Systems and methods for determining the concentration of an analyte - Patents.com
Hybrid III-V/IV semiconductor-on-chip photonic systems provide compact, efficient, and scalable solutions for real-time analyte concentration monitoring, addressing the impracticality of large spectroscopic devices and enabling accurate wearable healthcare applications.
Patent Information
- Application Number
- JP2022534193
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Priority Date
- 2019-12-06
- Filing Date
- 2020-12-03
- Publication Date
- 2024-11-07
- Estimated Expiration
- 2040-12-03
AI Technical Summary
Existing spectroscopic non-invasive analyte measurement technologies, such as those using near-infrared spectroscopy, are cumbersome and impractical for field or home use due to their large and complex structures, limiting their application in consumer electronics and wearable healthcare devices.
Hybrid III-V/IV semiconductor-on-chip photonic systems (SoC) that integrate light detection and generation functions, enabling compact, efficient, and scalable analyte concentration monitoring by combining III-V semiconductor optical gain and sensing with Group IV semiconductor photonic integrated circuits for optical routing and filtering.
Enables real-time, calibrated concentration level monitoring of biological analytes like glucose and lactate, suitable for wearable devices, with high accuracy and scalability, overcoming the size and complexity limitations of traditional systems.
Smart Images

Figure 0007579341000001 
Figure 0007579341000002 
Figure 0007579341000003
Abstract
Description
[Background technology]
[0001] CROSS-REFERENCE TO RELATED APPLICATIONS This application claims priority to and the benefit of U.S. Provisional Patent Application No. 62 / 944,644, entitled "SYSTEM AND METHOD FOR MEASURING THE CONCENTRATION OF AN ANALYTE," filed December 6, 2019, the entire contents of which are incorporated herein by reference.
[0002] FIELD OF THEINVENTION The present invention relates to a method for acquiring data from a target biological material by optical communication between the target material and a III-V / IV semiconductor photonic sensor, and a method for processing the data to obtain absolute concentration levels of target molecules within the material. The present invention is applicable, but not limited to, for transcutaneous detection and monitoring of glucose, urea, lactate, creatinine, ethanol, and other constituent molecules in blood by spectroscopic detection of wavelength absorption at variable wavelengths. The technology described herein is compatible with consumer electronics technology platforms in terms of manufacturing techniques and size, weight, power, and cost requirements, providing crucial advantages in terms of usability for technology of wearable healthcare devices. The technology may be utilized by people suffering from chronic diseases such as diabetes, for which no non-invasive detection solution currently exists. Furthermore, a novel approach is provided for non-invasive and continuous monitoring of vital physiological markers for which only point-of-care solutions currently exist.
[0003] background Many techniques for the spectroscopic noninvasive measurement of analytes, such as the measurement of blood glucose using near-infrared spectroscopy, employ broadband light sources, such as halogen lamps. The electromagnetic radiation (EMR) emitted by such sources, as well as the electromagnetic radiation received from (e.g., diffusely reflected by or propagated through) the medium being analyzed, have many wavelength components. The components of the EMR received from the medium are typically separated using diffraction techniques to obtain a spectrum. Spectrometers with broadband light sources and diffraction mechanisms are typically large and complex, making them cumbersome or impractical for use in the field or at home.
[0004] Photonic systems on chip (P-SoC) offer the possibility of the ultimate size reduction required for mass-production applications such as the consumer electronics market, automotive, home medical devices, and others. The P-SoC concept allows combining all or most of the functions of a typical photonic system and realizing them in a single chip assembly, which can typically be realized as a monolithic photonic integrated circuit (PIC) based on III-V semiconductors or a combination of III-V and IV semiconductors. The first approach is to allow all active and passive optical components to be realized inside the same wafer, resulting in a completely monolithic device. This is ideal since all sources and detectors are inherently aligned with the waveguides and do not require any assembly steps. However, the inherent properties of III-V materials such as high absorption and low optical confinement in the waveguides, as well as the large bending radius of the waveguides to reduce bending losses, combined with complex technologies requiring multiple epitaxial growths, limit the possibility of scaling to very large markets such as consumer electronics, which require a very low cost per chip. As a compromise solution, III-V / IV hybrid P-SoCs provide a solution where the light emission function is realized inside the III-V semiconductor chip, and the light routing, filtering and other functions are realized inside the group IV semiconductor chip. The EMR-based light detection can be realized inside either the group III-V or group IV semiconductor chip. The hybrid approach proves to be beneficial for the mass market, since group IV semiconductor fabrication technologies, e.g. CMOS, offer unparalleled scalability. However, techniques for analyte measurement using P-SoCs are generally unknown.
[0005] overview Hybrid integration of III-V semiconductor chips with group IV semiconductor photonic integrated circuits may provide a best-of-world combination, where light detection and light generation functions are realized in the direct bandgap of III-V semiconductors for ultimate efficiency, performance, cost and yield, while passive functions such as light filtering, routing, locking and feedback control are realized inside the photonic integrated circuit (PIC) inside group IV semiconductors, e.g., silicon-on-insulator, or silicon-on-silicon-nitride, or silicon on silicon nitride or silicon on insulator. In various embodiments, on-chip photonic systems based on wavelength-swept lasers with integrated emission wavelength tuning (sweeping), wavelength shift tracking and absolute wavelength calibration functions are used to remotely acquire relevant data from biological objects such as living organisms. In several different embodiments, the acquired data is then processed to provide absolute values specific to the biological molecules, such as concentration levels and / or concentration levels as a function of time (trends). The combination of hybrid III-V / IV semiconductor platforms and on-chip processing techniques for acquired data opens new opportunities for wearable device platforms, e.g. smart watches for real-time monitoring of important physiological parameters.
[0006] Techniques are described for acquiring and processing data and providing calibrated concentration levels of analytes (e.g., constituent molecules within biological material) in real time in combination with an on-chip photonic sensor system. The biological material may be blood, interstitial fluid, tissue, or a combination of materials. The on-chip photonic sensor system (SoC) assembly includes a hybrid assembly of III-V and Group IV semiconductors, where the III-V semiconductor elements provide optical gain and detection functions, and optical feedback, optical routing, filtering, locking, and other passive functions are provided within the Group IV semiconductor photonic integrated circuit.
[0007] In use, the assembly is in optical communication with the biological material and the sensor may be remote from the material (in vivo) or embedded within the material (implant). The sensor interacts with the target material via optical communication, i.e. light from the sensor interacts with the material and the optical signal is modulated based on the light-molecule interaction, which is molecule specific. After interaction, the signal is collected by the sensor tip by diffuse reflection or propagation.
[0008] In practical scenarios where such photonic sensors perform direct transcutaneous measurements on or are implanted in living organisms, the raw signals collected by the sensors are highly complex due to the complex nature of typical biological materials such as whole blood and / or tissue. The various data analysis techniques described herein, in combination with hardware (e.g., SoCs), can be used to obtain calibrated concentration level values from most of the complex biological materials. This is particularly important for transcutaneous / implanted monitoring of important metabolites such as glucose, lactate, urea, ethanol, serum albumin, creatinine, and others, both for patients suffering from chronic diseases such as diabetes, renal or hepatic failure, and acute clinical events such as sepsis, as well as for monitoring fitness levels and diets of athletes and the general public.
