also in this case, the previous definition of being in contact applies.

[0135] For the purposes of this disclosure, the term “a response characteristic” refers to any measurable physical or optical property or behavior exhibited by the nanostructured sensor when it is interrogated by optical radiation while in contact with a sample, which can include either direct contact or close proximity between the nanostructured sensor and the sample. This response characteristic can inherently depend, and can therefore be indicative of, the physical, chemical, or biological properties of the sample, particularly its refractive index. A response characteristic can manifest itself in various forms, including but not limited to: the intensity of redirected or transmitted radiation; parameters associated with spectral resonance features (such as spectral position, spectral width, spectral shape, quality factor (Q), group delay, spectral derivative, or peak asymmetry); phase shifts between incident and redirected radiation; changes in polarization; integrated intensity of reflectivity over specific spectral bands; temporal responses to pulsed radiation; scattering angle distributions; or the formation of interferometric patterns. This broad definition encompasses both single measurable values and complex spectral or angular datasets that collectively characterize the sensor's interaction with the sample. In this disclosure, the terms “radiation” and “light” are used interchangeably and the term “optical radiation” refers to electromagnetic radiation with vacuum wavelengths in a range of visible to infrared, e.g., in a range of about 450 nm to about 1500 nm. For example, the interrogating optical radiation can have a vacuum wavelength in a range of about 400 to about 750 nm (visible range), or in a range of about 750 nm to about 1700 nm (NIR / SWIR), in a range of about 2 micrometers (microns) to about 5 microns or in a range of about 8 microns to about 12 microns (MIR). It should be understood that the sizes of the nanostructured elements can vary depending on the wavelength of the radiation. In general, as the wavelength of the interrogating radiation deceases, so does the sizes of the nanostructured elements of the sensor. Such adjustments are taken into account by various methods of designing a nanostructured sensor according to various embodiments, as discussed in more detail below.

[0136] In various embodiments, the nanostructured metasurface can include sub -wavelength resonant elements whose characteristic lateral dimension is selected as a fraction of the effective wavelength of the interrogating optical radiation within the surrounding medium, typically in a range of about X / 20 to X / 2. For operation in the visible to near-infrared region (approximately 400 nm - 1.7 pm), the individual nano-antennas or resonators may have in-plane dimensions of about 50 - 500 nm; for mid-infrared operation (approximately 2 - 12 pm), the corresponding dimensions may be about 0.5 - 6 pm. The array period is generally between about 0.3 - 1.3 times the effective wavelength to support localized or lattice resonances while avoiding unwanted diffraction orders. In various embodiments, the conductive or high-index resonators (for example gold, silver, aluminum, copper, doped semiconductors such as ITO or AZO, or high-index dielectrics such as Si, Ge, TiO2, SisN4, or GaAs) can have a thickness sufficient to exceed several optical skin depths or to confine Mie-type modes, typically about 20 - 200 nm for visible-NIR operation and up to about 300 nm for mid-IR operation.

[0137] A plasmonic sensor is a device that exploits the interaction between electromagnetic radiation and free electrons at the surface of a metal-dielectric to detect and / or analyze various samples. Its main operational principle is based on phenomena such as surface plasmon resonance (SPR) or localized surface plasmon resonance (LSPR), where collective oscillations of conduction electrons in metallic nanostructures (e.g., nanostructures, such as nanoparticles or thin films of gold, silver, or other plasmonic metals) are excited by incident light. In various embodiments, the sensor’s operational principle may combine additional optical phenomena such as Fabry-Perot resonance, Fano-type coupling, or any other multi -resonant coupling. A sensor based on a photonic crystal refers to a nanostructured sensor that utilizes photonic crystals, which are periodic optical nanostructures designed to affect the motion of photons in a similar way that semiconductor crystals affect electrons. These structures can exhibit lower signal loss and support high-Q resonances in different spectral windows, providing versatility in sensing applications. In some embodiments, a sensor fabricated based on 2D materials can be utilized in the practice of the present teachings. A key property of such materials that render them suitable for use in various embodiments of the present teachings is their ability to provide light confinement (in any direction, planar, one-dimensional, etc.), which can be obtained in various ways - 2D materials, thin layers of semiconductors, dielectric or metal thin layers (thin here means less than 400 nm).

[0138] As used herein, the term “tissue” refers to an organized ensemble of cells, and often multiple cell types, along with their extracellular matrix, that collectively form a structural and functional unit within a biological organism. A tissue is, therefore, fundamentally a multicellular structure, thereby distinguishing it from an individual, isolated cell. This definition is intended to be broad and encompasses a wide variety of biological materials, including but not limited to, healthy tissue, diseased tissue (such as cancerous or non-cancerous tissue), margin tissue, and different tissue types like muscle, fat, or connective tissue.

[0139] Without any loss of generality, in the embodiments discussed below, it is assumed that the sample under examination is a tissue sample, such as a margin tissue sample. It should, however, be understood that methods, systems, and devices according to the present teachings can also be employed for characterizing other types of samples, such as semiconductor samples. For example, the present teachings can be used to detect defects and / or contaminants on a semiconductor surface.

[0140] With reference to the flow chart of FIG. 1, in one embodiment, a method for characterizing a tissue sample is disclosed, which includes placing at least a portion of at least one nanostructured sensor in contact with the tissue sample. The sensor is designed to include at least one response characteristic that exhibits a dependency on the refractive index of the tissue sample when in contact therewith, making it sensitive to the material properties of the tissue sample. Subsequently, interrogating optical radiation is directed at the nanostructured sensor at a plurality of distinct frequencies, e.g., across a frequency band. Subsequently, for each of these optical radiation frequencies, the radiation that is either redirected by the sensor or transmitted through the nanostructured sensor (and the sample that is in contact with the sensor) is detected. The detected radiation is then analyzed to determine the aforementioned response characteristic of the nanostructured sensor at each of the plurality of interrogating optical radiation frequencies. Finally, based on these measurements of the response characteristic at the various optical frequencies, at least one attribute of the tissue sample can be identified, providing crucial information about the tissue's characteristics. In some cases, the interrogating optical beam can be directed to different spatial locations of the sensor such that at each location a subset of the nanostructured elements is within the beam’s diameter. The response characteristic of the sensor for each localized illuminated portion (i.e., each pixel) can be determined and processed to obtain a spatial distribution of said at least one attribute. In this manner, a spatial map of the at least one attribute of the sensor can be determined (e.g., a spatial map of cancerous and non- cancerous tissue portions).

[0141] As noted above, in various embodiments, the wavelength of the interrogating optical radiation can be in a range of about 400 nm to about 750 nm, or in a range of about 750 nm to about 1700 nm, or in a range of about 2 microns to about 5 microns, or in a range of about 8 microns to about 12 microns. As the refractive index of a given tissue sample or other samples can depend on the wavelength (i.e., the refractive index can exhibit dispersion), measuring response characteristics of a sensor in contact with various samples at two or more different wavelength regimes (e.g., at 500 nm and 1200 nm) can produce different (e.g., complementary) results (e.g., likely distinct noise characteristics), though typically with some correlation between the results. In this manner, additional information regarding the response characteristics may be obtained from measurements at multiple frequencies, increasing the effective signal -to-noise ratio.

[0142] The methods according to various embodiments for characterizing a sample, such as a tissue sample, are highly adaptable, utilizing a broad class of nanostructured sensors for enhanced detection capabilities. These nanostructured sensors can be fabricated from diverse materials and designs, including plasmonic, dielectric, 2D-materials and semiconductor-based nanostructures, each offering unique advantages. For instance, plasmonic nanostructured sensors, particularly those made from gold, can be advantageous, e.g., due to their high sensitivity to changes in refractive index and superior resolution due to the strong localization and enhancement of the electric field. Gold nano-antennas or nano-cavities support surface plasmons whose resonance frequency is acutely sensitive to the antenna's shape, geometry, and its surrounding tissue.

[0143] In contrast, dielectric-based nanostructures, such as dielectric meta-surfaces and photonic crystals (which can include materials like silicon-based metamaterials), offer lower signal loss and support high-Q resonances in different spectral windows, providing additional robustness; however, this usually comes at the expense of sensitivity to small changes in the environment. This flexibility in sensor design allows for optimization based on specific application needs and desired spectral responses, ensuring robust and accurate detection across various scenarios.

[0144] By way of example, FIGS. 2A and 2B schematically depict an example of such a plasmonic sensor 200 that includes a plurality of nanosized electrically conductive structures (herein also referred to as nanosized antennas) 202 that are distributed periodically on an underlying substrate 204. By way of example, and without limitation, the nanosized conductive structures have at least one dimension in a range of about 10 nm to about 200 nm, though the sizes of the nanostructures are not limited to this range and can be determined using the design methods according to various embodiments as described below for a given application of the sensor. Further, in some embodiments, the nanosized antennas can have a uniform or tapered shape along their longitudinal dimension and variety of cross-sectional shapes, such as circular or polygonal. The dimension and the spacing between the nanostructured antennas can be designed, in a manner discussed in more detail below, to impart desired optical characteristics to the plasmonic sensor so that it can optimized for use in a particular application, e.g., in classification of tissue samples. The nanosized antennas can be formed of a variety of different electrically conductive materials, such as, gold, and silver. Further, in various embodiments, a nanostructured sensor can be formed of a stack of thin layers of materials, where each thin layer exhibits a thickness (herein also referred to as the size along the Z-axis) of about 1 micron or less, e.g., in a range of about 20 nm to about 1 microns, and macroscopic dimensions, e.g., greater than 50 microns, along the length and the width dimensions (herein also referred to, respectively, as the X and Y dimensions). Such stacked layers of these materials can exhibit resonance / confinement of radiation that can be sensitive to small changes in the environment, thereby allowing sensing attributes of a sample in contact therewith. By way of example, as shown in FIG. 5, a nanostructured sensor according to various embodiments, can include an SiCh spacer layer disposed on the top of a sensing surface of a plasmonic sensor.

[0145] FIG. 2C shows the plasmonic sensor 200 in contact with a tissue sample 206 via top surfaces of the nanosized antennas. An optical beam 201 at a plurality of frequencies (e.g., in the form of a broadband beam or multiple narrow-band, e.g., laser beams) is directed to illuminate a localized portion of the sensor, which contains a subset of the nanosized antennas 202 that are in contact with a portion of the tissue sample 206 directly beneath it. The light 203 reflected by this illuminated portion is then detected and processed, e.g., via Al engines, to determine at least one response characteristic (herein also referred to as at least one attribute) of the respective tissue portion. Each such localized measurement, corresponding to the area illuminated by the laser beam, is referred to as a "pixel" in a resulting data map generated via interrogation of different localized portions of the sensor’s surface, providing a high-resolution insight into the tissue's properties

[0146] For each pixel, the sensor can be interrogated at a plurality of frequencies using various methods. For instance, a broadband optical beam can be used to illuminate each sensor portion with a plurality of radiation frequencies, allowing for the detection and analysis of reflected light across a wide spectral range to determine the tissue characteristics. Alternatively, data for a pixel can be obtained by illuminating different portions of the sensor (i.e., different subsets of the nanosized antennas, where the subset can be partially overlapping or disjoint), or the same portion sequentially, with multiple narrow-band beams, such as laser beams, each corresponding to a specific frequency. In some embodiments, rather than interrogating different spatial locations of a sensing surface of a nanostructured sensor in different temporal intervals, all locations of interest are interrogated substantially concurrently, e.g., by concurrent illumination of those locations by a plurality of light beams.

[0147] The reflections corresponding to the different frequencies are then detected and analyzed to determine one or more response characteristics of the tissue sample relevant to that pixel. To generate a comprehensive map of the tissue sample, the interrogating beam can be systematically swept or scanned over the entire array of nanostructured elements of the sensor, allowing for the acquisition of pixel-by-pixel information and the creation of a detailed map of the tissue's characteristics.

[0148] The reflected light from each illuminated portion (pixel) of the plasmonic sensor is analyzed to determine a specific response characteristic of the sensor. This response characteristic provides critical information about the underlying tissue. For instance, a response characteristic can encompass the central frequency, the spectral width, and / or the overall spectral shape of one or more optical resonances exhibited by the sensor. These resonance properties are highly sensitive to the local dielectric environment, i.e., the tissue sample in contact with the nanosized antennas. By varying the frequency of the interrogating beam, an optical spectrum of the reflected light can be measured for each pixel, which can allow precise determination of such response characteristics.

