Spectral detection method, system, device, equipment and medium for detecting free hemoglobin content of plasma tube braid

Abstract

The invention relates to the technical field of blood detection, and particularly discloses a spectrum detection method, system, device, equipment and medium for detecting the free hemoglobin content of a plasma braid, and the method comprises the following steps: taking out a to-be-detected sample of the plasma braid to obtain an original optical measurement object to obtain the to-be-detected sample; performing multi-wavelength alternate irradiation on the to-be-detected sample and collecting continuous time sequence photoelectric response to obtain an original time sequence signal set; obtaining an absorbance set; a normalized absorption characteristic subjected to scattering and temperature dual compensation is obtained; obtaining an initial free hemoglobin concentration estimated value; the final free hemoglobin concentration and a judgment result are obtained; by directly detecting the separated plasma tube braid, the sealing performance and safety of the main blood bag are not influenced at all, and the pollution risk caused by puncture sampling is thoroughly avoided. The method is objective and accurate, results are quantified, manual subjective errors are thoroughly eliminated by adopting a spectral analysis technology, and particularly, judgment on critical value samples is extremely accurate.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of blood detection, and particularly relates to a spectrum detection method, system, device, equipment and medium for detecting the content of free hemoglobin in a plasma tube braid. BACKGROUND

[0002] In the field of blood detection, accurately determining the content of free hemoglobin in plasma is crucial for evaluating blood quality, judging blood transfusion safety, and monitoring changes during blood storage. Abnormal increase in the content of free hemoglobin may indicate the occurrence of red blood cell rupture (hemolysis), which can cause serious adverse effects on the recipient, such as kidney damage, allergic reactions, etc. Therefore, developing a reliable, accurate and convenient method for detecting free hemoglobin has always been a research hotspot in the field of blood detection.

[0003] Some traditional detection methods rely on manual operation and subjective judgment. For example, in some colorimetric detection, the detection personnel need to observe the color change with the naked eye and compare it with the standard colorimetric card to judge the content of free hemoglobin. However, there are differences in visual perception among different detection personnel, and environmental light and other factors will also affect color observation, which leads to a large subjective error in the detection result. Especially when dealing with critical value samples, this subjective error may lead to false judgment, which may cause waste of blood or safety hazards of blood transfusion.

[0004] To solve the above problems, the application provides a spectrum detection method, system, device, equipment and medium for detecting the content of free hemoglobin in a plasma tube braid. SUMMARY

[0005] The purpose of the application is to provide a spectrum detection method, system, device, equipment and medium for detecting the content of free hemoglobin in a plasma tube braid to solve the problems raised in the background.

[0006] To achieve the above purpose, the application provides the following technical solutions:

[0007] A spectrum detection method for detecting the content of free hemoglobin in a plasma tube braid, comprising:

[0008] S1, taking out a section of sample to be measured of the plasma tube braid, for obtaining an original optical measurement object, to obtain a sample to be measured;

[0009] S2, alternately irradiating the sample to be measured at multiple wavelengths and collecting continuous time sequence photoelectric responses to obtain a set of original time sequence signals;

[0010] S3. Perform dark current subtraction and blank reference normalization on the original time series signal set, and denoise and determine the steady state of the time series signal to obtain the steady state transmission value Tλ and convert it into absorbance Aλ to obtain the absorbance set.

[0011] S4. Calculate the differential absorbance based on the absorbance set and perform scattering and temperature compensation to obtain the normalized absorption characteristics after dual compensation of scattering and temperature.

[0012] S5. The normalized absorption characteristics and auxiliary measurement parameters are combined into a feature vector, which is used to input the pre-calibrated correction model to map the free hemoglobin concentration estimate, thereby obtaining the initial free hemoglobin concentration estimate.

[0013] S6. Perform a quality assessment on the initial free hemoglobin concentration estimate and, based on the quality assessment results, select direct output or adaptive retesting and backup regression to obtain the final free hemoglobin concentration and judgment result.

[0014] Preferably, the formula for calculating the initial free hemoglobin concentration estimate is as follows:

[0015]

[0016] Wherein, Cest: estimated free hemoglobin concentration (unit: g / L).

