Hybrid optical coherence tomography and raman spectroscopy sensor

Abstract

Embodiments of the present disclosure a forward looking hybrid Raman spectroscopy and optical coherence tomography sensor, having the same field of view for both sensing modalities, and suitable for deployment endoscopically. The sensor comprises a sensor tip that integrates the Raman spectroscopy fibers and the OCT scanning fiber, as well as the actuator to permit scanning movement of the OCT scanning fiber, together with a common field of view. The primary challenge is to provide sufficient scanning movement space for the OCT fiber (as controlled by the actuator) whilst arranging the Raman spectroscopy fibers so that they have the same field of view i.e. view the same tissue. This problem is met by spatially arranging the Raman spectroscopy fibers about or around the OCT imaging fiber so as to have the same forward looking field of view as the OCT fiber whilst permitting space for the OCT fiber to vibrate in at least one, and in some embodiments two, lateral dimensions. The spatial arrangements that achieve this have multiple Raman excitation light fibers, but with a larger number of Raman collection fibers, with the fibers carrying the Raman excitation light being substantially evenly spatially distributed within the plurality of Raman spectroscopy fibers around or to either side of the OCT fiber.

Description

[0001] P608857PC00

[0002] HYBRID OPTICAL COHERENCE TOMOGRAPHY AND RAMAN SPECTROSCOPY SENSOR

[0003] Field

[0004] The present disclosure relates to a hybrid optical coherence tomography (OCT) and Raman spectroscopy sensor that integrates both sensors into a forward looking sensor package with the same field of view for both sensing modalities. With such a sensor then both morphological (via OCT) and molecular (via Raman spectroscopy) analysis of the same tissue can be performed at the same time. In some examples of the disclosure the hybrid sensor is packaged in a manner suitable for in vivo endoscopy.

[0005] Background

[0006] Head and neck cancers including oral cancers are severe life-limiting diseases, particularly when diagnosed in the later stages and remain stubbornly resistant to outcome improvements with surgical technique improvements, makes it an ideal first model for us to use. Earliest possible detection in precancer (dysplasia) or early cancer stages is the single most important measure for reducing oral cancer patients' morbidity and mortality rates. Conventional diagnosis relies on visual inspection guiding the biopsy of suspicious oral lesions, which has resulted in 5 year survivals of up to 90% for early tumours or dysplasia. This survival rate however declines to ~50% for advanced cancers, highlighting the necessity of early diagnosis as a primary factor in determining patient outcomes. Visual guided tissue sampling suffers from inter-observer dependence, sampling errors and difficulty in identifying and targeting flat dysplastic lesions in particular. The differentiation between inflamed / scarred tissue and dysplasia, establishing cancer depth, local / distant spread assessment (staging) and grading (aggression), and margin assessment of malignant lesions all represent monumental clinical challenges. Further, biopsy also remains a significant deterrent to patient attendence. There is a significant unmet clinical need for a rapid, non-invasive, outpatient-deployable and patient acceptable diagnostic method to improve surveillance, early diagnosis, intra-operative margin assessment and follow up I long -term management of oral cancer patients.

[0007] Optical imaging and spectroscopy have recently offered great promise to address unmet clinical needs since they are non-invasive and can capture molecular / structural information without prior tissue preparation. These technologies offer new, clinically relevant biomarkers such that essential point-of-care decisions can be made with the use P606567PC00 of safe (non-ionizing) levels of radiation, and at a much lower cost than with MRI, CT and PET imaging.

[0008] Raman spectroscopy (RS) is a label-free, rapid and minimally invasive optical technique using laser light that provides a point-wise optical fingerprint of the myriad of inter- and intra-cellular building blocks (i.e., proteins, lipids and DNA) of tissue at the biomolecular level. In the last two decades there has been accumulating evidence on the accurate diagnostic capability of Raman spectroscopy through comprehensive in vitro, ex vivo and in vivo studies. However, Raman spectroscopy offers no morphological information about the tissue preventing any grading or depth of invasion assessment. Figure 1 illustrates a typical prior art RS arrangement, where a flexible probe suitable for endoscopy has multiple optical fibers running in parallel therethrough. A central one of the fibers provides laser light to act as Raman excitation light, which is focussed onto a tissue sample via a ball lens at the distal end of the arrangement. Multiple collection fibers are provided around the excitation fiber, with a field of view of the illuminated sample. Raman light emitted from the sample is then collected by the collection fibers and fed back to a spectroscope, which produces a Raman spectrum of the reflected light from which molecular analysis of the illuminated tissue can be made.

[0009] Optical coherence tomography (OCT) provides cross-sectional images of tissue / cell morphology on submicron scales. OCT images are created by differences in the way tissues or materials reflect and scatter light, revealing variations in their structure and composition. OCT can be used in real-time to visualize tissue layers, assess margins, and ensure they remove or treat diseased tissue precisely while minimizing damage to healthy surrounding structures. Figure 2 illustrates a typical prior art OCT sensor, which comprises a single mode fiber which passes through a piezoelectric actuator, with a gradient index lens focussing lights from the fiber on the sample (not shown). The piezoelectric actuator can move the fiber in one or two dimensions to allow the field of view of a the fiber to be scanned, to allow morphological images of the tissue to be built up.

[0010] Combining RS and OCT in fibre-optics has been a long-standing objective for researchers. Raman spectroscopy and OCT represent two distinct yet mutually highly reinforcing diagnostic methodologies. Wang et al in Optics Letters, Vol 41, Nol3, 1 July 2016 report on the development of a sideview handheld hybrid Raman spectroscopy (RS) and optical coherence tomography (OCT) technique for real-time in vivo tissue measurements, as shown in Figure 3. The sideview handheld RS-OCT optical probe is designed to coalign the optical paths of RS and OCT sampling arms, whereby a compact long-pass dichroic mirror (LPDM) is utilized to transmit the OCT signal through a gradient index rod lens and a reflection mirror, whereas the LPDM deflects the tissue Raman signal by 90°, leading to P606567PC00 coaligned RS / OCT optical samplings on the tissue. Wang shows that such a hybrid RS and OCT technique developed is capable of simultaneously acquiring both morphological and biochemical information about the oral tissue in vivo, facilitating real-time, in vivo tissue diagnoses and characterizations in the oral cavity.

[0011] The arrangement of Wang et al has significant drawbacks, however, in deploying hybrid RS and OCT imaging more broadly. Firstly, the solid and inflexible arrangement of Wang, necessary because of the use of the gradient index (GRIN) rod lens, means that the whole arrangement is stiff and cannot be bent. This is acceptable for the use case of Wang, which was designed for exploring the oral cavity only, but would be unacceptable for internal endoscopic use, for example where insertion into the body can occur through the oesophagus or rectum of the subject, or via trocar ports in the abdomen. Secondly, Wang et al. failed to solve the primary problem of integrated different modality imaging sensors, that of providing a forward facing common field of view. Instead, Wang et al were only able to come up with an arrangement that has a sideways facing common field of view, with the resultant geometry of the sensors being that they effectively lie side by side looking sideways along a common field of view.

[0012] In addition, there also exist forward viewing Raman and OCT devices that are combined via a single lens ( see e.g. Patil et al Opt Lett. 2008 May 15;33(10): 1135-1137, Klemes et al Biophotonics 2017, 10, 1442., and Mazurenka, et al Rev. Sci. Instrum. 2017, 88, 105103.). However, all of these arrangements are very bulky benchtop designs, are not optical fiber based, and cannot be used endoscopically.

[0013] In view of the above it is understood that the integration of RS and OCT imaging techniques into a single endoscope-compatible forward-viewing system represents a monumental technical challenge, but holds the promise of creating a cohesive optical biopsy platform capable of providing insights into both cell morphology and tissue biochemistry in vivo.

[0014] Summary of the Disclosure

[0015] In the Label-free Bioimaging Laboratory located at Guy's Hospital, London, UK we have recently established a state-of-the-art Fiber-optic Probe Manufacturing Facility (FMF) (described at https: / / bergholtlab.eom / #FMF). This facility equips us with the unique capability to develop highly innovative, miniaturized probes with forward viewing technology— a feat that was previously unattainable elsewhere in the field. Using this facility, we have developed a forward looking hybrid Raman spectroscopy and optical coherence tomography sensor, having the same field of view for both sensing modalities, and suitable for deployment endoscopically. The sensor comprises a sensor tip that P606567PC00 integrates the Raman spectroscopy fibers and the OCT scanning fiber, as well as the actuator to permit scanning movement of the OCT scanning fiber, together with a common field of view. The primary challenge is to provide sufficient scanning movement space for the OCT fiber (as controlled by the actuator) whilst arranging the Raman spectroscopy fibers so that they have the same field of view i.e. view the same tissue. The further challenge is also to ensure that the distribution of Raman fibers is such that the collection fibers are not located overly far from the excitation fiber, else insufficient Raman light will be collected by the collection fibers for good signal to noise ratios and spatial resolution to be maintained. Both of these problems are met by spatially arranging the Raman spectroscopy fibers about or around the OCT imaging fiber so as to have the same forward looking field of view as the OCT fiber whilst permitting space for the OCT fiber to vibrate in at least one, and in some embodiments two, lateral dimensions. The spatial arrangements that achieve this have multiple Raman excitation light fibers, but with a larger number of Raman collection fibers, with the fibers carrying the Raman excitation light being substantially evenly spatially distributed within the plurality of Raman spectroscopy fibers around or to either side of the OCT fiber.

[0016] For example in one embodiment where the OCT fiber scans laterally in one dimension, two rows of Raman sensing fibers may be provided either side of the lateral movement space of the OCT fiber, forming a movement channel for the OCT fiber. Every nth fiber in the row is a Raman excitation fiber, with n greater than 1, but less than x / a, where x is the number of fibers in each row, and a is 2 or more. This means that there are at least two excitation fibers in each row evenly spaced within the row, such that no collection fiber is overly distanced from an excitation fiber.

[0017] In another embodiment where the OCT fiber scans laterally in two dimensions, for example to permit a 2D OCT image to be made, the Raman sensing fibers are distributed in a circle, around the edge of the scanning movement space of the OCT fiber. In this example, every nth fiber around the circle of fibers is a Raman excitation fiber, with n greater than 1, but less than x / a, where x is the number of fibers in the circle, and a is 2 or more. This means that there are at least two (but preferably more) excitation fibers in the circle evenly spaced around the circle, such that no collection fiber is overly distanced from an excitation fiber.