[0009] Thus, in one embodiment, a method for calibrating a sensor for measuring the concentration of an analyte in a medium is provided. The method includes collecting a number of raw spectra from an object (e.g., a medium or a sample) having an analyte using a III-V / IV hybrid semiconductor on-chip photonic system (SoC). The method also includes dividing the raw spectra into a set of clusters according to their respective spectral shapes, where each cluster includes a group of raw spectra. The method further includes, within each cluster: (i) applying a localized scattering correction (LSC) to each of the raw spectra belonging to the cluster to obtain a group of locally corrected spectra; and (ii) deriving a cluster-specific set of optimized pre-processing parameters and a cluster-specific calibration vector. The optimized set of pre-processing parameters and the calibration vector are derived using absolute standard analyte concentration values corresponding to the locally corrected spectra and the group of raw spectra belonging to the cluster.
[0010] In some embodiments, deriving a cluster-specific optimized set of pre-processing parameters and a cluster-specific calibration vector for a particular cluster includes: (i) evaluating each of a number of candidate sets of pre-processing parameters, respectively, where evaluating a particular candidate set includes: (A) pre-processing each of the locally corrected spectra belonging to the particular cluster using the particular candidate set; (B) deriving a candidate calibration vector by applying a multivariate regression calibration to the pre-processed locally corrected spectra using absolute standard analyte concentration values corresponding to the group of raw spectra belonging to the particular cluster; and (C) calculating a corresponding measurement accuracy for the candidate calibration vector via cross-validation. The candidate set and corresponding candidate calibration vector associated with the greatest measurement accuracy are then designated as the cluster-specific optimized set of pre-processing parameters and the cluster-specific calibration vector, respectively.
[0011] The set of cluster-specific optimized pre-processing parameters may include a set of data processing parameters such as a) the order of filtering, b) the type or nature of the filter used for smoothing, c) the order of derivative used for baseline removal, etc. The set of optimized parameters may be stored in memory and may be used later to pre-process data in detection mode.
[0012] The subject may include tissue, and the analytes may include blood glucose, blood lactate, ethanol, creatinine, keratin, collagen, urea, serum albumin, globulin, troponin, acetone, acetate, hydroxybutyrate, cholesterol, albumin, globulin, ketone-acetone, or water, among others.
[0013] In some embodiments, the step of dividing the raw spectra according to their respective spectral shapes includes applying a global scatter correction (GSC) to each of the raw spectra to obtain several globally corrected spectra. The dividing step may also include clustering the several globally corrected spectra according to: (A) a specific number of clusters, (B) a specific maximum distance of the globally corrected spectra from the centroids of the clusters, or (C) both a specific number of clusters and a specific maximum distance of the globally corrected spectra from the centroids of the clusters. The dividing step may further include, within each cluster, assigning to that cluster each raw spectrum corresponding to the globally corrected spectrum belonging to that cluster. The clustering may include k-means clustering, affinity propagation, or agglomerative clustering.
[0014] In some embodiments, the method further includes storing in the SoC the GSC reference spectrum generated as part of the global scatter correction. The global scatter correction may be implemented as a global multiplicative scatter correction, a global standard normal variate (SNV) correction, a global mean centering and normalization correction, a Kubelka-Munk (KM) correction, a Sanderson correction, or a combination thereof. The local and / or global scatter correction may incorporate a correction for particle size differences and / or a correction for path length differences, and may use a KM correction, a Sanderson correction, a multiplicative scatter correction, or a combination thereof. In some embodiments, the method may include storing in the SoC, for each of the clusters: (i) a corresponding LSC reference spectrum, and / or (ii) a corresponding calibration vector, (iii) a cluster centroid, and / or (iv) a set of pre-processing parameters optimized for each cluster. Local scatter correction may also be implemented as a local multiplicative scatter correction, or a local standard normal variate (SNV) correction, a local mean centering and normalization correction, a KM correction, a Sanderson correction, or a combination of the above correction techniques to achieve a linearization effect. If appropriately chosen, the global and local scatter corrections allow for taking into account the effect of differences in particle size on light scattering as well as for taking into account corrections for differences in optical path length in tissues, for example linearizing the raw spectra so that both the linear Beer-Lambert absorption law and linear regression techniques, including multivariate partial least squares, can be applied.
[0015] In some embodiments, determining the spectral shape of each of the several raw spectra includes pre-processing the raw spectra by applying a linear transformation and baseline correction based on a reference spectrum of the selected analyte to the raw spectra, which may include a Kubelka-Munk correction, a Sanderson correction, a multiplicative scatter correction, or a combination of any two or all three of these.
[0016] In another embodiment, a method for measuring a concentration of an analyte is provided, the method including acquiring a raw spectrum from an object (e.g., a medium or sample) having an analyte using a hybrid III-V / IV semiconductor photonic system on chip (SoC) and identifying a cluster to which the raw spectrum belongs from a number of clusters of spectra, where the cluster is identified based on a spectral shape of the raw spectrum. The method also includes applying a local scatter correction (LSC) to the raw spectrum to obtain a locally corrected spectrum, pre-processing the locally corrected spectrum with a set of cluster-specific optimized pre-processing parameters, and multiplying the pre-processed locally corrected spectrum with a cluster-specific calibration vector to obtain a corresponding calibrated concentration value for the analyte.
[0017] In some embodiments, obtaining the raw spectrum includes directing tunable electromagnetic radiation (EMR) at several different wavelengths from the SoC to the object, measuring the intensity of the EMR received from the object at each of the different wavelengths using the SoC, and converting the intensity to an absorbance value such that the raw spectrum comprises an absorbance spectrum. The several different wavelengths may be selected from a range of 1000-3500 nm or a range of 1900-2500 nm.
[0018] In some embodiments, the clusters of spectra correspond to spectra previously collected using the SoC, and each of the clusters may be represented via a respective LSC criterion, a respective cluster centroid, and / or a respective calibration vector, where the respective LSC criterion, the respective cluster centroid, and the respective calibration vector for each cluster may be stored in the SoC. Identifying the cluster to which the raw spectrum belongs from the several clusters of spectra may include deriving a globally corrected spectrum using a global scatter correction (GSC) criterion. Identifying the cluster to which the raw spectrum belongs may also include, within each of the several clusters, comparing the globally corrected spectrum to the respective LSC criterion to obtain a distance corresponding to the cluster, and selecting the cluster with the smallest corresponding distance.
[0019] The global scatter correction may be performed as a global multiplicative scatter correction, a global standard normal variate (SNV) correction, a global mean centering and normalization correction, a KM correction, a Sanderson correction, or a combination thereof. The local and / or global scatter correction may incorporate a correction for particle size differences and / or a correction for path length differences. The local scatter correction may be performed as a local multiplicative scatter correction, or a local standard normal variate (SNV) correction, or a local mean centering and normalization correction, a KM correction, a Sanderson correction, or a combination thereof. LSC and GSC may include performing a linearization transformation on the raw spectra to take into account scattering and absorption in the tissue / object and to facilitate further data processing based on linear extinction techniques such as Beer-Lambert-based extinction, where the spectra are decomposed into individual components and / or further processed using PLS linear regression or other similar techniques.
