Sample quality check methods and apparatus

Abstract

In some embodiments, a method is provided that includes creating a simulation model of spectral response of a sample container and a sample contained within the sample container. The method further includes employing the simulation model to generate a plurality of simulated spectral responses based on at least one of different sample container properties and different sample properties. The method also includes generating a training dataset based on the plurality of simulated spectral responses, and training, via a processor, a machine-learning model using the training dataset. The method further includes selecting at least a portion of the trained machine-learning model for use in a diagnostic laboratory system. Numerous other aspects, methods, and systems are disclosed.

Description

SAMPLE QUALITY CHECK METHODS AND APPARATUSCROSS-REFERENCE TO RELATED APPLICATION

[0001] This application claims benefit under 35 USC § 119(e) of U.S. Provisional Patent Application No. 63 / 548,570, filed on February 1 , 2024, the disclosure of which is hereby incorporated by reference herein in its entirety.FIELD

[0002] This disclosure relates to diagnostic laboratory systems, and more particularly to sample quality check methods and apparatus for use in diagnostic laboratory systems.BACKGROUND

[0003] Diagnostic laboratory systems conduct clinical chemistry tests to identify analytes or other constituents in biological samples such as blood serum, blood plasma, urine, interstitial liquid, cerebrospinal liquids, and the like. Samples are stored and transported to testing equipment (e.g., analyzers) of a diagnostic laboratory system within sample containers.

[0004] Sample quality checks perform an important role in the efficiency and throughput of diagnostic laboratory systems. For example, sample quality checks may identify problematic samples before they are transported throughout, and processed by analyzers of, a diagnostic laboratory system. This may prevent erroneous test results from being generated and improve system efficiency by reducing the amount of time and reagents used analyzing unusable samples.

[0005] To reduce the time and complexity associated with sample quality checks, it is preferable to test sample quality without requiring a sample to be removed from a sample container. For example, some sample quality check systems directly image a sample stored within a sample container. However, extracting sample quality information by direct imaging is complicated by variations in sample condition or type, sample container properties (e.g., materials, geometry, etc.), labels placed on sample containers, lighting conditions, etc. Accordingly, systems and methods that provide improved sample quality checks within a diagnostic laboratory system are desired.SUMMARY

[0006] In some embodiments, a method is provided that includes creating a simulation model of spectral response of a sample container and a sample contained within the sample container. The method further includes employing the simulation model to generate a plurality of simulated spectral responses based on at least one of different sample container properties and different sample properties. The method also includes generating a trainingdataset based on the plurality of simulated spectral responses, and training, via a processor, a machine-learning model using the training dataset. The method further includes selecting at least a portion of the trained machine-learning model for use in a diagnostic laboratory system.

[0007] In some embodiments, a method of identifying properties of a sample includes acquiring a spectral response of a sample container and a sample contained within the sample container. The method further includes inputting the spectral response into a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties. The method also includes determining, via a processor, at least one sample property based on an output of the machine-learning model.

[0008] In some embodiments, a sample quality check module is provided that includes an imaging location, a light source configured to direct light toward a sample container and a sample within the sample container positioned at the imaging location, a light detector configured to detect light from the light source that travelled through the sample and at least one side of the sample container positioned at the imaging location, a processor, and a memory coupled to the processor. The memory includes a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties. The machine-learning model is configured to input a spectral response and output at least one sample property based on the spectral response. The memory further includes computer executable instructions stored therein that, when executed by the processor, cause the processor to: (a) employ the light source to direct light toward a sample within a sample container positioned at the imaging location; (b) employ the light detector to detect light from the light source that travelled through the sample and at least one side of the sample container positioned at the imaging location so as to measure a spectral response of the sample container and the sample contained within the sample container to light travelling through the sample container and the sample; (c) input the spectral response into the machine-learning model; and (d) determine at least one sample property from an output of the machine-learning model.

[0009] Still other aspects, features, and advantages of this disclosure may be readily apparent from the following description and illustration of a number of example embodiments, including the best mode contemplated for carrying out the disclosure. This disclosure may also be capable of other and different embodiments, and its several details may be modified in various respects, all without departing from the scope of the disclosure.BRIEF DESCRIPTION OF THE DRAWINGS

[0010] The drawings described below are provided for illustrative purposes and are not necessarily drawn to scale. Accordingly, the drawings and descriptions are to be regardedas illustrative in nature, and not as restrictive. The drawings are not intended to limit the scope of the disclosure in any way.

[0011] FIG. 1A illustrates an example flow diagram of a method of training and deploying a machine-learning model based on simulated spectral response data in accordance with embodiments provided herein.

[0012] FIG. 1 B illustrates an example computer in which the method of FIG. lA may be implemented in accordance with one or more embodiments provided herein.

[0013] FIG. 2A illustrates a cross section of a sample container having a sample stored therein in accordance with embodiments provided herein.

[0014] FIG. 2B illustrates transmitted and reflected light beams at the interfaces of a stack that includes a first medium (medium A) and a second medium (medium B) in accordance with embodiments provided herein.

[0015] FIGS. 3A and 3B illustrate example plots of real refractive index and absorption constant versus wavelength, respectively, for polyethylene terephthalate (PET) in accordance with one or more embodiments provided herein.

[0016] FIG. 3C illustrates an example plot of refractive index versus wavelength for plasma in accordance with one or more embodiments provided herein.

[0017] FIG. 3D illustrates an example plot of molar extinction coefficient of hemoglobin (Hb) versus wavelength in accordance with one or more embodiments provided herein.

[0018] FIG. 3E illustrates an example plot of molar extinction coefficient of oxygenated hemoglobin (HbO2) versus wavelength in accordance with one or more embodiments provided herein.

[0019] FIG. 3F illustrates an example plot of refractive increment factor of Hb versus wavelength in accordance with one or more embodiments provided herein.

[0020] FIG. 3G illustrates an example plot of real refractive index of H1 plasma versus wavelength in accordance with one or more embodiments provided herein.

[0021] FIG. 3H illustrates an example plot of absorption constant of H1 plasma versus wavelength in accordance with one or more embodiments provided herein.

[0022] FIG. 3I illustrates an example plot of absorption constant of a manufacturer label versus wavelength in accordance with one or more embodiments provided herein.

[0023] FIG. 3J illustrates an example plot of absorption constant of a barcode label versus wavelength in accordance with one or more embodiments provided herein.

[0024] FIG. 3K illustrates an example plot of transmission spectra of Normal, H1 and H2 plasma versus wavelength in accordance with one or more embodiments provided herein.

[0025] FIG. 3L illustrates an example plot of absorption spectra of Normal, H1 and H2 plasma versus wavelength in accordance with one or more embodiments provided herein.

[0026] FIG. 4 illustrates an example machine-learning (ML) model that may be trained using a training dataset formed with a spectral response simulation model in accordance with embodiments provided herein.

[0027] FIG. 5A illustrates an example sample container having a sample contained therein in accordance with embodiments provided herein.