[0149] As discussed in more detail below, in various embodiments, an Artificial Intelligence (Al) engine is employed to process these determined response characteristics for each pixel. The Al engine can implement a supervised, a semi -supervised or an unsupervised algorithm. For example, the Al engine can be trained to correlate specific response characteristics (e g., spectral signatures, such as those associated with the optical resonances of the sensor) with various attributes of the tissue sample. For example, for a tissue sample, a key attribute that can be determined is the likelihood of a tissue portion to be cancerous or non-cancerous. In various embodiments, the ultimate output of this process can be a high-resolution map illustrating the spatial distribution of the target attribute across the tissue sample, with each pixel on the map corresponding to a specific measurement point on the tissue. By way of example, to achieve this, an Al engine trained on response characteristics of the sensor for each pixel, when the sensor is in contact with a simulated or a real reference tissue sample, can be employed. For example, such reference characteristics can be obtained either from previous measurements using the same sensor (or a similarly -fabricated sensor) on known tissue types, or from simulated responses of the sensor in contact with reference tissue samples with known properties. This comparative analysis, facilitated by the Al engine, enables accurate classification and mapping of tissue attributes.

[0150] In some embodiments, the data acquired for characterizing a tissue portion extends beyond spectral information to include angular dependence of the reflected light as the angle of the incident light is varied (e.g., continuously over a range or at a plurality of discrete angles). For at least one pixel corresponding to at least one spatial location interrogated by the interrogating optical beam, the reflection data is obtained not only for multiple frequencies of the incident light, but also for different angles corresponding to the incidence of the interrogating optical beam on that pixel. In other words, for at least one pixel and at least for one frequency, the angular dependence of the reflected light — for example, in terms of the intensity of the reflected light and / or the frequency of the reflected light (e.g., a frequency shift between the incident and the reflected light) as a function of the angle of incidence — can be acquired and subsequently analyzed.

[0151] In some embodiments, for each pixel and for each frequency of the incident optical radiation, the angular distribution of the reflected light can be measured. This comprehensive angular data, whether derived from varying incidence angles or detection angles, provides a significantly richer dataset for each pixel. This detailed angular and spectral information, when processed by the Al engine, allows for a more nuanced and accurate determination of the tissue characteristics, further enhancing the system's diagnostic capabilities. In some embodiments, the nanostructured elements of a sensor may be positioned on a flexible substrate and hence may present different orientations to the incoming interrogating optical radiation. In some such cases, the data utilized for training an Al engine can include data obtained at multiple illumination angles. Further, in some embodiments, the optical interrogating radiation is applied to one or more localized portions of a sensor at different angles and the redirected or transmitted radiation is detected and measured at those angles to provide additional data for a robust characterization of a sample under investigation, such as a tissue sample. For example, such angular data can be useful for compensating for unknown effects of light polarization, noise, etc.

[0152] To ensure the system's accuracy and reliability in clinically realistic settings, such as an active surgical field, certain embodiments employ a specific training methodology to make the Al engine robust against intraoperative contamination. Such a methodology can involve simulating such conditions through a process of training data augmentation. In an exemplary implementation, a thin film of an interfering substance, such as blood or dyed saline, is intentionally applied to the surface of a tissue specimen being used for training or calibration. A first set of hyperspectral data is acquired from the contaminated specimen. Subsequently, the contaminant is partially or fully removed, for instance by wiping the surface, and a second set of hyperspectral data is acquired from the same location.

[0153] The spectral data collected from the tissue in its pristine state, its contaminated state, and its post -wiped state are all incorporated into the training dataset for the Al engine. By training on this augmented and diverse dataset, the Al engine learns to identify and deconvolve the specific spectral signatures associated with common contaminants from the desired spectral signature of the underlying tissue. The algorithm is thereby configured to computationally normalize, filter, or otherwise account for the artifacts introduced by these superficial fluid layers. Consequently, the resulting classification model exhibits significantly enhanced stability and accuracy, capable of rendering a reliable determination of the tissue's attributes even in the presence of confounding signals from residual fluids. This capability is critical for the practical and effective deployment of the system in real-world surgical environments where pristine sample conditions cannot be guaranteed.

[0154] In various embodiments, the system's operational flexibility further extends to dynamically adjusting the interrogating optical beam. By way of example, the beam’s shape, the beam’s size (e.g., its diameter), the beam’s polarization, and frequency bandwidth can be adjusted, e.g., based on a sensor design or in real-time between different scans of a nanostructured sensor or even during a single scan, e.g., based on processing of the acquired measurements of the at least one response characteristic.

[0155] In various embodiments, an optical interrogating radiation beam can be scanned over the nanostructures of a sensor, where for each subset of the nanostructures within the diameter of the optical radiation beam at a given time, at least one respective response characteristic of the nanostructured sensor can be measured and then analyzed, e.g., via an Al engine, to determine at least one attribute of the sample under examination corresponding to the sample’s portion in contact with that subset of the nanostructures can be determined. In this manner, a pixel-by-pixel determination of the sample’s attribute can be achieved. In this manner, a map of said at least one attribute (e.g., a map indicative of the spatial distribution of the refractive index of the tissue sample) can be obtained. In some embodiments, the scanning of the optical interrogating beam over the nanostructures of the sensor is configured such that there is no overlap between two consecutive positions of the optical beam over the sensor surface. The scanning of the radiation beam over the nanostructures can be performed in a variety of different ways, e.g., corresponding to different patterns of the radiation beam over the sensing surface of the nanostructured sensor.

[0156] In various embodiments, the shape of the optical beam (e g., circular, elliptical, etc.), its polarization, its frequency range, and an illumination pattern of the sensor’s sensing surface can be determined by the system’s control algorithm or by user selection. By way of example, in some embodiments, the system’s algorithm can define the spatial locations for illumination and the order in which those locations will be illuminated to acquire redirection or transmission data. Further, in various embodiments, the spatial sampling locations can be determined adaptively in real-time based on preliminary measurements. Further, an operator can manually select specific spatial locations and / or regions to sample (i.e., to illuminate and acquire data), allowing targeted verification within a larger tissue area. This flexible control enables optimization of acquisition speed, resolution, and sensitivity for various tissue types and surgical workflows.

[0157] In various embodiments, the methods and systems according to the present teachings can be used in environments where sampling locations are spatially or mechanically constrained, for example, when using a motorized micro-needle, an endoscopic probe of other volumetric measurement devices for accessing a tissue sample. In such cases, it may not be feasible to freely select arbitrary spatial locations for data acquisition. In such cases, a control algorithm can determine an optimized sequence of spatial sampling locations, including their depths and orientations below a reference surface based on prior measurements, signal confidence, or expected tissue heterogeneity. This adaptive selection can minimize the number of sampling actions while maximizing diagnostic coverage and accuracy. Although particularly useful in needle-based or endoscopic systems, this approach may be applied in any measurement configuration where sequential or spatially limited sampling is performed.

[0158] In certain embodiments, an initial pass over the sensor can be performed using a wider optical beam to acquire a coarse dataset. Such a coarse dataset can allow for rapid identification of broader regions of interest within the tissue sample. Once these specific regions of interest are identified, subsequent passes can then be conducted using a narrower beam, enabling interrogation of these areas at a significantly higher resolution. This multi -resolution scanning approach can optimize the data acquisition process, focusing a detailed analysis where it is most needed. Similar to other embodiments, for each pixel, regardless of the beam size used, the optical interrogation can still involve multiple radiation frequencies, allowing for the capture of a comprehensive spectral signature from each point of measurement. Further, in some embodiments, during a single scan, the beam size can be adjusted. For example, such an adjustment of the beam size can help resolve measurement uncertainty, e.g., by examining smaller subsets of the nanostructure elements for each pixel measurement to enhance data resolution (e.g., in some cases a micron-scale resolution may be achieved).

[0159] In some embodiments, the nanostructured sensor can include a non-periodic pattern of nanostructured elements, offering enhanced flexibility and customization compared to the periodic arrangements described for other embodiments. For example, such a plasmonic sensor can include nanostructured elements in the form of metallic nanosized rods, but distributed over an underlying substrate according to a non-periodic pattern. This non-periodic arrangement can involve varying the surface density of the nanostructured elements (i.e., the number of nanostructured elements per unit area) across the sensor's surface. For instance, the nanostructured elements can be more closely packed in the center of the sensor than towards its periphery, or vice versa. Alternatively, different surface densities of nanostructured elements can be implemented in distinct portions of the sensor to achieve specific sensing characteristics or tailored detection capabilities for different regions of the sample. As discussed in more detail below, in some embodiments, a nanostructured sensor can be designed by using different figure- of-merit functions for different portions thereof to achieve a desired characterization goal. For example, for characterizing a tissue sample including both a margin tissue portion as well as a portion of a tumor left in the sample as a guide for examination of the margin tissue portion, the surface density of the nanostructured elements may be different in areas of the sensor that are configured to be in contact with a portion of the tissue margin sample close to the tumor portion than those that are located farther from the tumor portion.

[0160] For example, FTG. 3 schematically depicts a plasmonic sensor 300 that similar to the plasmonic sensor 200 discussed above includes a plurality of nanostructured conductive elements 302 distributed over an underlying substrate 304. Unlike the plasmonic sensor 200, the nanostructured conductive elements 302 are not distributed in a periodic manner across the substrate. Rather, the surface density of the nanostructured elements 302 (i.e., the number of the nanostructured elements per unit area) varies in different portions of the sensor. For example, in this embodiment, the surface density of the nanostructured elements 302 in a central region (designed by a solid circle) is greater than the respective density of the nanostructured elements 302 in a peripheral portion of the sensor. The pattern of variation in the surface density of the nanostructured elements is not limited to that shown for sensor 300 but can be implemented in a variety of different ways, e.g., depending on a particular application. By way of example, in certain tissue classification applications in which a margin tissue surrounding a tumor is under investigation, it may be important to have a higher resolution map of the attributes of the margin tissue closer to the tumor than farther away. In some embodiments, by adjusting the size of the interrogating optical beam (making the optical beam smaller) and / or the scanning rate of the interrogating optical beam (reducing the scan rate) a higher resolution map of at least one attribute of the tissue margin (e g., whether the tissue margin is cancerous) in the regions closer to the tumor portion can be obtained. Instead of or in addition, the sensor itself can be designed with nanostructured elements more closely packed in some portions of the sensor compared to others. For example, in an application in which the sensor is intended to be used to examine tissue samples containing a margin tissue portion and a cancerous tissue portion along an edge of the margin tissue portion, the sensor can be designed to have a different surface density (i.e., the number of nanostructured element per unit surface area) of the nanostructured elements in a region of the sensor that is configured to be in contact with a region of the margin tissue portion close to the edge.

[0161] The plasmonic nanostructure sensors that can be employed in various embodiments of the present teachings are not limited to those described above. By way of example, FIGS. 4A and 4B show a plasmonic sensor 400 that includes a free-standing sheet of metal 402, such as gold, in which a plurality of nano-sized holes 404 are formed. The free-standing metal sheet 402 is supported at its two sides by two silicon ridges 403a / 403b. These nano-sized holes act as the nanostructured elements of the sensor, similar to the nano-antenna arrays described above in connection with sensors 200 and 300. The gold sheet, being a plasmonic metal, supports the excitation of surface plasmons when interrogated by optical radiation. The geometry, size, and arrangement of these holes are critical design parameters, as their resonance frequency is highly sensitive to the cavity shape and the surrounding material. When the free-standing metal sheet with its array of nano holes is brought into contact with a tissue sample, the optical properties of the tissue sample influence the plasmonic resonances within and around these holes. By detecting and analyzing the reflected or transmitted light, changes in these resonances can be correlated to the characteristics of the tissue sample. Such a free-standing configuration can be particularly useful in using the sensor in transmission mode. In various embodiments, the freestanding configuration can support sensing when the illuminating beam is coming from either side of the sensor, and in either reflection, scattering or transmission mode (4 different sensing modes).

[0162] In some embodiments, the nanostructured sensor can be a hybrid sensor that includes a plurality of different nanostructured elements and / or thin layers formed of different materials, such as nanoantenna arrays, nanohole arrays, dielectric spaces, TMD layer, semiconductor layers, among others. In certain such embodiments, the metasurface can include one or more two-dimensional (2D) materials, such as graphene, hexagonal boron nitride (h-BN), or transition-metal dichalcogenides (TMDs) including M0S2, WS2, MoSe2, or WSe2, integrated as an ultrathin layer (for example 0.5 - 10 nm in total thickness) on or between the resonant elements or spacer layer. These 2D materials may enhance field confinement, enable active electro-optic or chemical tuning, or provide additional excitonic or vibrational resonances in the visible-mid-infrared range.

[0163] When a metal-insulator-metal or dielectric-spacer-reflector configuration is used, the spacer layer (for example SiCh, AI2O3, S1 N4, Caf2, AIN, or sapphire) may have a thickness in a range of about 5 - 100 nm for visible-NIR operation or 20 - 300 nm for mid-IR operation to achieve near-field confinement at the sensing interface. These dimensional relationships can provide strong electromagnetic confinement and high refractive-index sensitivity across a spectral range of about 400 nm - 12 pm, encompassing visible, near-infrared, and mid-infrared regimes.