[0020] In some embodiments, determining the spectral shape of the raw spectrum includes pre-processing the raw spectrum by applying a linear transformation and baseline correction based on a reference spectrum of the selected analyte to the raw spectrum, which may include a Kubelka-Munk correction, a Sanderson correction, a multiplicative scatter correction, or a combination of any two or all three of these.
[0021] In another embodiment, a system for measuring a concentration of an analyte includes a hybrid III-V / IV semiconductor on-chip photonic system (SoC) for acquiring a raw spectrum from an object (e.g., a medium or a sample) containing an analyte, and a processing unit, the processing unit including a processor and a memory, configured to perform predetermined operations such as measuring the concentration of the analyte and storing information. Specifically, the processing unit is configured to acquire a raw spectrum from the object having the analyte using the hybrid III-V / IV semiconductor on-chip photonic system (SoC), and to identify a cluster to which the raw spectrum belongs from a number of clusters of spectra based on a spectral shape of the raw spectrum. The processing unit is also configured to apply a cluster-specific local scattering correction (LSC) to the raw spectrum to obtain a locally corrected spectrum. The processing unit is further configured to pre-process the locally corrected spectrum using a set of cluster-specific optimized pre-processing parameters, and to multiply the pre-processed locally corrected spectrum by a cluster-specific calibration vector to obtain a calibrated concentration value of the analyte.
[0022] In some embodiments, to obtain the raw spectrum, the SoC is configured to direct tunable electromagnetic radiation (EMR) at several wavelengths to the object and measure the intensity of the EMR received from the object at each wavelength. The processing unit is programmed to convert the intensity to an absorbance value, and thus the raw spectrum comprises or is represented as an absorbance spectrum. The SoC may be configured to emit EMR at wavelengths in the range of 1900-2500 nm or in the range of 1000-3500 nm.
[0023] Some clusters of spectra may correspond to spectra previously collected using the SoC. Each of the clusters may be represented via a respective LSC reference and a respective calibration vector. The SoC may include a memory for storing, for each of the clusters, the respective LSC reference and the respective calibration vector, as well as a global scatter correction reference (also referred to as a global scatter correction vector). The memory of the SoC may also store, for each cluster, an optimized set of corresponding pre-processing parameters.
[0024] In some embodiments, to identify the cluster to which the raw spectrum belongs among several clusters of spectra, the processor is programmed to derive a globally corrected spectrum using a global scatter correction (GSC) criterion stored in the memory. The processor may also be programmed to: (i) compare the globally corrected spectrum with each LSC criterion to obtain a distance corresponding to the cluster, and (ii) select the cluster with the smallest corresponding distance. The global scatter correction may include a global multiplicative scatter correction, or a global standard normal variate (SNV) correction, or a global mean centering and normalization correction. Similarly, the local scatter correction may include a local multiplicative scatter correction, or a local standard normal variate (SNV) correction, or a local mean centering and normalization correction. The local and / or global scatter correction may incorporate a linearization transformation for correction of particle size differences and / or correction of path length differences.
[0025] In some embodiments, the SoC includes a wavelength shift tracker for tracking a shift in the wavelength of the radiation emitted by the SoC, and / or a wavelength tracker for tracking the absolute wavelength of the radiation emitted by the SoC, and / or a temperature sensor for measuring the temperature of the chip, and / or an SoC output power monitor for monitoring or measuring the intensity of the EMR emitted by the SoC during a wavelength sweep to obtain a power curve.
[0026] In some embodiments, to determine the spectral shape of each of the several raw spectra, the processor is configured to pre-process the raw spectra by applying a linear transformation and baseline correction based on a reference spectrum of the selected analyte to the raw spectra. To perform this pre-processing, the processor may be configured to apply a Kubelka-Munk correction, a Sanderson correction, a multiplicative scatter correction, or a combination of any two or all three of these correction techniques. [Brief description of the drawings]
[0027] BRIEF DESCRIPTION OF THE DRAWINGS FIG. 1 is a schematic block diagram of a photonic SoC used to perform remote sensing of an object in accordance with an embodiment of the present invention; FIG. 2 is a simplified schematic diagram of a photonic sensor system used to perform detection experiments according to an embodiment of the present invention; FIG. 3 is a simplified schematic block diagram of an algorithm used in combination with the hardware shown in FIGS. 1 and 2 to generate a calibration algorithm for a sensor in accordance with an embodiment of the present invention; FIG. 4a is a graph of a large set of pooled raw absorbance spectra from a piglet with the global MSC vector bolded, and FIG. 4b is a graph of the spectra after baseline correction (MSC) and before the clustering procedure according to an embodiment of the present invention; Figure 5 is a graph of the baseline-corrected spectra from Figure 4b, clustered into six different clusters (by definition N = 6) using the k-means algorithm, and the maximum distance to the centroid within each cluster is shown in the graph; FIG. 6 is a block diagram outlining an algorithm for constructing an individual calibration model according to an embodiment of the present invention; Figure 7a is a graph of raw spectra collected from a subject - a piglet - in one cluster before applying MSC baseline correction, the bolded black spectrum is the calculated local MSC reference, and Figure 7b is a graph of the spectra in the same cluster after baseline correction (MSC); FIG. 8 is a graph of individual component concentration calibration vectors obtained for glucose molecules using a transcutaneous scattering detection geometry for a piglet; FIG. 9 is a schematic block diagram of a detection algorithm used with a hybrid III-V / IV semiconductor photonic sensor SoC according to an embodiment of the present invention; FIG. 10 is a graph of the transcutaneous blood glucose detection performance of the on-chip photonic sensor system using the data processing method of an embodiment of the present invention on a sedated piglet; FIG. 11 is a graph of the transcutaneous blood lactate detection performance of the on-chip photonic sensor system using the data processing method of an embodiment of the present invention on a sedated piglet; and FIG. 12 is a graph of the transcutaneous blood ethanol detection performance of the on-chip photonic sensor system using the data processing method of an embodiment of the present invention on a sedated piglet. FIG. 13 is a graph showing the effect of Kubelka-Munk pretreatment on transdermal tissue spectra in the analysis of ethanol. FIG. 14a shows the observed signal decomposed using the Beer-Lambert model without preprocessing the observed signal using the Kubelka-Munk correction. FIG. 14b shows the same transdermal observed transdermal signal decomposed using the same Beer-Lambert model after preprocessing the observed signal using Kubelka-Munk correction.
[0028] Detailed Description Optical remote sensing is a well-developed technology with a wide range of applications. Measurements can be performed as a form of ranging, i.e. by measuring distances by time-of-flight or frequency modulated continuous wave (FMCW) techniques, or by spectroscopy to remotely detect, identify and quantify the presence or absence of one or more molecules in an object.