[0017] β0: Constant term (bias, unit: g / L).

[0018] β1: Linear sensitivity coefficient of normalized absorption characteristic X (unit: g / L after combining with X).

[0019] X: Normalized absorption characteristic, defined as the absorbance after differential, scattering, and temperature compensation divided by the equivalent optical path length Leff (if Leff is in cm, the physical dimension of X is cm). -1 ).

[0020] β2: Linearity coefficient of the scattering / turbidity term SCT (used to correct for bias caused by scattering, unit: g / L).

[0021] Sct: Scattering / turbidity reference (e.g., absorbance at 850 nm, or its normalized value, dimensionless or in absorbance units).

[0022] β3: Temperature linearity coefficient (unit: g·L) -1 ·℃ -1 ), used for first-order temperature correction.

[0023] Tenv: Sample / ambient temperature (unit: °C, typical range 15–40 °C).

[0024] ψj(V) and γj: nonlinear correction term and its weight, ψj is a basis function (could be neuron activation, tree model output, or polynomial interaction term, etc.) with input of feature vector V = [X, A577, A600, Sct, Tenv, Leff], and γj scales it to g / L.

[0025] J: number of nonlinear basis functions (determined by training complexity, typical small model J is 5-50);

[0026] The above formula combines the physically interpretable linear correction (absorption, scattering, temperature) with the nonlinear machine learning correction for complex coupling effects, retaining interpretability while improving quantitative accuracy in real samples.

[0027] Preferably, the multi-wavelength alternating irradiation and acquisition of continuous time-series photoelectric response comprises:

[0028] Alternately irradiating the sample to be measured with a first wavelength of 577 nm, a second wavelength of 600 nm, and a third wavelength of 850 nm, and simultaneously acquiring N frames of time-series data for each wavelength, to obtain the blood hemoglobin absorption channel signal and the scattering and turbidity reference signal, respectively, to obtain three-wavelength original signals S577, S600, and S850.

[0029] Preferably, the preprocessing step for the original time-series signal set comprises:

[0030] The time-series signal for each wavelength is first subtracted by the corresponding dark current Dλ and normalized by the blank reference I0λ, to calculate the normalized transmission intensity Iλ(t) = (Sλ(-t) - Dλ) / I0λ, to obtain Iλ(t);

[0031] Then, Iλ(t) is denoised using sliding median filtering or Savitzky-Golay filtering and averaged in the steady-state segment, to obtain the steady-state transmission value Tλ and calculate the absorbance Aλ = -log10(Tλ), to obtain the absorbance Aλ, λ ∈ {577, 600, 850}.

[0032] Preferably, the differential, scattering, and temperature compensation step comprises:

[0033] The differential absorbance ΔA_basic = A577 - A600 is calculated to preliminarily suppress background pigments, to obtain the differential absorbance ΔA_basic;

[0034] The A850 or normalized A850 is used as the scattering reference Sct to quantify the turbidity effect, to obtain the scattering reference Sct;

[0035] The linear correction ΔA_corr1 = ΔA_basic - α1·Sct is used for scattering compensation, to obtain the linear correction ΔA_corr1;

[0036] The sample ambient temperature T_env is measured, and a temperature correction ΔA_corr2 = ΔA_corr1 - α is made by temperature correction

[0037] 2·(T_env-T_ref) is used for temperature compensation to obtain a temperature correction ΔA_corr2;

[0038] Finally, the sample equivalent optical path L_eff is determined or calculated, and ΔA_corr2 is normalized to an absorption feature X = ΔA_corr2 / L_eff, which is used for subsequent quantitative mapping to obtain the absorption feature X.