[0164] By way of illustration, FIG. 5 schematically depicts such a sensor 500 having a plasmonic array of nanostructured elements 502 deposited on an underlying electrically insulating substrate 504, e.g., a SiCh substrate. The sensor 500 further includes a WS2 / TMD layer 506 that is separated from the plasmonic array 502 by a thin layer of an electrically insulating layer 508, e.g., a SiCh layer. As noted above, in various embodiments, instead of or in addition to using a plasmonic sensor, a nanostructured sensor based on a dielectric metasurface or a photonic crystal can be used. By way of example, FIG. 6 schematically depicts such a sensor 600 formed as periodic layers of dielectric materials that include a plurality of nano-sized elements 602 with different indices of refraction designed to control and manipulate the flow of light, much like semiconductor crystals control electrons. With regards to dielectric metasurfaces, these are structured materials comprising subwavelength dielectric nano-elements engineered to support localized optical resonances, enabling controlled light manipulation and high- precision sensing applications. With regard to photonic crystals, their fundamental principle of operation relies on creating a "photonic bandgap," a range of frequencies within which light cannot propagate through the crystal. The properties of this bandgap, including its position and width, are highly sensitive to the refractive index and other optical characteristics of the materials forming the crystal and its surrounding environment. When a tissue sample is brought into contact with a photonic crystal -based sensor, any changes in the tissue's properties will alter the effective refractive index experienced by the crystal, thereby shifting or modifying its photonic bandgap. This change can be detected by interrogating the sensor with optical radiation and observing shifts in the reflected / scattered or transmitted spectrum. In various embodiments, dielectric-based nanostructures like photonic crystals offer distinct advantages, such as lower signal loss compared to plasmonic counterparts and the ability to support high-Q resonances across different spectral windows, providing additional versatility in sensing applications. This makes them highly effective for detecting subtle variations in tissue properties to characterize the sample.

[0165] In various embodiments, Artificial Intelligence (Al) engines and various algorithms are employed to process the data acquired from the nanostructured sensors, enabling the determination of one or more attributes of a sample, such as a tissue sample. As noted above, in various embodiments, the data can correspond to reflected, scattered, and / or transmitted radiation from each pixel obtained at a plurality of frequencies associated with illumination of the pixel.

[0166] In various embodiments, the Al engine can employ supervised, semi-supervised, or unsupervised learning algorithms to analyze the response characteristics of the nanostructured sensor. For example, classification algorithms, utilizing machine learning approaches together with Al technologies, can be utilized that are robust to data inaccuracies and systematic measurement bias.

[0167] A range of Artificial Intelligence (Al) algorithms and techniques to process data derived from nanostructured sensors can be employed for analysis of data acquired from a nanostructured sensor in contact with a sample under investigation, enabling the determination of various attributes from diverse samples. These Al engines can leverage different learning paradigms, including supervised, semi -supervised, and unsupervised algorithms, to analyze the complex response characteristics of the sensors. For example, algorithms such as Convolutional Neural Networks (e.g., ResNet architecture), Support Vector Machines (SVM), and Logistic Regression can be utilized for classification tasks allowing for the accurate categorization of various samples (e.g., classifying margin tissue as cancerous or non-cancerous) or other materials.

[0168] Beyond classification, as discussed in more detail below, in various embodiments, Al can play an important role in optimizing sensor design, with methods like Genetic Algorithms, gradient-based optimizers, and topology -optimization routines being employed in inverse design processes to engineer nanostructures for optimal performance. As discussed in more detail below, Al can also play a role in co-design feedback loop for optimizing the sensor design, for example, based on feature importance and / or performance evaluation. Furthermore, advanced Al techniques, including generative Al methods and computer vision algorithms like image segmentation and anomaly detection, can be utilized for tasks such as personalized recommendations, image enhancement, and identifying regions of interest, ensuring that the system is robust, adaptive, and capable of thoroughly characterizing a wide array of samples.

[0169] Referring to FIG. 7A, a system 700 according to an embodiment for characterizing a sample, such as a tissue sample, can include a nanostructured sensor 702 that can be placed in contact with a sample 704 (e.g., a margin tissue sample) and a light source 706 that can generate optical radiation for illuminating various locations of the sensor. A variety of light sources can be employed. For example, the light source can be, without limitation, a broadband source (such as a quartz-tungsten-halogen lamp) or a narrow-band light source, such as a tunable laser source or a supercontinuum laser. In some embodiments, the frequency of a light source can be scanned over a frequency range of interest, and the reflected, scattered and / or transmitted radiation can be collected over that frequency range for analysis.

[0170] The system further has a data acquisition module 708, which includes a spectrometer 708a and a data processing module 710 that is in electrical communication with the spectrometer 708a. As different spatial locations of the sensor are illuminated, the spectrometer 708a collects the light that is either reflected or scattered by the nanostructured sensor, or transmitted through the nanostructured sensor and the sample, and generates data signals in response to such detection. The data signals generated by this detector, which can define a hyperspectral image, are then transmitted to the data processing module 710 for analysis.

[0171] In various embodiments, the data processing module 710 can be configured (e.g., programmed) to utilize various Al algorithms, such as those described above, to process the raw data to characterize the sample, e.g., a tissue sample, under investigation. By way of example, the data processing module 710 can process the acquired data using a classification algorithm to classify a tissue sample as being cancerous or non-cancerous. In various embodiments, the data processing module can include a permanent data storage module that can store the acquired data and the data analysis results. In various embodiments, a user interface 712, a display, can present the analysis results, e.g., in the form of a false-color map, of the sample to a user.

[0172] In some embodiments, the data processing module 710 can execute a classification algorithm, which can be implemented by training an Al model on reference spectral datasets with collected and / or simulated labeled spectral data, allowing it to predict the correct label for new, previously unseen data. For example, the reference spectral dataset can include simulated or real reflected data obtained from a simulated or fabricated sensor in contact with a simulated or real reference tissue sample, when the nanostructured sensor is illuminated by simulated or real optical radiation having multiple frequencies.

[0173] Moreover, in some embodiments, the Al engine can employ advanced Al-based and probabilistic modeling techniques to assign uncertainty to the analysis results. In various embodiments, uncertainty quantification techniques such as Bayesian posterior inference, ensemble averaging, conformal prediction or hybrid methodologies may be used for assignment of uncertainties. In various embodiments, training data from the measurements at the calibration (fine-tuning) step can be used to calculate uncertainty by well-established techniques such as Monte-Carlo Dropout. In some embodiments, the uncertainty estimates will be calibrated and validated for correct coverage on held-out datasets. In various embodiments, low-confidence pixels may be flagged for surgeon review or additional measurement.

[0174] With reference to FIG. 7B, in some embodiments, the processing module 710 can include a database 714 that can receive and store the reflected / scattered, or transmitted data collected by the data acquisition module 708, as well as simulated data 714a, suitable for training an Al engine for processing these data. An Al engine 716 can be configured to execute a classification algorithm. For example, the Al engine 716 can include a training model 716a, which can be based on the simulated data and a measurement analysis module 716b that can receive the measured data from the database 714 and can utilize the training module 716a to process the measured data to characterize a sample, e.g., a tissue sample, under examination. The measurement analysis module 716b can present the analysis results to an output device 712 for presentation to a user, e.g., a surgeon.

[0175] As discussed above, a sensor according to various embodiments can have a variety of different types of nanostructures and different types of distribution of such nanostructures. By way of example, FIG. 8A schematically depicts a plasmonic sensor 800a according to an embodiment, which includes a plurality of nanostructured elements 802a, which are distributed as a combination of different co-located periodic lattices. FIG. 8B schematically shows a sensor 800b according to another embodiment that includes a plurality of nanostructured elements of the same type 802b, where the nanostructured elements have an irregular shape. FIG. 8C shows a sensor 802c according to another embodiment that includes two sets of nanostructured elements distributed as different arrays on two different parts of the sensor.

[0176] The interrogation of the nanostructured elements of a sensor by an optical radiation beam can be executed through various scanning patterns to acquire spatial data. The system allows for flexibility in how the optical beam addresses the sensor's surface, particularly in defining the individual "pixels" of data collection. For instance, the pixel size, which is determined by the objective spot size and the stage step size, can be configured to achieve different scanning characteristics. In some scanning patterns, the beam spots corresponding to adjacent pixels may be configured to partially overlap, as shown schematically in FIG. 9A, which can be beneficial, e.g., for achieving a higher spatial resolution or signal averaging. Conversely, other patterns may employ disjointed beam spots, where adjacent pixels do not overlap, as shown schematically in FIG. 9B, which may be preferred for faster acquisition or to cover a larger area with distinct measurements. The choice of scanning pattern is adaptable and can be tailored to the specific application requirements, balancing factors such as resolution, acquisition speed, and the nature of the tissue being analyzed.

[0177] In various embodiments, a nanostructured sensor according to the present teachings can be designed using an iterative procedure driven primarily by an inverse design approach. In various embodiments, the goal is to design a nano structured sensor that exhibits a specific, desired response characteristic (e.g., a specific optical response) when in contact with a sample. For example, in some embodiments, the design goal is that the sensor would exhibit multiple high-Q spectral resonances over a specific wavelength range of electromagnetic radiation such that a change in one or more of those resonances, such as a frequency change, can be employed to derive information about the sample.

[0178] With reference to the flow chart of FIG. 10, in various embodiments, the design process begins with defining a "figure-of-merit" (FOM) function, which quantifies the desired performance characteristics of the sensor, such as the number of spectral resonances over a specified spectral range, the sharpness of resonances, spectral separations between the spectral resonances, the largest resonance shift when discriminating between different refractive indices of different portions of a sample or different samples when in contact with the sensor, among others. In some embodiments, the FOM can be defined as a weighted function of such performance characteristics. By way of example, in some embodiments, more weight may be given to the number of resonances and their frequencies, rather than their shapes, or vice versa. The definition of a FOM function depends generally on the type of application for which the sensor is designed.

[0179] More specifically, in various embodiments, the FOM (figure-of-merit) reflects systemlevel goals, not just a single pixel. For example, the FOM can include a mapping objective (spatial diagnostic map), not only per-pixel performance. Further, in various embodiments, noise models and information regarding available fabrication techniques can be used for defining the FOM. For example, the FOM can be defined, at least partially, based on having an accurate classification of a tissue sample in presence of a given noise model. For example, and without limitation, when Electron Beam Lithography / Deep UV is employed for sensor fabrication, the shapes do not have sharp corners. This fact can have an effect both on the parameter space (e.g., the parameter space can include circular shapes) and the FOM function itself (e.g., the field intensity is maximized at the center of the shape and not on its boundaries, even if the shape has sharp comers).

[0180] Such an approach provides a link between the optical behavior of the sensor, the algorithmic performance, and the measured results. In addition, external and internal databases can be used, for example pathology (PT) image databases, together with newly collected datasets that include spectral measurements, clinical annotations, and relevant medical records. These combined resources allow the optimization process to reflect both physical and biological variability observed in real samples. In general, the FOM aims to produce spectra that show the highest sensitivity to the contrast between the conditions being measured, within a defined spectral range or across several wavelength ranges. Depending on the application, it can emphasize simple, narrow resonances or more complex spectral responses that contain many distinguishable features. In all cases, the goal is to design a spectrum that maximizes useful contrast while remaining stable under realistic measurement and fabrication noise.

[0181] In various embodiments, several families of FOMs can be used for design optimization. For example, range-based FOMs evaluate the spectral difference between two conditions (for example, healthy and cancerous tissue) over multiple wavelength intervals and weight the overall norm by the weakest sub-range. This ensures consistent contrast across the full wavelength band. Resonance-based FOMs quantify the prominence or sharpness of dips or peaks within a wavelength window, rewarding sensors that show strong and distinct resonances. The score can be taken as the largest prominence, the sum over all peaks, or the geometric mean of all resonance strengths. Aggregate or hybrid FOMs combine several factors such as number of resonances, Q-factors, inter-resonance spacing, or asymmetry, together with algorithmic indicators such as classification accuracy or uncertainty.

[0182] In various embodiments, the FOM can also incorporate information about the measurement process, including expected noise, spectral distortions, and known artifacts such as edge effects, bubbles, or non-uniform contact with tissue. In some cases, the optimization is not limited to identical pixels but considers the spatial role of each pixel within the map. The FOM can depend on pixel location, allowing certain regions to be optimized differently — for example, by designing specific pixel geometries at the sample edges to minimize boundary artifacts, or by including pixels dedicated to detecting particular types of noise or interference. These specialized pixels can provide additional reference information that can improve processing of the full image or tissue map. In this way, the optimization addresses both the optical design of each sensing element and the overall performance of the complete imaging or mapping system.

[0183] As discussed in more detail below, during the co-design process, the FOM can be revised in two ways. In the sequential approach, the sensor is optimized, fabricated (or simulated), and measured; the results are then used to redefine the FOM and start a new optimization cycle. In the integrated approach, the optimization and FOM updates occur together, using feedback from multiple databases and previous measurements to adjust FOM weights during runtime. This allows direct comparison with earlier data and progressive improvement of both the metric and the sensor design. This adaptive definition of the FOM is central to the co-design concept. It ensures that the optimization target evolves together with the physical sensor, the data processing algorithm, and the accumulated experimental and clinical knowledge, leading to designs that are better matched to real measurement conditions and application needs.