[0029] In this specification, the term spectroscopy refers to the installation of a hybrid III-V / IV semiconductor photonic system on chip (P-SoC) that generates tunable laser radiation and communicates with a remote target object. The wavelength change and absolute value are monitored and taken into account with every sweep, thus allowing the SoC to be auto-calibrated for absolute wavelength and wavelength shift, as well as the output spectrum.
[0030] Light impinges on the object and penetrates to a depth defined by the optical path length, which depends on the object's individual specifics, such as scattering substrate, contents, etc. For example, when using tunable laser radiation in the 1900-2500 nm spectral range to perform transdermal detection experiments on biological objects, the light penetrates approximately 1 mm below the skin surface, where it is scattered and partially absorbed by tissue, blood, and interstitial fluid. Such absorption is molecule-specific, with each constituent molecule modifying the spectrum of light with a unique spectral absorption signature. After interacting with the object, the transmitted, scattered, or reflected light is collected and detected by a photodetector.
[0031] A schematic block diagram illustrating an embodiment of the present invention is shown in Figure 1, where the on-chip photonic system includes a hybrid III-V / IV semiconductor chip 1 and control and signal processing electronics 2, which form the hardware portion of the photonic sensor chip. The on-chip photonic sensor is in optical communication with a target 3, which may be a living organism, isolated material, etc. In this configuration, the on-chip photonic system is remote from the target.
[0032] In the illustrated embodiment, hybrid III-V / IV semiconductor chip 1 includes a hybrid III-V / IV external cavity laser 100, which generates wavelength-swept laser radiation via optical path 10. A portion of this beam is split off via path 11 and fed to wavelength-shift tracker 120 via path 11, to absolute wavelength reference 130 via optical path 14, to laser output curve monitoring block 140 via optical path 17, and to the output via optical path 19. Chip 1 may also include a temperature sensor 110 for detecting the temperature of the chip, which can then be used for absolute wavelength reference calibration.
[0033] The wavelength shift tracker 120 can be any kind of unbalanced interferometer, such as Mach-Zehnder, Michelson, Fabry-Perot, or others. The unbalanced interferometer produces a beat signal at the output of the tracker 120 via optical path 12, and the photodetector (PD) block 121 registers an oscillation signal, where the oscillation period depends on the optical path difference inside the interferometer and the wavelength. The optical path difference is specified by design and is a known parameter. The wavelength shift value can thus be extracted if the absolute value of the wavelength at any given time is known. This is provided by the absolute wavelength reference block 130 coupled to the monitoring photodetector 131 via optical path 15. The absolute wavelength reference can be a distributed Bragg grating (DBR), a microring resonator (MRR), a distributed feedback grating (DFB), or any other optical cavity structure that has a unique characteristic transmission or reflection property within the spectral range covered by the sweep of the hybrid laser 100. In this manner, photodetector blocks 121 and 131 cooperate to provide information regarding the absolute wavelength value and the wavelength shift value at any given time during the sweep.
[0034] Tracking wavelength shift and absolute wavelength value are often necessary to separate system effects from object-related effects. For example, emission wavelengths may change in a nonlinear manner on the system side, so without precise knowledge of absolute wavelength shift and value information, it may be difficult to perform signal conversion from time domain to wavelength (or frequency domain). Another embodiment is that the collected spectrum changes based on changes on the object side, such as moisture movement based on temperature, or other strong baseline contributors. Without knowing the system output at all times, it is impossible to separate whether the spectrum collected from the object shifts based on changes in the system output or is affected by changes inside the object. Therefore, tracking wavelength shift and absolute wavelength information on a sweep-by-sweep basis allows the system-specific modulation of the collected spectrum to be separated from the object-specific modulation, which is the useful signal.
[0035] In practical cases, target molecules such as glucose, lactate, ethanol, etc., have very small concentrations compared to the major baseline contributors, such as major proteins (collagen, albumin, keratin) and water in the case of transdermal detection. These major contributors result in signals 10,000 times or more stronger than the target molecules, and thus baseline changes that may be introduced by small changes in water movement due to temperature effects, if unnoticed, would drown out any useful signal that could be attributed to glucose. Thus, the ability to track wavelength shifts and absolute values from sweep to sweep allows access to tracking baseline changes from sweep to sweep.
[0036] The wavelength shift may be monitored as a beat signal during the sweep, whereas the absolute value is measured once per sweep, and information from both the wavelength shift and the absolute wavelength is used to calibrate the recorded information immediately after the sweep is completed. The accuracy of the wavelength shift determination depends on the design of the system, such as the optical path difference inside the wavelength shift tracker, which then provides the beat signal. In a realistic case scenario, this depends on the sophistication of the absorption characteristics of the target molecular species inside the object. When the object is a biological material and the molecules exhibit a liquid phase, characterized by a very broad spectral signature, the wavelength shift tracker can have an accuracy of 0.1 nm to several nm, with a typical value of 3-5 nm.
[0037] In the case of gas detection, the width of the absorption line of interest is in the range of 100 MHz, and the wavelength shift tracking needs to be designed to have a better resolution, and the absolute wavelength reference needs to be designed to provide a sufficiently high resolution absolute wavelength. In practical cases, this can be achieved with very good accuracy. For example, typical group IV semiconductor fabrication techniques rely on fine node sizes of 160 nm or even 7 nm, which are three orders of magnitude larger than typical emission wavelengths. The length of a single sweep is determined by the system architecture and can last from several minutes, when the tuning mechanism is performed by mechanical action of the tuning element, to several microseconds, when the tuning action is performed electronically. In practical cases for hybrid III-V / IV sensor chips, the sweep speed can range from tens of Hz to MHz, depending on the actual practical system design and application conditions.
[0038] Depending on the design of the sensor and the range requirements of the spectral bandwidth, a single sweep can include tens to hundreds of individual wavelengths. In a typical practical example for glucose detection, roughly 100 or more individual wavelengths are required to make an accurate prediction. Based on the existing state-of-the-art concept of widely tunable (swept wavelength) lasers, the sweep can be almost continuous when a Vernier filter is operated in combination with phase control. In some embodiments, the absolute value of the emission wavelength is variable within a certain range, for example, from 1000 to 3000 nm, from 1900 nm to 2500 nm, etc. Thus, the value of the variable emission wavelength at a certain time may be 1898 nm, 1905 nm, etc. The corresponding wavelength shift can be 1 nm, 2 nm, 10 nm, etc.
[0039] The EMR received from the medium of interest is converted from the optical domain to an electrical signal inside the photodetectors 121 and 131, and the electrical signal from the photodetectors is routed via conductive paths 13 and 16 to the drive and control electronics block 2 and conductive path 30 connected to the analog-to-digital converter (ADC) and amplifier block 210 therein. Here, the analog signal from the photonic chip is amplified and digitized. The digitized signal is provided to the CPU 220, which filters, averages, and otherwise processes the signal. The CPU 220 includes a memory block with a calibration model. The calibration model is applied to the collected data to retrieve a calibrated concentration level value, which is then provided to an output port, e.g., a display 240, via conductive path 39. Another function of the CPU 220 is to provide control signals via conductive path 38 to the driver and digital-to-analog converter (DAC) block 230, which then provides control and drive signals to the SoC via conductive path 40. The entire sensor system is powered by a power supply 200 via electrical buses 31, 32, 33, 34, 35, 36.