[0039] Preferably, the correction and quality assessment step includes:

[0040] The feature vector V = [X, A577, A600, Sct, T_env, L_eff] is input into a fused correction model f_calib for mapping to an estimated free hemoglobin concentration C_est, the correction model f_calib being a combination of linear terms and machine learning nonlinear correction for consideration of interpretability and nonlinear compensation, to obtain the estimated free hemoglobin concentration C_est;

[0041] Then, the signal-to-noise ratio SNR and the model prediction uncertainty σ_pred are calculated, and a quality index Q = g(SNR, σ_pred) is formed for judging the measurement reliability, if Q ≥ Q_min, C_final = C_est is output and a determination result R is given; if Q < Q_min, adaptive retesting is performed:

[0042] The number of samplings M is increased on the original sample, and steps S2-S5 are repeated to obtain a free hemoglobin concentration retest value C_est2, if the retest still does not meet Q_min, a linear least squares multi-wavelength regression is performed with A577, A600, and A850 as independent variables to obtain a backup concentration C_backup, and the final concentration C_final and the determination result are selected according to the pre-set rule from the free hemoglobin concentration retest value C_est2 and the backup concentration C_backup.

[0043] A spectral detection system for detecting the free hemoglobin content of a plasma tube braid, comprising:

[0044] A sampling module: a section of a to-be-measured sample of the plasma tube braid is taken out, for obtaining a to-be-measured sample;

[0045] A multi-wavelength irradiation and collection module: the to-be-measured sample is alternately irradiated with multiple wavelengths, and N frames of time series data of each wavelength are collected, for obtaining an original time series signal;

[0046] A preprocessing module: dark current deduction, blank reference normalization, denoising and steady state determination are performed on the original time series signal, to obtain the steady state transmittance T and calculate the absorbance A, to obtain the absorbance set;

[0047] A compensation module: differential absorbance is calculated based on the absorbance set, and scattering and temperature compensation is performed using 850nm scattering reference and measured temperature, to obtain normalized absorption characteristics and normalize to absorption characteristics;

[0048] A quantitative mapping module: absorption characteristics and auxiliary measurement parameters are combined into a feature vector and input into a pre-calibrated correction model for mapping, to obtain an estimated concentration;

[0049] A quality determination and output module: the estimated concentration is evaluated for signal-to-noise ratio and prediction uncertainty, and direct output or adaptive retest, backup regression is determined accordingly, to obtain the final concentration and determination result.

[0050] A spectral detection device for detecting the free hemoglobin content of a plasma tube braid, comprising:

[0051] A multi-wavelength light source for alternately irradiating the plasma tube braid, with irradiation wavelengths of 577nm, 600nm and 850nm, to obtain a multi-wavelength irradiation beam;

[0052] An optical acquisition assembly for acquiring N consecutive frames of photoelectric response at each wavelength to obtain a set of original time series signals;

[0053] A signal preprocessing unit for dark current deduction and blank reference normalization on the original time series signal set, and denoising and steady state determination on the time series signal, to obtain steady state absorbance A;

[0054] A compensation processor for calculating differential absorbance based on the steady state absorbance A and performing scattering and temperature compensation, to obtain normalized absorption characteristics;

[0055] A correction estimation unit for combining normalized absorption characteristics and auxiliary parameters into a feature vector and inputting the feature vector into a pre-calibrated hybrid correction model to map to a free hemoglobin concentration estimate;

[0056] A quality evaluation and control unit for evaluating quality based on signal-to-noise ratio and model prediction uncertainty and deciding to directly output or trigger adaptive retest, backup regression to obtain the final concentration and determination result.

[0057] An electronic device, comprising:

[0058] At least one processor; and a memory connected in communication with the at least one processor; wherein:

[0059] The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method as described above.

[0060] A computer-readable storage medium, the medium being a computer-readable storage medium, the medium storing a computer program, the computer program being executed by a processor to implement the method as described above.

[0061] Compared with the prior art, the present application has the following beneficial effects:

[0062] (1) The present application directly detects the separated plasma tube braid without affecting the sealing and safety of the main blood bag, completely avoiding the pollution risk caused by puncture sampling. The objective and accurate quantitative results: using spectral analysis technology, the artificial subjective error is completely eliminated, especially for the judgment of critical value samples, which is extremely accurate, can prevent blood waste, and can eliminate the hidden danger of blood transfusion safety. By taking the normalized absorption characteristics after differential, scattering and temperature compensation as the dominant quantity, and on this basis, linear and nonlinear joint correction is carried out, so that the mapping of free hemoglobin concentration is closer to the physical law and can compensate for the nonlinear interference of the actual sample. Through a special scattering reference channel and scattering compensation, the optical deviation caused by particles, chyle or turbidity matrix can be directly absorbed and corrected, thereby reducing the error caused by matrix difference.