[0018] In view of the above, the present invention provides a forward looking hybrid Raman spectroscopy and optical coherence tomography (OCT) sensor, comprising respective sensing optical fibers arranged to have the same forward looking field of view for both sensing modalities, the arrangement being sufficiently sized and flexible so as to be suitable for deployment endoscopically. P606567PC00

[0019] From another aspect, there is further provided a hybrid Raman spectroscopy and optical coherence tomography (OCT) probe tip for use with a forward looking hybrid Raman spectroscopy and optical coherence tomography (OCT) sensor, the probe tip comprising: an OCT imaging fiber coupled to a vibrational drive system arranged in use to vibrate the OCT imaging fiber in at least one lateral dimension to provide for scanning of the OCT imaging fiber across a forward-looking field of view to be imaged; and a plurality of Raman spectroscopy fibers spatially arranged about or around the OCT imaging fiber so as to have the same forward looking field of view as the OCT fiber whilst permitting space for the OCT fiber to vibrate in the at least one lateral dimension; wherein the plurality of Raman spectroscopy fibers comprise multiple Raman excitation light fibers and a larger number of Raman collection fibers, with the fibers carrying the Raman excitation light being substantially evenly spatially distributed within the plurality of Raman spectroscopy fibers around or to either side of the OCT fiber.

[0020] In one example embodiment the OCT imaging fiber is arranged to scan laterally in one dimension, and two or more rows of Raman sensing fibers are provided either side of the lateral movement space of the OCT fiber, forming a movement channel for the OCT fiber. In a preferred example, every nth fiber in the row is a Raman excitation fiber, with n greater than 1, but less than x / a, where x is the number of fibers in each row, and a is 2 or more.

[0021] In another example embodiment, the OCT imaging fiber scans laterally in two dimensions, for example to permit a 2D OCT image to be made, and the Raman sensing fibers are distributed in at least one arc, around the edge of the scanning movement space of the OCT fiber. In a preferred example, every nth fiber around the arc of fibers is a Raman excitation fiber, with n greater than 1, but less than x / a, where x is the number of fibers in the arc, and a is 2 or more. In one example there are at least two or more excitation fibers on the at least one arc, evenly spaced around the arc, such that no collection fiber is overly distanced from an excitation fiber. In another example there are a plurality of arcs of Raman spectroscopy fibers surrounding the scanning movement space of the OCT fiber.

[0022] In one example the Raman sensing fibers are distributed in at least one circle, around the edge of the scanning movement space of the OCT fiber.

[0023] Further examples of the present disclosure include an actuator to drive the OCT fiber to scan in one or two dimensions, the actuator preferably being a piezoelectric tube scanner. In some examples the piezoelectric tube scanner has a plurality of segmented electrodes, P606567PC00 the piezo tube scanner being arranged in use to scan the fiber tip, the scan direction being controllable by the phases and amplitudes of voltage sine wave control signals applied to the plurality of segmented electrodes.

[0024] In some examples the plurality of Raman spectroscopy fibers are angled inwards and / or are bevel cut at the distal end of the probe tip so as to focus any light emitted from the fibers to a sensing point just beyond the distal end of the probe tip.

[0025] Some examples further comprise a transparent window disposed at the distal end of the probe tip, the sensing point of the Raman spectroscopy fibers being located outside the transparent window just beyond the distal end of the probe tip. In some examples the transparent window is formed of Raman transparent materials, such as any one of magnesium fluoride, calcium fluoride, or sapphire.

[0026] In one example the OCT imaging fiber comprises: a length of single-mode fiber (SMF) for light delivery; a no-core fiber (NCF1) for beam expansion; a gradient index fiber (GIF) for focusing; and an angle-cleaved no-core fiber (NCF2) for suppressing reflection. The fiber segments may be fused into a monolithic OCT fiber using a fusion splicer.

[0027] From another aspect the present disclosure also provides a forward looking hybrid Raman spectroscopy and optical coherence tomography (OCT) sensor further comprising a computer system arranged to process the Raman spectroscopy data and OCT sensor data, the computer system being provided with software arranged to control the computer system to: manage an OCT fiber scan trajectory; generate synchronization signals to control OCT imaging and Raman spectroscopy; read any generated OCT image and Raman spectra data; and preprocesses the OCT image and Raman spectra data.

[0028] In one embodiment the Raman data preprocessing includes one or more of dark current noise subtraction and intensity scaling. In addition, the OCT data preprocessing may include one or more of: background removal, k-domain linearization, fast Fourier transformation (FFT), and / or normalization. In a preferred arrangement the OCT preprocessing tasks are offloaded to a GPU, enabling low-latency, video-rate rendering of the OCT images.

[0029] Moreover, in one example a multithreading design is employed to run the fast OCT data acquisition in parallel with the slower Raman acquisition.

[0030] In one example the software standardizes the Raman spectra before multivariate statistical analysis by performing post-processing steps including one or more of: background removal, system response correction, truncation, and / or baseline correction. P606567PC00

[0031] In one example the system further includes a display, wherein the software controls the computer to display a real-time OCT image fused with a corresponding Raman spectrum, both derived from the same field of view, on the display.

[0032] In addition, in a further example the OCT fiber scanner and OCT spectrometer are hardware-synchronized to ensure precise image construction; and / or the Raman spectrometer is software-synchronized with the OCT spectrometer, allowing for temporal co-registration of the two modalities.

[0033] In a further example the computer system is further provided with software to implement a deep learning model to process the Raman spectra and OCT images, wherein the deep learning model: ingests the input data comprising Raman spectra and OCT images; applies multiple layers of convolutional filters to extract features including spatial patterns in the OCT images and spectral signatures in the Raman data; and applies a softmax layer or another probabilistic output layer to assign probabilities to different tissue types based on the extracted features. In a further example the deep learning model applies a fusion layer to combine the Raman features and OCT features prior to the probabilistic output layer assigning probabilities to different tissue types based on the combined features.

[0034] ]

[0035] Further features and advantages will be apparent from the appended claims.

[0036] Further features and advantages of the present disclosure will become apparent from the following description, presented by way of example only, and with reference to the accompanying drawings, wherein like reference numerals refer to like parts, and wherein:

[0037] Figure 1 is a diagram of a prior art Raman spectroscopy probe;

[0038] Figure 2 is a diagram and photograph of a prior art optical coherence tomography probe;

[0039] Figure 3 is a diagram and photograph of a prior art hybrid Raman spectroscopy and optical coherence tomography sensor;

[0040] Figure 4 is a diagram of a hybrid Raman spectroscopy and optical coherence tomography sensor of the present disclosure;

[0041] Figure 5 is a diagram of the OCT fiber used in embodiments of the present disclosure;

[0042] Figure 6 is a photograph of the OCT fiber used in embodiments of the present disclosure; P606567PC00

[0043] Figure 7 is a diagram illustrating how the OCT fibre can move in the sensor of the present disclosure;

[0044] Figure 8 is a series of diagrams illustrating how the sensor of the present disclosure is manufactured, as well as its manufactured configuration;

[0045] Figure 9 is a diagram illustrating the arrangement of optical fibres at a distal end of the sensor in a first embodiment of the present disclosure;

[0046] Figure 10 is a diagram illustrating the arrangement of optical fibres at a distal end of the sensor in a second embodiment of the present disclosure;

[0047] Figure 11 is a diagram illustrating the signal processing that is undertaken in an embodiment of the present disclosure;

[0048] Figure 12 is an example Raman spectrum and OC T image obtained from the hybrid sensor of the present disclosure;

[0049] Figure 13 is a diagram illustrating how the two types of data are fused together in embodiments of the disclosure;

[0050] Figure 14 is a diagram illustrating classification results obtained from embodiments of the disclosure;

[0051] Figure 15 is a first example combined OCT image (upper) and Raman spectrum (lower) obtained by an embodiment of the disclosure;

[0052] Figure 16 is a second example OCT image obtained by an embodiment of the disclosure;

[0053] Figure 17 is a second example Raman spectrum obtained by an embodiment of the disclosure;

[0054] Figure 18 is a photograph of a computer system forming an embodiment of the disclosure;

[0055] Figure 19 is a photograph of a hybrid probe according to an embodiment of the disclosure, illustrating the typical size of the probe;

[0056] Figure 20 is a diagram of a hybrid Raman spectroscopy and optical coherence tomography sensor of a second embodiment of the present disclosure;

[0057] Figure 21 is a series of Figures showing morphomolecular in vivo monitoring mouse model of H8d\l carcinogenesis used in the second embodiment; P606567PC00

[0058] Figure 22 is a series of figures showing longitudinal quantitative profiling of head and neck carcinogenesis; and

[0059] Figure 23 is a series of Figures showing Al-driven multi-modal RS-OCT data fusion and classification.

[0060] Overview of Embodiments of the Disclosure

[0061] The present disclosure relates to a hybrid Raman spectroscopy and optical coherence tomography imaging probe. Such a probe is already known per se in the prior art, in the Wang et al. paper in Optics Letters, volume 41, number 13, July 1, 2016 referred to above. As described previously, this describes a side viewing hybrid Raman spectroscopy and optical coherence tomography probe, wherein the optical coherence tomography probe and the Raman probe are co-located side-by-side in a side viewing arrangement. The probe is solid, and whilst the OCT beam can be steered from side to side, the driving of such is via an undisclosed arrangement disposed at the proximal end of the instrument, rather than the distal imaging end. This results in a non-flexible probe suitable only for side viewing.

[0062] In contrast to the above, the present disclosure provides a flexible combined Raman spectroscopy and OCT probe, which should be suitable for use as an endoscope. The probe is forward-looking, and the OCT scanning beam is steered via a small piezo tube located at the distal end of the probe. This differs from the Wang et al. prior art arrangement, wherein the OCT scanning beam is steered from the backend. In order to provide for distal end steering, however, there must be space around the OCT fibre to allow the OCT fibre to move. Given the need to provide this vibrational space around the OCT fibre, there then arises the problem of how to arrange the fibres of the Raman probe to have the same field of view as the OCT fibre. In this respect, typical Raman spectroscopy probes provide a bundle of optical fibres with the Raman excitation signal being provided by the fibres in the centre of the bundle i.e. the center of the field of view, and the Raman collection fibres being arranged around the outer edges of the bundle. With such an arrangement it is not immediately apparent how a prior art Raman probe can be integrated with the OCT fibre to provide the same field of view, given the need to provide space around the OCT for it to be able to vibrate from side to side to provide scanning images.