[0040] A simplified version of the sensor system of FIG. 1 is presented in FIG. 2, where several internal blocks, such as analog-to-digital converter (ADC) block 210, wavelength-shift tracker photodetector block 121, absolute wavelength reference photodetector 131, laser output curve photodetector 140, signal photodetector 150 and CPU 220, are individually highlighted for clarity.
[0041] When used in the field, the on-chip photonic sensor 1 transmits a tunable wavelength signal to a remote target 3 via an optical path 20. The intensity I of this signal can be expressed as an arbitrary function of frequency ω (or wavelength) and time t: I = f(ω,t) (1)
[0042] This light interacts with the object 3 and undergoes numerous scattering and absorption events within the object. A portion of the scattered and diffusely reflected light is collected by a signal photodetector 150 via optical path 21. The intensity of this optical signal can be expressed as a function of frequency and time I': I'=f'(ω,t) (2)
[0043] This signal is modulated due to interaction with the object and carries object-specific information, e.g. the concentration level of a component. This latter can be evaluated as an absorbance A, which is the individual absorbance A i can be expressed as a linear superposition of: A = -ln(I' / I) = Σ i A i =Σ i ε(ω) i c i l = ε(ω) 1 c 1 l+ε(ω) 2 c 2 l+(ω) 3 c 3 l+ (3)
[0044] In the formula, ε(ω) i is the frequency-dependent molar absorptivity of individual component i, c iis the molar concentration of the individual component i, and l is the effective optical path length within the object.
[0045] In the practical case where the object is a biological organism, the individual absorbance contributions can be expressed as contributions from different components, e.g.: 1-keratin, 2-glucose, 3-lactate, 4-urea, 5-collagen, etc. This provides a way to decompose complex matrices into their components, thus providing the possibility of detection. The procedure for collecting and processing the data to derive calibrated concentration values is shown in block diagram form in Figures 3, 6 and 9.
[0046] The basic working method for performing detection involves first using a calibration algorithm in combination with the hardware to generate a calibration model and storing it in the memory of the CPU. This model can be considered generic and can be adopted for each sensor in the field without the need for modifications during use. The next step is then to use the algorithm according to Figure 9 in combination with the hardware and the calibration model stored in the system memory or CPU.
[0047] According to an embodiment of the present invention, when used in a detection configuration, the on-chip photonic system provides several output channels containing information about the state of the photonic chip, such as wavelength shift values via photodetector 121, absolute wavelength reference values via photodetector 131, laser intensity curves via laser output curve monitoring block 140, and / or reflected signals containing object-specific information via signal photodetector 150. These electrical signals are routed to the control and signal processing electronics block 2, where they are fed to an analog-to-digital converter and amplifier block 210.
[0048] System calibration for analyte measurements The algorithm for processing the acquired analog signal received from the photonic SoC1 starts by first amplifying and digitizing the received signal in the ADC and amplifier block 210. At this stage, the signal is still processed as a time domain signal. The thus amplified and digitized signal is then fed to a central processing unit (CPU) 220, where the object-specific signal 22 is processed and converted from the time domain to the frequency domain using the wavelength shift received via the conductive path 13 and the absolute wavelength calibration information received via the conductive path 16, and normalized with respect to the laser power curve received via the conductive path 18. This procedure allows to first obtain the signal in the time domain and further process the signal carrying mainly object-specific data, dealing with system-related nonlinearities, and is shown as step 2210 in FIG. 3.
[0049] Multiple spectra are collected, averaged, and filtered to reduce noise. For example, in FIG. 4A, each individual curve represents an averaged spectrum. The corrected intensity is then converted to absorbance in step 2220 by equation (3), and multiple raw absorbance spectra are accumulated as shown in step 2230. The raw spectra typically have a wide variety of spectral shapes due to different tissue physiology (e.g., due to different tissue samples with different scattering particle sizes, etc.), intensity based on optical path length differences, and / or particle size differences. To correct for scattering effects, a multiplicative scattering correction (MSC) is applied to the raw absorbance data in step 2240, and a global MSC reference file (or average spectrum) is extracted in step 2250 (shown by the bold line approximately in the center of the graph in FIG. 4a) and then stored in the system memory. This global MSC spectrum is later used to assign the raw data to the correct clusters based on the exact same baseline correction procedure. Then, all the data (see FIG. 4b) that have been baseline corrected in step 2240 (e.g., after MSC) are then grouped into clusters based on the similarity of their spectral shapes in step 2260. Other types of scatter correction techniques, such as standard normalization (SNV) correction, Kubelka-Munk correction, Sanderson correction, or mean centering and normalization correction, may be applied as an alternative to the multiplicative scatter correction.
[0050] Referring to Figure 5, the baseline-corrected spectra of Figure 4b are clustered into six different clusters (by definition N=6) using the k-means algorithm. The maximum distance to the centroid within each cluster is shown in the graph.
[0051] As illustrated, the global MSC corrected data is used only to assign the raw spectra to their respective clusters. The clusters thus assigned contain raw, i.e. unprocessed, data. Clustering can be done in a number of ways. Two possible methods are shown in FIG. 3. In the first method, i.e., step 2270, the spectra are grouped into a fixed, defined number N of clusters based on the similarity of their spectral shapes. The drawback of this approach is that the error, i.e., the distance σ from the spectrum to the assigned cluster centroid, where σ is the array of cluster centroids and k is the number of clusters, is very large. k However, the difference between the number of clusters can vary widely between different clusters (see FIG. 5). In a potentially better method, shown as step 2275, clustering is performed by specifying the maximum distance from any spectrum to a cluster centroid, resulting in an arbitrary number of clusters, which may be large in practical cases. Thus, an intermediate method, shown as 2276, may be used, in which both the number of clusters N and the maximum allowable distance to the cluster centroid are specified. In such a case, spectra that meet the distance to the centroid criterion within the specified number of clusters are considered outliers and are not used in the data analysis. The predetermined number of clusters may be arbitrarily large, and in practical cases may be 10-50 depending on the analyte and the detection geometry. The set of cluster centroids is stored in the CPU memory and the detection function is later used to assign the GS corrected spectra.
[0052] Once clustering is complete, individual calibration models within each cluster are generated in step 2280. The individual calibration models assign calibrated concentration level values to each spectrum within each cluster as measured by the absolute standard as described above. This set of calibration models is then stored in CPU memory next to the MSC reference vector in step 2300.