[0063] (2) The present application reduces the matrix interference by scattering and temperature compensation, improves the quantitative accuracy of complex samples by mixed correction model, reduces the risk of misjudgment and sample rejection by introducing uncertainty and retest strategy, and effectively eliminates the optical interference caused by turbidity, temperature fluctuation and other non-target factors in plasma samples by introducing differential absorbance calculation in the detection process and combining scattering reference and temperature compensation, so that the detection result of free hemoglobin is more stable and reliable. By establishing a mapping relationship combining physically interpretable normalized absorption characteristics and nonlinear correction model, not only the traceability and transparency of the result are retained, but also the nonlinear deviation under complex samples can be captured, thereby significantly improving the accuracy and robustness of the detection. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 A flow chart of a spectral detection method for detecting the free hemoglobin content of a plasma tube braid according to the present application;

[0065] Figure 2 A schematic diagram of a spectral detection device for detecting the free hemoglobin content of a plasma tube braid according to the present application;

[0066] Figure 3 A block diagram of the composition of a spectral detection system for detecting the free hemoglobin content of a plasma tube braid according to the present application. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present application will be clearly and completely described with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0068] Embodiment one:

[0069] Please refer to Figure 1 As shown in FIG. 1, a spectral detection method for detecting free hemoglobin content of a plasma tube braid includes:

[0070] Data acquisition:

[0071] Wavelength: λ1=577nm (main absorption of hemoglobin), λ2=600nm (reference absorption), λ3=850nm (scattering / turbidity reference).

[0072] Frame number: N=50 (collect 50 consecutive frames, and take the steady-state mean value).

[0073] The original signal Sλis collected (the dark current Dλand the blank I0λmay be measured and recorded in advance).

[0074] Preprocessing:

[0075] Normalization: Iλ=(Sλ-Dλ) / I0λ.

[0076] Steady-state transmission is averaged and denoised (Savitzky-Golay or sliding median) to obtain Tλ.

[0077] Absorbance: Aλ=-log10(Tλ).

[0078] Difference and compensation:

[0079] Difference absorbance: ΔAbasic=A577-A600

[0080] Scattering reference: Sct=A850

[0081] Scattering compensation (the method): ΔAcorr1=ΔAbasic-α1·Sct.

[0082] (Note: α1is a calibration coefficient; the control method takes α1=0)

[0083] Temperature compensation: ΔAcorr2=ΔAcorr1-α2(Tenv-Tref). In the embodiment, α2is small or 0.

[0084] Equivalent optical path: Leff (measured from transmission channel geometry, unit: cm).

[0085] Normalized feature: X = ΔAcorr2 / Leff.

[0086] Correction mapping:

[0087]

[0088] In this example, for demonstration, we take β0=0, β2=0, and only keep one small nonlinear correction γ1ψ1(V) (ψ1=Sct·X, γ1 is very small).

[0089] Baseline: only take linear term Cbaseline=βbXbasic (no scattering compensation, no ML_corr).

[0090] Quality assessment:

[0091] Calculate prediction error σ_pred (estimated from training residuals or historical data), if σ_pred exceeds threshold, trigger retest or backup regression.

[0092] Main parameters (values used in test):

[0093] Further, this example is specifically applied to laboratory batch:

[0094] Leff=0.5 cm; N=50.

[0095] Absorption constant (implicit, equivalent representation) ε=0.8 (so that ΔAtrue=0.8·c·Leff).

[0096] Scattering impact coefficient (real) ksc=0.30 (i.e. scattering causes absorbance about 0.30·Sct); control method is considered uncompensated (α1=0); this method fits compensation coefficient α1_real=0.20 (there are calibration residuals, reflecting that the actual cannot completely eliminate scattering).

[0097] Correction coefficient: this method β1=1.25, γ1=0.01 (ψ1=Sct·X); control method βb=1.20.

[0098] Temperature: Tenv=22℃ (this example does not explicitly use α2 / β3).