[0063] The present disclosure solves this problem by providing a particular configuration of fibres to act as the Raman excitation and Raman collection fibres. In Figures 9 and 10, two arrangements are shown with the upper arrangement permitting one-dimensional scanning movement of the OCT fibre, and the lower arrangement permitting two P606567PC00 dimensional scanning movement of the OCT fibre, in either case under the control of the piezo tube. Whilst arranging the Raman spectroscopy fibres around the OCT scanning fibre as shown permits movement of the OCT scanning fibre, the inventors have found that it is not enough simply to have one or two of the fibres providing Raman excitation light, or to cluster the excitation fibers together. Instead, when having such an arrangement to provide movement room for the OCT scanning fibre, while still having the same field of view, the Raman excitation fibres must also be relatively evenly distributed along the row of Raman sensing fibres, so that no single sensing fibre in the array is located overly far from a Raman excitation fibre. By having this arrangement of Raman excitation and collection fibres disposed around the OCT fibre but with enough room for the OCT fibre to vibrate (in one or two dimensions) a combined Raman spectroscopy and OCT probe having the same field of view for each imaging modality is obtained. Moreover, the probe can be implemented in a flexible package, for use as an endoscope.

[0064] As well as the hybrid OCT / RS probe design, the present disclosure also relates to the control of the probe, and the processing of the imaging and spectroscopy data that is obtained. In this respect, a comprehensive software package has been developed to control the RS / OCT system, manage the fiber scan trajectory, generate synchronization signals, and preprocess the data. This software enables real-time acquisition and fusion of OCT images and Raman spectra. In addition, we also use deep learning algorithms, specifically convolutional neural networks (CNNs), to fuse features from Raman spectroscopy and OCT images for tissue classification. This approach enhances the diagnostic capabilities by providing complementary molecular and morphological information.

[0065] The software development for the RS / OCT system involves creating a comprehensive realtime software package that manages various aspects of the system. The software controls the fiber scan trajectory, generates synchronization signals, reads the OCT and Raman spectra, and preprocesses the data. The fiber scanner and OCT spectrometer are hardware-synchronized to ensure precise image construction, while the Raman spectrometer is software-synchronized with the OCT spectrometer, allowing for temporal co-registration of the two modalities. The software makes use of a multithreading design, for the reason that the Raman spectrometer operates much more slowly than the OCT spectrometer, and hence a multithreading design is employed to run the fast OCT data acquisition in parallel with the slower Raman acquisition. This design ensures efficient data processing and minimizes latency. P606567PC00

[0066] Raman data preprocessing is also performed by the software, which includes dark current noise subtraction and intensity scaling. In contrast, OCT data preprocessing performed by the software is more complex, involving background removal, k-domain linearization, fast Fourier transformation (FFT), and normalization. These OCT preprocessing tasks are offloaded to a GPU, enabling low-latency, video-rate rendering of the OCT images.

[0067] In addition, the software provides a real-time display of the OCT images and Raman spectra. A screenshot of the software in Figure 15 shows the grayscale OCT image of a polystyrene sheet sample overlapped by its Raman spectrum. The entire workflow, from data acquisition to data processing, is managed by the software.

[0068] In a further development, the system uses data fusion of OCT and Raman data to improve its results. In particular, when trying to classify tissue using the obtained Raman spectrum, a deep learning model is provided that ingests the input data from both input modalities (Raman spectra and OCT images) and applies multiple layers of convolutional filters to extract features, such as spatial patterns in the OCT images and spectral signatures in the Raman data. The features extracted by the CNN are then passed through fully connected neural network layers, where for example a softmax layer or another probabilistic output layer assigns probabilities to different tissue types, based on the learned features. This results in a probabilistic tissue classification, allowing the system to assign likelihoods to various tissue types, increasing confidence in the predictions. By using this dual-modality approach, the system can overcome limitations inherent in using either Raman spectra or OCT imaging alone, as each provides unique insights into tissue composition and structure.

[0069] Overall, the software development for the RS / OCT system is crucial for the real-time acquisition, synchronization, and processing of OCT images and Raman spectra, enabling the integration of these two modalities for enhanced diagnostic capabilities. Moreover, the RS / OCT system as a whole has demonstrated its potential in pro of- of- co nee pt studies for early detection, diagnosis, and prognosis of cancers across various organ systems. The technology holds promise for identifying malignancies in the oral cavity, larynx, nasopharynx, esophagus, stomach, and colorectal region, amongst others.

[0070] Detailed Description of Embodiments

[0071] First Embodiment

[0072] A first embodiment of the present disclosure will now be described with respect to Figure 4, and the following Figures 5 to 19. P606567PC00

[0073] Using our inhouse Fiber-optic Manufacturing Facility (FMF) we have developed a miniaturised 3.4 mm compact RS / OCT probe that is compatible with colonoscopy examinations (Figure 4).

[0074] The RS / OCT technology (Figure 4) is based on a custom-built fibre-optic OCT scanning platform using a miniaturized 2.2 mm piezo tube scanner 42 with four electrodes (Physike Instrumente) and high-speed NIR spectrometer (Wasatch Photonics) controlled using our image acquisition device (NI PCIe-1433, National Instruments). We use a sandwich-like fiber arrangement 44, where the Raman fibers are placed in the top and bottom layers, while the OCT fiber scans across the middle slit (Figure 4). This symmetrical arrangement of the OCT and Raman fibers allows for correlative imaging and spectroscopy. A field of view (FOV) of 1-2 mm is clinically needed. OCT have been integrated by using a singlemode fiber (SMF) for light delivery, a no-core fiber (NCF1) for beam expansion, a gradient index fiber (GIF) for beam focusing for fully correlative RS / OCT. We use a 785 nm diode laser 46 for RS with multiple laser light delivery fibers. In the RS fiber path, in-line optical filters are used to separate the Raman scattered light from the Rayleigh scattering and couple it to the input slit of a fibre-coupled scientific grade spectrograph. The inline filter accepts a round fiber bundle as input and provides a linear fiber array as output. The round fiber bundle input consists of 7 multimode fibers, each with a core diameter of 200 pm, allowing for enhanced light throughput. The linear fiber array contains 19 multimode fibers, each with a core diameter of 100 pm, which ensures high spectral resolution. For RS fiber optics, we have deposited thin film bandpass (48) and longpass (50) filters on the distal tip of the excitation and collection fibers respectively (Shenzhen Photonstream Limited). Assembly has been performed with stainless steel 316, medical grade glue and a biocompatible specially manufactured thin CaF2 window. To be clinically viable, each tissue RS spectrum and OCT image must be acquired with a light irradiance not exceeding the European laser regulation (IEC / EN 60825) maximum permissible exposure limit for 540 nm, 785 nm and 1300 nm. We obtain 25 frame per second for OCT and 1 spectrum per second of RS. Comprehensive risk assessments including but not limited to light exposure, electrical safety and biological safety will be performed in line with EU and UK standards. We have benchmarked any crosstalk effects between the two modalities.

[0075] Figures 5, 6, and 7 show the OCT fiber and piezo tube scanner with four-fold segmented electrodes. As shown in Figure 5, the OCT fiber consists of a length of single mode fiber (SMF) for light delivery, a no core fiber (NCF1) for beam expansion, a gradient index fiber (GIF) for focusing and an angle-cleaved no core fiber (NCF2) for suppressing reflection. Figure 6 shows a photo of the actual real-life embodiment. Figure 7 demonstrates that the piezo tube scanner can scan the fiber tip in one dimension with sine voltage signals. And P606567PC00 the scan direction is controllable by the phases and amplitudes of the sine waves applied to the four electrodes.

[0076] In more detail, OCT imaging requires its distal optics to focus the beam to a spot size matching its axial resolution, usually around 10 pm. However, in a miniature fiber probe, space constraints limit beam expansion, and achieving high lateral resolution results in a very short working distance (typically less than 0.5 mm). To address this, we target a lateral resolution of 20 pm. The OCT fiber consists of several components (as shown in Figures 5 to 7): a length of single-mode fiber (SMF) for light delivery, a no-core fiber (NCF1) for beam expansion, a gradient index fiber (GIF) for beam focusing, and a short angle-cleaved fiber (NCF2) as the end cap for anti -reflection. The lengths of the NCF1 and GIF are optimized to achieve a maximum working distance of 0.6 mm, with a focus diameter of 20 pm. Each fiber segment is fused into a monolithic OCT fiber using a fusion splicer.

[0077] The OCT fiber Is mounted to a 2.2 mm piezo tube scanner. The piezo tube scanner is applied with sine voltage signal at the mechanical resonant frequency of the OCT fiber cantilever. With amplitude of 200 Volts, the deformation of the piezo scanner is around 10 pm, which results in a vibration amplitude of > 1 mm for the fiber tip at the resonant mode. By manipulating the amplitudes and phases of the sine waves applied to the four electrodes, the vibration direction can be fine-tuned after the assembly of the probe, which greatly reduces the alignment effort in the probe fabrication.

[0078] Figure 8 shows the fabrication steps and ultimate configuration of the hybrid probe. With reference to Figure 8, firstly (step 1), the piezo tube with four-fold segmented electrodes is mounted to the stainless-steel tube by a 3D-printed piezo holder. The piezo holder has two side holes, which allows for the optical fibers passing through. Secondly (step 2), two linear fiber bundles, each of them consisting of 5 multimode fibers, are mounted to the fiber bundle holder. The fiber bundle holder is a double wedge design, which forces the two linear fiber bundles bending toward the central axis at the distal end. It will be seen that we are actually bending the Raman fibers so that they point inwards at the tip (82) so they are essentially acting like a "lens" i.e. weakly focusing the light to a line just beyond the CaFz window (once fitted, see next). This focusing effect resulting from the bent fiber tip geometry 82 increases the Raman collection efficacy of the probe.

[0079] As an alternative embodiment to bending the fibers, we can instead cleave the fibers so they have a bevelled angle, rather than bending them. More particularly, we understand that in an extremely small implementation (e.g. < 2.2 mm in diameter), there might not be sufficient space to bend the linear fiber bundles so that they are angled to form a weakly P606567PC00 focused line. Therefore, as an alternative, we can instead use bevel cutting of the fiber tips, with typical bevel angle of around 25°, which allows for the light emitting from the linear fiber bundles at a tilt angle typically of 14°. Other bevel angles can be used, depending on the desired tilt angle and the geometry of the probe tip. We estimate that bevel angles of between 10° to 45° may be used, depending on the amount of space within the probe tip. In addition, a combination of bending of the fiber tip and using a bevelled end may be used, depending on the precise geometry of the probe tip. Where a fiber tip is bent, then typically a smaller bevel angle will be needed to bring the light to a weakly focused imaging line.

