Time-resolved spectroscopy

Abstract

A method of identifying a characteristic of a sample (124) comprising receiving time resolved spectral data obtained by exposing the sample to a pulsed excitation beam, detecting photons emitted from the sample (124) and recording a time of each photon detection from a corresponding pulse of the excitation beam; determining an estimate of Raman emissions (160) and / or fluorescence (161); and determining the characteristic from the estimate.

Description

[0001]

[0002] TIME-RESOLVED SPECTROSCOPY

[0003] Field of Invention

[0004] This invention relates to time-resolved spectroscopy and, in particular, but not exclusively, time correlated single photon counting (TCSPC) spectroscopy and a method of estimating photon detections due to fluorescence and / or Raman emissions from a sample.

[0005] Background

[0006] The Raman Effect is a phenomenon in which a sample scatters excitation light of a given frequency into a frequency spectrum, which has characteristic peaks caused by interaction of the excitation light with the molecules making up the sample. Different molecular species have different characteristic Raman peaks, and so the effect can be used to analyse the molecular species present.

[0007] For some samples, the excitation radiation gives rise to fluorescent radiation together with the Raman radiation. Since the Raman signals are emitted from the sample in a shorter time frame (less than a picosecond) compared to the fluorescence signals (a few thousand picoseconds or even tens of milliseconds), a fast-time resolved detector can be used to distinguish spectra from the two processes. It is known to use time correlated single photon counting (TCSPC) to separate out the Raman and fluorescence signals. The Raman and fluorescence signals will typically overlap because of the statistical nature of the processes and the finite width of the excitation pulse.

[0008] US2012 / 0194815 Al discloses apparatus comprising a single-photon detector, which may be a single photon avalanche diode (SP D) array, that functions in a “Geiger-mode”, wherein a single photon impacting on the photodetector causes an output pulse to be generated (likened to a “click” of a Geiger counter). Detecting Raman radiation in a pulsed mode effectively filters the background noise and fluorescence out, since outside the detecting or registration period no optical nor electric pulses are taken into account.

[0009] US2015 / 0369666 Al discloses apparatus for measuring Raman radiation from an object comprising a detector having a plurality of single-photon avalanche diode (SP D) elements. A counter measures the individual timing of each detection made in the detector array with respect to an excitation pulse. The counter counts a number of detections in a predetermined Raman time window and one or more predetermined fluorescence time windows. The counter forms a number of Raman detections by eliminating an estimate of a number of detections of fluorescence photons in the Raman time window on the basis of a number of detections in the at least one fluorescence window.

[0010] Summary of Invention

[0011] According to a first aspect of the invention there is provided a method of identifying a characteristic of a sample comprising receiving time resolved spectral data obtained by exposing the sample to a pulsed excitation beam, detecting photons emitted from the sample and recording a time of each photon detection from a corresponding pulse of the excitation beam; determining an estimate of Raman emissions and / or fluorescence; and determining the characteristic from the estimate.

[0012] The estimate of the of Raman emissions and / or the fluorescence may be determined by analysing the time resolved spectral data.

[0013] The estimate of Raman emissions may be determined from a (first) subset of photon detections of the time resolved spectral data deemed to correspond to Raman emissions. The estimate of fluorescence may be determined from a (second) subset of photon detections of the time resolved spectral data deemed to correspond to the fluorescence.

[0014] The first and / or second subset of photon detections may be determined from a temporal form of the time resolved spectral data. For example, the first subset of photon detections may be a subset of photon detections of the time resolved spectral data falling within a first (temporal) interval. The second subset of photon detections may be a subset of photon detections of the time resolved spectral data falling within a second (temporal) interval. The first and / or second intervals may be determined from a temporal form of the time resolved spectral data.

[0015] The second subset of photon detections may be determined by fitting a Raman emissions curve having an expected temporal form for the Raman emissions to the temporal resolved spectral data. A lower endpoint for the second interval may be set dependent on the fitting the Raman emissions curve to the temporal resolved spectral data.

[0016] By using the temporal form of the time resolved spectral data to determine which subset of the photon detections are deemed to correspond to the Raman emissions and / or fluorescence, a portion of the time resolved spectral data used for the estimate is adapted based on a how the emissions from the sample change with time. This may allow a greater contribution of the time resolved spectral data corresponding to the Raman emissions and / or fluorescence to the estimate compared to using predetermined Raman and / or fluorescence windows (windows at fixed time periods). For example, time resolved spectral data recorded for a greater range of times may be used for the estimate. Using more of the time resolved spectral data may result in a better estimate.

[0017] The first and / or second subset of photon detections may be photon detections recorded at times in accordance with a function, the function defining the times in dependence on the temporal form of the time resolved spectral data. The function may determine the times in dependence on a temporal maximum of the time resolved spectral data.

[0018] The first subset of photon detections may be recorded at times before the temporal maximum of the time resolved spectral data. The first subset of photon detections may be defined, at least in part, by a (statistically significant) local maximum or inflection point (from concave to convex) in the time resolved spectral data before a global maximum. For example, the local maximum or inflection point may be a point on a smoothing curve (a curve to smooth out noise) fitted to the time resolved spectral data. The method of fitting a smoothing curve may be as described in W02020 / 079439, which is incorporated herein in its entirety by reference.

[0019] The method may comprise fitting, e.g. constructing or scaling, a Raman emissions curve having an expected temporal form for the Raman emissions to the time resolved spectral data, for example, to the first subset of photon detections. The estimate of Raman emissions and / or fluorescence may be based on the fitted Raman emissions curve.

[0020] The Raman emissions curve may be empirically determined, for example from a Raman profile generated from a non-fluorescent sample, such as diamond. The method may comprise fitting the Raman emissions curve to the time resolved spectral data such that a half width of the Raman emissions curve matches an initial temporal profile of the time resolved spectral data.