[0028] FIG. 5B illustrates a sample container supported by a sample carrier and including a cap in accordance with embodiments provided herein.

[0029] FIG. 5C illustrates a top view of the cross section of the sample container of FIG. 5A in accordance with embodiments provided herein.

[0030] FIG. 5D illustrates a top view of the cross section of a sample container showing an example position of a light source and detector that may be employed for measuring transmittance through and absorption by the sample container and a sample stored therein in accordance with embodiments provided herein.

[0031] FIG. 5E illustrates a side view of the light source and detector in relation to the sample container of FIG. 5D in accordance with one or more embodiments provided herein.

[0032] FIGS. 5F-5H illustrate side views of example reflective arrangements of a light source and a light detector relative to a sample container in accordance with one or more embodiments provided herein.

[0033] FIGS. 5I and 5J illustrate top views of a cross section of a sample container showing example lateral placements of a light source and a detector in a reflective arrangement in accordance with one or more embodiments provided herein.

[0034] FIG. 6A illustrates a first example sample check module provided in accordance with one or more embodiments provided herein.

[0035] FIG. 6B illustrates a second example sample quality check module in accordance with one or more embodiments provided herein.

[0036] FIG. 6C illustrates a third example sample quality check module in accordance with one or more embodiments provided herein.

[0037] FIG. 7 illustrates a flowchart of a method of training a machine-learning model for determining sample properties based on spectral response in accordance with one or more embodiments provided herein.

[0038] FIG. 8 illustrates a flowchart of a method of identifying properties of a sample in accordance with one or more embodiments provided herein.DETAILED DESCRIPTION

[0039] Independent of the grammatical term usage, individuals with male, female or other gender identities are included within the term.

[0040] As stated previously, sample quality checks are employed within diagnostic laboratory systems to identify problematic samples before they are processed. This improves system efficiency by preventing system resources (e.g., analyzer time, reagents, etc.) from being used on samples that are not suitable for testing and that produce unusable results.

[0041] Interpreting the spectral response of a sample within a sample container is difficult as the resultant spectral response is dependent on numerous variables such as sample container material and thickness, lighting conditions, whether labels are present, the type and thickness of labels employed, the type and volume of the sample employed, the amount of interferents present in the sample, and the like. Because of the large number of variations possible, acquiring and testing samples and sample containers with all possible variations is not practical.

[0042] In accordance with embodiments provided herein, a physics-based model of the spectral response of a sample within a sample container is determined. For example, transmission and absorption of light travelling through a sample container and a sample stored therein may be modelled. In some embodiments, this modelling may take into account transmission and absorption of light (e.g., across the visible spectrum) due to any label present on the sample container. In one or more embodiments, sample characteristics of interferents such as concentration of hemolysis, icterus, and lipemia, sample container properties such as material type, thickness (e.g., tube material and sidewall thickness) and diameter, label properties such as label material, label thickness, ink type employed, number of layers of labels present, lighting conditions, and the like may be included in the physicsbased spectral response model.

[0043] Once the spectral response model is created for a sample container with a sample stored therein, simulated spectral responses may be generated for any combination of different sample properties and different sample container properties. Example sample properties that may be varied include sample type such as blood serum, blood plasma, urine, interstitial fluid, etc., sample condition such as centrifuged, clotted, etc., and sample interferents. Example sample container properties that may be varied include sample container thickness, material, shape, diameter, etc., and properties of labels attached to a sample container such as label type, thickness, material, surface finish, ink properties, adhesive properties, etc. The simulated spectral responses generated for different sample and sample container properties may then be used to train a machine-learning model which predicts sample and / or sample container properties based on the spectral response measured for a sample within a sample container, such as within a sample check module of a diagnostic laboratory system. For example, at least a portion of the trained machinelearning model (e.g., a trained inverse model portion of the overall machine-learning modelin some embodiments as described further below) may be deployed within a sample check module of a diagnostic laboratory system and used to identify problematic samples prior to analysis (e.g., by translating a measured spectral response of a sample and sample container into an interferent concentration of the sample, sample type, sample condition, or another indicator of sample quality). Such a sample quality check may be performed rapidly and without requiring the sample to be removed from the sample container. In some embodiments, sample container and / or label properties may also be determined based on a measured spectral response. These and other embodiments are described below with reference to FIGS. 1A-8.

[0044] FIG. 1A illustrates an example flow diagram 100 of a method of training and deploying a machine-learning model based on simulated spectral response data in accordance with embodiments provided herein. With reference to FIG. 1 A, and as described in further detail below, the method of flow diagram 100 includes developing a spectral response simulation model that models the spectral response of a sample within a sample container to a light source. The spectral response simulation model is then employed to generate a training dataset (spectral response training dataset 104) by varying sample and sample container properties and determining a corresponding simulated spectral response for the sample and sample container for each variation (e.g., spectral responses 105). The spectral response training dataset 104 is employed to train an initial machine-learning model 106, and at least a portion of the trained, initial machine learning model 106 (e.g., all or a portion of trained, initial learning model 106) may be used within a diagnostic laboratory system as a deployed machine-learning model 108 to predict sample and / or sample container properties based on a measured spectral response. In some embodiments, the deployed machine-learning model 108 may be employed within a sample quality check module of a diagnostic laboratory system.

[0045] FIG. 1 B illustrates an example computer 120 in which the method of FIG. 1 A may be implemented in accordance with one or more embodiments. With reference to FIG. 1B, computer 120 includes a processor 122 coupled to a memory 124. Memory 124 may include spectral response simulation model 102, spectral response training dataset 104, and initial machine-learning model 106. Memory 124 may also include one or more programs 126 for carrying out the methods described herein when executed by processor 122, such as creating spectral response training dataset 104 by varying sample and / or sample container properties and computing a simulated spectral response 105 for each variation using spectral response simulation model 102. In some embodiments, processor 122, executing one or more of programs 126, may train initial machine-learning model 106 based on spectral response training dataset 104. Memory 124 may include multiple memory units and / or types of memory. In some embodiments, all or a portion of memory 124 may beexternal to and / or remote from computer 120. Additionally, in some embodiments, multiple processors may be employed.SPECTRAL RESPONSE SIMULATION MODEL

[0046] As mentioned, in accordance with one or more embodiments, a simulation model may be developed to simulate the spectral response of a sample within a sample container to light transmitted therethrough. In particular, a physics-based model may be developed that directly models spectroscopic light transmission / absorption along the light path through the sample and sample container. By employing known transmission / absorption characteristics of hemolysis, icterus, and lipemia (HIL) or other interferents, plastic or other materials employed for sample containers, paper materials and inks used for labels, etc., a physics-based model may be developed by solving the forward and backward pass of light through a sample container and sample, considering reflection at each surface, transmission at each surface, and absorption within each medium, to obtain the transmittance at wavelengths across a desired wavelength range (e.g., 300 to 1550 nanometers, the visible spectrum, or another wavelength range). Once the physics-based model is determined, the model may be employed to simulate samples with a wide range of variations of impact factors such as different sample properties (e.g., different sample types, conditions, interferents, and the like) and different sample container properties (e.g., different sample container thicknesses, materials, shapes, diameters, label properties, etc.).