[0080] Thirdly at step 3, a CaFz window is attached to the linear fiber bundle holder with an angle. The tilt angle mitigates the back reflection from the window surfaces. The OCT fiber typically consists of a length of single mode fiber with distal optics for light focusing and is inserted through the inner hole of the piezo tube, so as to protrude to the window as close as possible and is fixed to one end of the piezo tube by epoxy. The distal end of the OCT fiber is not in touch to the window or the linear fiber bundles surrounding it, so that it can vibrate freely in the chamber. Lastly (step 4), the central fibers of the two linear fiber bundles are interfaced to the Raman excitation laser in the proximal end. Seven of the rest fibers in the linear bundles are grouped to a round fiber bundle, which is interfaced to the Raman spectrometer via a round-to-linear fiber bundle. The last multimode fiber is not in use. The proximal end of the OCT fiber is interfaced to the OCT system.

[0081] In slightly more detail, to assemble the dual-modality probe, a piezo tube with four electrodes is first mounted to a stainless steel tube (Figure 8). The piezo tube is connected to a voltage source via four thin wires and secured to the stainless tube using a piezo holder. The motion of the free end of the piezo tube is controllable by voltage. The piezo holder has two side holes that allow the optical fibers to pass through. Next, ten multimode fibers for Raman excitation and collection are cleaved flat and grouped into two linear fiber bundles, each containing five fibers at the distal end. The linear fiber bundles are mounted onto a holder with a wedge design, forcing the two bundles to bend inward to increase the overlapping volume of excitation and collection, and to act as a lens i.e. focussing the light to a point beyond the calcium fluoride window. In the third step, a CaF2 or sapphire window is attached to the end facet of the linear fiber bundle holder. The holder is angled at the end facet to mitigate back reflection. The window seals the probe and prevents contamination. Once the window is attached, an OCT fiber— consisting of a single-mode fiber and distal fiber optics— is inserted through the inner hole of the piezo tube. The OCT fiber must be positioned as close as possible to the window (typically within 0.1 mm) due to its limited working distance. The distal tip of the OCT fiber must not touch the window or surrounding fiber bundles to ensure proper scanning. Alignment of the OCT P606567PC00 fiber is performed under a microscope for lateral positioning and using OCT imaging for axial positioning. Once aligned, the OCT fiber is secured to the free end of the piezo tube using epoxy adhesive. Finally, the proximal ends of the fibers are interfaced with the Raman / OCT system. Since there is only one OCT fiber, its connection to the OCT system is straightforward. For the Raman fibers, the central fibers of the two linear fiber bundles are used for excitation, where the Raman laser light is coupled in. The remaining Raman fibers are grouped into a round fiber bundle and connected to the spectrometer via an inline filter. One Raman fiber at the edge of the linear fiber bundle is unused, as it collects less Raman signal due to its distance from the excitation fiber. The signal loss from discarding the edge fiber is minimal.

[0082] Figures 9 and 10 show alternative configurations of the Raman fibers at the distal end of the sensor. In Figure 9 the OCT fiber scans backwards and forwards in one dimensions (laterally across the page in this example), and the Raman excitation and collection fibers are arranged in two rows either side of the scanning movement direction of the OCT fiber, and parallel thereto. As will be seen, in the two rows of Raman fibers, every nth fiber is a an excitation fiber, with the excitation fibers substantially equally distributed along the rows of fibers, with 2 or 3 collection fibers located therebetween. Using more excitation fibers gives increased Raman signal efficiency, as the collection fibers are then all located either next to, or no more than one or two fiber widths away from an excitation fiber.

[0083] In Figure 10 the OCT fiber can undertake 3D volumetric scanning through spiral or Lissajous scanning within the space created by the circle of Raman fibers. In this respect, as shown the Raman fibers are arranged in a circle, with typically every 3rdor 4thfiber being an excitation fiber, so that again a collection fiber is no more than one or two fiber widths away from an excitation fiber. The arrangement of Figure 10 allows for the OCT fiber to undertake two dimensional scanning, for example to allow an OCT 2D image to be built up.

[0084] Turning to Figure 11, we have also developed prototype software to control our benchtop RS / OCT system. Figure 11 shows the elements of this software. This comprehensive realtime software package manages the fiber scan trajectory, generates synchronization signals, reads the OCT and Raman spectra, and preprocesses the data. The fiber scanner and OCT spectrometer are hardware-synchronized to ensure precise image construction, while the Raman spectrometer is software-synchronized with the OCT spectrometer, allowing for temporal co-registration of the two modalities.

[0085] Since the Raman spectrometer operates much more slowly than the OCT spectrometer, a multithreading design is employed to run the fast OCT data acquisition in parallel with the P606567PC00 slower Raman acquisition. Raman data preprocessing includes simple dark current noise subtraction and intensity scaling. In contrast, OCT data preprocessing is more complex, involving background removal, k-domain linearization, fast Fourier transformation (FFT), and normalization. These OCT preprocessing tasks are offloaded to the GPU, enabling low- latency, video-rate rendering of the OCT images.

[0086] A screenshot of the software Is shown In Figure 15, where the real-time OCT image (the grayscale image) is fused with the Raman spectrum (the curve at the bottom of the Figure) below. To standardize the Raman spectra before multivariate statistical analysis, postprocessing steps such as background removal, system response correction, truncation, and baseline correction are required. The entire workflow, from data acquisition to data processing, is shown in the right panel of Figure 11. In Figure 11, the screenshot of the software shows the grayscale OCT image of a polystyrene sheet sample (top) overlapped by its Raman spectrum (bottom). The light coloured arrows are the front and back surfaces of the CaF2 window.

[0087] Turning now to Figures 12 to 14, these illustrate how a machine learning model is used to fuse features from Raman spectroscopy and OCT images for tissue discrimination and classification based on complementary molecular and morphological information. Figure 12 shows another example set of OCT imaging data (bottom) and Raman spectroscopy data (top) from the same sample. To process this data efficiently, we utilize deep learning algorithms including convolutional neural networks (CNNs), which are particularly well- suited for handling complex image data due to their ability to automatically extract hierarchical features. The CNNs are trained on labeled datasets, learning patterns from both Raman spectra and OCT images to create a probabilistic model for tissue classification.

[0088] During the classification process, the deep learning model ingests the input data (Raman spectra and OCT images) and applies multiple layers of convolutional filters to extract features, such as spatial patterns in the OCT images and spectral signatures in the Raman data. The features extracted by the CNN are then passed through fully connected layers, where a softmax layer or another probabilistic output layer assigns probabilities to different tissue types, based on the learned features. This results in a probabilistic tissue classification, allowing the system to assign likelihoods to various tissue types, increasing confidence in the predictions. By using this dual-modality approach, the system can overcome limitations inherent in using either Raman spectra or OCT imaging alone, as each provides unique insights into tissue composition and structure. As an example, in Figure 12 to 14 we demonstrate classification of tissues based on complementary information (Raman: biomolecular & OCT: morphology) for discriminating keratinized tissue from non-keratinized tissues (see Figure 12). The model was trained for 3 epochs P606567PC00 with a leaning rate of 0.0001, leading to convergence of 100% accuracy (100% sensitivity and 100% specificity - see Figure 14). This highlights the model's ability to effectively integrate complementary information from both Raman and OCT data, leading to highly reliable tissue classification.

[0089] As a final example of the advantages of the combined OCT imaging and Raman spectroscopy that embodiments of the present disclosure provide, Figures 16 and 17 show an OCT image and a Raman spectra taken by the multimodal Raman-OCT fiber-optics probe of the present disclosure. The OCT image of human skin in Figure 16 clearly shows the layered structure and the sweat duct feature. The Raman spectra of human skin and nail in Figure 17 demonstrate the capability of chemical sensing with good signal. The OCT images, with dimensions of 400 pixels x 1024 pixels (lateral x depth), were acquired at a frame rate of 185 Hz. The lateral and axial resolutions are calibrated to approximately 20 pm and 10 pm, respectively. The lateral field of view is 2 mm, while the axial field of view is 5 mm. However, due to the limited penetration depth of light in tissue, only the first 2 mm of depth provides useful information. The Raman spectra were acquired with an integration time of 1 second. The light power applied to the tissue is 7 mW for the 1310 nm OCT and 90 mW for the 785 nm Raman. The OCT image of human finger skin, along with the corresponding Raman spectra, as shown in Figure 16, demonstrates excellent signal quality from the novel Raman-OCT fiber-optic probe.

[0090] In terms of the physical implementation of the system, Figure 18 shows the hardware that comprises the experimental implementation of embodiments of the disclosure. The hardware includes a computer system running control software to control RS and OCT sensors, and sensor processing software to process the RS and OCT signals to obtain the Raman spectroscopy data and OCT imaging data described above. A display is also provided to display the processed data, in the form of a Raman spectrum and OCT image of the sample. Figure 19 shows another photograph of the prototype probe, measuring 3cm in length.

[0091] Second Embodiment

[0092] A second embodiment of the present disclosure will now be described with respect to Figures 20 to 23.

[0093] Head and neck squamous cell carcinoma (HNSCC) remains a global health challenge, imposing substantial health, social and economic burdens1. Despite advances in treatment, the prognosis for HNSCC remains poor, as majority of cases are diagnosed at advanced stages. This makes early detection crucial, as it creates a window of opportunity for timely disease management, better clinical outcomes, and the preservation of vital functions such P606567PC00 as speech and swallowing, ultimately enhancing patients' quality of life. Some precancerous lesions remain latent without ever progressing, while others (10-25%) are more likely to advance to malignancy or develop resistance to treatment2-4. Leaving high- risk lesions in situ leads to evolution of the tumour microenvironment and exacerbation of disease, with devastating consequences for the patient. Therefore, monitoring the disease trajectory of lesions is of significant clinical importance to initiate targeted and aggressive treatment (e.g., resection, ablation or neoadjuvant chemotherapy etc.). Conventional diagnostic techniques are based on visual examination, white light or narrowband endoscopy. When combined with biopsy and histopathology this is inherently invasive and hinder continuous longitudinal monitoring due to the need for repeated tissue sampling.