[0184] By way of example, for a given sensor type, e.g., a plasmonic sensor formed of a plurality of electrically conducting nanostructured antennas (e.g., in the form of nano-rods), the inverse design algorithm can systematically search a vast parameter space, encompassing geometric parameters like antenna shape, size (e.g., length, width, thickness), periodicity, and spacing between the antennas to arrive at an optimal set of design parameters, i.e., a set of design parameters that optimizes the figure-of-merit function.

[0185] In design approaches according to various embodiment, a parameter space can be selected that can guide the selection of one or more of the design parameters of the nanostructured sensor. For example, and without limitation, a parameter space for a height of nanoantenna can provide a range within which the height of the nanoantenna can be selected. Such a parameter space can inform the FOM function for optimizing the sensor’s design. By way of example, the FOM can be adjusted during various iterations of the design process by adjusting the parameter space, e.g., changing the lower and / or an upper limit of a range of values corresponding to the parameter space. In various embodiments, the search of the parameter space is often facilitated by optimization algorithms, such as Genetic algorithms, or other methods like topology optimization, to identify the optimal configuration that best meets the defined FOM. In various embodiments, simulations, including FDTD (Finite-Difference Time-Domain) simulations and generative / physics-based Al simulation tools, are extensively used to predict the spectral response of various designs and accelerate the exploration of this design space.

[0186] In various embodiments, an important aspect of the sensor design is the codesign of the sensor-algorithm system, in which a simulated or a real sensor generated based on a set of design parameters can be tested with a simulated or real sample, e.g., a tissue sample. Such testing of design parameters as various stages of a design process can provide a feedback loop between simulated or real performance of a sensor formed based on a set of design parameters and the design process, including the performance of algorithms employed for the design, providing insights that can guide adjustments to the sensor's design and / or the FOM itself.

[0187] For example, if certain wavelengths employed for optical interrogation of a sensor that is in contact with a sample (such as a tissue sample) prove to be unsuitable due to interference, the sensor's spectral response can be adjusted to avoid them. This iterative refinement, supported by performance evaluation and feature importance computation, ensures that the final sensor design is not only highly sensitive but also robust and reliable in practical applications, even under noisy conditions.

[0188] Once a design is finalized, the actual sensor, e.g., including an array of nanostructured elements, can be fabricated using a variety of known nano-fabrication techniques, such as Electron Beam Lithography (EBL), Deep UV lithography, or FIB (focused ion beam).

[0189] The systems and methods described herein are highly versatile, designed for deployment in both in-vivo and ex-vivo (in-vitro) clinical settings. This dual capability allows for real-time guidance during active surgical procedures to preserve healthy tissue and critical structures (in- vivo), as well as for comprehensive assessment of excised tissue or pathological samples (ex- vivo). The flexibility of the system ensures its applicability across a wide range of surgical workflows and diagnostic stages. In various embodiments, a nanostructured sensor according to the present teachings can be integrated into distinct device configurations, such as a stationary ex vivo system with automated stage control, a handheld probe, a flexible patch to which the nanostructured sensor is coupled, or an endoscopic or robotic surgical attachment. Each configuration enables measurements for real-time tissue characterization, applicable across in-vivo and ex-vivo workflows.

[0190] A system according to various embodiments can be implemented across a range of surgical and diagnostic applications, including but not limited to the following:

[0191] • Breast surgery: Real-time margin detection during lumpectomy to prevent re-excision and in selected mastectomy cases to guide tissue removal while preserving skin and subcutaneous layers for improved aesthetic and reconstructive outcomes.

[0192] • Prostate surgery: Intraoperative guidance to achieve complete tumor removal while preserving neurovascular bundles.

[0193] • Bladder surgery: Integration into an endoscopic head for real-time visualization of residual tumors in hard-to-reach regions.

[0194] • Gastrointestinal procedures: Detection and delineation of polyps or early neoplasia during endoscopic or laparoscopic resections.

[0195] • Skin (Mohs) surgery: Optical feedback during margin assessment to reduce operative stages and duration.

[0196] • Brain surgery: Visualization of tumor margins in eloquent regions to enable maximal safe resection.

[0197] • Lymph-node dissection: Localization of involved nodes to reduce unnecessary dissections and postoperative complications.

[0198] • Head and neck (ENT): Precise margin assessment near critical nerves and vessels to preserve function. • Pathology laboratory applications: Analysis of fresh or fixed tissue specimens prior to histology, providing rapid optical feedback on cancer presence or margin status and generating a digital record that can guide or verify microscopic examination.

[0199] In ex-vivo applications, in various embodiments, the system serves as a powerful tool for margin assessment of excised tissue. For example, as discussed in more detail below, a dedicated "microscope-like" device can be utilized for stationary ex-vivo measurements.

[0200] In some embodiments, such a device can be used in an operating-room setting, where the operator inserts a disposable sensor into the device, places an excised specimen on a motorized stage in such a way that the sensor is in contact with the specimen, and initiates data acquisition. The stage performs an automated X-Y scan (optionally with overlapping pixels for averaging), streams spectral data to a processing unit configured to implement an Al analysis algorithm, and the Al algorithm renders a color-coded map with an adjustable confidence threshold for “extend resection” versus “send to pathology.” This real-time workflow allows immediate assessment of margin status without interrupting the surgical procedure.

[0201] In various embodiments, this device can be designed to be integrated seamlessly into the pathology workflow, enabling the analysis of fresh or formalin-fixed tissue specimens prior to traditional histological processing. Pathological samples, including biopsies or larger excised tissue portions, are placed on a stage within the device, where a nanostructured sensor according to various embodiments makes contact with the tissue.

[0202] The system can acquire hyperspectral data, which is processed by the integrated algorithms to generate a dielectric map of the tissue, identifying cancerous margins. This feedback helps surgeons make precise decisions regarding additional excisions without the delays associated with conventional pathology, thereby minimizing healthy tissue removal and potentially reducing the need for repeat surgeries.

[0203] In general, a pathological sample containing a margin tissue sample to be investigated can also include a portion of a tumor (i.e., tissue suspected of being cancerous). This can help an Al engine configured to process the optical data generated via illumination of the nanostructured sensor to distinguish between cancerous and healthy tissue. In various embodiments, the information gleaned from the tumor portion within these samples can serve as robust additional training data for the Al engine to accurately identify and differentiate the subtle response characteristics of the nanostructured sensor when interacting with various tissue types. Such additional training on diverse pathological samples ensures the algorithm's accuracy and reliability in classifying tumor margins.

[0204] FIG. 11 schematically depicts an example of a system 1100 according to an embodiment for performing ex-vivo characterization of a tissue sample, such as a margin tissue sample. The system 1100 includes a light source 1 that can generate light at a plurality of optical frequencies. For example, the light source can be a broadband light source, a tunable laser, or any other suitable light source. An optical fiber bundle 2 transmits the light generated by the light source to a collimator 3 that generates a collimated light beam. The collimated light beam is received by a polarizer 4, which is mounted on a rotatable mount, to allow selecting a desired polarization for the light beam. The polarized light beam passes through an iris 5, which determines the size (e.g., the diameter) of the light beam. Abeam splitter 6 reflects the light beam received from the iris onto an objective lens 7, which in turn focuses the light beam onto a nanostructured sensor 8, which is held in a sensor holder 8a and that has been placed in contact with a tissue sample 9. More specifically, in this embodiment, the tissue sample 9 is placed in a sample holder 9a, which is in turn supported by a movable stage 10. The movable stage 10 can be moved in X and Y directions to expose different spatial locations of the sensor to the light beam. By changing the size of the iris 1, the diameter of the beam can be adjusted, if needed, during a single scan of the nanostructured detector or for different scans of the nanostructured detector via the light beam.

[0205] In this embodiment, the light reflected from each illuminated portion of the nanostructured sensor passes through the objective 7 and the beam splitter 6 to be received by a second iris 11. A portion of the reflected light passing through the second iris 11 is reflected by a second beam splitter 12 into a fiber coupler 13, which focuses the reflected light into an optical fiber 14, which in turn transmits the reflected light to a spectrometer 15, which detects the reflected light and generates data corresponding to an electromagnetic spectrum of the reflected light. The spectrometer further transmits the data to a computer 18, which is configured to process the data in accordance with various embodiments of the present teachings to characterize the sample under investigation. With continued reference to FIG. 11, a portion of the reflected light passing through the second beam splitter 12 is focused by a lens 16 onto a CCD imaging device 17 to generate a visible image of the upper surface (i.e., the surface opposed to the sensing surface) of the nanostructured sensor.

[0206] As noted above, in various embodiments, a surgical device is provided that can be utilized for in-vivo characterization of a tissue sample, e.g., a margin tissue sample, based on various embodiments of the present teachings. By way of example, FIG. 12 schematically depicts such an endoscopic surgical device (for example a resectoscope) 1200 in which an optical sensing system according to various embodiments of the present teachings, including a nanostructured sensor configured to be placed in contact with a tissue sample, is incorporated. The resectoscope is particularly advantageous for procedures such as Transurethral Resection of Bladder Tumors (TURBT) or Transurethral Resection of the Prostate (TURP), where precise tissue discrimination and real-time feedback are critical. The resectoscope itself functions as a medical instrument for visualizing, removing, or treating tissue within cavities such as the bladder or prostate, often including a built-in working channel.

[0207] In this example, the resectoscope includes a housing 1202 to a distal end of which a nanostructured sensor 1204 according to various embodiments of the present teachings, such as those discussed above, is coupled. Optical illumination light is generated by a suitable illumination source 1206, which can include, for example, a broadband light source, a tunable laser, or any other suitable light source capable of generating optical radiation at a plurality of frequencies. An optical fiber 1208 receives the light generated by the light source to a collimating lens 1210, which generates a collimated light beam. This collimated light beam then encounters a beam splitter 1212, through which the optical illumination light passes to reach a beam scanning and conditioning module 1214. This module serves multiple functions, including actively scanning the illumination beam across a desired area of the nanostructured sensor, optically conditioning the beam, and ensuring a de-scanned return path for the light reflected back from the nanostructured sensor when in contact with a tissue sample.

[0208] The scanned and conditioned beam subsequently passes through a focusing lens 1216, which focuses the light onto various positions on the nanostructured sensor 1204, allowing for localized optical interaction with a meta-surface of the nanostructured sensor in contact with a tissue sample 1218. In this embodiment, the nanostructured sensor is a disposable sensor that can be discarded after a single use. In other embodiments, the nanostructured sensor can be a multiuse sensor that can be disinfected after each use for reuse.

[0209] Reflected radiation from the metasurface of the nanostructured sensor 1204 in contact with the tissue sample 1218 then travels back along substantially the same optical path as that of the interrogating optical radiation. This reflected radiation is effectively de-scanned by the beam scanning and conditioning module 1214, maintaining spatial correlation. The de-scanned reflected radiation is then directed by the beam splitter 1212 towards a collection pathway in which a collection lens 1220 is positioned to focus this reflected radiation into a collection fiber 1222. The collection fiber 1220 transmits the focused light to an external detection / spectrometer module 1224, where the reflected radiation is detected and its spectral characteristics analyzed. This integrated system thus enables real-time acquisition of spectral data from tissue surfaces during surgical procedures, facilitating a rapid and accurate tissue characterization. In particular, subsequent to resection of a cancerous tissue, the nanostructured sensor coupled to the distal end of the resectoscope 1200 can be used, in a manner discussed herein, to characterize the margin tissue to determine in real-time, for example, whether it is likely to be healthy or cancerous.

[0210] By way of further illustration, FIGS. 13A and 13B schematically depict another surgical device 1300 in which a nanostructured sensor 1302 according to the present teachings, such as the nanostructured sensors described above, is incorporated. While the nanostructured sensor incorporated in the resectoscope 1200 can be used for surface analysis of a tissue sample, the nanostructured sensor 1302 incorporated in the surgical device 1300 can be employed for volumetric tissue interrogation. The device 1300 includes an optical needle probe 1304, which can be configured as a standalone device or integrated with robotic surgical platforms for minimally invasive tissue interrogation.

[0211] The probe 1300 includes a motorized needle, that allows for precise insertion into tissue. The distal end of the needle is equipped with a sapphire tip 1306, which offers durability and optical transparency, facilitating light interaction with the tissue. The nano structured sensor 1302, e.g., a plasmonic nanostructured sensor having plurality of gold nano-antennas, is integrated at the sapphire tip, functioning as the primary optical sensor. In this embodiment, optical communication with the sensor is achieved via a double-clad fiber (DCF) 1308 housed within the needle. The DCF includers a DCF core 1308a, an inner cladding 1308b, and an outer cladding 1308c.