[0047] Propagation of electromagnetic waves through a stratified medium, such as a sample container and a sample, may be modeled using a transfer-matrix method. Initially, a simplified model without reflection may be determined, followed by a full model with reflection.

[0048] Considering light moving through a stack of different materials, the incident light can be transmitted or reflected at each interface, and the light may be absorbed when passing through the different materials. FIG. 2A illustrates a cross section 200 of a sample container 202 having a sample 204 stored therein in accordance with embodiments provided herein. The sample container includes a label 206.

[0049] As seen in FIG. 2A, light 208 originating from a light source 210 passes through the sample container 202 and sample 204. The light 208 passes through air, followed by a first side 212 of the sample container 202 (e.g., a first side wall), the sample 204, a second side 214 of the sample container (e.g., a second side wall), and the label 206. The transmitted light is collected by a detector 216.

[0050] Because the light 208 may be considered to originate from a single light source (light source 210), the reflection terms or the backward travelling light terms may be ignored in the equations. This is because the backward light component in the final transmission coefficient term is small for low concentrations of interferents such as hemolysis and icterus(e.g., the reflected light is at least 50 times less than the forward transmitted light). Such a simplification has the following implications: (1) the calculation of spectra of the forward transmitted light becomes easier (involving mathematically multiplying the incident light by various components that account for transmission and absorption of light as it travels through the cross section 200), and (2) it may enable the determination of the true transmission spectra of the sample 204 (e.g., blood) when the light source intensity is not constant over wavelength. The transmission coefficient of the sample 204 may be determined from the transmission spectra of an empty sample container or one that includes any non-absorbing liquid such as water.

[0051] FIG. 2B illustrates transmitted and reflected light beams at the interfaces 220a and 220b of a stack 222 that includes a first medium (medium A) and a second medium (medium B) as provided herein. With reference to FIG. 2B, assume that E ' is the transmitted light at the end of medium B, and E is the light incident to the interface between medium A and medium B. The transmitted light and incident light are related by equations (1) and (2) as follows:wherein tABis the transmission coefficient from medium A to medium B and e" / ,Bis the attenuation factor within medium B as the light travels through medium B.

[0052] Equations (1) and (2) can be solved to determine the transmission spectra for stack 222. Here, transmission at interface 220a is governed by transmission coefficient tAB, which is dependent on the polarization of the incident light and the angle of incidence of the light. Considering normal incidence of light, the transmission coefficient tAB, when moving from material A to 8 is2Wj4, irrespective of polarization of incident light, wherein nA, nBarecomplex refractive indices of the mediums A and B, respectively. Similarly, when considering the light moving through the medium 8, the transmitted light is attenuated according to a wave equation by a factor of e" / ,B, wherein fa =2TTn^dstdBis the thickness of medium 8, and A is the wavelength of incident light.

[0053] For the sample container and sample cross section 200 of FIG. 2A, there are six mediums through which the light travels, two of which are air: (0) air, (1) first side 212 of sample container 202, (2) sample 204, (3) second side 214 of sample container 202, (4) label 206, and (5) air. Therefore, for the cross section 200 of FIG. 2A, the final transmitted light, ET, for each wavelength as calculated from Equations (1) and (2) for each medium, is given by Equation (3):(3)T= t45ei^it34ei‘f>3t23ei<l>2t12ei^1t01EI

[0054] In the above formulation, the backward travelling light is not considered. The backward travelling light is produced due to reflection at the interfaces. Referring to FIG. 2B, the reflection coefficient, rAB, when moving from medium A to B is given byWhenbackward travelling light is considered, there are two formulations at each interface to account for reflection and transmission of the forward and backward travelling light. As shown in FIG. 2B, the forward travelling light EBis incident on the interface between mediums A and B, and EBis the light transmitted into medium B. Similarly, backward travelling light Ef is incident on the interface between mediums A and B, and E is transmitted into medium A. The relationship between incident and transmitted light within mediums A and B is governed Equations (4) and (5). Furthermore, the backward and the forward light travelling through the medium B is governed by Equations (6) and (7).(4) ERB= tABE + rBAE(5) Ef = tBAE + rABE$(6) Ef' = Efe^

[0055] The interface and medium transmittance condition can be written in matrix form as Equations (8) and (9):FRA=J_ [ 1 FRB(8) FA1 ] LCFLB(9)

[0056] At the last interface, since there is no backward source of light, E 'is 0. The final transmission coefficient through the medium is calculated as Ef ' / EBas E 'is the final transmitted light and Ef is the incident light. For a stack of different materials, such as shown by cross section 200 of FIG. 2A, the components of Equations (8) and (9) are multiplied with each other as in Equation (3) to obtain the solution of the stack. As previously described, the reflected light from each interface is far less than the transmitted light in most cases. As such, the reflected (backward travelling) light is ignored in the embodiments described below.

[0057] To calculate the energy of the light transmitted through a sample container, one must determine the material properties (e.g., refractive index) and geometry of the various components of the sample container and sample stored therein. The refractive index of absorbing materials is a complex number, where the imaginary part is referred to as the absorption constant (or coefficient), K. The absorption constant accounts for attenuation oflight as it travels through the material. The real, n(A), and imaginary parts, IK(2), of the refractive index are also a function of the wavelength of the light travelling through the material:

[0058] The most commonly used sample container material for blood and similar fluids is polyethylene terephthalate (PET). Other materials that may be used include polyetherimide (PEI), polycarbonate (PC), polystyrene (PS), poly-vinyl chloride (PVC), glass-crown glass and flint glass. The real and imaginary index versus wavelength for each of these materials is known (see, for example, www.refractiveindex.info). As an example, FIGS. 3A and 3B illustrate example plots of real refractive index and absorption constant versus wavelength, respectively, for PET in accordance with one or more embodiments (numerical values based on Xiaoning Zhang, Jun Qiu, Xingcan Li, Junming Zhao, and Linhua Liu, "Complex refractive indices measurements of polymers in visible and near-infrared bands," Appl. Opt. 59, 2337- 2344 (2020)).