[0053] An algorithm 2280 for constructing individual calibration models is shown in Figure 6. Referring also to Figures 7a and 7b, according to an embodiment of the present invention, step 2280 for constructing individual calibration models in each cluster begins by applying a scatter correction (e.g., MSC) to the raw spectra in the cluster. This results in an individual, local MSC reference for each cluster, which is stored in the CPU memory (see bold lines in Figure 7a). This local reference is used to process the raw data acquired in the detection mode. Other types of scatter correction techniques, such as standard normalization (SNV) correction, Kubelka-Munk correction, Sanderson correction, or mean centering and normalization correction, may be applied as an alternative to the multiplicative scatter correction.
[0054] The local criteria from step 2281 are then used to build a partial least squares (PLS) model within each cluster, and optimal model parameters such as noise filtering parameters, order of derivative, and number of PLS latent vectors are obtained using cross-validation in step 2282. This task results in an optimal set of data pre-processing parameters in step 2283, which are then applied to all clusters containing raw spectra to build individual calibration models in step 2284. In other words, within each cluster, the raw spectra are corrected using a local scatter correction criterion. This ensures that all data is treated in the same way with the same set of parameters. The calibration model then assigns to each of the locally corrected spectra a calibrated concentration level(s) of the analyte(s) of interest, measured by a selected reference technique (also called an absolute reference). The calibration model maps the absorbance represented by the spectrum at a particular wavelength to the concentration level of the analyte. Referring to Figure 8, the resulting individual calibration vectors are then stored in the CPU memory. The calibration vectors are the output of the multivariate regression calibration. After training the model using all the spectra in the cluster, at each wavelength, weights are determined for the locally corrected pre-processed absorbance spectral values. In prediction, the pre-processed absorbance is multiplied by the corresponding weight for every ith wavelength value, and then by summing over all wavelengths, the predicted concentration is: c=w 1 A 1 +w 2 A 2 +···+w n A nwhere n is the number of wavelengths in the spectrum. In some cases, when the sample is associated with a relatively simple scattering matrix and when the sample contains fewer components, reasonable concentration predictions can be obtained by simply pre-processing the spectral data obtained from the sample to correct for the nonlinear effects of scattering using Kubelka-Munk correction, MSC, Sanderson correction, or a combination thereof, and then removing the baseline to obtain the spectrum of the component of interest. For greater accuracy, especially for more complex samples such as biological tissues, a scattering correction (or linearization transformation) may be used in combination with a multivariate linear regression such as PLS or the like.
[0055] Typically, during calibration, EMR is directed towards a sample (also called a medium) where it is swept across a range of wavelengths. In response, EMR is received from the sample where it is diffusely reflected by or transmitted through the sample. The received EMR has different wavelength components and is converted into a raw absorption spectrum (also called a raw spectrum). This process may be repeated several times to obtain several raw spectra, which are then averaged to obtain an averaged raw spectrum. In the following description, the term "averaging" is omitted for simplicity. Such raw spectra are referred to as X raw i where the subscript i indicates each averaged raw sample and can range from 1 to M, where M can be any number such as 50, 100, 2000, 10000 or more. The above process may be repeated at different time points, where the analyte concentration in the sample may be different at different time points, and different regions of the sample or different samples may be used, where the concentration of the analyte in the different regions of the sample or different samples may be different.
[0056] Then, scatter correction (MSC, Kubelka-Munk correction, Sanderson correction, etc.) is applied to the raw spectrum X raw iis applied to X G ref The global reference and global correction spectra X, denoted as GC i The global criterion X G ref is stored in memory. The clustering is then performed to obtain the globally corrected spectrum X GC i This is done using x to identify N clusters. The number N (e.g., 4, 5, 6, 10, etc.) may be specified for the clustering operation, or alternatively, the clustering itself may determine the optimal N. GC i For each of the corresponding clusters C k , k∈[1,N] are identified and then the corresponding raw spectrum X raw i are assigned to the same cluster. After clustering, the optimal number of clusters, the cluster centroids, and the maximum allowable distance to the cluster centroids are stored in memory and used for the detection function.
[0057] Once all of the raw spectra have been assigned to the respective clusters, the above process is repeated within each cluster. Specifically, within a particular cluster k, scatter correction is applied to the raw spectrum X raw i Local criteria X applied to Lk ref is obtained and stored in memory. The scatter correction is applied to the raw spectrum X rawk i By applying it locally to the kth cluster, we obtain the local corrected spectrum X LCk i This process is repeated for all clusters, and for each k∈[1,N], the respective local criterion X Lk ref and local correction spectrum X LCk i is obtained.
[0058] Again, different raw spectrum X raw imay correspond to different levels of analyte concentration. i These concentration levels, denoted as , are obtained from the sample using the selected absolute standard technique. Finally, via multivariate linear regression calibration, a calibration vector V k The calibration vector V k , local standard X Lk ref , and the data pre-processing set used to generate the calibration vector may be stored in the memory module of the SoC for each cluster. The data pre-processing set specifies whether the calibration vector was obtained using absorbance, absorbance processed with nth derivative, the order of filtering, Kubelka-Munk correction, Sanderson correction, multiplicative scattering correction, etc. This is necessary to ensure that all raw data are processed in exactly the same way when the sensor is used for detection. Global criterion X G ref may also be stored in a memory module of the SoC.
[0059] One exemplary process for obtaining an optimal data preprocessing set is as follows:
[0060] Within a cluster, signal smoothing (noise filtering) is applied iteratively to the locally corrected spectra using a selected filter and degree (e.g., Savitzky-Golay, Fourier transform filter, percentile, moving average). In addition, first or second order differential baseline removal may be applied.
[0061] The locally corrected and preprocessed spectra and corresponding concentrations are randomly split into training and testing sets.
[0062] A multivariate regression calibration algorithm is applied to the training set and, after the model is trained, the test set is used to predict concentrations and the accuracy of the predictions is assessed.
[0063] In a process called cross-validation, steps 2 and 3 are repeated a number of times (eg, n iterations) to obtain an average prediction accuracy for the current data pre-processing set.
[0064] Steps 1-4 may be repeated with a different set of parameters selected in step 1. The optimal set of parameters is the set that results in the highest average prediction accuracy.
[0065] Multivariate regression algorithms model the relationship between predictors and response variables. Thus, the calibration spectral matrix X∈R d may be considered as predictors, where d is the number of wavelengths, and the analyte concentration vector Y∈R is considered as the response. Each of the i-th rows of the spectral matrix corresponds to a locally corrected preprocessed spectrum (e.g., a Savitzky-Golay filter and a second derivative applied to the locally corrected absorbance spectrum), and each of the i-th rows of the response vector corresponds to the concentration of the analyte measured in an absolute standard. Once the relationship between the predictors and the response has been determined, unknown values of the analyte concentration can be predicted based on the new locally corrected preprocessed spectra. Multivariate regression may include partial least squares regression and its modifications, multiple linear regression, support vector regression, artificial neural networks, and / or principal component regression.