[0099] Example one (laboratory batch)

[0100] Leff=0.5 cm, ε=0.8→ΔAtrue=εCrefLeff=0.8·Cref·0.5=0.4Cref.

[0101] Scatter Bias (True) = ksc- Sct = 0.30- Sct

[0102] Collected ΔAbasic = ΔAtrue + ksc- Sct

[0103] The method of scattering compensation uses aΐ = 0.20 (actual calibration value), and ΔAcorr1 = ΔAbasic - aΐSct is obtained.

[0104] X = ΔAcorr1 / Leff.

[0105] The method predicts: Cest = 1.25- X + 0.01- (Sct- X).

[0106] Baseline: Cbaseline = 1.20- Xbasic (here Xbasic = ΔAbasic / Leff).

[0107] Sample 1 (low concentration):

[0108] Cref = 0.05 g / L, Sct = 0.02

[0109] ΔAtrue = 0.4- 0.05 = 0.02

[0110] scatter_bias = 0.30- 0.02 = 0.006

[0111] ΔAbasic = 0.02 + 0.006 = 0.026

[0112] Xbasic = 0.026 / 0.5 = 0.052

[0113] Baseline concentration: Cbaseline = 1.20 x 0.052 = 0.0624 g / L (error + 0.0124 g / L).

[0114] The method: ΔAcorr1 = 0.026 - 0.20 x 0.02 = 0.026 - 0.004 = 0.022

[0115] X = 0.022 / 0.5 = 0.044

[0116] Cest = 1.25 x 0.044 + 0.01 x (0.02 x 0.044) = 0.055 + 0.0000088 = 0.05501 g / L (error + 0.00501 g / L).

[0117] Sample 2 (medium concentration):

[0118] Cref = 0.15 g / L, Sct = 0.08

[0119] ΔAtrue = 0.4 · 0.15 = 0.06

[0120] scatter_bias = 0.30 · 0.08 = 0.024

[0121] ΔAbasic = 0.06 + 0.024 = 0.084

[0122] Xbasic = 0.084 / 0.5 = 0.168

[0123] Control: Cbaseline = 1.20 · 0.168 = 0.2016 g / L (error + 0.0516 g / L).

[0124] Method: ΔAcorr1 = 0.084 - 0.20 · 0.08 = 0.084 - 0.016 = 0.068

[0125] X = 0.068 / 0.5 = 0.136

[0126] Cest = 1.25 · 0.136 + 0.01 · (0.08 · 0.136) = 0.17 + 0.0001088 ≈ 0.17011 g / L (error + 0.02011 g / L).

[0127] Sample 3 (high concentration):

[0128] Cref = 0.35 g / L, Sc = 0.15

[0129] ΔAtrue = 0.4 · 0.35 = 0.14

[0130] scatter_bias = 0.30 · 0.15 = 0.045

[0131] ΔAbasic = 0.14 + 0.045 = 0.185

[0132] Xbasic = 0.185 / 0.5 = 0.37

[0133] Control: Cbaseline = 1.20 · 0.37 = 0.444 g / L (error + 0.094 g / L).

[0134] Method: ΔAcorr1 = 0.185 - 0.20 · 0.15 = 0.185 - 0.03 = 0.155

[0135] X = 0.155 / 0.5 = 0.31

[0136] Cest = 1.25 x 0.31 + 0.01 x (0.15 x 0.31) = 0.3875 + 0.000465 = 0.38797 g / L (error + 0.03797 g / L).

[0137] Example one statistical indicators (three samples);

[0138] Baseline error (absolute value): 0.0124, 0.0516, 0.094 → MAE = 0.0527 g / L; RMSE = 0.0623 g / L.

[0139] Method error (absolute value): 0.00501, 0.02011, 0.03797 → MAE = 0.0210 g / L; RMSE = 0.0250 g / L.

[0140] Improvement: RMSE is reduced from 0.0623 to 0.0250, a decrease of about 59.9%.

[0141] As can be seen from the above, by directly detecting the separated plasma tube braid, the sealing and safety of the main blood bag are not affected at all, and the pollution risk caused by puncture sampling is completely avoided.