[0094] HNSCC is a multistep disease driven by alterations in genomic, cellular, molecular, and tissue microarchitecture5. The diverse oncogenomic landscape largely stems from the selection and clonal expansion of tumour-initiating cells that acquire various molecular and genetic alterations6. This includes frequent mutations of tumour suppressor genes such as TP53, FAT1, NOTCH1, PIK3CA and CASP8. Many of these mutations disrupt cellular processes that regulate proliferation and differentiation, resulting in a progressively altered molecular profile within the tissue7. Concurrently, these cellular and molecular changes are accompanied by microscopic morphological transformations. Dysplastic tissues often exhibit loss of epithelial cohesion, disrupted stratification, basal cell hyperplasia, epithelial thickening, and irregularly shaped rete pegs. These structural abnormalities are strongly associated with malignant transformation and increased risk of lesion recurrence8. As dysplasia advances to carcinoma, tumour-initiating cells invade the surrounding stroma, breaking down normal structural boundaries and leading to highly disorganised tissue architecture.

[0095] Fibre-based optical imaging techniques have emerged as promising tools, offering a unique combination of label-free, non-invasive modalities that deliver high-resolution morphological imaging alongside molecular analysis, paving the way for more dynamic and informative tissue assessment. Techniques such as Raman spectroscopy (RS)9-11, optical coherence tomography (OCT), confocal laser endomicroscopy (CLE)12, and multiphoton microscopy13-15including second harmonic generation (SHG), two-photon excited fluorescence (TPEF), and coherent anti-Stokes Raman spectroscopy (CARS), have demonstrated considerable potential for cancer diagnostics. RS exploits the inelastic scattering of light to provide detailed vibrational signatures of molecular bonds, thereby enabling the identification of biomolecular fingerprints and disease markers. RS provides rich molecular fingerprints of tissue pathology, yet is blind to the morphological context in which these changes occur. OCT complements this gap by offering high-resolution, cross- P606567PC00 sectional reconstructions of tissue microarchitecture, enabling the visualisation of subtle structural perturbations of the disease onset and progression16.

[0096] Integrating RS and OCT into a single, miniaturised endoscopic probe presents major engineering challenges. RS depends on efficient photon collection and distal optical filtering. Meanwhile, OCT relies on beam scanning and integrated focusing optics to generate high-resolution images. These fundamentally different optical demands create a complex set of design trade-offs, making it difficult to merge both modalities into one compact fibre-optic probe compatible with endoscope instrument channels. Another critical challenge is balancing a sufficient optical field of view (FOV) with the strict mechanical constraints required for miniaturisation within standard endoscopic channels. Historically, pioneering efforts to combine RS and OCT in a fibre-optic format have therefore been limited to large or side-viewing designs with dimensions that exceed requirements needed in clinical endoscopy17'19. Temporal co-registration remains another challenge: RS acquires spectra on subsecond timescales, whereas OCT provides video-rate images. This temporal discrepancy demands novel fusion strategies that can integrate the disparate data into a joined, coherent representation.

[0097] In this second embodiment we introduce a unified, forward-viewing endoscopic platform that integrates Raman spectroscopy and optical coherence tomography (RS-OCT). The system incorporates a miniaturized, high throughput fibre-optic probe with a spliced-fibre architecture and distal beam scanning. Designed to be compatible with standard clinical endoscopes, we applied this probe in a longitudinal in vivo mouse model of H&N carcinogenesis. Our approach enables high-resolution, video-rate longitudinal OCT imaging of tissue microarchitecture alongside subsecond RS-based molecular profiling. To interpret the multimodal data, we developed a quantitative analysis pipeline and an explainable deep learning-based fusion framework that classifies tissue by integrating biomolecular and morphological signatures. Collectively, this technology provides a powerful, non-invasive strategy for comprehensive temporal tissue characterisation in internal organs, with strong translational potential to inform precision diagnostics and guide personalised therapeutic decision-making in oncology.

[0098] Unified RS-OCT fibre-optic endoscopy

[0099] We developed a compact, integrated endoscopic platform that unites RS and OCT within single, miniaturised fibre-optic probes (2.2 mm and 3.4 mm diameter versions; Fig. 20 a,b). RS was implemented with 785 nm laser excitation and high-throughput signal collection across the 800-1800 cm-1fingerprint window, while OCT imaging employed a custom-built ~1300 nm spectral-domain system. The developed miniaturised probe P606567PC00 utilised a sandwich-like fibre configuration: two arrays of RS excitation / collection fibres were angled toward the optical axis flanking a centrally positioned single-mode fibre for OCT (Fig. 20d). Since RS is not compatible with distal micro lenses, the OCT light focusing was achieved by splicing the single mode fibre with a no core fibres (NCFs) and a gradient index fibre (GIF) maintaining focusing of ~0.5 mm in tissue (Fig. 20e). This ensures that no parasitic Raman signals were generated from any distal optics. This design enabled OCT B-scanning across a central opening, providing a 1 mm field of view with axial and transverse resolutions of 10 pm and ~25 pm, respectively. By maintaining a consistent spatial relationship between the structural and molecular sensing components, this symmetric design allows for precise co-localisation of morphological features with their corresponding biomolecular signatures. The internal structure of the fibre probe was 3D printed to house a piezo tube enclosed by biocompatible stainless steel and sealed with a 10-degree tilted CaFz window to reduce OCT back reflections (Fig. 20f).

[0100] Processing real-time multimodal datasets acquired at different timescales remains a major challenge. We constructed a parallel data acquisition and processing framework implemented with CPU based RS analysis and GPU based OCT image processing. The OCT images were reconstructed using background removal, k-domain linearisation, Fast Fourier Transform (FFT) and offset / scaling. Raman spectra were processed using background subtraction, autofluorescence removal using a polynomial background subtraction and vector normalisation. This integrated framework allows us to measure real-time colocalised high quality OCT images (Fig. 20g) and tissue Raman spectra (Fig. 20h). Temporal coregistration of RS and OCT data was achieved by synchronising each one-second RS acquisition with a randomly selected frame from the corresponding OCT sequence, acquired at 46 frames per second. An OCT frame within the RS acquisition window was randomly selected to avoid consistently sampling the same phase of endoscope-induced motion, ensuring that the co-registered morphological and molecular data represented the same tissue state in time.

[0101] In vivo imaging in a murine model of head and neck carcinogenesis

[0102] In a murine model of head and neck carcinogenesis, we demonstrated real-time, in vivo imaging, providing a powerful view of the disease process at both molecular and microstructural levels. Oral tissue was used as test bed to represent an inherently challenging multiphenotypic tissue environment. We employed a 4-Nitroquinoline N-oxide (4NQO) oral carcinogenesis mouse model, which is well established for studying disease progression and exhibits molecular and histological changes similar to those seen in human HNSCC7. RS-OCT data were collected longitudinally at predefined anatomical locations in both the 4NQO-treated group (n= 10) and control group (n= 10) at multiple time points P606567PC00

[0103] (weeks 10, 12, 14, 16, 18, 20 and 22) (Fig. 21a). At the study endpoint (week 22), visual inspection revealed lesions at multiple sites on the tongues of treated mice. Histological evaluation of these tissues was then performed as the gold standard to determine the ground truth disease outcome (Fig. 21b). The lesions were histologically categorised as hyperplasia, various grade of dysplasia, squamoproliferative lesions (SPL) resembling carcinoma in situ, papilloma and invasive OSCC (Fig. 21b). Exposure to 4NQO resulted in the development of multifocal hyperplastic and / or mild dysplastic lesions in 100% of treated mice, with each mouse presenting with at least four lesionsOf these, 70-80% progressed to moderate-to-severe dysplasia or squamous papillary lesions (SPL), while 40-60% developed microinvasive or invasive OSCC within 22 weeks. OCT imaging enabled clear visualisation of epithelial architecture, with control tissues displaying well-defined, stratified epithelium (Fig. 21c). We observed progressive epithelial thickening in carcinogen-treated tissues, along with increasing architectural disarray, a hallmark of dysplastic and malignant progression as confirmed by endpoint histopathology (Fig. 21b). Co-registered high-fidelity Raman spectra exhibited distinct biochemical signatures, including prominent peaks at 1301 cm-1(CH2 vibrations from lipids / proteins), 1450 cm-1(CH2 / CH3 bending), and 1650 cm-1(Amide I, C=O stretching), aligning with recognised spectral markers of biological tissue (Fig. 21d)9'20. RS revealed spectral changes between pathologies, as visualised in average difference spectra ±1 standard deviation (SD), reflecting underlying biochemical remodeling across disease stages. These RS-OCT results establish the integrated system as a label-free, multimodal endoscopic platform for tracking carcinogenesis with both spatial and molecular fidelity

[0104] Longitudinal quantitative profiling of head and neck carcinogenesis

[0105] We developed an analytical framework to quantify morphological and molecular changes from multimodal data, with OCT image analysis measuring epithelial thickness and grading stratification and architectural disorganisation (Fig. 22a). Hyperplastic lesions and lesions with histologically confirmed dysplasia and to invasive OSCC exhibited progressive epithelial thickening as compared to control tissue (Fig. 22b), consistent with histopathological grading (Fig. 21b). Both dysplastic and malignant tissues showed progressive architectural disorganisation as the primary morphological indicator (Fig. 22c). This could be a result of loss of basal cell polarity together with a breakdown in the orderly maturation and differentiation of epithelial cells, a hallmark of neoplastic transformation. Crucially, our platform captured these dynamic changes noninvasively and without labels. Longitudinal analysis in a single mouse revealed continuous remodeling of epithelial architecture, with quantifiable changes in thickness and stratification over time (Fig. 22d), demonstrating the system's capability for real-time monitoring of disease trajectory. P606567PC00

[0106] For RS molecular characterisation, we applied principal component analysis (PCA) to Raman spectra from H&N tissues, capturing dominant biomolecular variance. PCA loadings highlighted vibrational signatures of such as 850, 1301 and 1650 cm’1associated with collagen, lipids and proteins (Fig. 22e). PCA score plots showed separation between control and treated tissues, reflecting disease-associated molecular shifts (Fig. 22f-g). In a single mouse, longitudinal PCA mapping revealed a continuous trajectory in spectral space, in which successive timepoints traced a directional shift along principal component axes associated with specific biochemical changes. This gradual migration reflected coordinated biochemical remodeling of the tissue microenvironment, with the spectral trajectory serving as a biomolecular fingerprint of disease progression (Fig. 22h).