[0059] The refractive index of blood plasma may be modelled as a real number, as plasma without any impurities has negligible absorption in the visible light region. FIG. 3C illustrates an example plot of refractive index versus wavelength for plasma in accordance with one or more embodiments (based on formula of Moritz Friebel and Martina Meinke, "Model function to calculate the refractive index of native hemoglobin in the wavelength range of 250-1100 nm dependent on concentration," Appl. Opt. 45, 2838-2842 (2006) (hereinafter “Friebel”) - see, also, Liu S, Deng Z, Li J, Wang J, Huang N, Cui R, Zhang Q, Mei J, Zhou W, Zhang C, Ye Q, Tian J. Measurement of the refractive index of whole blood and its components for a continuous spectral region. J Biomed Opt. 2019 Mar;24(3):1-5. doi: 10.1117 / 1 .JBO.24.3.035003. PMID: 30848110; PMCID: PMC6403469.). For hemolysis or presence of hemolysis interferent, the presence of hemoglobin in both deoxygenated (Hb) and oxygenated (HbO2) forms may be considered. The absorption coefficient of a solution of Hb and plasma depends on the concentration of Hb and may be calculated from the molar extinction coefficient of Hb. For example, FIG. 3D illustrates an example plot of molar extinction coefficient of Hb versus wavelength in accordance with one or more embodiments (numerical values based on Scott Prahl, “Tabulated Molar Extinction Coefficient for Hemoglobin in Water,” https: / / omlc.org / spectra / hemoglobin / summary.html (hereinafter “Prahl”)). Likewise, the absorption coefficient of a solution of HbO2and plasma depends on the concentration of HbO2and may be calculated from the molar extinction coefficient of HbO2. For example, FIG. 3E illustrates an example plot of molar extinction coefficient of HbO2versus wavelength in accordance with one or more embodiments (numerical values based on Prahl).

[0060] The real part of the refractive index of a solution of Hb and plasma is calculated as per the concentration of Hb and the refractive increment factor ( ) of Hb which is shown in FIG. 3F (an example plot of refractive increment factor of Hb versus wavelength based on formula of Friebel). The real part of the refractive index of a solution of HbCh and plasma may be similarly calculated.

[0061] Assuming Hb is present, the real component of the resultant blood sample is thus calculated asnHb ~nplasma [P(A)cHb+ 1] (in accordance with the formula provided in Friebel, which uses water as the solvent instead of plasma). For the imaginary component of the solution, the total absorption can be presented as the sum of absorption of Hb and solvent (e.g., plasma) that leads to K = KHI, + Kplasma. Since plasma absorption is much weaker than that of Hb, K « KHb(per Sydoruk, Oleksiy & Zhernovaya, Olga & Tuchin, Valery & Douplik, Alexandre, Refractive index of solutions of human hemoglobin from the nearinfrared to the ultraviolet range: Kramers-Kronig analysis, Journal of biomedical optics, 17, 115002, 10.1117 / 1. JBO.17.11.115002 (2012)). Now for H1 type of hemolysis, where the concentration of Hb « 50mg / dl, the refractive index and absorption curves of the net hemolyzed plasma are shown in FIGS. 3G and 3H, respectively (using the above formula). For icteric blood, in some embodiments, the extinction coefficient is found from the optical absorption spectrum of bilirubin in chloroform (see, for example, “Optical absorption spectrum of Bilirubin in chloroform,” https: / / omlc.org / spectra / PhotochemCAD / data / 119- abs.txt). The real refractive index of bilirubin is considered constant at n = 1 .722 (see, for example, https: / / www.chemspider.com / Chemical-Structure.4444055.html) The final refractive index of the resultant plasma is hence determined as a combination of concentration of hemoglobin (Hb) and bilirubin (bil).

[0062] A manufacturer or barcode label applied to a sample container is typically formed of a white paper with writing in black ink. The material of the paper may be glossy, and adhesives may be used to attach the label to a sample container. The refractive properties of these adhesives are not known, nor is the percentage of print on the paper. As such, the transmission spectra may be measured for the labels to be employed. In some embodiments, the real refractive index may be 1 .4 or 1 .6, which is considered constant over A (see, for example, Bakker, Jim & Bryntse, G. & Arwin, H., Determination of refractive index of printed and unprinted paper using spectroscopic ellipsometry, Thin Solid Films, 455-456, 361-365, 10.1016 / j.tsf.2004.01 .024 (2004) (hereinafter “Bakker”) and Pages H, Piombini H, Enguehard F, Acher O, Demonstration of paper cutting using single emitter laser diode and infrared-absorbing ink, Opt Express, 2005 Apr 4;13(7):2351-7, doi: 10.1364 / opex.13.002351 , PMID: 19495124) (hereinafter “Pages”). Thereafter, a curve is fit for the absorption constant of a manufacturer label as a function of 2 from real transmission spectra of plasma and areference label. Likewise, a curve is fit for the absorption constant of a barcode label as a function of A from real transmission spectra of plasma and a reference label. The absorption constant versus wavelength of a manufacturer label is shown in FIG. 3I, while the absorption constant versus wavelength of a barcode label is shown in FIG. 3J (based on Bakker and Pages). The transmission and absorption spectra from simulation using reported barcode and manufacturer label properties are shown in FIGS. 3K and 3L, respectively (based on Bakker and Pages). The simulation curves for Normal (N), H1 , and H2 (Hb « 170 mg / dL) plasma transmission exhibit very similar characteristics to that of real readings (e.g., the spectral response measured in a laboratory for sample containers with normal, H1 , H2 plasma and manufacturer and barcode labels).

[0063] The above and other sample, sample container, and label properties may be employed within the spectral response simulation model 102 to generate expected spectral responses for any variation in these properties. That is, once the spectral response model is created for a sample container with a sample stored therein, simulated spectral responses may be generated for any combination of different sample properties and different sample container properties. Example sample properties that may be varied include sample type such as blood serum, blood plasma, urine, interstitial fluid, etc., sample condition such as centrifuged, clotted, etc., and sample interferent type or concentration (e.g., different hemolysis, icterus, and lipemia concentrations). Example sample container properties that may be varied include sample container thickness, material, shape, diameter, etc., and properties of labels attached to a sample container such as label types, thicknesses, materials, surface finishes, ink properties, adhesive properties, number of labels employed, etc. These simulated spectral responses may then be used to generate a training dataset (e.g., spectral response training dataset 104 of FIGS. 1A-1 B). The training dataset may be employed to train a machine-learning model (e.g., initial machine-learning model 106 of FIGS. 1A-1B). At least a portion of the trained machine-learning model (e.g., deployed machine learning model 108 of FIGS. 1A-B) may be used to predict sample and / or sample container properties based on the spectral response measured for a sample within a sample container (e.g., within a sample check module of a diagnostic laboratory system). In some embodiments, the trained machine-learning model may predict sample interferent concentration (e.g., of hemolysis, icterus, and lipemia concentration) and / or level of interference (e.g., H1 , H2, etc.).

[0064] FIG. 4 illustrates an example machine-learning model, referred to as initial ML model 400, that may be trained using a training dataset formed with a spectral response simulation model in accordance with embodiments provided herein. For example, initial ML model 400 may be similar to initial ML model 106 of FIGS. 1A-1 B which is trained usingspectral response training dataset 104 generated by spectral response simulation model 102.