[0066] Detection or measurement of analytes 9, the individual calibration vectors, global MSC reference, and local MSC vectors stored in memory enable the detection algorithm to perform the detection function of the hybrid photonic SoC. In particular, once deployed in the field, the hybrid III-V / IV photonic SoC collects the scattered reflection signal, which is then amplified and digitized in the ADC+amplifier section 210 along with the absolute wavelength reference, wavelength shift value, and laser power curve signal. This time domain signal is then converted to the frequency domain in step 2210 and averaged and calibrated with respect to absolute wavelength, wavelength shift, chip temperature, and laser power curve. The reflected intensity is then converted to absorbance in step 2220.
[0067] Next, in step 2221, the collected absorbance spectra are baseline corrected using a global scatter correction GSC criterion retrieved from the CPU memory to begin the clustering procedure. To cluster the collected spectra, the cluster centroids and the maximum allowable distance to the cluster centroid are provided from the CPU memory, and the data are classified accordingly in step 2223. If the proposed distance to the cluster centroid exceeds the maximum allowable distance, the CPU issues an error message and prompts the user to adjust the sensor position, and restarts data collection in step 2224 until the error does not exceed the maximum allowable value. If the collected data after baseline correction has a distance to the cluster centroid within the allowable range in step 2225, the corresponding collected raw spectrum is assigned to the cluster with the smallest distance to the centroid in step 2226.
[0068] Next, in step 2227, within the newly assigned cluster, the raw spectra are baseline corrected using a local scatter correction criterion from the CPU memory, and the data is pre-processed in step 2228 using a data processing set from the CPU memory suitable for a data prediction step 2229, where the data is converted to an individual calibration vector V from the CPU memory obtained by multivariate regression calibration. kMultiplying the row vector of the spectrum by the column vector of regression weights gives a single value for the concentration of the analyte. Each different analyte will have a different calibration vector and therefore different weights, i.e., different wavelength specificities for the particular analyte. For example, 2100 nm may be associated with both lactate and glucose, but the weights will be different. The concentration of the analyte is given by c=w 1 *A 1 +w 2 *A 2 +···+w n *A n Here, w n is the calibration weight at the nth wavelength, and A n is the locally corrected preprocessed absorbance at the nth wavelength, and the output is the calibrated concentration level of the analyte of interest.
[0069] In general, the detection process begins in a similar manner to the calibration process. Specifically, EMR is directed to a sample (also called a medium) from which the concentration of an analyte is to be determined. The EMR is swept across a range of wavelengths. In response, EMR is received from the sample, where it is diffusely reflected by or transmitted through the sample. The received EMR has different wavelength components and is converted into a raw absorption spectrum (also called a raw spectrum). This process may be repeated several times to obtain several raw spectra, which are then averaged to obtain Y raw This results in an averaged raw spectrum, denoted as: Again, in the following discussion, the term "averaging" will be omitted for simplicity.
[0070] Next, X G ref Scatter correction is performed on the raw spectrum Y using a global reference (generated during the calibration process) denoted as raw is applied to the global correction spectrum Y GC Then, the clustering is performed by using the cluster centroid values σ kand the distance to the centroid value. The cluster is C k where k∈[1,N], where the number N is specified for the clustering operation, or alternatively, is determined while performing the clustering as part of the calibration process. raw Then, the same cluster C k is specified for
[0071] Then, X Lc ref The selected cluster C is then processed using the corresponding local criterion, denoted as k In, the raw spectrum Y raw Scatter correction is applied again to cluster C. k In, the raw spectrum Y raw A locally corrected and preprocessed spectrum Y is obtained by locally applying a set of scatter correction and data preprocessing parameters to LC This spectrum Y LC Absorbance values of and selected cluster C k Calibration vector V for k Using the α-α signal, the concentration level for the analyte of interest is predicted. This entire process may be repeated several times to obtain several predictions of the analyte concentration, resulting in an averaged predicted analyte concentration.
[0072] Examples of transdermal sensor performance for three different analytes in piglets, namely blood glucose, blood lactate and blood ethanol, according to an embodiment of the present invention are shown in Figures 10 to 12. Here, for all experiments, a sow weighing approximately 40 kg was sedated for a period of 8 hours and a buffered analyte solution, glucose solution, was injected intravenously to raise the analyte level in the pig's blood. In the case of glucose in Figure 10, blood glucose levels were raised by injecting a buffered glucose solution and insulin was administered to lower the blood glucose level. In the case of lactate in Figure 11, blood lactate levels were raised intravenously and lowered by discontinuing the administration of buffered lactate and allowing the pig to naturally clear the lactate level. In the case of ethanol, blood ethanol concentrations were again raised by intravenously injecting a buffered ethanol solution and lowered by discontinuing the administration and allowing the body to naturally clear the ethanol. In all cases, the III-V / IV sensor was in contact with the pig's skin on the sedated pig's abdomen. The sensors sampled the pigs at a frequency of 40 Hz (40 sweeps per second or 40 spectra per second). Blood samples were taken from the pigs' arteries every 6 minutes and analyzed on clinical analyzers as absolute standards. In the embodiment described in this application, two Abaxis Piccolo Xpress analyzers were used for blood glucose calibration, an EKF Biosen C-line analyzer was used for lactate calibration, and an Agilent 8860 gas chromatograph was used for blood ethanol calibration as clinical absolute standards. The collected spectra were then assigned to calibrated glucose concentration levels measured with the absolute standards, and the data was processed according to the procedures described in the embodiments of the present invention.
[0073] In FIG. 10, data points 1002 represent the data points used to form the calibration model, and the red data points 1004 represent multiple predictions using that model for the particular pig under study. In this case, the model and calibration used data obtained from the same pig. The pig's blood glucose levels were ramped up and down over a daytime period, and calibration data was measured every 6 minutes using an absolute standard. Optical spectra collected between two calibration points were interpolated and assigned absolute glucose concentration values.
[0074] Representative results showed excellent sensor performance over a wide dynamic range of glucose concentration levels from 75 mg / dl (4.16 mmol / l) to 400 mg / dl (22.2 mmol / l), with a coefficient of determination of 97.2%, root mean square error of prediction (RMSEP) of 14.7 mg / dl (or 0.8 mmol / l), and a mean absolute relative deviation of 6.7% over the entire range.
[0075] In Figure 11, the green data points 1006 represent the data points used to form the calibration model, and the red data points 1008 represent multiple predictions using that model for the particular pig under study. In this case, the model and calibration use data obtained from the same pig. Representative results show a coefficient of determination of 92.4% and an RMSEP of 0.954 mmol / l for transcutaneous blood lactate detection at concentration levels ranging from 1 mmol / l to 15 mmol / l.
[0076] In Figure 12, the green data points 1010 represent the data points used to form the calibration model, and the red data points 1012 represent multiple predictions using that model for the particular pig under study. In this case, the model and calibration use data obtained from the same pig. Representative results show a coefficient of determination of 96.4% and an RMSEP of 0.217‰ for transdermal blood ethanol detection over a range of concentration levels from 0.2‰ to 4.2‰.