[0021] The Raman emissions curve may be a symmetrical curve, such as a Gaussian or Lorentzian curve. The method may comprise fitting the symmetrical curve to the time resolved spectral data such that a half width of the symmetrical curve matches an initial temporal profile of the time resolved spectral data. A temporal location of a maximum of the Raman emissions curve may be based upon, for example coincide with, a temporal location of a temporal maximum, such as a global temporal maximum, of the time resolved spectral data. A width of the Raman emissions curve, such as a full width half maximum, may be located in dependence on a location of a point before the global temporal maximum that is a preset fraction, such as between 10% and 30%, of a maximum number of detections of the global temporal maximum.

[0022] The first subset of photon detections may be a subset of photon detections up to a predetermined point, such as the local maximum or inflection point or a time at which the number of photon detections is a predetermined percentage, such as a predetermined percentage between 10% and 30%, of a maximum number of photon detections at a global maximum. The local maximum or inflection point may be determined by removing or discounting noise in the time resolved spectral data.

[0023] It will be understood that “fitting” of a curve may comprise construction of a curve or scaling of a (predetermined) curve. The term “construction” of the curve as used herein means determining a curve (for example, a mathematical function defining the curve), including the shape / form of the curve. Construction of the curve may comprise fitting a polynomial, spline, Lorentzian or Gaussian curve to the data (i.e. determining parameters). This term can be contrasted with the term “scaling” of a curve which means scaling (i.e. globally altering an amplitude / intensity of) a curve having a predetermined shape / form to fit data, such as scaling of an empirically determined Raman emissions curve to the time resolved spectral data. In both cases, the fitting may comprise finding a curve that minimises a merit function, such as goodness of fit.

[0024] Fluorescence from the sample may be estimated by subtracting a number of photon detections deemed to be from Raman emissions from the temporal resolved spectral data based on the fitted Raman emission curve and determining the characteristic from the estimate of fluorescence. The method may comprise estimating a temporal profile for photon detection due to fluorescence from the sample by subtracting a number of photon detections deemed to be due to Raman emissions from the temporal resolved spectral data at each time based on an estimate of a number of photon detections due to Raman emissions at that time from the fitted Raman emission curve. The Raman emissions curve may be fitted to the temporal resolved spectral data by scaling the Raman emission curve by a factor that minimises a ratio of the temporal resolved spectral data to the scaled Raman emission curve.

[0025] The second subset of photon detections may be recorded at times after the temporal maximum of the time resolved spectral data. The second subset of photon detections may be at times after the temporal maximum, wherein a number of photon counts is less than a predetermined amount below a maximum number of counts at the temporal maximum, for example, the predetermined amount may be between 80% and 99%, preferably 90% and 98%, and most preferably around 95% of the maximum number of counts. The temporal maximum may be a global temporal maximum.

[0026] The function may define an interval (a Raman and / or fluorescence window) of the time resolved spectral data to be used for the estimate, wherein a location of the interval along the time resolved spectral data is defined by the temporal form of the time resolved spectral data. A location of a start of the interval may be set in dependence on a temporal maximum, e.g. the global temporal maximum, of the time resolved spectral data. A start of a fluorescence interval may be set to be a time after the temporal maximum, wherein a number of photon counts is less than a first predetermined amount below a maximum number of counts at the temporal maximum, for example, the first predetermined amount may be between 80% and 99%, preferably 90% and 98%, and most preferably around 95% of the maximum number of counts. An end of the fluorescence interval may be set to be a time after the temporal maximum, wherein a number of photon counts is less than a second predetermined amount below the maximum number of counts, for example, the second predetermined amount may be between 1% and 20%, preferably 2% and 10%, and most preferably around 5% of the maximum number of counts.

[0027] The time resolved spectral data may comprise separate temporal data for each of a plurality of wavenumbers and a corresponding estimate of fluorescence and / or Raman emissions may be determined for each wavenumber of the plurality of wavenumbers. As such, the terms “temporal maximum” and “global temporal maximum” as used herein includes a maximum for a number of photon detections in time allocated to a particular wavenumber for which the corresponding estimate of fluorescence and / or Raman emissions is determined (e.g. the “temporal maximum” or “global temporal maximum” may not be such a maximum when all the wavenumbers are considered).

[0028] The method may comprise determining a Raman spectrum from Raman emissions estimated for each wavenumber. The method may comprise determining a fluorescence spectrum from photon detections of the time resolved spectral data unaccounted for by the Raman spectrum.

[0029] Determining the estimate of the fluorescence may comprise constructing a fluorescence spectrum by, for each wavenumber, determining a number of fluorescence counts from the second subset of photon detections. Determining the number of the fluorescence counts may comprise summing the counts of the second subset of photon detections. Determining the number of the fluorescence counts for each wavenumber may comprise fitting, for example scaling, the fluorescence spectrum to a total number of counts for each wavenumber. The number of fluorescence counts for each wavenumber may be the number of counts at the wavenumber represented by the fitted fluorescence spectrum. The method may comprise estimating the Raman emissions by removing the number of fluorescence counts from the time resolved spectral data for each wavenumber. The counts that remain may be deemed to be due to Raman emissions.

[0030] The method may further comprise determining a Raman spectrum by determining a number of counts deemed to correspond to Raman emissions for each wavenumber. Determining the Raman spectrum may comprise removing photon detections from the time resolved spectral data in dependency on the estimate of the fluorescence on a per wavenumber basis.

[0031] Determining the estimate of the Raman emissions may comprise constructing a Raman spectrum by, for each wavenumber, determining a number of Raman emission counts from the first subset of photon detections. The number of Raman emission counts for each wavenumber may be determined from the Raman emissions curve fitted to the first subset of photon detections. Determining the number of the Raman emission counts for each wavenumber may comprise fitting, for example scaling, the Raman spectrum to a total number of counts for each wavenumber. The method may comprise estimating the fluorescence emissions by removing the number of Raman counts from the time resolved spectral data. The counts that remain may be deemed to be due to fluorescence.