[0212] Illumination light from a light source 1310 is delivered via a coupler 1312 into the DCF core 1308a. This illumination light is directed to the nanostructured sensor 1302 via the sapphire tip 1306. Reflected and scattered light from the nanostructured sensor when in contact with a volumetric tissue sample propagates back through the inner cladding 1308b of DCF 1308 to reach the coupler 1312, which then routes this collected light from the DCF inner cladding to an data acquisition module having a detector and a spectrometer (not shown in the figures) for spectral analysis. This design allows for both illumination and collection of light through a single fiber, enabling a compact and efficient optical sensing mechanism for detecting properties of the surrounding tissue.

[0213] This motorized needle device can be operated in various configurations. In the configuration depicted schematically in FIG. 13A, it functions as a standalone device. Here, the optical needle probe 1304 is coupled to a robotic manipulator 1314 for precise control over the needle's insertion depth, trajectory, and scanning motion within the tissue. The entire system is controlled by a controller 1316, and the output, typically a real-time map or analysis of tissue properties, is displayed on a dedicated display 1318. This standalone configuration offers a controlled environment for tissue analysis, providing high precision for various applications.

[0214] With reference to FIG. 13C, in an alternative and highly advanced embodiment, the motorized needle 1304 can be integrated into an existing robotic surgery device, such as a Da Vinci™ platform. This integration leverages the sophisticated control and dexterity of robotic surgical arms. In such a system 1320, the motorized needle 1304 and a camera 1322 for visual guidance are mounted on robotic arms 1314 and 1324. A controller 1326 controls the operation of the system and an analysis module 1328 receives and processes data generated via illumination of the nanostructured sensor in contact with a tissue sample.

[0215] The robotic system provides enhanced control and real-time analysis capabilities, allowing for in-vivo application of the motorized needle for volumetric tissue characterization. This integration enables surgeons to precisely target and analyze tissue regions deep within the body, potentially providing real-time feedback during complex surgical procedures and enhancing the accuracy of interventions. In some embodiments, the optical sensing principles of systems such 1100, 1200 and 1300 can be used in different packaging and system sizes for additional use cases, both in-vivo and ex-vivo. Furthermore, in various embodiments, free space beam propagation can be replaced with an appropriate optical-fiber component to ensure compact and noise-free operation and allow flexible device packaging.

[0216] For example, in one embodiment, a compact handheld probe for ex-vivo applications such as breast cancer margin detection can be provided for rapid scanning of excised tissue surfaces. The handheld probe can house the illumination and collection optics within a small enclosure and can be coupled to a spectrometer and processing module, for example, similar those discussed above in connection with the devices 1200 or 1300. This configuration allows surgeons or pathologists to scan larger tissue areas directly on the back table or in the pathology suite without repositioning samples inside the microscope-like box. The handheld format preserves the same contact-based sensing principle and provides real-time spectral feedback and classification results through the shared software interface.

[0217] In further embodiments, the handheld probe configuration can also be utilized for in-vivo intraoperative sensing. In such use, the probe tip, which can include the disposable sensor cartridge, can be brought into gentle contact with exposed tissue within the surgical field, allowing real-time assessment of margins or tissue composition without removal of the specimen. In some embodiments, the probe can be sterilized or fitted with single-use sterile covers and may be introduced through a surgical access port or laparoscopic trocar. This in-vivo probe implementation extends the same contact-based sensing principle to living tissue, enabling immediate optical feedback during surgery. The external packaging and the physical dimension of the in-vivo probe may vary according to specific use cases and cancer types. For example, breast cancer probe will be large, while Mohs surgery probe will be much smaller.

[0218] In additional embodiments, a nanostructured sensor according to various embodiments can be integrated within robotic surgical systems, enabling localized real-time sensing directly from robotic arms or cutting tools. This allows automatic margin detection synchronized with the robotic workflow, not limited to needle or endoscopic integration. Various embodiments of methods, systems and devices according to the present teachings offer a flexible use model that allows customization, for example, based on demographics, ethnicity, gender, type of tissue, skin properties, etc.

[0219] In some embodiments, the Al engines are configured with algorithms that employ a two- layer training approach. The initial stage involves pre-training on a diverse array of samples, allowing the system to learn from various populations before analyzing a specific sample under examination. This pre-training can be customized to enhance the model's effectiveness for particular groups by selecting relevant datasets for training. Subsequently, the fine-tuning process adjusts the model based on the specific characteristics of the sample being measured. For example, for surgical tissue samples or pre-surgery biopsy, the fine-tuning process can be applied to the information relevant to the specific patient. For example, the Al algorithm may be trained on characteristics of a healthy tissue sample of that patient, or on lesion shapes typical for that patient, as obtained from their pathology samples.

[0220] This dual training strategy ensures that the sensor enhances its applicability to diverse populations while increasing overall performance. By enabling users to select specialized configurations, a product is provided for various user groups, addressing different types of cancers or other health conditions.

[0221] Additionally, in various embodiments, a system according to the present teachings and in particular an Al engine integrated in the system can receive software updates and algorithm enhancements, ensuring that users benefit from the latest advancements in technology and research, further improving performance and adaptability over time.

[0222] During sensor fabrication and before delivery, each sensor can undergo a calibration process, such as an optical response testing in air or on a reference target. This calibration data can be associated with the sensor, e.g., it can be cached with a barcode employed to provide a unique identification of the sensor.

[0223] A system in which the nanostructured sensor is incorporated can include dedicated modes for quality control and verification to ensure optical and mechanical integrity, before surgery, and during clinical use. For example, before each procedure, the system can perform an automated self-test that verifies signal stability, optical alignment, and system connectivity. The test may be performed in air or using a built-in reference standard to confirm that the sensor is functioning within acceptable limits. If the sensor is dropped, damaged, or exposed to contamination, the software initiates a diagnostic routine that evaluates signal quality and alerts the user if deviations exceed tolerance, preventing further use until the issue is resolved.

[0224] For sensors designed for reuse, a controlled cleaning protocol followed by automatic revalidation can be provided. A variety of cleaning materials, such as ethanol, or approved enzymatic detergents can be employed. After washing and drying, the system can repeat the verification routine.

[0225] All calibration, verification, and cleaning data records can be linked to the sensor’s barcode and stored automatically, enabling full traceability and compliance with quality and regulatory requirements.

[0226] As noted above, in various embodiments, the nanostructured sensor can incorporate a unique digital identifier, e.g., a unique barcode. For example, each nanostructured sensor can be associated with its calibration data and a unique barcode record. By way of example, such a unique digital identifier can be used for authentication, activation for performing measurements, quality control activation, and data tracking. Upon initialization, a system incorporating such a nanostructured sensor, such as the system described below, can read the digital identifier retrieve a calibration record associated with that nanostructured sensor, and activate the sensor for data collection. This mechanism advantageously ensures traceability, prevents reuse of single-use nanostructured sensors, and links the collected data to a verified, traceable unit.

[0227] In some embodiments, a tissue map generated, e.g., in-vivo during a surgical procedure, using methods according to the present teachings can be automatically registered to a corresponding anatomical location of measurement, ensuring spatial precision and enabling integration with other imaging or navigation systems. The map can be visualized on a standard display, in virtual reality, or through augmented reality glasses (such as Ray -Ban Meta). Such hands-free visualization allows the surgeon to view diagnostic feedback and margin information while continuing the surgical procedure, without diverting attention from the operating field. All data, including calibration history, verification results, and mapping outputs, can stored under the sensor’s digital ID for traceable data management and compliance. Further, in some embodiments, a system in which a nanostructured sensor according to various embodiments is incorporated, can be equipped with downloadable software modules selected by the surgeon or other user according to the target tissue type, cancer subtype, or surgical procedure. A single sensor or a family of sensors may support multiple operational modes optimized for different applications. Measurement data collected during authorized use may be anonymously linked to a unique sensor barcode and employed to improve or retrain subsequent algorithm versions. Updated, validated software releases can be distributed to users to enhance system performance without hardware modification.

[0228] In some embodiments, a method is provided for correlating the optical properties of fresh and fixed tissue states to enhance model accuracy and versatility. In this method, a tissue specimen is first characterized in its fresh state using a nanostructured sensor according to the present teachings. Subsequently, the same specimen undergoes a standard fixation or preservation procedure, for example, by immersion in saline, a short-chain alcohol, or a formaldehyde solution. During this process, the same spatial region of the specimen is repeatedly interrogated by the sensor at predefined time intervals. This sequential measurement protocol generates a time-resolved dataset that quantifies the spectral evolution of the tissue as it transitions from a fresh state to partially fixed and fully fixed states.

[0229] This dataset establishes a direct, empirical correspondence between the optical response of the tissue before, during, and after fixation. This correspondence enables an Al model to be trained to learn the temporal and state-dependent transformations of the tissue's optical signature. Consequently, a classification model trained primarily on data from fixed pathological samples can be accurately applied to characterize fresh tissue samples in real-time during a surgical procedure (and vice-versa), a process also known as model transfer. This capability significantly enhances the versatility and clinical applicability of the system by allowing data from extensive archives of fixed tissues to inform real-time, in-vivo diagnostics, thereby improving the robustness and calibration of the classification algorithm.

[0230] In certain embodiments, the present teachings can be applied to the critical task of differentiating healthy anatomical structures, such as nerves, vessels, and connective tissue, from tumor margins in ex-vivo specimens. The system can leverage not only the distinct refractive- index contrasts and unique spectral signatures of these different tissue types but also their morphological features. By generating a high-resolution spatial map of the tissue's optical properties, an Al engine can be trained to perform a multi-level analysis. First, the Al engine identifies the unique spectral fingerprint of each tissue type on a pixel-by-pixel basis. Second, by processing the spatial arrangement of these classified pixels across the map, the Al engine is further configured to recognize the characteristic morphological patterns — such as the linear structure of a nerve or the tubular cross-section of a blood vessel. This dual -analysis, which combines both spectral and inferred morphological data, enables a highly reliable classification, providing surgeons with crucial guidance for the preservation of vital structures during surgical resection.

[0231] In some embodiments, a system in which a nanostructured sensor according to various embodiments is incorporated can be configured to account for thermal variations that may affect the sensor's optical response. The performance of the nanostructured sensor can exhibit sensitivity to temperature, which may arise from temperature-dependent changes in the refractive index of the sample and / or from the thermal expansion or contraction of the sensor substrate itself. To address this sensitivity and ensure robust performance across various conditions, a temperature characterization process is performed. This process involves measuring identical reference samples, such as specific tissue types (e.g., muscle and adipose tissue) or calibration liquids, at a plurality of controlled temperatures, for instance, within a physiologically relevant range of approximately 4 °C to 40 °C. During these measurements, the corresponding changes in the sensor's response characteristic, such as shifts in resonance wavelengths and variations in overall spectral intensity, are systematically recorded. This acquired temperature-dependent data is subsequently incorporated into the training dataset for the Al engine. By training on this comprehensive dataset, the Al engine learns to distinguish thermal-induced spectral variations from those indicative of the sample's intrinsic properties. This enables the system to perform active temperature compensation or normalization during analysis, thereby ensuring robust and accurate classification and mapping across a range of physiological and environmental conditions. Furthermore, this inherent robustness to thermal effects demonstrates the platform’s suitability for applications beyond the medical field, such as in industrial process-monitoring environments where temperature control is a critical operational parameter.

[0232] While many embodiments described herein focus on the characterization of cancerous versus non-cancerous tissue, the utility of the present teachings is not so limited. The fundamental capability of the system is to distinguish between different material states based on subtle variations in their optical properties, such as refractive index. This allows for a broad range of applications where high-resolution material classification is required. An exemplary application demonstrating this versatility is the assessment and mapping of thermal damage in tissue, such as in the clinical evaluation of burns.

[0233] In various embodiments, to configure a system according to the present teachings for such an application, a specific training methodology can be employed. In an exemplary embodiment, controlled levels of thermal damage are created on ex-vivo tissue analogues, such as meat samples, to emulate a gradient of burn severity (e.g., from superficial first-degree burns to full -thickness necrotic tissue). The nanostructured sensor is then used to acquire hyperspectral data from these samples. The measured spectral features, or response characteristics, are systematically compared across the different levels of damage. These spectral variations are correlated with the known changes in the optical constants of key biological components, such as denatured proteins and lipids, which occur as a result of thermal injury.

[0234] The Al engine is trained on this correlated data to recognize the unique spectral signature corresponding to each degree of tissue damage. Once trained, the system can be used to generate a rapid, high-resolution spatial map of a burn-affected area on a patient. This map provides clinicians with an immediate and quantitative assessment of burn depth and extent, clearly delineating zones of varying severity, which is a significant advancement over traditional, subjective visual inspection. Such a map can guide critical treatment decisions, such as determining which areas require debridement, which may heal spontaneously, and which necessitate skin grafting, thereby facilitating more precise interventions and improving patient outcomes.