[0142] Objective and accurate, quantitative results: using spectral analysis technology, completely eliminates human subjective error, especially for critical value sample judgment is extremely accurate, both to prevent waste of blood, but also to eliminate the hidden danger of blood transfusion safety.

[0143] Fast detection speed and high efficiency: single detection can be completed in a short time, much faster than manual comparison, very suitable for rapid screening of large quantities of plasma in blood stations, significantly improving work efficiency.

[0144] High integration, simple operation: the device is a dedicated desktop terminal, compact structure, "one-key" operation, greatly reducing the skill requirements for operators, easily integrated into existing workflows.

[0145] Example two:

[0146] This embodiment is specifically applied to portable / field analysis;

[0147] Leff = 0.3 cm; N = 30 (reduced sampling times).

[0148] ε = 0.8 (as above).

[0149] ΔAtrue = ε·Cref·Leff = 0.8·Cref·0.3 = 0.24Cref

[0150] Scattering true coefficient ksc = 0.05 (the absolute amount of scattering of this field device is small). Compensation coefficient α

[0151] 1 real = 0.04. Correction factor same as control (this method β1 = 1.25, baseline βb = 1.18)

[0152] Scatter coefficient ksc = 0.05 (absolute amount of scattering is small due to short optical path and different device design); this method α1 real = 0.04 (partially compensated).

[0153] Correction factors: this method β1 = 1.25, γ1 = 0.01; control method βb = 1.18.

[0154] Temperature: Tenv = 25°C.

[0155] Test sample:

[0156] Low concentration: Cref = 0.05 g / L; Sct = 0.02

[0157] Medium concentration: Cref = 0.15 g / L; Sct = 0.08

[0158] High concentration: Cref = 0.35 g / L; Sct = 0.15

[0159] Note: The Sct values above represent different degrees of turbidity / chyle, and in actual testing Sct is measured directly from the absorbance at 850 nm.

[0160] Sample 1 (low concentration):

[0161] Cref = 0.05, Sct = 0.03

[0162] ΔAtrue = 0.24 x 0.05 = 0.012

[0163] scatter_bias = 0.05 x 0.03 = 0.0015

[0164] ΔAbasic = 0.0135

[0165] Xbasic = 0.0135 / 0.3 = 0.045

[0166] Control: Cbaseline = 1.18 x 0.045 = 0.05310 g / L (error + 0.00310).

[0167] This method: ΔAcorr1 = 0.0135 - 0.04 x 0.03 = 0.0135 - 0.0012 = 0.0123

[0168] X = 0.0123 / 0.3 = 0.041

[0169] Cest = 1.25 x 0.041 + 0.01 x (0.03 x 0.041) = 0.05125 + 0.0000123 = 0.05126 g / L (error + 0.00126).

[0170] Sample 2 (medium concentration):

[0171] Cref = 0.15, Sct = 0.10

[0172] ΔAtrue = 0.24 x 0.15 = 0.036

[0173] scatter_bias = 0.005

[0174] ΔAbasic = 0.041

[0175] Xbasic = 0.1366667

[0176] Control: Cbaseline = 1.18 x 0.1366667 = 0.16133 g / L (error + 0.01133).

[0177] Method: ΔAcorr1 = 0.041 - 0.04 x 0.10 = 0.041 - 0.004 = 0.037

[0178] X = 0.1233333

[0179] Cest = 1.25 x 0.1233333 + 0.01 x (0.10 x 0.1233333) = 0.15429 g / L (error + 0.00429).

[0180] Sample 3 (high concentration):

[0181] Cref = 0.35, Sct = 0.18

[0182] ΔAtrue = 0.24 x 0.35 = 0.084

[0183] scatter_bias = 0.009

[0184] ΔAbasic = 0.093

[0185] Xbasic = 0.31

[0186] Control: Cbaseline = 1.18 x 0.31 = 0.3658 g / L (error + 0.0158).

[0187] Method: ΔAcorr1 = 0.093 - 0.04 x 0.18 = 0.093 - 0.0072 = 0.0858

[0188] X = 0.286

[0189] Cest = 1.25 x 0.286 + 0.01 x (0.18 x 0.286) = 0.35802 g / L (error + 0.00802).