[0032] The method may further comprise determining a fluorescence spectrum by determining a number of counts deemed to correspond to fluorescence for each wavenumber. Determining the fluorescence spectrum may comprise removing photon detections from the time resolved spectral data in dependency on the estimate of the Raman emissions on a per wavenumber basis.

[0033] The method may comprise decomposing the time resolved spectral data into component spectral data defining component spectra including at least one fluorescence spectrum, and temporal intensity data defining a contribution (e.g. time dependent scaling) of each of the component spectra to a number of single photon detections at different times. The component spectra may include at least one Raman spectrum. Decomposing the time resolved spectral data may comprise carrying out an iterative resolution of the component spectral data and the temporal intensity data, wherein the component spectral data and the temporal intensity data are progressively developed in each iteration from an initial estimate for at least a first component spectra used for a first iteration. The initial estimate of the first component may an initial estimate of the Raman emissions and / or the fluorescence. The initial estimate of the Raman emissions may be determined from the first subset of photon detections deemed to correspond to Raman emissions. The initial estimate of the fluorescence may be determined from the second subset of photon detections deemed to correspond to fluorescence. Alternatively, the initial estimate for the first component may be an equal number of counts for all wavenumbers. The equal number of counts may be based on a normalisation of the time resolved spectral data.

[0034] The iterative resolution of the component spectral data and the temporal intensity data may comprise an optimisation algorithm. The optimisation algorithm may comprise an alternating optimisation algorithm comprising, alternately:

[0035] 1) fixing a current estimate of the component spectral data and finding the temporal intensity data that optimises a first merit function, and

[0036] 2) fixing a current estimate of the temporal intensity data and finding the component spectral data that optimises a second merit function.

[0037] The first and / or second merit functions may comprise minimising a least squares error.

[0038] The method may comprise constraining the temporal intensity data based on an expected time frame for the Raman emissions and / or fluorescence from the sample. For example, the temporal intensity data may be constrained such that one or more of the component spectra have a zero-intensity value at times outside of those deemed to comprise photon detections corresponding to the Raman emissions and / or the fluorescence.

[0039] The estimate of fluorescence may be at least one fluorescence spectrum. The estimate of Raman emissions may be at least one Raman spectrum.

[0040] The method may further comprise scaling the at least one Raman spectrum and the at least one fluorescence spectrum as a linear combination to best fit the time resolved spectral data in accordance with the first, second or a further merit function. The scaling may comprise a least squares fit. Constraints may be imposed on the least squares fit. For example, the constraints may comprise non-negative scalings. The method may comprise determining a concentration of a components based on the scaling of the at least one Raman spectrum and / or the at least one fluorescence spectrum in the linear combination.

[0041] The characteristic may be a material component, such as a chemical component, of the sample. The characteristic may be a concentration of a component in the sample.

[0042] The method may be computer-implemented. The method may comprise generating a computer output in response to the identified characteristic. The computer output may be a display of the characteristic on a computer display or the encoding of the characteristic in a data carrier.

[0043] The method may comprise controlling a process and / or carrying out further processing on the sample based upon the identified characteristic. For example, the process may be a manufacturing process. The sample may be a sample of one or more manufactured products and the identified characteristic may be used to determine if the manufactured products meet a required specification. Failure to meet the required specification may require an adjustment of the process such that products are manufactured to the required specification. The sample may be a tissue sample and the process may be treatment of a patient providing the tissue sample. The method may be used as part of a checking procedure, for example a security procedure or a quality control procedure), comprising generating an alarm based upon the analysis of the spectral data.

[0044] According to a second aspect of the invention there is provide a method of calibrating a detector comprising an array of pixels, each pixel comprising at least one singe photon detector, the method comprising recording a histogram for each pixel when illuminated with the light having a known intensity with time, the histogram comprising a number of detections recorded within certain time bins, and, for each pixel, determining a correction for variations in time bin width by comparing the number of detections in each time bin with an expected number of detections given the known intensity with time of the light.

[0045] It has been found that for known single photon detectors, the bin width of the resultant histogram is not uniform. To correct for this discrepancy, a bin width can be determined. By illuminating the detector with a light of known intensity with time, a number of counts recorded by the detector can be compared to the expected number of counts given for the expected bin width. Variations in the number of counts is indicative in a difference between the actual bin width and the expected bin width.

[0046] The known intensity with time may be a constant intensity with time. It may be more accurate to determine a correction from such a light source than one whose intensity varies with time.

[0047] Illumination of the detector may comprise dispersing a spectrum of the light across the array of pixels. In such an arrangement, it may be advantageous for the light to have an intensity that is constant or slowly varying across the range of wavelengths directed on the array of pixels.

[0048] According to a third aspect of the invention there is provided a method of calibrating a detector comprising an array of pixels, each pixel comprising at least one single photon detector, the method comprising recording a histogram for each pixel when illuminated with the pulsed laser, the histogram comprising a number of detections recorded within certain time bins, and, for each pixel, determining a correction for variations in timing offsets of the pixels by comparing a timing of the detected laser pulse between the pixels.

[0049] It has been found that for known single photon detectors, the timing of different pixels may be offset. To correct for this discrepancy, a correction for variations in timing offsets of the pixels can be determined. These corrections are stored as one per pixel and applied to shift all future histograms into time alignment.

[0050] Corrections determined in accordance with the second and / or third aspects of the invention may be applied to the photon detections before carrying out the method of the first aspect of the invention. According to a fourth aspect of the invention there is provided a data carrier having instructions stored thereon, which, when executed by a processor cause the processor to carry out the first, second and / or third aspect of the invention.

[0051] The data carrier may be a non-transient data carrier, such as volatile memory, e.g. RAM, non-volatile memory, e.g. ROM, flash memory and data storage devices, such as hard discs, optical discs, or a transient data carrier, such as an electronic or optical signal.

[0052] According to a fifth aspect of the invention there is provided a processor configured to carry out the method of the first, second and / or third aspect of the invention.