[0235] In further embodiments, the nanostructured sensing platform is applied to the characterization of nanoparticles and other materials in liquid or surface-bound states. The sensor can be configured to interrogate metallic (e.g., Au, Ag, Al), semiconductor (e.g., Si, Ge), and dielectric nanoparticles dispersed in aqueous or organic media, for example, at concentrations ranging from approximately 106M to 103M. To provide an optical ground truth for calibrating the system, fluorescent nanoparticles may be employed, allowing for direct correlation between external imaging modalities and the plasmonic sensing results. Furthermore, by performing time- resolved measurements, the system can assess the kinetics of dynamic processes such as the adsorption and desorption of nanoparticles on the sensor surface.

[0236] To validate the system's utility for industrial applications, such as quality control in the semiconductor industry, engineered reference samples may be utilized. These test structures, fabricated using methods such as 3-D printing or clean-room nanofabrication techniques (e.g., e- beam lithography, Deep UV lithography, or Focused Ion Beam), are designed to emulate known surface patterns, voids, or contamination. Corresponding high-resolution images, such as those from a Scanning Electron Microscope (SEM), provide ground truth data for evaluating and confirming the spatial accuracy and diagnostic capability of the nanosensor mapping.

[0237] The application of the present teachings also extends to environmental and food-quality sensing, for example, for the early detection of spoilage. In one embodiment, the optical response of a food surface (e.g., bread, cheese, or fruit) is monitored over time. The system is configured to detect subtle changes in the reflectance spectra indicative of the onset of mold growth or humidity-induced changes. The spectral data acquired by the sensor can be correlated with standard microbiological assays to train an Al engine to recognize the signatures of spoilage, enabling a rapid, non-invasive method for quality assessment.

[0238] As discussed above, the methods, systems and devices according to the present teachings can exhibit robustness and reliability across real-world conditions by systematically characterizing and learning from diverse noise sources. This can be achieved, for example, by deliberately incorporating data from both simulated and measured noise conditions into the Al training datasets. This process can enhance classification accuracy, improve uncertainty estimation, and enable the algorithm to distinguish meaningful signals from confounding factors.

[0239] This noise characterization can include several categories, such as the following:

[0240] (a) Biological and Tissue-Related Noise: Tissues exhibit significant local heterogeneity due to factors such as blood perfusion, extracellular matrix composition, the presence of microbubbles, and variations in hydration. In various embodiments, to account for this, reference databases are created by performing measurements on tissues of different origins and states — including fresh, frozen, and fixed samples — thereby teaching the Al algorithm to recognize and normalize for these natural biological fluctuations. (b) Optical and Environmental Noise: Instrumental and environmental noise sources, such as photon shot noise, detector dark current, spectral drift, and illumination instability, are characterized. These noise profiles can be artificially added to clean training data through modulation of parameters like exposure time, spectral resolution, or operating temperature. Training the Al engine with and without these perturbations improves its stability under variable field conditions.

[0241] (c) Fabrication and Sensor Variability: To account for minor inconsistencies inherent in fabrication processes, sensor arrays fabricated under slightly different process parameters (e.g., variations in metal thickness, period, or dielectric coating) are characterized. This creates a dataset of sensor-induced variations, which enables the Al algorithm to normalize for sensor-to-sensor differences, ensuring consistent performance across different sensor batches.

[0242] (d) Dynamic and Brownian Noise: For samples involving particles in motion, such as colloidal nanoparticles in a droplet, spectra are collected as a function of time. This captures the temporal spectral fluctuations caused by Brownian motion. These temporal patterns are then incorporated as an additional input dimension to train advanced models, such as temporal neural networks, enabling the system to characterize dynamic processes.

[0243] The following examples are provided for further elucidation of various aspects of the present teachings and are not provided to indicate necessarily the optimal ways of practicing the present teachings or optimal results that may be obtained.

[0244] Example 1

[0245] A plasmonic nanostructured sensor was designed using an initial design space that included a set of plasmonic sensors as shown in FIGS. 14A and 14B. A plasmonic sensor 1400a illustrated in FIG. 14A includes a plurality of gold nano-antennas 1402a according to a periodic array and deposited on an underlying chromium (Cr) layer 1404a, which is in turn disposed on an underlying SiCh substrate. FIG. 14B shows another plasmonic sensor 1400b that has the same structure as that of the plasmonic sensor 1400a with the addition of an SiCh spacer 1402b covering the nanostructured elements. FIGS. 14C and 14D schematically illustrate the sensors 1400a and 1400b in contact with a sample (A), and FIGS. 14E and 14F schematically illustrate the sensors 1400a and 1400b in contact with a sample (B).

[0246] With reference to the schematic diagrams illustrated in FIGS. 14G and 14H, each of the nanostructured elements was parametrized using the following geometric parameters:

[0247] • xh and yh: These parameters represent the dimensions of the nano-antenna in the X and Y directions, respectively.

[0248] • b and a: These parameters denote the periodicity or spacing between adjacent nanoantennas in the array along the X and Y directions, respectively.

[0249] • t: This parameter refers to the thickness of the gold (Au) layer forming the nanoantennas.

[0250] AFigure-of-Merit (FOM) function was defined to guide the optimization process, giving preference to multi -resonant spectra with sharp peaks. The design parameters were varied within the ranges and at increments presented in Table 1 below:

[0251] Table 1

[0252] A genetic algorithm was utilized for this design optimization, starting with a population size of 20 initial designs, each corresponding to a different set of these geometric parameters. The algorithm iteratively refined these designs by evaluating their respective FOMs, and through processes of "mutation" and "cross-breeding," it converged towards an optimal configuration that produced the desired spectral response. This inverse design approach systematically explored the vast design space to find the most suitable nanostructure configuration for the sensor. A Scanning Electron Microscope (SEM) image of the designed sensor used for the following measurements is shown in FIG. 141, This sensor was used to create a map of beef meat tissue containing both fat and muscle portions in order to find the boundary between the two portions. Initially, the sensor was characterized with different mixtures of glycerol and water and then the sensor was used with the beef meat sample.

[0253] FIG. 15A shows reflection data acquired from the sensor in the form of a normalized intensity of the light reflected from the sensor when in contact with different mixtures of glycerol in water as the sensor was illuminated with an interrogating optical radiation with the wavelength of the radiation scanned over a range of 600 nm to 1000 nm. The refractive index values of the different mixtures are as follows: water - 1.333, gt - 1.3412, gO - 1.3486, gl - 1.361, g2 - 1.3797, g4 - 1.4263, glyc - 1.473. The vertical dashed lines in FIG. 15A correspond to the top 10 spectral features ranked by their importance - in this case, the absolute value of the coefficient in the corresponding binary classifier. These features can be used for further improvement of the FOM, for example, by weighting the spectral components according to the importance value.

[0254] FIG. 15D shows a confusion matrix associated with the classification of the different mixtures as shown in FIG. 15A by one embodiment of the classification algorithm (here, a Generalized Linear Model classifier, one-vs-all, with 2-fold cross-validation).

[0255] FIG. 15B, in turn, shows the respective normalized reflection data when the sensor was placed in contact with muscle and fat portions of a meat sample, as shown in FIG. 15C.

[0256] To obtain training data for an Al logistic regression analysis, labeled tissue samples of fat and muscle, such as that shown in FIG. 15C, were measured using the acquisition system. Each measurement yielded 256 pixels, with a total of 512 training samples, comprising 341 background pixels, 108 fat pixels, and 63 muscle pixels. These precisely labeled samples provide the ground truth necessary for supervised learning. FIGS. 16A and 16B present the measured spectra of these fat and muscle labeled samples, respectively, used for training the classifier.

[0257] Next, a logistic regression classification model was trained using this labeled data. The training process resulted in a 10-fold cross-validation test score of 0.990 ± 0.016, with a specificity of 99.5% (99.8% for muscle) and a sensitivity of 98.1% (96.8% for muscle) for fat and muscle, respectively. These metrics indicate the high accuracy and reliability of the trained algorithm in distinguishing between the two tissue types.

[0258] FIGS. 17A, 17B, and 17C show confidence scores for the test sample.

[0259] Subsequently, this trained logistic regression classifier was applied to classify an unknown sample containing a mixed margin (i.e., a sample with unlabeled regions). For each predicted pixel label in the unknown sample, the classifier outputs a decision function value, which can be interpreted as a confidence score.

[0260] FIG 17D shows the predicted map of the unknown sample, which is the predicted class multiplied by its confidence score, so that white areas indicate low confidences. This map visually represents the likelihood of each pixel being a particular tissue type. To finalize the tissue designation, a threshold of 0.8 was applied to the probability values. Pixels exceeding this threshold were assigned a high-confidence prediction, effectively categorizing them as corresponding to either muscle or fat. In various embodiments, this threshold can be fully adjustable by the user, allowing for customization based on specific application requirements and desired confidence levels. In some embodiments, the system employs probabilistic or Bayesian uncertainty quantification methods (e.g., ensemble averaging, Monte-Carlo dropout, or conformal prediction) to estimate confidence levels for each pixel classification. The uncertainty map provides calibrated coverage for clinical decision support.

[0261] With reference to FIG. 17E, the importance of various spectral features for the classification task can be quantified by inspecting the mean regression coefficients. High values of these coefficients indicate that corresponding frequencies are more critical for accurate classification, providing insights for potential sensor design improvements.

[0262] Example 2

[0263] A plasmonic nanostructured sensor, such as the sensor described in the above example was used to simulate response characteristic of the sensor when in contact with a healthy tissue sample and when in contact with a cancerous tissue sample. More specifically, FIG. 18A schematically shows the sensor in contact with a simulated healthy tissue sample with a radiation beam incident on the sensor (i.e., on an opposed surface of the sensor relative to its meta surface that is in contact with the healthy tissue sample) as well as a reflected radiation beam generated in response to the illumination of the sensor. FIG. 18B schematically depicts the sensor in contact with a simulated cancerous tissue sample with a radiation beam incident on the sensor as well as a reflection beam generated in response to the illumination of the sensor.

[0264] The spectra depicted in FIG. 18C exhibit a multi-channel spectral signature, which functions analogously to a key with multiple pins. Each “pin” or channel represents an independent or partially correlated spectral feature, such as a distinct resonance mode, wavelength band, polarization response, or spatial sensing element within the array. The combination of multiple channels produces a complex optical signature that should align across several dimensions for accurate tissue identification. Partial correlations between channels capture systematic noise and cross-talk inherent to the measurement environment, allowing the associated algorithm to learn and compensate for these effects. This multi-channel, partly- correlated design provides a robust spectral fingerprint, maintaining reliable classification even when individual resonances are perturbed by noise, fabrication tolerances, or tissue variability.

[0265] Example 3

[0266] In the above Example 2, the simulated data did not include any noise. However, in real in-vivo and ex-vivo measurements, some level of noise is inevitably present. FIG. 19A shows the simulated spectra depicted in FIG. 18C, but with some simulated noise added to the spectra. Even with the added noise, it was feasible to distinguish the simulated healthy tissue spectrum from the simulated cancerous tissue spectrum using an Al classification engine. Further, FIG. 19B shows a confusion matrix computed for the classification, indicating the classification was statistically robust.

[0267] FIG. 19C is a ROC (Receiver Operating Characteristics) curve providing true positive rate as a function of false positive rate, indicating that the classification can be performed with high sensitivity and specificity. FIG. 19D, in turn, is a precision recall curve that shows how well the classification can identify true cases (the cancerous tissue sample in this example) designated as “recall” and how accurate its positive results are (precision) as the classification threshold changes. The depicted precision recall curve shows that the above classification can identify true cases with high precision (low false discovery rate).

[0268] Example 4 This example illustrates that in some cases, it is advantageous to use subject-specific training data (referred to as a "calibration dataset") to supplement previously-obtained training data of an Al engine (e.g., an Al classification engine) to obtain accurate classification results for that subject. In some cases, such an approach, termed herein as "Fine-tuned," can show superior performance compared to using only a general pre-trained model or only the calibration data.

[0269] FIG. 20A displays a high-resolution histological image of a tissue sample, in which the different shades of gray represent distinct tissue types, namely, a non-cancerous issue (herein also referred to as a healthy tissue) versus a cancerous tissue.

[0270] FIG. 20B presents the "Ground Truth" classification for the tissue shown in FIG. 20A. It is a simplified, block-like representation where each block corresponds to a region of the tissue. The shaded areas represent the non-cancerous issue while the solid areas represent the cancerous tissue. This "Ground Truth" serves as the benchmark against which the Al model's predictions are compared.

[0271] FIG. 20C shows the prediction of a "Pretrained #1" Al model. This model has been trained on a general dataset of tissue samples. For this particular subject, the pretrained model performs well, as indicated by a low Mean Squared Error (MSE=2.449) and a high Correlation (Corr=0.985) with the Ground Truth. The visual representation closely matches the pattern in FIG. 20B. The "Analysis Parameters" indicate that this prediction was made by including only the pretrained model, without additional calibration data for this specific subject.