[0065] Referring to FIG. 4, with the data generated from the physics-based simulation model, an inverse model 402 may be constructed that uses simulated spectroscopy data as input and sample properties and sample container properties (including label properties, in some embodiments) as the target output. In addition, a forward model 404 may be constructed that decomposes the sample, sample container and / or label properties (generated by the inverse model 402) into spectroscopy data because the ground truth for the simulated data is known. In such an embodiment, ML model 400 operates as an autoencoder network in which the inverse model and forward model are combined together as shown in FIG. 4 and wherein the input signal is the simulated spectroscopy data (e.g., simulated spectral responses). Each latent variable represents a specific property of a sample or sample container holding the sample. Once ML model 400 is trained, the inverse model 402 may be used (e.g., deployed) to estimate one or more sample or sample container properties (including label properties) based on the measured spectral response for an arbitrary sample and sample container. For example, the trained inverse model 402 may be used (e.g., as deployed ML model 402’) to estimate sample properties, sample container properties, and / or label properties within a diagnostic laboratory system.

[0066] In an auto-encoder embodiment of ML model 400, simulated spectroscopy data (simulated spectral responses generated by spectral response simulation model 102 for FIG. 1A, for example) may be input to an encoder network (e.g., inverse model 402) that generates latent features in a latent variable LAin a multidimensional latent space. In some embodiments, the inverse model 402 generates latent features in low-dimensional space. The features of the latent variable LAmay be partitioned into multiple groups of features such as W, X, Y, and Z as shown in FIG. 4. Each of the features may correspond to a specific one of the attributes of the sample and sample container (and / or label) corresponding to the simulated spectral response input to the inverse model 402. In some embodiments, at least one of the features (e.g., feature Z) may be reserved for an intrinsic property of the sample or sample container in which variations cannot be controlled by a setup. In some examples, all sample containers may have a cylindrical structure such that it is expected to have all sample containers share the same values in feature Z.

[0067] The forward model 404 may be implemented as a decoder network that uses the features of the latent variable LAto reconstruct any simulated spectral response input to the inverse model 402 (e.g., as similar to the input spectral response as possible). A reconstruction loss module 406 may compare the input spectral response to the output (reconstructed) spectral response to determine whether the input and output spectral responses are close to each other. For example, the reconstruction loss module 406 mayanalyze the attributes of the input spectral response and the output spectral response to determine whether the latent variable l_Ais correct and whether the inverse model 402 and forward model 404 are trained correctly.

[0068] In some embodiments, the latent space representation may encode data across different scales. For example, normalizing flows, autoregressive models, variational autoencoders (VAEs), and deep energy-based models are examples for deep generative learning. Furthermore, the latent space may be constrained to be a known parametric distribution (e.g., Gaussian or mixture-of-Gaussian) or a non-parametric distribution, such as with a vector quantized variational autoencoder (VQ-VAE).

[0069] FIG. 5A illustrates an example sample container 500 having a sample 502 contained therein. In the embodiment of FIG. 5A, sample container 500 has a tube-shape. Other sample container shapes may be employed. FIG. 5B illustrates sample container 500 supported by a sample carrier 504 and including a cap 506. Cap 506 may be of different types and / or colors (e.g., red, royal blue, light blue, green, grey, tan, yellow, or color combinations), which may indicate what test the sample container 500 is used for, the type of additive included therein (e.g., for preventing clot formation), whether the container includes a gel separator, whether the sample is provided under a vacuum, or the like. Other colors may be used.

[0070] The sample container 500 may be provided with one or more labels 508 that may include identification information 510 (i.e., indicia) thereon, such as a barcode, alphabetic characters, numeric characters, or combinations thereof. Example identification information 510 may include or be associated with patient information (e.g., name, date of birth, address, and / or other personal information), tests to be performed, time and date the sample was obtained, medical facility information, tracking and routing information, etc. Other information may also be included. As shown in FIGS. 5A and 5B, the label 508 may not extend all the way around the sample container 500 or all along a length of the sample container 500 such that from the particular lateral front viewpoint shown, some or a large part of sample 502 (e.g., a serum or plasma portion 502SP) is viewable and unobstructed by the label 508.

[0071] The sample 502 may include any fluid to be tested and / or analyzed (e.g., blood serum, blood plasma, urine, interstitial fluid, cerebrospinal fluid, or the like). In some embodiments, the sample 502 may include the serum or plasma portion 502SP and a settled blood portion 502SB contained within sample container 500. Air 512 may be provided above the serum and plasma portion 502SP and a line of demarcation between them is defined as the liquid-air interface. The line of demarcation between the serum or plasma portion 502SP and the settled blood portion 502SB is defined as a serum-blood interface. An interface between the air 512 and cap 506 is defined as a tube-cap interface.

[0072] FIG. 5C illustrates a top view of sample container 500 of FIG. 5A in accordance with embodiments provided herein. FIG. 5D illustrates a top view of sample container 500 showing an example position of a light source 514 and detector 516 that may be employed for measuring transmittance through and absorption by sample container 500 and sample 502 in accordance with embodiments provided herein. FIG. 5E illustrates a side view of the light source 514 and detector 516 in relation to sample container 500.

[0073] As shown in the embodiment of FIGS. 5D and 5E, in some embodiments, light 520 from light source 514 may travel through a first side 518a of sample container 500, through sample 502, through a second side 518b of sample container 500, through label 508, and to detector 516. In some embodiments, light source 514 may emit light within the visible spectrum. In other embodiments, light source 514 may emit light having a wavelength within the range from about 300 to 1550 nanometers. Other wavelengths and / or wavelength ranges may be employed. In some embodiments, light source 514 may include one or more light emitting diodes, lasers, or the like. Detector 516 may include a photodetector, a photodiode, a phototransistor, etc. Other light sources and / or detectors may be used.

[0074] FIGS. 5F-5H illustrate side views of example reflective arrangements of light source 514 and light detector 516 relative to sample container 500 in accordance with one or more embodiments. As shown in FIG. 5F, light 520 from light source 514 travels through the first side 518a of sample container 500 and through sample 502. Light 520 then reflects of second side 518b of sample container 500 and / or off of label 508 and travels back through sample 502 and first side 518a of sample container 500 to detector 516. In the arrangement of FIG. 5F, detector 516 is positioned to detect the maximum amount of light reflected off of second side 518b of sample container 500 and / or label 508. In the arrangements of FIG. 5G and 5H, detector 516 is positioned to detect only scattered light from the second side 518b of sample container 500 and / or label 508.

[0075] FIGS. 5I and 5J illustrate top views of sample container 500 showing example lateral placements of light source 514 and detector 516 in a reflective arrangement in accordance with one or more embodiments. In the arrangement of FIG. 5I, detector 516 is positioned to detect the maximum amount of light reflected off of second side 518b of sample container 500 and / or label 508. In the arrangement of FIG. 5J, detector 516 is positioned to detect only scattered light from the second side 518b of sample container 500 and / or label 508. Other arrangements of light source 514 and detector 516 may be employed.