[0053] According to a sixth aspect of the invention there is provided spectroscopy apparatus comprising a detector having an array of single photon detector elements and a processor according to the fifth aspect of the invention.

[0054] The term “single photon detector element” is used herein to mean a photodetector element arranged to generate an output signal when a single photon is detected by the photodetector such that detection of each photon can be distinguished. For example, the single photon detector element may generate a signal pulse for each photon that is detected by the photodetector. The single photon detector may comprise a single photon avalanche diode (SPAD).

[0055] The array of single photon detector elements may be a line sensor (a onedimensional array of pixels). The line sensor may be oriented to only have a single pixel width in the spatial direction and a plurality of pixels in the spectral direction. For example, the line sensor may be as described in “A CMOS SPAD Line Sensor With Per-Pixel Histogramming TDC for Time-Resolved Multispectral Imaging”, IEEE Journal of Solid State Circuits, Volume: 54, Issue: 6, June 2019, incorporated herein in its entirety by reference.

[0056] Description of Drawings

[0057] FIGURE 1 is a schematic representation of spectroscopy apparatus according to an embodiment of the invention;

[0058] FIGURE 2 is a schematic representation of a detector used in the spectroscopy apparatus shown in Figure 1;

[0059] FIGURE 3 is a histogram exemplifying detections made by the singlephoton detector;

[0060] FIGURE 4 is a flow chart detailing a method of identifying a characteristic of a sample according to an embodiment of the invention;

[0061] FIGURE 5 is a flow chart detailing a method of identifying a characteristic of a sample according to another embodiment of the invention;

[0062] FIGURE 6 is a flow chart detailing a method of identifying a characteristic of a sample according to yet another embodiment of the invention;

[0063] FIGURE 7 is a flow chart detailing a method of identifying a characteristic of a sample according to yet another embodiment of the invention; and

[0064] FIGURE 8 is a flow chart detailing a method of calibrating a detector comprising an array of pixels, each pixel comprising a plurality of single photon detectors, according to an embodiment of the invention.

[0065] Description of Embodiments

[0066] Referring to Figure 1, the spectroscopy apparatus comprises a source 110 of excitation light, a confocal microscope 118 for delivering the laser beam to a sample 124 and spectral light, such as Raman-shifted light, to a spectral analyser 128 of a spectrometer. The spectral analyser spatially disperses the received spectral light by wavelength into a spectrum and delivers the spectrum to a detector 130. As shown in Figure 2, detector 130 comprises a one-dimensional array of pixels Pi, P2,...Pn. Each pixel Pi, P2,...Pncomprises a two-dimensional array of singlephoton detector elements 133, for example a two-dimensional array of single photon avalanche diodes (SPADs). Detections are recorded by the detector 130 on a per pixel basis, i.e. detections made by SPADs 133 of a pixel Pi, P2,...Pnare counted together.

[0067] In this embodiment the light source 110 is a pulsed laser configured to generate laser pulses along laser optical input path 113 to an optical device. The optical device comprises a mirror 114 and a Rayleigh filter 116. In this embodiment, the Rayleigh filter 116 is a notch or edge filter, which acts as a dichroic beam splitter. The mirror 114 and Rayleigh filter 116 reflect the laser beam directed along the optical input path 113 to a sample 124. The laser beam directed to the sample 124 enters into the microscope 118, wherein the laser beam is deflected by an optic 120 through an objective lens 122 and focused on to a sample 124. Raman scattering and fluorescence takes place at the sample, producing Raman-shifted light and fluorescence at different wavenumbers from the incident laser line.

[0068] The sample is mounted on a stage 123 that can be moved in two dimensions (x,y) perpendicular to the laser beam. Movement of the stage 123 allows the laser beam to be positioned at different locations on the sample 124 enabling spectroscopy data to be gathered at each of these locations.

[0069] The Raman-shifted light and fluorescence is collected by the objective lens 122 and passed back along a spectroscopy optical path 115 via the optic 120 to the Rayleigh filter 116. Whereas the Rayleigh filter 116 reflects light of the laser wavelength, it transmits the Raman-shifted wavenumbers and fluorescence. While doing so, it rejects the much more intense laser line. The Raman-shifted light and fluorescence then passes through the spectral analyser 128 to the detector 130. The spectral analyser 128 disperses the light into a spectrum across pixels of the detector 130. Typically, a spectral analyser comprises a collimating lens, a spectral dispersive optic, in this embodiment a diffraction grating, and at least one focussing optic for focussing the dispersed light onto the detector 130.

[0070] The detector 130 records and stores time resolved spectral data in the form of a histogram of the photon detections on a per pixel basis detected over a number of pulses (exposures) of the sample to the laser beam. The detector 130 allocates detections by each pixel to appropriate bins of a corresponding histogram based on the measured time of arrival of the photon. Hence, the resultant histogram for each pixel plots time of arrival versus number of detections (photon detections / counts). Each pixel corresponds to a wavenumber based on the spectral dispersion at the detector 130. An example of a histogram produced is shown in Figure 3 (only one histogram is shown but it will be understood that a histogram per pixel is produced as illustrated by the wavenumber axis). A Raman spectrum and / or fluorescence spectrum can then be determined, for example by controller 150, from the histograms based on time of arrival of the photons after an excitation pulse. In one embodiment, the controller 150 sends the spectral data to another computer, separate from the spectrometer, and the Raman spectrum and / or fluorescence spectrum is determined by this other computer.

[0071] Not all single photon detector elements 133 of each pixel Pi, P2,. . Pnmay be used because some of the single photon detector elements 133’ may produce too many dark counts to be useable. Accordingly, for each pixel Pi, P2,. . Pn, detections from only a subset of the total available single photon detector elements 133 may be used for forming the time resolved spectral data. The subset may be selected based on a figure of merit relating to a measured number of dark counts, such as a predetermined number, such as four, of the single photon detector elements 133 that have the lowest number of dark counts.