[0107] To investigate how molecular alterations relate to morphological remodeling, we correlated the PCA scores from Raman spectra with epithelial thickness and stratification metrics extracted from OCT. Plotting PC2 scores against epithelial thickness revealed clustering, in which progressive thickening was accompanied by coordinated shifts in molecular composition (Fig. 22h). Similarly, PC stratification plots demonstrated that increasing architectural layering and disorganisation tracked with distinct molecular signatures along specific PC axes (Fig. 22i). These relationships suggest that the same biochemical pathways captured in the spectral PCs may underlie both hyperproliferative thickening and loss of orderly epithelial structure. The ability to link molecular trajectories with quantitative OCT imaging metrics highlights the integrated nature of morphological and biochemical remodeling during carcinogenesis.

[0108] Al-driven multi-modal RS-OCT data fusion and classification

[0109] We then developed a convolutional neural network (CNN)-based Al framework to more efficiently extract, fuse and classify multi-modal RS and OCT data (Fig. 23a), effectively integrating the complementary strengths of both modalities. In our architecture, each modality was initially processed independently through a dedicated series of convolutional and maxpooling layers, allowing the network to learn hierarchical, modality-specific feature representations. The extracted features from the RS and OCT branches were then concatenated and passed through a set of shared fully connected convolutional layers to perform feature fusion and joint representation learning. This fusion strategy enabled the model to learn synergistic patterns across compositional and structural domains. Probabilistic classification to normal group and dysplasia / cancer group showed overall aggregated accuracy of 95.8% ± 3.4% (95% CI) (93.9% sensitivity and 97.8% specificity) for RS-OCT (Fig. 23b-c). By deactivating each modality, this enabled us to benchmark RS and OCT individually. Their integration translated into a synergistic improvement in diagnostic accuracy. P606567PC00

[0110] We interrogated the decision-making process of our Al framework to ensure biological plausibility. Gradient-weighted Class Activation Mapping (Grad-CAM) was used to visualise the regions within the RS and OCT feature space that most strongly influenced individual classification outcomes (Fig. 23d-e). Exemplar Grad-CAM overlays were generated for correctly classified samples, revealing clear and modality-specific activation patterns. For RS inputs, Grad-CAM showed broad, distributed activations across the Raman spectral range, indicating that the model did not rely on a single peak or narrow feature but instead integrates information from the entire biomolecular fingerprint (Fig. 23d). For OCT inputs, Grad-CAM localised the most salient features within the epithelium across different tissue states. Interestingly, in control and hyperplastic tissue, attention was focused on epithelial structures, in particular normal shaped papillae in accordance with histology (Fig 21b) with minor emphasis along the basal layer. In contrast, dysplastic and cancerous regions showed strong Grad-CAM signals at structurally abnormal areas, particularly in the underlying stroma, consistent with sites of early pathological remodeling and tumour invasion (Fig.23e). Therefore, the network leverages tissue microarchitecture and boundary integrity, key hallmarks used by pathologists, to drive its classification, effectively mimicking human diagnostic reasoning. Together, these findings demonstrate that the network learns representations across modalities, structural cues from OCT and molecular signatures from Raman spectroscopy, with high-intensity Grad-CAM regions aligning with biologically and diagnostically meaningful features. This multi-modal RS-OCT framework underscores its potential as a powerful tool for tissue classification and diagnostic decision support. The fusion allows for contextualised interpretation, linking molecular alterations to their corresponding architectural changes in vivo.

[0111] Discussion

[0112] Current diagnostic methods for head and neck cancer primarily utilise white-light endoscopy and biopsy, which provide only intermittent snapshots rather than continuous monitoring. This approach misses the evolving molecular and microscopic changes in tumours over time, making early detection of progression from dysplasia to cancer challenging and potentially delaying crucial treatment decisions. Despite advances in widefield endoscopic imaging, in vivo longitudinal assessment of tissue at both molecular and microstructural levels remains a critical unmet need in oncology.

[0113] We presented a fully integrated dual-modality RS-OCT endoscopic platform that enables real-time, co-registered microstructural and biomolecular characterisation. This approach allowed dynamic monitoring of tissue changes during carcinogenesis. By leveraging a symmetric fibre geometry and a custom spliced-optic configuration, we achieved precise spatial and temporal co-registration of Raman and OCT signals, ensuring that biomolecular P606567PC00 and architectural information is captured from the same tissue location effectively creating a dynamic snapshot of the tissue state. This opens new avenues for real-time, in vivo tracking of tumour evolution and response to therapy, with potential to fundamentally shift how early cancer is detected and monitored.

[0114] Applying this platform in a longitudinal 4NQO-induced oral carcinogenesis mouse model, we were able to temporally monitor disease progression from hyperplasia to dysplasia and eventually neoplasia. OCT imaging captured progressive epithelial thickening and loss of tissue stratification, while RS revealed molecular signatures indicative of increased nucleic acid and protein content. The combination of RS and OCT thus enables dynamic tracking of neoplastic transformation, allowing us to not only detect cancerous changes but also to observe the transitional states that precede malignancy. By longitudinally tracking these changes, the platform facilitates a deeper understanding of tumour evolution and heterogeneity, with RS effectively identifying molecular alterations and OCT robustly assessing lesion invasion.

[0115] The deep learning fusion model we developed enabled multi-modal representation learning, leveraging the microstructural and biomolecular contrast of each modality to enhance diagnostic classification. Grad-CAM visualisations consistently emphasised structurally relevant tissue regions while excluding common imaging artifacts such as speckle noise or motion distortions. This indicates that the model is not relying on acquisition-related artifacts but instead is focusing on meaningful tissue architecture across multiple depths, aligning with established histopathological principles where subsurface changes are critical for identifying dysplasia and cancer. A similar interpretability approach was applied to our RS branch, where importance mapping across the spectral domain revealed that the classifier prioritised specific Raman shifts corresponding to biologically specific molecular peaks such as 850, 1301 and 1650 cm'1associated with collagen, lipids and proteins, respectively. These findings suggest that the model captures clinically relevant biochemical alterations associated with disease progression. Taken together, these explainability analyses reinforce that the deep learning models are leveraging biologically and clinically interpretable features, structural in OCT and biochemical in Raman spectroscopy, to make accurate predictions. This not only increases confidence in the model's reliability but also provides a bridge between Al-driven insights and established diagnostic knowledge.

[0116] While the RS signal spatially is primarily derived from the region imaged by OCT, the Raman collection area extends slightly beyond the OCT B-scan field of view due to the wider spatial sampling of the collection fibres. Nevertheless, given the ~l mm axial imaging depth of the OCT and the spatial overlap in the probe design, the majority of the Raman signal corresponds to the region visualised by OCT. Future iterations of the P606567PC00 platform may incorporate circular configuration with spiral or Lissajous scanning mechanisms to enable full en-face 3D OCT acquisition, thereby improving spatial coregistration with RS.

[0117] From a broader perspective, this work exemplifies how integrating advanced complementary optical modalities with Al can fundamentally reshape our diagnostic capabilities. Its compatibility with existing clinical endoscopes opens avenues not only for oncology, but also for inflammatory and degenerative diseases across the gastrointestinal tract where tissue morphological and molecular changes precede clinical symptoms. Looking ahead, we envision the integration of this dual-modality system with additional contrast mechanisms such as photoacoustic imaging, multi-photon imaging or fluorescence lifetime imaging to capture microvascular, functional and metabolic cues in parallel under widefield guidance21or correlation with ex vivo spatial biology22. Hence combining these advances with real-time Al-driven analysis could yield a comprehensive in vivo biopsy. Further pragmatic developments could leverage Al enhanced imaging, wherein OCT image reconstruction with < 100 ms RS clinical acquisition will be possible23,24. As optical technologies converge with computational advances, multi-modal endoscopy like RS-OCT may ultimately redefine the future of diagnostic medicine by enabling earlier, deeper, and more accurate temporal insight into disease state transitions and trajectory.

[0118] Conclusion

[0119] We presented a compact, unified endoscopic platform integrating Raman spectroscopy and OCT within a miniaturised forward-viewing fibre-optic probe, enabling real-time, coregistered molecular and structural imaging in vivo. Combining subsecond RS acquisition with video-rate OCT and distal scanning, this system delivers unprecedented spatial and biomolecular resolution within a single optical field. Applied longitudinally in a head and neck carcinogenesis mouse model, it dynamically captured epithelial remodeling alongside molecular alterations driving malignant progression. This multimodal strategy transcends the limitations of individual techniques, providing a comprehensive, quantitative assessment of disease evolution. By fusing morphology with intrinsic molecular contrast, our approach pioneers label-free, real-time diagnostics and can revolutionise early cancer detection and guide precision interventions during clinical endoscopy.

[0120] RS-OCT instrumentation and fibre-optic probe fabrication

[0121] The RS-OCT instrumentation (Fig. 20a) was based on an (2D) spectral domain OCT scanning platform and an inhouse developed Raman system. For OCT we used a 1325 nm superluminescent light emitting diode (SLED) (SLD1325, Thorlabs, Inc: 1325 nm / 10 mW / 100 nm), a short-wave infrared Cobra 1300 spectrometer (C1300-1310 / 150, Wasatch P606567PC00

[0122] Photonics, 147 kHz max line rate) and a fibre polarisation controller (FPC032, Thorlabs, Inc). The spectrometer was interfaced to a Camera Link image acquisition device (PCIe- 1433, National Instruments Corp.). A visible laser (488 nm Toptica Ibeam Smart) was coupled in the same light path of the OCT to identify the location of the OCT output beam. The acquisition of OCT frames was synchronised to the beam scanning system using an analog output device (USB-6343, National Instruments). We used a 200 pm fibre (FG200LEA, Thorlabs, Inc.) coupled 785 nm diode laser for RS (Cleanlaze 500 mW, IPS photonics). RS system comprised a high throughput spectrometer (Acton LS785, Teledyne) with a thermoelectric cooled deep depletion charged coupled device (CCD) (Pixis 400x1340, Teledyne). The spectrometer had no slit so that the linear array of input fibre array (FG105LCA, Thorlabs, Inc.) acted as a slit.