[0077] The effect of data processing / correction is highlighted in Figures 13 and 14. In Figure 13, a typical experimental raw absorption spectrum 1300 based on diffuse reflectance is shown collected from an infiltrated pig ear. The spectrum contains signals from the tissue-skin, its components (collagen, water, etc.) and the infiltrating solution, specifically 2% ethanol in water in this case. In this experiment, ethanol is the analyte of interest. The solution is injected into the ear artery and the return is collected from the vein. A sensor is attached to the ear skin surface and the diffuse reflectance of the tissue and the infiltrating solution is collected.
[0078] Due to the nonlinear nature of diffuse reflectance, one important step in the pre-processing of the data is to linearize the collected spectra and perform a diffusion correction, which, if properly applied, allows for further processing of the data, such as a Beer-Lambert absorbance based analysis, where the linearized and corrected spectra are resolved into their individual components. This post-stage analysis may be performed in combination with other linear regression techniques to obtain calibrated values for the concentration levels of the components / analytes of interest.
[0079] In Figure 13, Kubelka-Munk linearization has been performed to resolve the raw spectrum 1300, and by using the absorption spectrum 1400 of pure ethanol obtained from a calibrated transmission measurement, also referred to as the reference spectrum of the selected analyte, ethanol can be isolated / resolved in the observed transdermal spectrum 1500. Although noisy due to the lack of additional processing, the isolated spectrum 1500 shows three ethanol specific peaks.
[0080] Further processing of the isolated spectra can be performed as shown in Figure 14a and Figure 14b. Here, a 24-hour permeation cycle of different concentrations of ethanol ranging from 0.1% to 2% was performed. Arterial inlet and venous outlet control flow cuvettes were used to monitor the concentration of the permeation solution and its stability, shown as reference cuvette signal 1600. Detection was performed transcutaneously, based on diffuse reflectance geometry.
[0081] In FIG. 14a, the obtained raw spectrum 1300 was directly processed by applying -ln(x) for fitting to the Beer-Lambert model without any linearization transformation / correction. The components used for fitting included water, skin, ethanol, fat, slope, path length, and offset, so the spectrum was decomposed into water, skin, fat, ethanol, slope, path length, and offset. The obtained approximation was compared to the reference cuvette measurement, i.e., the baseline cuvette signal 1600. As can be seen, the ethanol trace 1700a has some correlation with the baseline trend 1600, but is mostly inconclusive and does not provide a reliable reading for detection applications.
[0082] In Figure 14b, the same diffuse reflectance spectrum is processed using the Kubelka-Munk correction for linearization and scatter correction, followed by the Beer-Lambert approximation (decomposition into individual components and fitting). In this case, the extracted transdermal ethanol trace 1700b is in good agreement with the reference cuvette signal 1600 over the entire range from 0.1% to 2%, including the sharp rise / fall profile.
[0083] The embodiments of the invention described herein are intended to be merely exemplary, and numerous variations and modifications are intended to be within the scope of the invention as defined in the appended claims.
Claims
1. 1. A method for calibrating a sensor to measure a concentration of an analyte, comprising: Collecting multiple raw spectra from a subject having an analyte using a hybrid III-V / IV semiconductor photonic system on a chip (SoC); dividing the raw spectra into a set of clusters according to their respective spectral shapes, each cluster including a group of the raw spectra; Splitting multiple raw spectra according to their respective spectral shapes: applying a global scatter correction (GSC) to each of the plurality of raw spectra to obtain a plurality of globally corrected spectra; clustering the plurality of globally corrected spectra according to: (A) a particular number of clusters, or (B) a particular maximum distance from the centroids of the clusters to the globally corrected spectra, or (C) both a particular number of clusters and a particular maximum distance from the centroids of the clusters to the globally corrected spectra; and within each cluster, assigning to that cluster each of the raw spectra corresponding to the globally corrected spectra that belong to that cluster; and Within each cluster: applying a respective local scattering correction (LSC) to each of the raw spectra belonging to the cluster to obtain a group of locally corrected spectra; and deriving a set of cluster-specific optimized pre-processing parameters and a cluster-specific calibration vector using the local correction spectrum and absolute reference values of analyte concentrations corresponding to the group of raw spectra belonging to the cluster; Deriving a set of cluster-specific optimized pre-processing parameters and a cluster-specific calibration vector for a particular cluster includes: evaluating each of a plurality of candidate sets of pre-processing parameters, where evaluating a particular candidate set includes: pre-processing each of the locally corrected spectra that belong to the particular cluster using the particular candidate set; deriving candidate calibration vectors by applying a multivariate regression calibration to the pre-processed local correction spectra using absolute reference values of analyte concentrations corresponding to groups of raw spectra that belong to specific clusters; and calculating corresponding accuracy measures for the candidate calibration vectors via cross-validation; and The method includes designating the candidate set and corresponding candidate calibration vector associated with the greatest accuracy measure as a cluster-specific optimized set of pre-processing parameters and a cluster-specific calibration vector, respectively.
2. The object includes a tissue; and 10. The method of claim 1, wherein the analyte comprises at least one of: blood glucose, blood lactate, ethanol, urea, creatinine, troponin, cholesterol, albumin, globulin, ketone-acetone, acetate, hydroxybutyrate, collagen, keratin, or water.
3. The method of claim 1 , wherein the clustering comprises at least one of: k-means clustering, affinity propagation, or agglomerative clustering.
4. The method of claim 1 , further comprising storing the GSC reference spectrum on the SoC.
5. 2. The method of claim 1, wherein the global scatter correction comprises a global multiplicative scatter correction, a global standard normal variate (SNV) correction, a Kubelka-Munk correction, a Sanderson correction, or a global mean centering and normalization correction.
6. 2. The method of claim 1, wherein the local or global scatter correction comprises a particle size difference correction or a path length difference correction, each of which comprises a Kubelka-Munk correction, a Sanderson correction, a multiplicative scatter correction, or a combination thereof.
7. 2. The method of claim 1, further comprising storing, for each cluster, in the SoC: (i) a corresponding LSC reference spectrum, (ii) a corresponding calibration vector, and (iii) a cluster centroid.
8. The method of claim 7 , further comprising: (iv) storing on the SoC for each cluster: a set of cluster-specific optimized pre-processing parameters.
9. The method of claim 1 , further comprising storing an optimized set of pre-processing parameters on the SoC for each cluster.
10. The method of claim 1 , wherein the local scatter correction comprises a local multiplicative scatter correction, a local standard normalization (SNV) correction, a Kubelka-Munk correction, a Sanderson correction, or a local mean centering and normalization correction.
11. Determining the spectral shape of each of the plurality of raw spectra includes:
10. The method of claim 1, comprising pre-processing the plurality of raw spectra by applying a linear transformation and a baseline correction to the plurality of raw spectra based on a reference spectrum for a selected analyte.
12. The method of claim 11 , wherein the pre-processing includes a Kubelka-Munk correction, a Sanderson correction, a multiplicative scatter correction, or a combination thereof.
Citation Information
Patent Citations
Method and Apparatus for Noninvasively Estimating a Property of an Animal Body Analyte from Spectral Data
US20070179367A1
Tunable hybrid iii-v / iv laser sensor system-on-a-chip for real-time monitoring of a blood constituent concentration level
WO2018215388A1