[0072] Referring to Figure 4, in one embodiment, a Raman spectrum is determined by first determining an estimate of fluorescence from the time resolved spectral data deemed to correspond to fluorescence and then subtracting a number of counts from the time resolved spectral data based on the estimate of fluorescence to determine a Raman spectrum.

[0073] Histograms are received 201 from the detector 130 and corrections 202 are applied to the histograms to correct for non-uniformity in the response of the pixels of the detector 130. In this embodiment, corrections are made for the variability in the sensitivity of the pixels of the detector 130, variations in time bins widths and relative timing delays between the pixels of the detector 130. An embodiment of how the corrections are calculated is described in more detail below with reference to Figure 6. The correction for the sensitivity of the pixels and variations in bin width may adjust a number of counts for each time bin of the histogram. The correction for variations in time delay is carried out by shifting the histogram of each pixel Pi, P2,...Pn in time in accordance with the previously determined time delay correction.

[0074] Each histogram comprises a number of counts 161 due to fluorescence and a number of counts 160 due to Raman emissions. The Raman emissions 160 are temporally offset from the fluorescence 161 but the distributions overlap. Accordingly, to determine a Raman spectrum for the sample, the counts due to fluorescence are separated from those due to Raman emissions.

[0075] In one embodiment, the estimate of fluorescence (e.g. a construction of a fluorescence spectrum) is made from a subset of photon detections of the histograms deemed to correspond to fluorescence. In each histogram, the subset of photon detections deemed to correspond to fluorescence is based on a temporal form of the histogram. Accordingly, a timing of the counts deemed to correspond to fluorescence can vary dependent on the temporal form of the histogram. In this embodiment, time bins after the time bin with the maximum number of counts are deemed to be counts of photon detections due to fluorescence. In one example, time bins with a number of counts between 95% and 5% of the maximum number of counts are deemed to be counts of photon detections due to fluorescence, however it will be understood that other cut-on and cut-off points may be chosen that are based on the temporal form of the histogram. The choice of cut-on and cutoff points may balance capturing as many fluorescence counts as possible versus including counts from other sources such as Raman emissions or background counts. These bins are summed 203 for each histogram to produce an estimate of a total number of fluorescence counts for each wavenumber / pixel Pi, P2,...Pn(in other words, a fluorescence spectrum).

[0076] A Raman spectrum is then determined 204 by summing all the time bins for each histogram to produce a spectrum of total counts (total spectral data) at each wavenumber and subtracting from the total spectral data a best fit (scaling) of the fluorescence spectrum to the total spectral data. For example, a best fit of the fluorescence spectrum to the total spectral data may be a non-negative least squares fit to the data.

[0077] The Raman spectrum can then be analysed 205 using standard techniques to find Raman components that produced the Raman spectrum. For example, Raman reference spectra corresponding to different components may be fitted (by linear combination) to the Raman spectrum to find the components present in the sample.

[0078] The Raman spectra of the identified components and the fluorescence spectrum are then scaled 206 (again by using non-negative least squares fitting) by ratios which provide the best fit to the total spectral data. The scaling of the Raman spectra of the identified components may provide a measure of the concentration of the components at the site of the excitation beam on the sample.

[0079] Optionally, an estimate of background radiation can also be subtracted from the total spectral data. For example, the estimate of background may be determined using a polynomial or spline fitting technique, such as disclosed in WO2012 / 150434 or WO2014 / 064447, which are incorporated herein in their entirety. The data may also be smoothed to account for noise, for example using the technique disclosed in W02020 / 079439, which is incorporated herein in its entirety by reference.

[0080] The Raman, fluorescence and background spectra and the remaining lack of fit to the collected histogram are then reported by computer 150 or the further computer. For example, a report may be generated on a display, stored in memory or communicated in electronic form on a data carrier. The lack of fit may be determined by reconstructing the histograms using the Raman, fluorescence and background spectra and measuring the RMS error between the actual histograms and the reconstructed histogram).

[0081] In another embodiment of the method shown in Figure 4, the subset of time bins deemed to correspond to counts due to fluorescence are time bins more than 200ps after the peak.

[0082] In a further embodiment, an attempt to improve the fit is made by adding random noise to the spectra and keeping the best spectral set so far in terms of the lack of fit described above. To do this, a scale for noise is defined. For example, the noise may be calculated using scale*random(0-l)xeach pixel of the spectral set, wherein random(O-l) is a random number between 0 and 1 and scale is a set value for this iteration of noise addition. A set number, such as 100, repetitions of adding noise are tried at this scale. While the scale is above a set threshold and below a set timeout:

[0083] If none of the fits are better than the current best fit, the scale of noise is decreased and process repeated.

[0084] If any of the fits are better than the current fit, the best fit with this scale of noise is kept and the scale of the noise is increased and process repeated.

[0085] Another embodiment of the invention is illustrated in Figure 5. This embodiment differs from that described with reference to Figure 4 in that, in step 303, a subset of time bins deemed to comprise counts corresponding to Raman emissions are identified based on a temporal form of the histogram and an estimate of a Raman spectrum is determined from the counts in the identified time bins. More particularly, an initial region of each histogram can be attributed to Raman emissions as, in this time frame, the stimulation of the sample by the pulsed laser has yet to cause fluorescence. This rise of Raman emissions before fluorescence results in an initial hump or peak in the rising number of counts, as shown in Figure 3. For Raman emissions induced by a laser pulse, the Raman emissions will typically have a symmetrical profile, for example a Gaussian or Gaussian-like profile. In this embodiment, the Raman emissions are estimated by fitting a curve, such as a half Gaussian or Lorentzian distribution, to the counts that form the initial hump / peak, wherein an extent of the curve is determined by a form of the temporal resolved spectral data (the histogram). The Raman emission is then estimated to be the number of counts represented by a full width of the (symmetrical) fitted curve.