[0076] FIG. 6A illustrates a first example sample check module 600a provided in accordance with one or more embodiments. With reference to FIG. 6A, sample check module 600a includes an imaging location 602 at which a sample container 500 (containing a sample 502) may be positioned during a sample check operation. A light source 514 isconfigured to direct light 520 toward sample container 500 positioned at the imaging location 602. A light detector 516 is configured to detect light 520 from the light source 514 that travelled through the sample 502 and the sample container 500 positioned at the imaging location 602. In the embodiment of FIG. 6A, light 520 travels through both sides of sample container 500, as well as through label 508.

[0077] Sample check module 600a includes a processor 604 coupled to a memory 606. Memory 606 may include deployed ML model 402’ (FIG. 4) and one or more programs 608 for carrying out the methods described herein when executed by processor 604, such as measuring the spectral response of sample container 500 and sample 502 by detecting light 520 transmitted therethrough from light source 514 and detected by detector 516, and feeding the spectral response into deployed ML model 402’ to determine one or more properties of sample 502 and / or sample container 500. Specifically, memory 606 may include a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties, the machine-learning model configured to input a spectral response and output at least one sample property based on the spectral response. Memory 606 may include multiple memory units and / or types of memory. In some embodiments, all or a portion of memory 606 may be external to and / or remote from processor 604. Additionally, in some embodiments, multiple processors may be employed.

[0078] Memory 606 may further include computer executable instructions stored therein that, when executed by the processor 604, cause the processor 604 to (1) employ the light source 514 to direct light 520 toward sample 502 within sample container 500 positioned at the imaging location 602; (2) employ the light detector 516 to detect light 520 from the light source 514 that travelled through the sample 502 and at least one side of the sample container 500 positioned at the imaging location 602 so as to measure a spectral response of the sample container 500 and the sample 502 contained within the sample container 500 to light travelling through the sample container 500 and the sample 502; (3) input the spectral response into the machine-learning model (e.g., deployed ML model 402’); and (4) determine at least one sample property and / or sample container property from an output of the machine-learning model. Example sample properties include interferent concentration of the sample, sample type, sample condition, or another indicator of sample quality. Example sample container properties include sample container type, thickness, label properties, etc. Other sample and / or sample container properties may be determined.

[0079] FIG. 6B illustrates a second example sample quality check module 600b in accordance with one or more embodiments. Sample quality check module 600b is similar to sample quality check module 600a but employs a reflective arrangement of light source 514 and detector 516 as described previously with reference to FIGS. 5F-5J.

[0080] FIG. 6C illustrates a third example sample quality check module 600c in accordance with one or more embodiments. Sample quality check module 600c is similar to sample quality check module 600b but employs an additional machine-learning model 610. In the embodiment of FIG. 6C, deployed ML model 402’ is trained to output at least one sample property as a feature vector (that includes information about the sample property). The additional machine-learning model 610 may be trained to input the feature vector from deployed ML model 402’ and output sample property information based on the feature vector. Example sample property information may include interferent concentration of the sample, sample type, sample condition, or another indicator of sample quality. Likewise, deployed ML model 402’ may output a feature vector relating to at least one sample container property. The additional ML model 610 may be trained to input the sample container feature vector from deployed ML model 402’ and output sample container property information based on the feature vector. Example sample container property information may include sample container type, thickness, label properties, etc. Other sample and / or sample container property information may be determined. In some embodiments, additional ML model 610 may include a convolutional neural network (CNN), a region-based CNN (R- CNN), a fully convolutional neural network (FCN), a region-based FCN (R-FCN), etc. Any suitable neural networks may be employed. Example architectures include Inception, ResNet, ResNeXt, DenseNet, or the like, although other architectures may be employed.

[0081] FIG. 7 illustrates a flowchart of a method 700 of training a machine-learning model for determining sample properties based on spectral response in accordance with one or more embodiments. With reference to FIG. 7, method 700 includes, in block 702, creating a simulation model of spectral response of a sample container and a sample contained within the sample container. In some embodiments, the spectral response simulation model, such as spectral response simulation model 102 of FIG. 1A, may be a physics-based model that models the transmission and absorption properties of a sample container, sample, and label.

[0082] Following creation of the spectral response simulation model, in block 704, method 700 includes employing the simulation model to generate a plurality of simulated spectral responses based on at least one of different sample container properties and different sample properties. In some embodiments, the sample container properties may include one or more of sample container thickness, material, shape, and diameter. The sample container properties may also include sample container label properties such as how many layers of labels are present, label type, thickness, material, and surface finish, ink properties and adhesive properties. In some embodiments, the different sample properties may include one or more of different hemolysis, icterus, and lipemia concentrations,interference level, sample type, sample condition, the presence of other interferents, etc. Other sample container properties and / or sample properties may be varied.

[0083] Method 700 then includes, in block 706, generating a training dataset (e.g., spectral response training dataset 104 of FIG. 1 A) based on the plurality of simulated spectral responses. Thereafter, method 700 includes, in block 708, training, via a processor, a machine-learning model using the training dataset. For example, processor 122 of FIG. 1 B may train training ML model 106 using spectral response training dataset 104. In some embodiments, the machine-learning model may include an inverse model that employs simulated sample spectral response as an input and sample container and sample properties as an output (e.g., inverse model 402 in FIG. 4). Likewise, the machine-learning model may include a forward model that employs sample container and sample properties as an input and that outputs a (reconstructed) sample spectral response based on the sample container and sample properties input (e.g., forward model 404 of FIG. 4).

[0084] Once the machine-leaning model is trained, in block 710, method 700 includes selecting at least a portion of the trained machine-learning model for use in a diagnostic laboratory system. For example, a portion of training ML model 400 of FIG. 4 may be selected (e.g., inverse model 402, which may serve as deployed ML model 402’ within a diagnostic laboratory system).

[0085] FIG. 8 illustrates a flowchart of a method 800 of identifying properties of a sample in accordance with one or more embodiments. With reference to FIG. 8, method 800 includes, in block 802, acquiring a spectral response of a sample container and a sample contained within the sample container. For example, the spectral response of a sample container having a sample stored therein may be determined at sample quality check module 600a, 600b or 600c of FIGS. 6A-6C. Thereafter, in block 804, method 800 includes inputting the spectral response into a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties (e.g., machinelearning model 402’ of FIGS. 6A-6C). Method 800 also includes, in block 806, determining, via a processor, at least one sample property based on an output of the machine-learning model (e.g., processor 604 and machine-learning model 402’ of FIGS. 6A-6C).

[0086] As described above with reference to FIGS. 1 A-8, in some embodiments, at least a portion of a trained machine-learning model may be deployed within a sample check module of a diagnostic laboratory system and used to identify problematic samples prior to analysis (e.g., by translating a measured spectral response of a sample and sample container into an interferent concentration of the sample, sample type, sample condition, or another indicator of sample quality).