[0123] The fibre-optic probe was constructed inhouse in a dedicated facility using motorised precision (< 5 mm) assembly (Fig. 20a-f). The fibre-probe employed a layered fibre arrangement resembling a sandwich, with two RS collection / excitation fibre arrays positioned on either side of a centrally aligned single-mode fibre (SMF-28 Ultra, Corning Inc.) dedicated to OCT. These RS fibres were angled slightly inward toward the optical axis. This configuration facilitated OCT B-scanning through a central slit, achieving a 1 mm field of view and delivering axial and transverse resolutions of 10 pm and ~25 pm, respectively (Fig. 20d). The symmetrical layout preserved a fixed spatial relationship between the structural (OCT) and molecular (RS) sensing elements, enabling accurate alignment of morphological structures with their corresponding biochemical markers. Since RS is incompatible with distal quartz lenses, the OCT focusing optics were developed by splicing a NCF (FG125LA, Thorlabs, Inc.) and GIF (GIF625, Thorlabs, Inc.) fibre to the cantilever single mode fibre. We developed both a 3.4 mm and 2.2 mm diameter fibre probe.

[0124] To assemble the dual-modality probe, a piezo tube with four electrodes is first mounted to a stainless-steel tube. The piezo tube was connected to a voltage source via four thin wires and secured to the stainless tube using a piezo holder. The motion of the free end of the 2.2 mm piezo tube (Physik Instrumente) was controllable by voltage. Next, ten 200 mm multimode low-OH silica fibres (FG200LEA, Thorlabs, Inc.) for Raman excitation and collection were cleaved flat and grouped into two linear fibre arrays, each containing five fibres. These were deposited with thin film bandpass / longpass filters (Shenzhen Photonstream Ltd). The linear fibre bundles were mounted onto a holder with a wedge design, forcing the two bundles to bend inward to increase the overlapping volume of excitation and collection with OCT single mode fibre. In the third step, a CaFz was attached to the end facet of the linear fibre bundle holder. The holder was angled at the end facet to mitigate OCT back reflection. The CaFz window sealed the probe and prevents P606567PC00 contamination. Once the window was attached, an OCT fibre, consisting of a single-mode fibre and distal fibre optics, was gently inserted through the inner hole of the piezo tube using motorisation. The OCT fibre was positioned close to the window (~150 mm) due to its limited working distance. Alignment of the OCT fibre was performed under a microscope for lateral positioning and using OCT imaging for axial positioning. Once aligned, the OCT fibre was secured to the free end of the piezo tube using epoxy adhesive. Finally, the proximal ends of the fibres were interfaced with the RS-OCT system. The clinically required OCT FOV of 1 mm is approximately equivalent or slight less that the size of biopsies. For the RS fibres, the central fibres of the two linear fibre bundles were used for excitation, where the Raman laser light is coupled in. The remaining Raman fibres were grouped into a round fibre bundle and connected to the spectrometer via an inline bandpass and longpass filters (BLP01-785R-25 & LL01-785-25, Semrock).

[0125] RS-OCT real-time acquisition framework.

[0126] We developed an acquisition framework in the Python environment. The software enables real-time RS and OCT data acquisition and analysis (Supplementary Fig. S2). This comprehensive real-time software package manages the fibre scan trajectory, generates synchronisation signals, reads the OCT and Raman spectra, and preprocesses the data. The fibre scanner and OCT spectrometer are hardware-synchronised to ensure precise image construction, while the Raman spectrometer is software-synchronised with the OCT spectrometer, allowing for temporal co-registration of the two modalities. Since the Raman spectrometer operates much more slowly than the OCT spectrometer, a multithreading design is employed to run the fast OCT data acquisition in parallel with the slower Raman acquisition. Raman data preprocessing includes simple dark current noise subtraction and intensity scaling. In contrast, OCT data preprocessing is more complex, involving background removal, k-domain linearisation, fast Fourier transformation (FFT), and normalisation. These OCT preprocessing tasks are offloaded to the GPU, enabling low- latency, video-rate rendering of the OCT images.

[0127] Preprocessing and analysis of Raman data

[0128] The obtained spectra undergo wavelength calibration employing an atomic lamp (Ocean Optics HG-1). The background Raman spectrum of the system was subtracted from the tissue Raman spectra. To eliminate autofluorescence background, a custom fifth-order polynomial fit function9, constrained to the lower segment of the Raman spectrum was applied. Finally, the spectra were normalised to their integrated area to enabled relative abundance estimation and reduce absolute intensity fluctuations. P606567PC00

[0129] Principal component analysis (PCA) was then applied to reduce dimensionality and identify dominant sources of spectral variance. The number of principal components (PCs) retained for downstream analysis was determined based on the cumulative variance explained and the inspection of scree plots. PCs that collectively captured >99.29% of the total variance were initially considered. Individual components were further evaluated for their relevance to tissue pathology by inspecting loadings and their correlation with histopathological classifications. This approach ensured that retained PCs represented meaningful biochemical variation rather than noise.

[0130] OCT image processing and quantitative morphological feature extraction

[0131] To extract morphometric parameters relevant to epithelial architecture and tissue organisation, we developed a quantitative OCT image analysis pipeline that processes B- scan images to derive two key metrics: epithelial thickness and an epithelial stratification / disorganisation index. These features were selected based on their clinical relevance to dysplasia and neoplastic progression in mucosal tissues. Raw OCT B-scans were first normalised to the fibre tip reflection (that appears as a horizontal bright line at the top of the supplementary OCT video) to exclude the fluctuation of light source. To mitigate sensitivity roll-off due to the finite-size of CCD pixel, depth-dependent normalisation was applied by dividing each A-line by a calibrated sensitivity curve. Following preprocessing, image segmentation was performed to delineate the epithelial surface and basal membrane. A semi-automated approach was employed to segment the epithelial layer:

[0132] 1. The epithelial surface was identified as the first significant intensity rise along each A-line (first derivative Dijkstra's-shortest-path-based edge detection).

[0133] 2. The stroma surface was defined as the second significant intensity rise below the epithelial surface, reflecting the transition between the lamina propria and muscle. (ResNet-50 model trained by a manually labelled dataset, aided by an automated data labelling based on a first derivative edge detection)

[0134] Epithelial thickness was computed per A-line as the Euclidean distance between the epithelial surface and basal membrane along the axial direction. The final value for each OCT B-scan was derived by averaging thickness measurements across all valid A-lines (typically 256 per frame after down sampling). This provided a robust representation of overall epithelial thickening, a hallmark of dysplasia and malignancy.

[0135] To quantify epithelial architectural organisation, we defined a stratification index based on the variance in OCT intensity between the epithelium and stroma. For each A-line, the P606567PC00 tissue depth was divided into two zones of equal thickness: a 21 pm lower epithelium region and a 21 pm upper stroma region. The zone thickness of 21 pm was chosen as a balance between two constraints: the minimum epithelial thickness in mouse tongues is approximately 42 pm, which sets the upper bound, and the OCT axial resolution is about 10 pm, which sets the lower bound. Selecting a midpoint value ensures that each zone is thick enough to capture relevant tissue structure while remaining within the resolution limits of the imaging system. We calculated the mean pixel intensities within these two zones for each A-line and computed a normalised stratification index as: si = — 201og ( / e / / s), where leis the mean intensity of the lower epithelium and lsis the mean intensity of the upper stroma. In well-organised (healthy) epithelium, stratification results in consistent reflectivity gradients between the basal layers and the muscle underneath, yielding predictable SI values. In contrast, dysplastic or disorganised epithelium exhibits disrupted polarity and reflectivity patterns, leading to altered SI ratios. All morphological features (mean epithelial thickness and stratification index) were computed for each imaging session per animal and temporally tracked across timepoints. These features were subsequently aligned with molecular (RS) features and endpoint histopathology labels for statistical analysis and multimodal classification.

[0136] Temporal co-registration of OCT and RS

[0137] Analysing large volumes of multimodal data acquired at different temporal resolutions poses a significant challenge, especially in ensuring accurate alignment and integration across modalities. In our study, we combined RS, which provides one-dimensional spectral data at a rate of one spectrum per second, with OCT, which produces two-dimensional imaging at a high video frame rate of 46 frames per second. To align these temporally mismatched datasets, we implemented a straightforward temporal co-registration approach. Specifically, for each RS spectral acquisition (1 Hz), we randomly selected one of corresponding OCT frames captured within the same one-second interval. This method assumes that all OCT frames within that second are sufficiently similar for the purposes of co-registration. We chose this approach for its simplicity and practicality in exploratory data analysis. It allows efficient pairing of OCT and RS data without the need for complex synchronisation algorithms or hardware-based triggers.

[0138] Al-based multimodal fusion and classification

[0139] To enable robust and generalisable morphomolecular classification from limited in vivo data, we implemented a deep learning framework for the fusion of RS and OCT data using a custom-designed multimodal CNN. The goal was to learn a shared representation that captures the synergistic diagnostic value of both biochemical and morphological features. P606567PC00

[0140] The raw dataset consisted of n = 196 Raman spectra and n = 196 OCT B-scan frames, each co-registered and acquired from temporally tracked tissue sites in the oral mucosa of mice. All RS and OCT data were synchronised temporally and spatially via acquisition timestamps and scanner positional metadata. Since the mice study offered limited data of some groups, to mitigate overfitting and maximise generalisation, we applied imbalanced analysis extensive validation to avoid overfitting. The dataset was randomly stratified into training (72%), validation (18%), and test (10%) sets, ensuring that each set contained spectra and OCT frames. This was performed using a cross-validation strategy to provide statistical confidence estimates. Reported benchmarks (ROC, accuracy, sensitivity and specificity) were aggregated measures from the cross validation due to limited data in a 20 mice experiment.

[0141] We designed a two-branch multimodal CNN to independently process Raman and OCT data streams before joint fusion and classification. The Raman branch consisted of a ID convolutional stack (5 layers, 32-128 filters, ReLU activations, kernel size = 5), followed by a ID global average pooling layer. The OCT branch was implemented using a 2D convolutional backbone (based on a truncated ResNet-18 architecture), optimised for OCT's texture-rich features. Both modality-specific feature embeddings were concatenated and passed through a series of fully connected layers (FC1: 64 nodes, FC2: 64 nodes) before a softmax output layer using a two-class weighted classification layer (control vs hyperplasia vs dysplasia vs cancer) with categorical cross-entropy loss to combat imbalanced datasets. Optimisation was performed using the Adam optimizer (learning rate = le-4, batch size = 512), with early stopping, choosing the most optimal model based on validation loss plateauing for 100 epochs. Model performance was evaluated only on the held-out test set using standard classification metrics, including accuracy, sensitivity, specificity, and area under the receiver operating characteristic curve (AUC). To assess the contribution of each modality, we deactivated either the RS or OCT branch, allowing us to quantify the added diagnostic value of multimodal fusion. All models were implemented in Matlab (v2023) and trained using a NVIDIA RTX A6000 GPU. Training took ~few minutes per fold, and all results were aggregated over 10 cross-validation folds. The Al fusion framework is available at GitHub (www.github.com / bergholtlab / ).