[0086] A Raman spectrum is determined, wherein a total number of counts (intensity) for each wavenumber is based on the time curve fitted to the histogram. A fluorescence spectrum is determined 304 by, for each histogram, summing the number of counts for all time bins and subtracting from the total number of counts, a number of counts deemed to correspond to Raman emissions (as determined in step 303).

[0087] The Raman spectrum can be analysed 305 using standard techniques to find Raman components that produced the Raman spectrum. For example, Raman reference spectra corresponding to different components may be fitted (by linear combination) to the Raman spectrum to find the components present in the sample. The Raman spectra of the identified components and the fluorescence spectrum are then scaled 306 (again by using non-negative least squares fitting) by ratios which provide the best fit to the total spectral data. The scaling of the Raman spectra of the identified components may provide a measure of the concentration of the components at the site of the excitation beam on the sample.

[0088] As with the embodiment of Figure 4, an estimate of background radiation and noise can also be made based on the temporal resolved spectral data.

[0089] Another embodiment of the invention is illustrated in Figure 6. This embodiment in similar to the embodiment described with reference to Figure 5, but counts of photon detections corresponding to Raman emissions are estimated by fitting a predetermined Raman emission curve to the time resolved spectral data, the predetermined emissions curve determined from measuring Raman emissions from a sample that does not fluoresce when exposed to the pulsed laser beam. The Raman emissions curve may be predetermined by recording Raman emissions from a diamond sample when exposed to the pulsed laser beam. An estimate of a Raman spectrum is then determined by combining the Raman emission estimates for each histogram, wherein an intensity for each wavenumber is based on the counts for Raman emissions in each histogram.

[0090] The fit comprises scaling the predetermined Raman emissions curve by a scale factor, a. The value used for the scale factor, a, is found from the minimum value for a ratio of the (normalised) temporal resolved spectral data to the (normalised) predetermined Raman emissions curve for all times, t (subject to the photon detections being >0 for all times). The fit of the Raman emissions curve to each histogram gives a Raman spectrum.

[0091] The number of photon detections in each histogram due to fluorescence is determined by subtracting the scaled Raman emissions curve, Raman(t) from the temporal resolved spectral data, Total (t), so Fluorescence(t) = Total(t) - a x Raman (t).

[0092] Fluorescence counts from each histogram can be used to generate a fluorescence spectrum. The fluorescence spectrum may be determined by summing the photon detections deemed to corresponding to fluorescence at each wavenumber and / or a plurality of fluorescence spectra may be generated, each spectrum for a different time bin or a group of time bins such that a user can determine how the fluorescence spectrum evolves with time. In times dominated by Raman emissions when there is little or no fluorescence, subtracting the scaled Raman emissions curve from the temporal resolved spectral data may result in small positive or negative residuals. To avoid these residuals being considered photon detections due to fluorescence, any residual counts before a set time may be set to zero. For example, the resultant fluorescence profile, Fluorescence(t), may be further modified by setting to zero all counts (a) before the index where the Raman emission curve is first non-zero and (b) where the fluorescence profile is < 0.

[0093] As with the previous embodiment, the Raman spectrum can be analysed 605 using standard techniques to find Raman components that produce the Raman spectrum. For example, Raman reference spectra corresponding to different components may be fitted (by linear combination) to the Raman spectrum to find the components present in the sample.

[0094] The Raman spectra of the identified components and the fluorescence spectrum are then scaled 606 (again by using non-negative least squares fitting) by ratios which provide the best fit to the total spectral data. The scaling of the Raman spectra of the identified components may provide a measure of the concentration of the components at the site of the excitation beam on the sample.

[0095] As with the embodiments of Figures 4 and 5, an estimate of background radiation and noise can also be made based on the temporal resolved spectral data. The above method may be further modified to guard against finding the minimum for the ratio at too early a time where both the counts for the temporal resolved spectral data and the Raman emissions curve are close to zero. This may be achieved by adding a condition that scale factor, a, is minimum subject to the Raman profile being significant at that point. Significant may be counts above a minimum value for the normalised Raman emissions curve. The minimum value may be in a range [0.1, 0.5], A further addition is to set all fluorescence contributions below the time at which the ratio is a minimum to zero as, before this time, all counts are assumed to be due to Raman emissions.

[0096] In a further embodiment, fluorescence counts are determined from Fluorescence(t) = Total(t) - P x Raman(t), wherein the method finds the largest value for p (to some tolerance) for which Fluorescence(t) is positive for all t. In this embodiment, no clean-up for the fluorescence profile is required for negative values of fluorescence, although setting counts to zero before the index wherein the Raman emission curve is first non-zero may be carried out.

[0097] These methods aim to maximally fit the Raman emission curve into the temporal resolved spectral data with a difference between the counts at the point of maximal fitting being attributed to fluorescence.

[0098] A further embodiment of the invention is illustrated in Figure 7. In this embodiment, steps 401 and 402 are like that for the embodiment described with reference to Figures 4, 5 and 6 and reference is made to the above description and Figure 8 for these steps. This embodiment uses an alternating least squares optimisation technique to find fluorescence and Raman spectra from the time resolved spectral data.

[0099] The time resolved spectral data comprises multiple spectra, one for each time bin of the plurality of histograms, as illustrated by Figure 3. The time resolved spectral data can be unfolded into a matrix X, so that each collected spectrum occupies a row of the data matrix (Fig. 1). The matrix has dimensions / < / where / is the total number of spectra from the data set, for example to number of time bins, and J is the number of variables, which are the frequencies at which the intensities were collected, for example the number of pixels Pi...Pn. This matrix X can be decomposed into the physically meaningful submatrices C and S:

[0100] X = C • ST+ E (1)

[0101] For a data set modelled by n components, C is an I n matrix, where each column corresponds to the concentration values of a component for each time bin, and S is a . / / / matrix, where each column represents a component spectrum. E is the residual matrix with the same dimensions as X.