[0087] While the disclosure is susceptible to various modifications and alternative forms, specific method and apparatus embodiments have been shown by way of example in thedrawings and are described in detail herein. It should be understood, however, that the particular methods and apparatus disclosed herein are not intended to limit the disclosure.NON-LIMITING ILLUSTRATVE EMBODIMENT

[0088] The following is a list of non-limiting illustrative embodiments disclosed herein:

[0089] Illustrative embodiment 1. A method comprising: creating a simulation model of spectral response of a sample container and a sample contained within the sample container; employing the simulation model to generate a plurality of simulated spectral responses based on at least one of different sample container properties and different sample properties; generating a training dataset based on the plurality of simulated spectral responses; training, via a processor, a machine-learning model using the training dataset; and selecting at least a portion of the trained machine-learning model for use in a diagnostic laboratory system.

[0090] Illustrative embodiment 2. The method of illustrative embodiment 1 further comprising deploying the at least one selected portion of the trained machine-learning model in a diagnostic laboratory system.

[0091] Illustrative embodiment 3. The method according to one of the preceding illustrative embodiments wherein creating the simulation model of spectral response comprises creating the simulation model of spectral response for light within a range from 300 to 1550 nanometers.

[0092] Illustrative embodiment 4. The method according to one of the preceding illustrative embodiments wherein the different sample container properties comprise at least one of different sample container thicknesses, materials, shapes, and diameters.

[0093] Illustrative embodiment 5. The method according to one of the preceding illustrative embodiments wherein the different sample container properties comprise different sample container label properties.

[0094] Illustrative embodiment 6. The method according to one of the preceding illustrative embodiments wherein the different sample container label properties comprise how many layers of labels are present.

[0095] Illustrative embodiment 7. The method according to one of the preceding illustrative embodiments wherein the different sample container label properties comprise one or more of different sample container label types, thicknesses, materials, surface finishes, ink properties, and adhesive properties.

[0096] Illustrative embodiment 8. The method according to one of the preceding illustrative embodiments wherein the different sample properties comprise one or more of interferent levels or different hemolysis, icterus, and lipemia concentrations.

[0097] Illustrative embodiment 9. The method according to one of the preceding illustrative embodiments wherein the different sample properties comprise one or more of sample type and sample condition.

[0098] Illustrative embodiment 10. The method according to one of the preceding illustrative embodiments wherein the machine-learning model comprises a forward model that employs sample container and sample properties as an input and simulated spectral response as a target output.

[0099] Illustrative embodiment 11 . The method according to one of the preceding illustrative embodiments wherein the machine-learning model comprises an inverse model that employs spectral response as an input and that outputs at least one of sample container properties and sample properties based on the spectral response input.

[0100] Illustrative embodiment 12. The method according to one of the preceding illustrative embodiments wherein the output of the inverse model comprises at least one sample interferent concentration.

[0101] Illustrative embodiment 13. The method o according to one of the preceding illustrative embodiments wherein at least one sample interferent concentration comprises at least one of hemolysis, icterus, and lipemia concentration.

[0102] Illustrative embodiment 14. The method according to one of the preceding illustrative embodiments wherein the output of the inverse model comprises one or more of sample type and sample condition.

[0103] Illustrative embodiment 15. The method according to one of the preceding illustrative embodiments wherein selecting at least a portion of the trained machine-learning model comprises selecting the inverse model.

[0104] Illustrative embodiment 16. A method of identifying properties of a sample, comprising: acquiring a spectral response of a sample container and a sample contained within the sample container; inputting the spectral response into a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties; and determining, via a processor, at least one sample property based on an output of the machine-learning model.

[0105] Illustrative embodiment 17. The method of illustrative embodiment 16 wherein the machine-learning model employs spectral response as an input and outputs at least one of sample container and sample properties based on the spectral response input.

[0106] Illustrative embodiment 18. The method according to one of the preceding illustrative embodiments wherein the output of the machine-learning model comprises at least one sample interferent concentration.

[0107] Illustrative embodiment 19. The method according to one of the preceding illustrative embodiments wherein at least one sample interferent concentration comprises at least one of hemolysis, icterus, and lipemia concentration.

[0108] Illustrative embodiment 20. The method according to one of the preceding illustrative embodiments wherein the output of the machine-learning model comprises at least one of sample container thickness, material, shape, and diameter.

[0109] Illustrative embodiment 21. The method according to one of the preceding illustrative embodiments wherein the output of the machine-learning model comprises at least one of sample container label type, thickness, material, surface finish, ink properties, and adhesive properties.

[0110] Illustrative embodiment 22. The method according to one of the preceding illustrative embodiments wherein the output of the machine-learning model comprises how many layers of labels are present.

[0111] Illustrative embodiment 23. The method according to one of the preceding illustrative embodiments wherein acquiring the spectral response comprises: transmitting light through a first side of the sample container, the sample stored in the sample container, and a second side of the sample container; and detecting light transmitted through the second side of the sample container.

[0112] Illustrative embodiment 24. The method according to one of the preceding illustrative embodiments wherein the light is transmitted through a label attached to the sample container prior to detecting the light.

[0113] Illustrative embodiment 25. The method according to one of the preceding illustrative embodiments wherein acquiring the spectral response comprises: transmitting light through a first side of the sample container and the sample stored in the sample container; and detecting light reflected off of a second side of the sample container or a label attached to the second side of the sample container and transmitted back through the sample and the first side of the sample container.

[0114] Illustrative embodiment 26. A sample quality check module comprising: an imaging location; a light source configured to direct light toward a sample container and a sample within the sample container positioned at the imaging location; a light detector configured to detect light from the light source that travelled through the sample and at least one side of the sample container positioned at the imaging location; a processor; and a memory coupled to the processor, the memory including a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties, the machine-learning model configured to input a spectral response and output at least one sample property based on the spectral response, the memory further including computer executable instructions stored therein that, when executed by the processor,cause the processor to: employ the light source to direct light toward a sample within a sample container positioned at the imaging location; employ the light detector to detect light from the light source that travelled through the sample and at least one side of the sample container positioned at the imaging location so as to measure a spectral response of the sample container and the sample contained within the sample container to light travelling through the sample container and the sample; input the spectral response into the machinelearning model; and determine at least one sample property from an output of the machinelearning model.

[0115] Illustrative embodiment 27. The sample quality check module of illustrative embodiment 26 wherein the at least one sample property comprises at least one sample interfere nt.

[0116] Illustrative embodiment 28. The sample quality check module according to one of the preceding illustrative embodiments wherein at least one sample property comprises at least one of hemolysis, icterus, and lipemia concentration.

[0117] Illustrative embodiment 29. The sample quality check module according to one of the preceding illustrative embodiments wherein at least one sample property comprises one or more of sample type and sample condition.

[0118] Illustrative embodiment 30. The sample quality check module according to one of the preceding illustrative embodiments wherein the light detector is configured to detect light transmitted through a first side of the sample container, the sample stored in the sample container, and a second side of the sample container.

[0119] Illustrative embodiment 31. The sample quality check module according to one of the preceding illustrative embodiments wherein the light detector is configured to detect light transmitted through a first side of the sample container and the sample stored in the sample container and reflected off of a second side of the sample container or a label attached to the second side of the sample container and transmitted back through the sample and the first side of the sample container.