[0190] Statistical indicators of Example Two (three samples):

[0191] Baseline error (absolute value): 0.00310, 0.01133, 0.0158 -> MAE = 0.0101 g / L; RMSE = 0.0114 g / L.

[0192] Method error: 0.00126, 0.00429, 0.00802 -> MAE = 0.00453 g / L; RMSE = 0.00536 g / L.

[0193] Improvement: RMSE is reduced from 0.0114 to 0.00536, with a reduction of about 53.0%.

[0194] The comprehensive comparison between Example One, Example Two and their respective baseline methods is shown in Table 1 below:

[0195] The RMSE and MAE in Table 1 are both in units of g / L; "Improvement (%) " is calculated according to the relative reduction of RMSE.

[0196] Table 1

[0197]

[0198]

[0199] As can be seen from the above, by taking the normalized absorption characteristics after differential, scattering and temperature compensation as the dominant quantity, and performing linear and nonlinear joint correction on this basis, the mapping of free hemoglobin concentration is closer to the physical law and can compensate for the nonlinear interference of actual samples.

[0200] By using a special scattering reference channel and implementing scattering compensation, the optical deviation caused by particles, chyle or turbid matrix can be directly absorbed and corrected, thereby reducing the error caused by matrix differences.

[0201] Taking temperature as an explicit correction quantity and correcting it in the model can reduce the deviation of measurement under different temperature conditions, so that the method remains stable within the room temperature and clinical / on-site temperature range.

[0202] The hybrid model of "physical linear term + machine learning nonlinear correction" not only retains the interpretable linear part based on Beer-Lambert and other physical relationships, but also captures complex coupling effects through nonlinear terms, facilitating bias diagnosis and regulatory compliance.

[0203] Example Three:

[0204] As shown in Figure 2 the device and initial configuration:

[0205] Hardware: Three-channel narrow-band LED light source (577 nm, 600 nm, 850 nm); silicon photodetector or small spectrometer; transmission sample cell (equivalent optical path Leff can be selected from 0.3-0.5 cm); temperature sensor (accuracy ±0.5℃); master CPU (or MCU + coprocessor); local storage and network interface.

[0206] Software: Embedded pre-processing module (dark current / blank normalization, filtering), compensation and correction module (a1, a2, hybrid model f_calib), quality assessment module (SNR, sigma_pred estimation), log and upload module (LIS / server).

[0207] Parameters: Sampling frame number N = 50; filter: Savitzky-Golay window 11; upper limit of retest number M = 3; quality threshold Qmin: SNR ≥ 10 and sigma_pred ≤ 0.05 g / L.

[0208] Process steps:

[0209] Step 1 - Sampling and loading:

[0210] Sampling: The technician takes a section of the sample to be tested from the plasma tube braid and places it in the device sample cell, which is used to obtain the sample to be tested, resulting in the sample to be tested (containing sample ID and metadata).

[0211] Key points: The barcode is associated with the sample information, and the sampling time and operator are recorded.

[0212] Step 2 - Multi-wavelength irradiation and raw acquisition:

[0213] Multi-wavelength irradiation and acquisition: The device alternately irradiates the sample to be tested at 577 nm, 600 nm, and 850 nm, and acquires N consecutive photovoltaic responses at each wavelength, which is used to obtain the original time series signal set Sλ(t), resulting in Sraw = {S577(t), S600(t), S850(t)}.

[0214] Parameters: N = 50, interval trigger to ensure frame synchronization and light source synchronization.

[0215] Dark current (Dλ) and blank reference (I 0λ) are measured periodically and cached simultaneously.

[0216] Step 3 - Preprocessing (dark current subtraction, normalization, denoising):

[0217] Preprocessing: Dark current subtraction and blank reference normalization are performed on the time series signal of each wavelength, and the time series is filtered and steady state judged to obtain steady state transmission T and calculate absorbance A, to obtain the absorbance set A = {A577, A600, A850}.

[0218] Equation:

[0219] I (t) = S (t) - D I 0, T = mean (steady state section of I (t))

[0220] A = -log10(T)

[0221] Denoising: Savitzky-Golay filtering (window 11, 3rd order polynomial) or sliding median filtering.