[0102] In the alternating least squares optimisation 404, an estimate of either S or C is used to solve Eq. 1 for the other unknown, with the solution constrained to ensure a physically meaningful result. This solution is then used to recalculate an improved value for the estimated matrix with physically meaningful constraints as before. This process is repeated until the model converges. A measure of fit is used to test for convergence, in this embodiment a percentage lack of fit (to ), which is calculated using:- lof = 100

[0103] The iteration is said to have converged, when the change in lack of fit between two consecutive iterations is smaller than a threshold value.

[0104] For the first iteration an initial estimate of either S or C is set 403, in this embodiment an initial estimate of at least one component for matrix C. The initial estimate of the first component of matrix C may be an “empty spectrum” having all intensities equal, with a value dependent on the normalisation of the spectra in X.

[0105] Alternatively, an initial estimate of components of matrix C is based on one or more portions (“windows”) of the time resolved spectral data, such as a particular time bin or set of time bins of the time resolved spectral data deemed to correspond to fluorescence or Raman emissions. The windows are based on a temporal form of the time resolved spectral data. For example, an initial guess at a component corresponding to Raman emissions may be based on time bins before the time bin with the maximum number of counts such as time bins having a number of counts within a certain percentage of the maximum number of counts, such as between 5% and 90% of the maximum number of counts. In another embodiment, the initial estimate of a Raman spectrum may be made in the manner described with reference to Figure 4, Figure 5 or Figure 6. A background signal may be estimated and subtracted from each spectrum within this time region and then the remaining counts for each wavenumber added together as an initial guess at the Raman signal.

[0106] An initial guess of a fluorescence component may be determined from time bins after the time bin with the maximum number of counts, such as time bins having a number of counts less than 95% of the maximum number of counts. An average of the spectra for these time bins may be used as an initial guess of a first fluorescence component. An initial guess may be made of a second fluorescence component using the final last time bin as a guess at the fluorescence signal, The fluorescence spectrum appears to be changing (beyond just intensity) with time and this second guess at a fluorescence component helps to correct the fit for such changes.

[0107] The lowest count at each wavenumber is used as an initial estimate of a background spectrum, i.e. a constant value in time at each wavenumber. These components are then used as initial components for the matrix C, which are then improved upon through the alternating least squares optimisation.

[0108] Constraints are applied to the components of the matrix C. For example, nonnegativity of concentrations and / or spectra. This can be achieved by obtaining a least squares solution for Eq. 1, which minimises E, and by setting all negative values in the solution to zero. Equation 1 may also be solved using a non-negative least squares algorithm, such as non-negativity-constrained linear least squares (NNLS) or fast non-negativity-constrained least squares (FNNLS).

[0109] The background component is forced to be constant in time.

[0110] The fluorescence components are forced to be zero outside of cut-off values. These cut-off values may be chosen based on the time window used to make the initial estimate of a fluorescence component (e.g. 5% and 90% of the maximum less one time bin).

[0111] In an alternative approach to produce an initial set of spectral components for the matrix C, the laser pulse width is used to define the time bins deemed to correspond to Raman emissions (for example, as a Gaussian starting at the 5% signal point). A constant background may be used. Time bins deemed to correspond to fluorescence are time bins that are beyond the end of the laser pulse width, such as more than twice the width of the Gaussian pulse width from the 5% signal point).

[0112] A combination of the empty model approach and the initial estimates based on “windowing” of the time resolved spectral data may be used. For example, empty model components may be used in addition to the initial estimates of components made from the time resolved spectral data. The empty model components may be refined into further fluorescence spectra by the alternating least squares optimisation. Once the Raman spectrum / spectra has been determined it may be used to determine properties of the sample, such as elements / substances in the sample and / or the concentrations of elements / substances in a sample. This may involve the fitting or comparison of reference spectra for those elements / substances to the determined Raman spectrum / spectra. Furthermore, the fluorescence spectrum / spectra may be used to determine properties of the sample, for example through Fluorescence Lifetime Imaging microscopy (FLIM).

[0113] The computer 150 or other computer carrying out the method may output the determined properties of the sample, for example on a display or as data of a data carrier. The computer 150 or the other computer may generate a command for another machine or an alert based on the determined properties.

[0114] An embodiment for determining corrections to the time resolved spectral data for variability in the sensitivity of the pixels Pi, P2. . Pn, variations in time bins widths and relative timing delays between the pixels Pi, P2. . Pn of the detector 130 will be described with reference to Figure 8. A number of dark counts generated by each SPAD 133 will vary and, for a typical detector 130, some of the SPADs 133’ will produce so many dark counts that the Raman counts are swamped out. Accordingly, as described above, a subset of the SPADs of each pixel Pi. . Pnis selected 501 for generating the time resolved spectral data.

[0115] It has been found the resulting sensitivity of the pixels Pi...Pncan vary pixel to pixel, as well as a width of time bins of each histogram and a timing of each pixel. Accordingly, a pixel sensitivity correction and time bin width correction value and a pixel timing offset value are determined for the detector 130 and these corrections are applied to the histograms recorded by the detector 130. Correction values are determined for each mode the detector is to be used in. The modes may comprise:-

[0116] 1) Single photon counting (SPC) mode - photons detected by each pixel are summed for a period of time, for example in firmware, then read-out to a computer;

[0117] 2) Time correlated SPC (TCSPC) mode - Each detected photon is time tagged and the time tags are passed to the firmware / computer every laser shot;

[0118] 3) On-chip Histogram accumulation (HIST) mode- A histogram of pixel number versus arrival time is built up on the chip for many laser shots before being transferred to the firmware / computer.

[0119] The detector 130 is illuminated 502 with uniform source of light. Firstly, the dark counts are measured to decide which SP DS can be used in each pixel. The uniform light is light of constant intensity in time and, as far as possible, constant (or at least slowly varying) with wavelength. The whole detector is illuminated as uniformly as possible. A histogram is recorded by each pixel Pi, P2...P11 for the illumination using the selected SPADs.