[0120] Illustrative embodiment 32. The sample quality check module according to one of the preceding illustrative embodiments wherein at least one sample property comprises a feature vector that includes information about the sample property.

[0121] Illustrative embodiment 33. The sample quality check module according to one of the preceding illustrative embodiments further comprising an additional machine-learning model configured to input the feature vector and output sample property information based on the feature vector.

[0122] Illustrative embodiment 34. The sample quality check module according to one of the preceding illustrative embodiments wherein the sample property information comprises a sample interferent present within the sample.

[0123] Illustrative embodiment 35. The sample quality check module according to one of the preceding illustrative embodiments wherein the memory further includes computer executable instructions stored therein that, when executed by the processor, cause the processor to determine at least one sample container property from an output of the machine-learning model.

[0124] Illustrative embodiment 36. The sample quality check module according to one of the preceding illustrative embodiments wherein at least one sample property comprises level of interference.

Claims

WHAT IS CLAIMED IS:

1. A method comprising: creating a simulation model of spectral response of a sample container and a sample contained within the sample container; employing the simulation model to generate a plurality of simulated spectral responses based on at least one of different sample container properties and different sample properties; generating a training dataset based on the plurality of simulated spectral responses; training, via a processor, a machine-learning model using the training dataset; and selecting at least a portion of the trained machine-learning model for use in a diagnostic laboratory system.

2. The method of claim 1 further comprising deploying the at least one selected portion of the trained machine-learning model in a diagnostic laboratory system.

3. The method of claim 1 wherein creating the simulation model of spectral response comprises creating the simulation model of spectral response for light within a range from 300 to 1550 nanometers.

4. The method of claim 1 wherein the different sample container properties comprise at least one of different sample container thicknesses, materials, shapes, and diameters.

5. The method of claim 1 wherein the different sample container properties comprise different sample container label properties.

6. The method of claim 5 wherein the different sample container label properties comprise how many layers of labels are present.

7. The method of claim 5 wherein the different sample container label properties comprise one or more of different sample container label types, thicknesses, materials, surface finishes, ink properties, and adhesive properties.

8. The method of claim 1 wherein the different sample properties comprise one or more of interferent levels or different hemolysis, icterus, and lipemia concentrations.

9. The method of claim 1 wherein the different sample properties comprise one or more of sample type and sample condition.

10. The method of claim 1 wherein the machine-learning model comprises a forward model that employs sample container and sample properties as an input and simulated spectral response as a target output.11 . The method of claim 10 wherein the machine-learning model comprises an inverse model that employs spectral response as an input and that outputs at least one of sample container properties and sample properties based on the spectral response input.

12. The method of claim 11 wherein the output of the inverse model comprises at least one sample interferent concentration.

13. The method of claim 12 wherein the at least one sample interferent concentration comprises at least one of hemolysis, icterus, and lipemia concentration.

14. The method of claim 11 wherein the output of the inverse model comprises one or more of sample type and sample condition.

15. The method of claim 11 wherein selecting the at least a portion of the trained machine-learning model comprises selecting the inverse model.

16. A method of identifying properties of a sample, comprising: acquiring a spectral response of a sample container and a sample contained within the sample container; inputting the spectral response into a machine-learning model trained on simulated spectral data for different sample container properties and different sample properties; and determining, via a processor, at least one sample property based on an output of the machine-learning model.

17. The method of claim 16 wherein the machine-learning model employs spectral response as an input and outputs at least one of sample container and sample properties based on the spectral response input.

18. The method of claim 16 wherein the output of the machine-learning model comprises at least one sample interferent concentration.

19. The method of claim 18 wherein the at least one sample interferent concentration comprises at least one of hemolysis, icterus, and lipemia concentration.

20. The method of claim 16 wherein the output of the machine-learning model comprises at least one of sample container thickness, material, shape, and diameter.

21. The method of claim 16 wherein the output of the machine-learning model comprises at least one of sample container label type, thickness, material, surface finish, ink properties, and adhesive properties.

22. The method of claim 16 wherein the output of the machine-learning model comprises how many layers of labels are present.

23. The method of claim 16 wherein acquiring the spectral response comprises: transmitting light through a first side of the sample container, the sample stored in the sample container, and a second side of the sample container; and detecting light transmitted through the second side of the sample container.

24. The method of claim 23 wherein the light is transmitted through a label attached to the sample container prior to detecting the light.

25. The method of claim 16 wherein acquiring the spectral response comprises: transmitting light through a first side of the sample container and the sample stored in the sample container; and detecting light reflected off of a second side of the sample container or a label attached to the second side of the sample container and transmitted back through the sample and the first side of the sample container.

26. A sample quality check module comprising: an imaging location; a light source configured to direct light toward a sample container and a sample within the sample container positioned at the imaging location; a light detector configured to detect light from the light source that travelled through the sample and at least one side of the sample container positioned at the imaging location; a processor; and a memory coupled to the processor, the memory including a machine-learning model trained on simulated spectral data for different sample container properties and differentsample properties, the machine-learning model configured to input a spectral response and output at least one sample property based on the spectral response, the memory further including computer executable instructions stored therein that, when executed by the processor, cause the processor to: employ the light source to direct light toward a sample within a sample container positioned at the imaging location; employ the light detector to detect light from the light source that travelled through the sample and at least one side of the sample container positioned at the imaging location so as to measure a spectral response of the sample container and the sample contained within the sample container to light travelling through the sample container and the sample; input the spectral response into the machine-learning model; and determine at least one sample property from an output of the machinelearning model.

27. The sample quality check module of claim 26 wherein the at least one sample property comprises at least one sample interferent.

28. The sample quality check module of claim 26 wherein the at least one sample property comprises at least one of hemolysis, icterus, and lipemia concentration.

29. The sample quality check module of claim 26 wherein the at least one sample property comprises one or more of sample type and sample condition.

30. The sample quality check module of claim 26 wherein the light detector is configured to detect light transmitted through a first side of the sample container, the sample stored in the sample container, and a second side of the sample container.31 . The sample quality check module of claim 26 wherein the light detector is configured to detect light transmitted through a first side of the sample container and the sample stored in the sample container and reflected off of a second side of the sample container or a label attached to the second side of the sample container and transmitted back through the sample and the first side of the sample container.

32. The sample quality check module of claim 26 wherein the at least one sample property comprises a feature vector that includes information about the sample property.

33. The sample quality check module of claim 32 further comprising an additional machine-learning model configured to input the feature vector and output sample property information based on the feature vector.

34. The sample quality check module of claim 32 wherein the sample property information comprises a sample interferent present within the sample.

35. The sample quality check module of claim 26 wherein the memory further includes computer executable instructions stored therein that, when executed by the processor, cause the processor to determine at least one sample container property from an output of the machine-learning model.

36. The sample quality check module of claim 26 wherein the at least one sample property comprises level of interference.

Images