[0272] FIG. 20D illustrates the prediction of a "Pretrained #2" Al model. While also generally pre-trained, this model performs poorly for the specific tissue shown in FIGS. 20A and 20B. This is evident from the high MSE (11.576) and low Correlation (Corr=0.585) compared to the Ground Truth. Visually, regions identified as non-cancerous and cancerous are scattered and do not accurately reflect the contiguous tissue areas seen in FIG. 20B. This demonstrates a scenario where a general pre-trained model may struggle with specific, potentially atypical, subject data. The "Analysis Parameters" confirm that this prediction also only included the pretrained model.

[0273] FIG. 20E shows the result when the Al engine is trained only on the "calibration dataset (subject-specific data) without incorporating a pre-trained model. The prediction is poor, with a very high MSE (14.071) and zero Correlation (Corr=0.000). The entire image is classified as a single tissue type, indicating that the limited subject-specific calibration data alone is insufficient to train a robust classification model. This highlights that while subject-specific data is valuable, it typically needs the knowledge base of a larger pre-trained model to be effective. The "Analysis Parameters" show that only the calibration dataset was included, and the number of positive and negative samples (27 each) is relatively small.

[0274] FIG. 20F demonstrates the benefit of "Fine-tuned" prediction. Here, the Al model leverages both the pretrained model and calibration dataset options (as indicated by the checked boxes in "Analysis Parameters"). This means a pre-trained model (Pretrained #2, which performed poorly initially) is fine-tuned or adapted using the subject-specific calibration data. The result shows a significant improvement in accuracy compared to Pretrained #2 (FIG. 20D) and Calibrated only (FIG. 20E). The MSE (4.123) is much lower than in FIG. 20D and 20E, and the Correlation (0.957) is substantially higher, indicating a good match with the Ground Truth (FIG. 20B). This visually demonstrates that combining the general knowledge from a pre-trained model with targeted, subject-specific calibration data yields highly accurate, personalized predictions.

[0275] Example 5

[0276] This example demonstrates the efficacy of the flexible sensor co-design methodology, wherein the figure-of-merit (FOM) function and the resulting optimal sensor parameters are adapted to different operating conditions. To illustrate this, two distinct operational environments, characterized by different measurement noise profiles, were simulated to represent challenges that might be encountered with different cancer types or surgical modalities.

[0277] The first simulated environment (Environment 1) was characterized by both low- and high-frequency spectral noise, while the second simulated environment (Environment 2) was characterized by localized spectral inhomogeneous broadening and frequency shifts in the midwavelength range, typical in the presence of non-uniform scattering medium. In accordance with the present teachings, two corresponding figure-of-merit functions were constructed to avoid the corresponding noise patterns. For Environment 1, the FOM was defined to prioritize the prominence of spectral resonances (i.e., high-Q factors), in the mid-spectral range. For Environment 2, the FOM was defined to prioritize the integrated spectral difference between healthy and cancerous tissue signatures over a broad wavelength range, a metric more robust against baseline drift.

[0278] Using the inverse design method described herein, two distinct, optimal sensor designs were generated: Sensor 1, optimized using the FOM for Environment 1; and Sensor 2, optimized using the FOM for Environment 2. Subsequently, simulated noisy training datasets were generated for both sensors in both environments. FIGs. 21A-21D depict these simulated datasets. Specifically, FIG. 21A shows the spectral response of Sensor 1 in Environment 1, while FIG. 21B shows the response of Sensor 2 in Environment 1. Similarly, FIG.

[0279] 21C and FIG. 21D show the responses of Sensor 1 and Sensor 2, respectively, in Environment 2.

[0280] A classifier model was trained on these datasets, and its predictive performance in classifying tissue as either cancerous or non-cancerous was evaluated for each sensorenvironment combination. The results are presented in FIG. 21E. As expected, Sensor 1, having been designed for sharp resonances, significantly outperformed Sensor 2 in the low- and high- frequency noise of Environment 1. Conversely, Sensor 2, designed for robust broad-range contrast, demonstrated superior classification accuracy compared to Sensor 1 in the baseline drift conditions of Environment 2. This example validates the claimed co-design methodology, proving that by adaptively adjusting the figure-of-merit function, a sensor can be specifically optimized for superior performance under distinct and challenging operational conditions.

[0281] Example 6

[0282] In some embodiments, a nanostructured sensor can operate through an evanescent-field interaction confined to a nanoscale region immediately adjacent to its surface. The effective sensing depth is typically less than about 100 nm, depending on the wavelength and nanostructure geometry. Accordingly, only the liquid or tissue layers within this distance contribute measurably to the detected optical response. As depicted schematically in FIG. 22A, for liquid-phase measurements, a nanostructured sensor (2101) can be positioned facing downward over a droplet (2103) containing suspended nanoparticles (2102) on a transparent substrate or coverslip (2104). This configuration enables the analysis of nanoscale refractive- index variations while minimizing optical interference from the bulk fluid volume.

[0283] In one exemplary implementation of this principle, an antenna array nanosensor was fabricated using Electron Beam Lithography. The sensor included a 5x5 grid of nanoantenna arrays, with each array having a size of 200 pm by 200 pm and separated by 10-20 pm spacings. Each individual antenna was designed with dimensions of 500 nm by 160 nm, a thickness of 50 nm, and periodic spacing of 100 nm in the X-direction and 120 nm in the Y-direction. The sample liquid was prepared by dispersing 30-micron diameter colloidal polystyrene microspheres in glycerol.

[0284] To perform the measurement, the sensor surface was scanned by moving the sample stage in 50 pm increments, and the reflected spectra were collected from each discrete location, or "pixel." The collection of spectra from this scan is presented in FIG. 22E. From the acquired hyperspectral data, the total reflected intensity was first calculated for each pixel to generate a spatial intensity map, as shown in FIG. 22B. To spatially localize the nanosphere, an unsupervised outlier detection algorithm, specifically the Isolation Forest algorithm, was employed. This Al-based approach identifies pixels whose spectral response deviates significantly from the background response of the glycerol medium. The algorithm produced a decision function map, presented in FIG. 22C, and identified the outlier pixels corresponding to the nanosphere's location, as shown in FIG. 22D. The spectra classified as outliers are highlighted in a distinct shade of grey within the full spectral dataset in FIG. 22E.

[0285] The spatial extent of the detected outlier region was consistent with the known 30-micron diameter of the microsphere, confirming the successful localization of the particle. To further validate the method, the same area was re-measured several minutes later. In the subsequent measurement, no outlier was detected at the original location, a result consistent with the expected Brownian motion of the nanoparticle within the liquid. This demonstrates the sensor's and algorithm's combined ability to perform real-time, high-sensitivity detection of nanoscale events.

[0286] Various computational devices disclosed herein including those that execute various Al algorithms can be implemented using software, hardware, and / or firmware using standard engineering techniques including using various digital processors, memory modules, etc., as informed by the present teachings.

[0287] Those having ordinary skill in the art will appreciate that various changes can be made to the above embodiments without deviating from the scope of the claimed invention.

Claims

1. What is claimed is:

1. A method for characterizing a tissue sample, comprising: placing at least one nanostructured sensor in contact with said tissue sample, wherein said nanostructured sensor includes at least one response characteristic exhibiting dependence on a refractive index of said tissue sample when said nanostructured sensor is in contact with said tissue sample; directing interrogating optical radiation at a plurality of frequencies to said nanostructured sensor; detecting, for each of said optical radiation frequencies, radiation redirected by or transmitted through said nanostructured sensor in response to said interrogating optical radiation; analyzing said detected redirected or transmitted radiation to determine said at least one response characteristic of said nanostructured sensor at each of said plurality of interrogating optical radiation frequencies; and determining at least one attribute of said tissue sample based on said determined at least one response characteristic at said interrogating optical radiation frequencies.

2. The method of Claim 1, wherein said determining at least one attribute of said tissue sample comprises comparing said determined at least one response characteristic with at least one respective reference simulation or measurement of said at least one response characteristic.

3. The method of any one of the preceding claims, wherein said detecting the redirected radiation comprises detecting said redirected radiation at a plurality of angular orientations relative to incidence direction of the interrogating optical radiation.

4. The method of Claim 3, further comprising determining an angular dependence of said redirected radiation as a function of said plurality of angular orientations, wherein optionally said angular dependence of said redirected radiation comprises a dependence of an intensity of said redirected radiation on said plurality of angular orientations.

5. The method of Claim 4, wherein said analysis of the angular dependence is used to compensate for effects of light polarization or noise.

6. The method of Claim 4, wherein said at least one response characteristic comprises said angular dependence of said redirected radiation.

7. The method of Claim 1, wherein said at least one nanostructured sensor comprises a plurality of nanostructured elements distributed over a metasurface of said nanostructured sensor.

8. The method of Claim 7, wherein said placing said at least one nanostructured sensor in contact with said tissue sample comprises placing said plurality of nanostructured elements in contact said tissue sample.

9. The method of Claim 8, wherein said directing the optical interrogating radiation comprises directing, for at least one of said optical frequencies, the interrogating optical radiation to a plurality of spatial locations of said metasurface, wherein each of said spatial locations contains one or more of said nanostructured elements, and detecting radiation redirected by or transmitted through said each of said spatial locations in response to illumination thereof by said interrogating optical radiation.

10. The method of Claim 9, further comprising: performing an initial scan of said metasurface using the interrogating optical radiation beam at a first beam size to acquire a coarse dataset and identify one or more regions of interest; and performing a subsequent scan of said one or more regions of interest using the interrogating optical radiation beam at a second beam size smaller than the first beam size to acquire a high -resolution dataset11 . The method of any one of Claims 9 and 10, further comprising analyzing said detected redirected or transmitted radiation associated with said plurality of spatial locations to generate a map indicating, for each of said spatial locations, a respective at least one response characteristic of said nanostructured sensor corresponding to that spatial location, wherein optionally said analyzing said detected redirected or transmitted radiation comprises analyzing an intensity of said detected redirected or transmitted radiation associated with each of said spatial locations.

12. The method of Claim 11, wherein said analyzing said detected redirected or transmitted radiation comprises analyzing an intensity spectrum of said redirected or transmitted radiation associated with each of said plurality of spatial locations.

13. The method of Claim 12, wherein said analyzing said detected redirected or transmitted radiation further comprises determining said intensity of the redirected or transmitted radiation for a plurality of angular orientations.

14. The method of Claim 11, wherein said determining at least one attribute of said tissue sample comprises processing said plurality of said at least one response characteristic, each corresponding to one of said spatial locations, to generate a spatial map of said refractive index of said tissue sample.

15. The method of Claim 14, wherein said spatial map has a resolution equal to or better than about 1 micron.

16. The method of Claim 15, wherein said resolution is in a range of about 200 nm to about 1 micron.

17. The method of Claim 15, wherein said resolution is a sub-cellular resolution.

18. The method of any one of Claims 14 - 17, further comprising determining said at least one attribute of said tissue sample based on said map.

19. The method of Claim 18, wherein said determining said at least one attribute of the tissue sample comprises classifying each of a plurality of said tissue sample portions as being cancerous or non-cancerous based on said spatial map of the refractive index of said tissue sample.

20. The method of any one of the preceding claims, wherein said at least one response characteristic comprises any of an intensity of said redirected or said transmitted radiation, at least one parameter associated with at least one spectral resonance feature, a phase shift between interrogating optical radiation incident on the nanostructured sensor and said redirected radiation, a change in polarization of the redirected radiation relative to the incident interrogating optical radiation, an integrated intensity or an integrated reflectivity over a spectral band, a temporal response of the redirected light when the interrogating optical radiation comprises a plurality of radiation pulses, a scattering angle distribution of said redirected radiation, or an interferometer pattern generated via interference between the interrogating optical radiation incident on the nanostructured sensor and said redirected radiation.

21. The method of Claim 20, wherein said parameter associated with said at least one spectral resonance feature comprises any of a spectral position, a phase angle, a quality factor (Q), a group delay, an integrated intensity, a spectral derivative, or a peak asymmetry.

22. The method of Claim 21, wherein said at least one spectral resonance feature comprises a plurality of spectral resonance features.

23. The method of Claim 22, wherein said parameter comprises any of at least one spectral spacing between said plurality of spectral resonance features, or at least one ratio of intensity peaks of said plurality of resonances.

24. The method of any one of the preceding claims, wherein said interrogating optical radiation comprises any of a broadband and a multi-band optical radiation containing said plurality of optical frequencies.

25. The method of Claim 24, wherein said multi -band optical radiation is generated by tuning a narrow-band tunable optical source.

26. The method of Claim 25, wherein said tunable optical source comprises any of a tunable laser and a supercontinuum laser.

27. The method of any one of the preceding claims, wherein said interrogating optical radiation comprises any of continuous, pulsed, and modulated radiation.