[0142] Modal specific interpretability with Grad-CAM

[0143] To interpret the modality-specific contributions to the final classification, Grad-CAM was applied independently to each branch of the multimodal CNN. For the Raman branch, gradients of the output class score were computed with respect to the final ID convolutional layer. These were used to generate a weighted activation map highlighting key spectral regions influencing the model's decision. For the OCT branch, Grad-CAM was P606567PC00 applied to the last convolutional block of the truncated ResNet-18 backbone. The resulting 2D heatmaps identified spatial regions within OCT images most relevant to the classification. By isolating Grad-CAM computation to each branch before fusion, we obtained interpretable, modality-specific visualisations of learned discriminative features.

[0144] 4NQ0 murine carcinogenesis model

[0145] All animal work was approved locally at King's College London (UK) and performed under a UK Government Home Office Project License (PP0313918). The animal experiment was designed with careful consideration of the 3Rs25. C57BL / 6 mice were treated with 4- Nitroquinoline 1-oxide (4NQO) diluted to 100 pg / ml in drinking water for 16 weeks. 4NQO models are known to replicate the alterations caused by tobacco mutagens, progressing through a series of premalignant stages and exhibiting histological features that closely resemble human HNC7. Exposure to 4NQO leads to development of multi-focal dysplastic lesions in 100% of the treated mice. Of these, ~50% will develop into invasive cancer within the 22 weeks. 4NQO-containing water was prepared and changed once a week for 16 weeks. During the experiments, the mice were maintained with regular mouse chow and water (with or without 4NQO) ad libitum (n=10 control and n=10 4NQO treated group). After that period, mice were given normal drinking water until the endpoint of 22 weeks or if mice losses > 15% of maximum weight. Over multiple time points treated, and control mice were sedated with inhaled isoflurane, and the oral cavities were screened for any visible lesions and RC-OCT data measurements.

[0146] Histology

[0147] At the experimental endpoint, mice were euthanised in accordance with institutional animal care and use protocols. Whole tongues were carefully resected from the oral cavity using sterile surgical instruments. Care was taken to preserve both the tumour-bearing and surrounding non-tumorous mucosa for comprehensive histological evaluation. For frozen sections, tongue tissues were embedded in OCT (optimal cutting temperature compound), cryosectioned and stained with haematoxylin and eosin (H8<.E) by conventional methods. H8<.E slides were scanned with the NanoZoomer 2.0RS Digital Slide Scanner (Hamamatsu Photonics K.K., Japan) with 0.23 pm / pixel, 40x high resolution (Brightfield) mode.

[0148] In vivo RS-OCT imaging of mice

[0149] For in vivo imaging, both treated and control mice were placed on a heating pad and sedated with inhaled isoflurane. The RS-OCT probe was gently positioned in direct contact P606567PC00 with the mucosal surface at predefined anatomical sites within the tongue. To ensure consistent and reproducible sampling, probe placement was guided by anatomical landmarks and maintained using a micromanipulator arm. Raman spectra were acquired with an integration time of 1.0 s per location, while OCT imaging was performed continuously at a video-rate acquisition speed, enabling structural assessment in real time.

[0150] A foot pedal was utilised to control whether to save the data to the disk. The optical power delivered to the tissue was 7 mW for the 1310 nm OCT channel and 60 mW for the 785 nm Raman excitation. To achieve temporal and spatial co-registration of Raman and OCT measurements, data acquisition was synchronised through a custom software interface that triggered simultaneous frame capture and spectral collection at each site, where the timestamp for each frame / spectrum acquisition was recorded. This ensured precise pairing of morphological and molecular data from the same tissue region across all imaging sessions. Repeated measurements were performed longitudinally at identical sites across timepoints (weeks 10-22) to enable morphomolecular tracking of carcinogenic progression in vivo.

[0151] Various modifications, whether by addition, deletion or substitution may be made to the above to provide further embodiments, any and all of which are intended to be encompassed by the appended claims.

Claims

P608486PC00Claims1. A forward looking hybrid Raman spectroscopy and optical coherence tomography (OCT) sensor, comprising respective sensing optical fibers arranged to have the same forward looking field of view for both sensing modalities, the arrangement being sufficiently sized and flexible so as to be suitable for deployment endoscopically.

2. A hybrid Raman spectroscopy and optical coherence tomography (OCT) probe tip for use in the sensor of claim 1, comprising: an OCT imaging fiber coupled to a vibrational drive system arranged in use to vibrate the OCT imaging fiber in at least one lateral dimension to provide for scanning of the OCT imaging fiber across a forward-looking field of view to be imaged; and a plurality of Raman spectroscopy fibers spatially arranged about or around the OCT imaging fiber so as to have the same forward looking field of view as the OCT fiber whilst permitting space for the OCT fiber to vibrate in the at least one lateral dimension; wherein the plurality of Raman spectroscopy fibers comprise multiple Raman excitation light fibers and a larger number of Raman collection fibers, with the fibers carrying the Raman excitation light being substantially evenly spatially distributed within the plurality of Raman spectroscopy fibers around or to either side of the OCT fiber.

3. A probe tip according to claim 2, where the OCT imaging fiber is arranged to scan laterally in one dimension, and two or more rows of Raman sensing fibers are provided either side of the lateral movement space of the OCT fiber, forming a movement channel for the OCT fiber.

4. A probe tip according to claim 3, wherein every nth fiber in the row is a Raman excitation fiber, with n greater than 1, but less than x / a, where x is the number of fibers in each row, and a is 2 or more.

5. A probe tip according to claims 3 or 4, wherein there are at least two excitation fibers in each row evenly spaced within the row, such that no collection fiber is overly distanced from an excitation fiber.

6. A probe tip according to claim 2, where the OCT imaging fiber scans laterally in two dimensions, for example to permit a 2D OCT image to be made, and the RamanP608486PC00 sensing fibers are distributed in at least one arc, around the edge of the scanning movement space of the OCT fiber.

7. A probe tip according to claim 6, wherein every nth fiber around the arc of fibers is a Raman excitation fiber, with n greater than 1, but less than x / a, where x is the number of fibers in the arc, and a is 2 or more.

8. A probe tip according to claim 7, wherein there are at least two or more excitation fibers on the at least one arc, evenly spaced around the arc, such that no collection fiber is overly distanced from an excitation fiber.

9. A probe tip according to any of claims 6 to 8, wherein there are a plurality of arcs of Raman spectroscopy fibers surrounding the scanning movement space of the OCT fiber.

10. A probe tip according to any of claims 6 to 9, wherein the Raman sensing fibers are distributed in at least one circle, around the edge of the scanning movement space of the OCT fiber.

11. A probe tip according to any of claims 2 to 10, and further comprising an actuator to drive the OCT fiber to scan in one or two dimensions, the actuator preferably being a piezoelectric tube scanner.

12. A probe tip according to claim 11, wherein the a piezoelectric tube scanner has a plurality of segmented electrodes, the piezo tube scanner being arranged in use to scan the fiber tip, the scan direction being controllable by the phases and amplitudes of voltage sine wave control signals applied to the plurality of segmented electrodes.

13. A probe tip according to any of claims 2 to 12, wherein the plurality of Raman spectroscopy fibers are angled inwards and / or bevel cut at the distal end of the probe tip so as to focus any light emitted from the fibers to a sensing point just beyond the distal end of the probe tip.

14. A probe tip according to claim 13, and further comprising a transparent window disposed at the distal end of the probe tip, the sensing point of the Raman spectroscopy fibers being located outside the transparent window just beyond the distal end of the probe tip, the transparent window being preferably formed of Raman transparent materials, optionally any one of magnesium fluoride, calcium fluoride, or sapphire..P608486PC0015. A probe tip according to any of claims 2 to 14, wherein the OCT imaging fiber comprises: a length of single-mode fiber (SMF) for light delivery; a no-core fiber (NCF1) for beam expansion; a gradient index fiber (GIF) for focusing; and an angle- cleaved no-core fiber (NCF2) for suppressing reflection.

16. A probe tip according to claim 15, wherein the fiber segments are fused into a monolithic OCT fiber using a fusion splicer.

17. A forward looking hybrid Raman spectroscopy and optical coherence tomography (OCT) sensor according to any of the preceding claims, and further comprising a computer system arranged to process the Raman spectroscopy data and OCT sensor data, the computer system being provided with software arranged to control the computer system to: a. manage the OCT fiber scan trajectory; b. generate synchronization signals to control the OCT imaging and Raman spectroscopy; c. read any generated OCT image and Raman spectra data; and d. preprocesses the OCT image and Raman spectra data.

18. The system of claim 17, wherein the Raman data preprocessing includes one or more of dark current noise subtraction and intensity scaling.

19. The system of claims 17 or 18, wherein the OCT data preprocessing includes one or more of: background removal, k-domain linearization, fast Fourier transformation (FFT), and / or normalization.

20. The system of claim 19, wherein the OCT preprocessing tasks are offloaded to a GPU, enabling low-latency, video-rate rendering of the OCT images.

21. The system of any of claims 17 to 20, wherein a multithreading design is employed to run the fast OCT data acquisition in parallel with the slower Raman acquisition.

22. The system of any of claims 17 to 21, wherein the software standardizes the Raman spectra before multivariate statistical analysis by performing post-processing steps including one or more of: background removal, system response correction, truncation, and / or baseline correction.P608486PC0023. The system of any of claims 17 to 22, and further comprising a display, wherein the software controls the computer to display a real-time OCT image fused with a corresponding Raman spectrum, both derived from the same field of view, on the display.

24. The system of any of claims 17 to 23, wherein: i) the OCT fiber scanner and OCT spectrometer are hardware-synchronized to ensure precise image construction; and / or ii) the Raman spectrometer is software-synchronized with the OCT spectrometer, allowing for temporal co-registration of the two modalities.

25. The system of any of claims 17 to 24, wherein the computer system is further provided with software to implement a deep learning model to process the Raman spectra and OCT images, wherein the deep learning model: a. ingests the input data comprising Raman spectra and OCT images; b. applies multiple layers of convolutional filters to extract features including spatial patterns in the OCT images and spectral signatures in the Raman data; and c. applies a softmax layer or another probabilistic output layer to assign probabilities to different tissue types based on the extracted features.

Images