[0120] A pixel sensitivity and time bin width correction value is determined for each pixel Pi, P2. . Pn from a variation in total counts recorded by the pixel Pi, P2. . Pnfrom a mean number of the total counts recorded per pixel Pi, P2...P11. This correction value is normalised (adjusting for a notionally common scale) and saved as a correction to be applied to all future histograms.

[0121] The detector 130 is illuminated by a laser pulse and a histogram of the laser pulse is collected for each pixel Pi, P2. . Pn. The laser pulse illumination is arranged such that the whole SPAD array is uniformly illuminated directly by the laser beam. The timing of the pulse in each histogram is compared and a pixel timing offset correction value is determined for each pixel Pi, P2...Pn to correct for offsets in timing between the pixels Pi, P2. . Pn. The pixel timing offset correction value is stored as one per pixel and applied to shift all future histograms for that pixel into time alignment.

[0122] It will be understood that alterations and modifications may be made to the abovedescribed embodiment without departing from the invention as defined herein. For example, the correction values may be determined using another method. Furthermore, for certain collection methods wherein spectral light for a given wavenumber is collected across multiple pixels Pi, P2.. Pn of the detector 130, such as “extended scanning” as described in GB2404528.8, which is incorporated herein in its entirety by reference, correction of the histograms for pixel sensitivity and time bin width may not be carried out as collection method averages variations between the pixels across the recorded histograms.

[0123] A subset of photon detections corresponding to Raman emissions may be determined by measuring the profile of the laser pulse on the detector 130 (either at the same time as collecting the Raman / fluorescence signal but in a different pixel or separately before or after collecting the Raman / fluorescence signal). The expected profile of the Raman emissions can be inferred from the profile of the laser pulse and this profile fitted to the time resolved spectral data, as described with reference to Figure 5. Fluorescence is then determined by subtracting the estimated number of counts for Raman emission from the total number of counts.

Claims

CLAIMS1. A method of identifying a characteristic of a sample comprising receiving time resolved spectral data obtained by exposing the sample to a pulsed excitation beam, detecting photons emitted from the sample and recording a time of each photon detection from a corresponding pulse of the excitation beam; determining an estimate of Raman emissions and / or fluorescence by fitting a Raman emissions curve having an expected temporal form for the Raman emissions to the temporal resolved spectral data; and determining the characteristic from the estimate.

2. A method according to claim 1, comprising estimating fluorescence from the sample by subtracting a number of photon detections deemed to be from Raman emissions from the temporal resolved spectral data based on the fitted Raman emission curve and determining the characteristic from the estimate of fluorescence.

3. A method according to claim 2 comprising estimating a temporal profile for photon detection due to fluorescence from the sample by subtracting a number of photon detections deemed to be from Raman emissions from the temporal resolved spectral data at each time based on an estimate of a number of photon detections due to Raman emissions at that time from the fitted Raman emission curve.

4. A method according to any one of the preceding claims, wherein the Raman emissions curve is fitted to the temporal resolved spectral data by scaling the Raman emission curve by a factor that minimises a ratio of the temporal resolved spectral data to the scaled Raman emission curve.

5. A method according to any one of the preceding claims, wherein the Raman emission curve is determined empirically.

6. A method according to any one of the preceding claims, wherein the time resolved spectral data comprises separate temporal data for each of a plurality ofwavenumbers and a corresponding estimate of the fluorescence and / or the Raman emissions is determined for each wavenumber of the plurality of wavenumbers.

7. A method according to claim 6, comprising determining a Raman spectrum from Raman emission estimated for each wavenumber.

8. A method according to claim 7 and determining a fluorescence spectrum from photon detections of the time resolved spectral data unaccounted for by the Raman spectrum.

9. A method according to any one of the preceding claims, comprising decomposing the time resolved spectral data into component spectral data comprising component spectra including at least one fluorescence spectrum, and temporal intensity data defining a contribution of each of the component spectra to a number of single photon detections at different times.

10. A method according to claim 9, wherein the component spectra include at least one Raman spectrum.

11. A method according to claim 9 or claim 10, wherein decomposing the time resolved spectral data comprises carrying out an iterative resolution of the component spectral data and the temporal intensity data, wherein the component spectral data and the temporal intensity data are progressively developed in each iteration from an initial estimate for at least a first component spectra of the component spectral data used for a first iteration.

12. A method according to claim 11, wherein the initial estimate for the at least first component is an equal number of counts for all wavenumbers.

13. A method according to claim 12, wherein the initial estimate for the at least first component is based on the second subset of photon detections deemed to correspond to fluorescence.

14. A method according to any one of claims 11 to 13, wherein the iterativeresolution of the component spectral data and the temporal intensity data comprises an optimisation algorithm.

15. A method according to claim 14, wherein the optimisation algorithm comprises an alternating optimisation algorithm comprising, alternately: i) fixing a current estimate of the component spectral data and finding the temporal intensity data that optimises a first merit function, and ii) fixing a current estimate of the temporal intensity data and finding the component spectral data that optimises a second merit function.

16. A method according to claim 15, wherein the first and / or second merit functions comprises minimising a least squares error.

17. A method of identifying a characteristic of a sample comprising receiving time resolved spectral data obtained by exposing the sample to a pulsed excitation beam, detecting photons emitted from the sample and recording a time of each photon detection from a corresponding pulse of the excitation beam; determining an estimate of fluorescence from a subset of photon detections of the time resolved spectral data deemed to correspond to the fluorescence, the subset of photon detections determined from a temporal form of the time resolved spectral data; and determining the characteristic from the estimate.

18. A method according to claim 17, wherein the photon detections of the subset fall within a temporal interval, the temporal interval determined from a temporal form of the time resolved spectral data.

19. A method according to claim 18, wherein the subset of photon detections is determined by fitting a Raman emissions curve having an expected temporal form for the Raman emissions to the temporal resolved spectral data and setting a lower endpoint for the temporal interval dependent on the fitting of the Raman emissions curve to the temporal resolved spectral data.

Images