28. The method of Claim 27, further comprising utilizing lock-in detection to detect any of said redirected and transmitted radiation generated in response to illumination of said at least one nanostructured sensor by said modulated radiation.

29. The method of any one of the preceding claims, wherein said interrogating optical radiation comprises a plurality of narrow-band optical beams each having one of said plurality of frequencies, wherein optionally said optical beams are directed to said at least one nanostructured sensor concurrently or in a plurality of different temporal intervals.

30. The method of Claim 29, wherein at least two of said narrow-band optical beams have different polarizations.

31. The method of any one of Claims 29 and 30, wherein said two of said narrow-band optical beams are directed to said at least one nanostructured sensor along different incidence angles.

32. The method of any one of Claims 29 - 31, wherein said plurality of narrow-band optical beams comprise a plurality of laser beams.

33. The method of any one of the preceding claims, where said at least one response characteristic comprises a plurality of response characteristics.

34. The method of Claim 33, wherein said plurality of response characteristics are independent of one another.

35. The method of Claim 33, wherein said plurality of response characteristics are interrelated.

36. The method of any one of the preceding claims, wherein said determining at least one attribute of said tissue sample comprises utilizing an artificial intelligence (Al) engine to process said determined at least one response characteristic of said nanostructured sensor.

37. The method of Claim 36, wherein said Al engine employs any of a supervised, a semisupervised and an unsupervised learning algorithm.

38. The method of any one of Claims 36 and 37, wherein said Al engine is configured to employ a classification algorithm for determining said attribute of the tissue sample, wherein optionally said Al engine is trained on both general reference data and patientspecific data.

39. The method of Claim 38, wherein training said Al engine comprises: pre-training said Al engine on a general reference dataset comprising a diverse array of samples to create a pre-trained model; and fine-tuning said pre-trained model using a subject-specific calibration dataset .

40. The method of any one of the preceding claims, further comprising utilizing any of an AI- based and a probabilistic modeling technique to assign uncertainty to said determined at least one attribute of said tissue sample.41 . The method of any one of the preceding claims, wherein said nanostructured sensor comprises a plasmonic sensor.

42. The method of Claim 41, wherein said plasmonic sensor comprises a plurality of nanosized electrically conductive elements, wherein optionally said nano-sized electrically conductive elements are distributed over an electrically insulating substrate.

43. The method of Claim 42, wherein said nano-sized electrically conductive elements are arranged relative to one another according to a periodic pattern.

44. The method of Claim 43, wherein said periodic pattern is characterized by a unit cell having any of a square or hexagonal geometry.

45. The method of Claim 44, wherein said unit cell has a dimension in a range of about 20 nm to about 1 micron.

46. The method of any one of Claims 42-45, wherein said nano-sized electrically conductive elements are distributed according to a non-periodic pattern, and wherein a surface density of said nanostructured elements varies across a surface of the sensor.

47. The method of any one of Claims 42 - 46, wherein said nano-sized electrically conductive elements comprise a metal, and wherein optionally said metal comprises any of gold and silver.

48. The method of any one of Claims 42 - 47, wherein said nano-sized conductive elements comprise a plurality of rods.

49. The method of Claim 48, wherein said plurality of rods have any of a cylindrical and a tapered shape.

50. The method of any one of Claims 1 - 40, wherein said nanostructured sensor comprises a photonic crystal.

51. The method of any one of Claims 1-40, wherein said nanostructured sensor is a hybrid sensor comprising a plasmonic array and at least one two-dimensional (2D) material layer, wherein said 2D material comprises any of graphene, hexagonal boron nitride (h- BN), or a transition-metal di chalcogenide (TMD).

52. The method any of one of said preceding claims, wherein said tissue sample includes a margin tissue sample at least partially surrounding a tissue suspected of being cancerous.

53. A surgical device for in-vivo characterization of a tissue sample, comprising: a housing; a nanostructured sensor coupled to a distal end of said housing and configured to be placed in contact with a tissue sample, said nanostructured sensor exhibiting at least one response characteristic dependent on a refractive index of said tissue sample; a light source configured to generate interrogating optical radiation at a plurality of frequencies; an interrogation optical path configured to direct said interrogating optical radiation to said nanostructured sensor; a detection optical path; and a data acquisition module configured to receive, via said detection optical path, radiation redirected from said nanostructured sensor and process said redirected radiation to determine at least one attribute of said tissue sample.

54. The surgical device of Claim 53, wherein the device is a resectoscope, further comprising a beam scanning and conditioning module configured to scan the interrogating optical radiation across an area of the nanostructured sensor.

55. The surgical device of Claim 53, wherein the device comprises an optical needle probe, and wherein said nanostructured sensor is integrated at a distal tip of said needle.

56. A robotic surgical system comprising the surgical device of Claim 55, wherein said needle probe is mounted on a robotic arm.

57. A method of designing a nanostructured sensor, comprising: generating one or more initial theoretical designs of a nanostructured sensor by defining an initial configuration of a plurality of nanostructured elements via a plurality of sets of design parameters, initializing each of said sets of plurality of design parameters via assignment of a respective set of initial parameter values thereto, theoretically determining at least one response characteristic of said one or more initial theoretical designs of the sensor when said sensor is in contact with a theoretically- defined tissue sample and illuminated with interrogating optical radiation at a plurality of frequencies, evaluating a figure-of-merit for said one or more initial theoretical designs based on a predefined figure-of-merit function and said determined at least one response characteristic for said one or more initial theoretical designs, accepting at least one of said sets of said initial design parameters associated with at least one of said one or more initial theoretical designs as at least one initial optimal sensor design when said figure-of-merit function evaluated based on said at least one of said sets of said initial design parameters is optimal, and iteratively adjusting one or more of said sets of initial design parameters to optimize said figure-of-mertic function and thereby generating at least one initial optimal sensor design when said evaluated function-of-merit for said sets of one or more initial designs is sub-optimal,simulating or fabricating at least one test nanostructured sensor based on said at least one initial optimal sensor design, utilizing said simulated or fabricated test nanostructured sensor to determine at least one response characteristic thereof when said test nanostructured sensor is placed in contact with a reference simulated or real tissue sample and illuminated via simulated or real interrogating optical radiation at a plurality of optical frequencies, analyzing said determined at least one response characteristic to perform at least one of (1) accepting said at least one initial optimal sensor design as a final sensor design, (2) adjusting at least one of said figure-of-merit function and one or more of sets of said design parameters to generate a revised initial optimal design of said nanostructured sensor.

58. The method of Claim 57, wherein said analyzing said determined at least one response characteristic comprises determining at least one attribute of said reference tissue sample based on said at least one response characteristic.

59. The method of Claim 58, further comprising adding a simulated or previously-observed noise pattern to said at least one response characteristic and determining said at least one attribute in presence of said noise pattern.

60. The method of any one of Claims 57 and 59, further comprising adjusting said function- of-merit based on said determined at least one attribute.

61. The method of any one of Claims 57 - 60, wherein said figure-of-merit function comprises at least one of a number of spectral resonances exhibited by said nanostructured sensor over a target optical frequency band, a sharpness of at least one spectral resonance exhibited by said nanostructured sensor, a spectral separation between at least two spectral resonances exhibited by said nanostructured sensors, a classification accuracy of said tissue sample based on an Al algorithm trained based on previously-obtained simulated or real tissue classification data, impact of simulated noise on classification accuracy of said tissue sample based on an Al classification algorithm.

62. The method of any one of Claims 57 - 60, wherein said figure-of-merit function is defined as a weighted function of a number of spectral resonances exhibited by said nanostructured sensor over a target optical frequency band, a sharpness of at least one spectral resonance exhibited by said nanostructured sensor, a spectral separation between at least two spectral resonances exhibited by said nanostructured sensors, a classification accuracy of said tissue sample based on an Al algorithm trained based on previously- obtained simulated or real tissue classification data, impact of simulated noise on classification accuracy of said tissue sample based on an Al classification algorithm.

63. The method of any one of Claims 57 - 62, wherein said theoretically-defined tissue sample is defined based on a refractive index thereof.

64. The method of any one of Claim 63, wherein said theoretically-defined tissue sample is defined based on a distribution of said refractive index at a plurality of spatial locations of said tissue sample.

65. The method of any one of the preceding claims, wherein said theoretically determining at least one response characteristic comprises determining, for a plurality of spatial locations associated with a metasurface of said initially designed nanostructured sensor in contact with a plurality of different spatial locations of said tissue sample, said response characteristic for each of said spatial locations of said nanostructured sensor.

66. The method of Claim 65, further comprising analyzing said response characteristic associated with said plurality of spatial locations of the nanostructured sensor to theoretically determine an attribute of said tissue sample at said plurality of spatial locations of said tissue sample.

67. The method of Claim 66, further comprising evaluating said figure-of-merit function via comparison of said theoretically-determined attribute at said plurality of spatial locations of the tissue sample with previously-defined attributes of said tissue sample at said plurality of spatial locations.

68. The method of Claim 67, further comprising adjusting said figure-of-merit function based said evaluation thereof.

69. A method for characterizing a sample, comprising: placing at least one nanostructured sensor in contact with said sample, wherein said nanostructured sensor includes at least one response characteristic exhibiting dependence on a refractive index of said sample when said nanostructured sensor is in contact with said sample, directing interrogating optical radiation at a plurality of frequencies to said nanostructured sensor, detecting, for each of said optical radiation frequencies, radiation redirected by or transmitted through said nanostructured sensor in response to said interrogating optical radiation , analyzing said detected redirected or transmitted radiation to determine said least one response characteristic of said nanostructured sensor at each of said plurality of interrogating optical radiation frequencies, and identifying at least one attribute of said sample based on said measurements of said at least one feature at said interrogating optical radiation frequencies.

70. The method of Claim 69, wherein said sample comprises a semiconductor sample.

71. A system for ex-vivo characterization of a tissue sample, comprising: a light source configured to generate interrogating optical radiation; a sample holder configured to hold the tissue sample,a sensor holder configured to hold the nanostructured sensor, and configured to attach to the sample holder to bring the sensor into contact with the sample; a movable stage configured to translate in at least two dimensions relative to the interrogating optical beam, on which said sample holder attached to the said sensor holder can be positioned, such that the beam is incident on different spatial locations on the said nanostructured sensor by translating the said movable stage; an interrogation optical path configured to direct the interrogating optical radiation from said light source onto a portion of said nanostructured sensor that is in contact with the tissue sample; a detection optical path configured to collect radiation reflected from said portion of the nanostructured sensor or transmitted through said portion of the nanostructured sensor; a spectrometer in said detection optical path configured to detect said reflected radiation and generate spectral data corresponding thereto; and a data processing module configured to receive said spectral data and process said data to determine at least one attribute of the tissue sample at a spatial location corresponding to said portion.

72. The system of Claim 71, wherein said movable stage is configured to translate in X and Y directions to expose different spatial locations of the sensor to the interrogating optical radiation, thereby enabling a scan of an area of the tissue sample.

73. The system of Claim 71, further comprising an iris disposed in the interrogation optical path, wherein a size of the interrogating optical radiation beam is adjustable by said iris.

74. The system of Claim 71, further comprising a polarizer disposed in the interrogation optical path, wherein a polarization of the interrogating optical radiation is selectable.

75. The system of Claim 71, wherein said light source is any of a broadband light source, a tunable laser, or a supercontinuum laser.

76. The system of Claim 71, further comprising a CCD imaging device configured to receive a portion of the reflected radiation via a beam splitter to generate a visible image of a surface of the nanostructured sensor.

77. The system of any one of Claims 71-76, wherein the tissue sample is an excised margin tissue sample obtained during a surgical procedure.

78. A method for ex-vivo characterization of a tissue sample in a surgical environment, comprising: placing a tissue sample on a sample holder of a measurement system, said sample holder being supported by a movable stage; bringing the tissue sample into contact with a nanostructured sensor supported by a movable stage, using the movable stage to expose different spatial locations of said nanostructured sensor to an interrogating optical beam, detecting, via a spectrometer, a spectrum of radiation redirected by or transmitted through said different spatial locations of said nanostructured sensor; processing said spectrum via a data processing module of said measurement system to determine at least one attribute of the tissue sample corresponding to said different spatial locations.

79. The method of Claim 78, wherein each of said spatial location is placed in contact with at least a portion of said sensor via translating said movable stage.

80. The method of Claim 79, wherein said at least one attribute is determined for said plurality of spatial locations, thereby generating a spatial map of said at least one attribute.

81. The method of Claim 80, further comprising displaying said map to a user in real-time, thereby providing immediate feedback during a surgical procedure.

82. The method of any one of Claims 78 - 81 , wherein said tissue sample comprises an excised tumor margin.

83. The method of Claim 82, wherein said tissue sample further comprises a portion of a tumor, and wherein said processing comprises training an Al engine based at least partially on spectral data obtained from the tumor portion to differentiate between cancerous and non-cancerous tissue.

Images