[0222] Step 4 - Differential calculation and scattering / temperature compensation

[0223] Differential and compensation: Calculate the differential absorbance based on A and compensate with 850 nm as the scattering reference and temperature measurement to obtain the double-compensated normalized absorption feature X, to obtain ΔAcorr2 and X.

[0224] Equation and steps:

[0225] Differential: ΔAbasic = A577 - A600

[0226] Scattering reference: Sct = A850

[0227] Scattering compensation: ΔAcorr1 = ΔAbasic - α1 · Sct (α1 is a calibration coefficient, stored in the device)

[0228] Temperature compensation: Measure the temperature Tenv, and perform ΔAcorr2 = ΔAcorr1 - α2 · (Tenv - Tref).

[0229] Normalization: X = ΔAcorr2 / Leff.

[0230] Explanation: α1, α2 are obtained by calibration and can be updated periodically.

[0231] Step 5 - Feature vector formation and quantitative mapping:

[0232] Feature formation and mapping: combine X with auxiliary parameters A577, A600, Sct, Tenv, Leff into a feature vector V, and input V into the pre-calibrated hybrid correction model fcalib (linear + non-linear terms) for mapping to concentration estimate Cest, resulting in an initial concentration Cest.

[0233] Model implementation: embed a light-weight ML model (e.g. regression tree or small feed-forward network), model file calibrated by engineers and deployed to devices or returned results after inference by central server.

[0234] Step 6 - Quality assessment, re-measurement and final output:

[0235] Quality assessment and output: compute quality indicator Q based on SNR and model output uncertainty σ_pred, for deciding whether to output directly or trigger adaptive re-measurement and fallback regression, final output Cfinal and decision result R.

[0236] Indicator computation:

[0237]

[0238] Re-measurement strategy: if Q is REMEASURE, device automatically re-acquire additional M times (preferably M=2), repeat steps 2-5; if still unqualified, enable fallback linear regression and mark result as "need human review".

[0239] Output content: Cfinal (g / L), σ_pred, Q, decision R (qualified / scrap / review), all intermediate quantities (Aλ, X, V) and timestamps, write to local log and upload to LIS / server.

[0240] Calibration and validation (implementation details)

[0241] Initial calibration: prepare no less than 8 reference samples with known concentrations (cover target detection interval), measure according to device procedure and fit linear part (solve β0, β1) and scattering coefficients α1, α2, then train ML correction term (least squares + L2 regularization or cross-validation to select model hyperparameters).

[0242] Validation: evaluate RMSE / MAE with 10-20 independent reference samples, and set σ_pred model to estimate single prediction uncertainty.

[0243] Periodic maintenance: device performs blank and dark measurement daily; calibration review every month or every batch; when offline-trained model passes regression test, central server deploys new model and records version change.

[0244] Quality control and audit records

[0245] Each measurement record: sample ID, operator, ambient temperature, I0, Dλ, Tλ, Aλ, Sct, X, V, C_est, σ_pred, Q, retest record, model version number, timestamp.

[0246] Log chain / upload: periodically aggregate logs and upload to database for long-term trend analysis and model retraining.

[0247] Abnormal handling: if a device has system drift (calibration offset exceeds the allowed range), it is automatically marked and enters maintenance mode, stopping the issuance of results until the engineer completes the repair.

[0248] As can be seen from the above, by reducing the matrix interference mixed correction model through scattering and temperature compensation, the quantitative accuracy under complex samples is improved; by introducing uncertainty and retest strategy, the risk of misjudgment and sample rejection is reduced;

[0249] By introducing differential absorbance calculation in the detection process and combining scattering reference and temperature compensation, the optical interference caused by non-target factors such as turbidity and temperature fluctuation in plasma samples can be effectively eliminated, making the detection results of free hemoglobin more stable and reliable.

[0250] By establishing a mapping relationship between the physically interpretable normalized absorption characteristic and the nonlinear correction model, not only the traceability and transparency of the results are preserved, but also the nonlinear deviation under complex samples can be captured, thereby significantly improving the accuracy and robustness of the detection.

[0251] Example Four:

[0252] As shown in Figure 3 , the on-site / blood bank level detection system:

[0253] Implementation purpose: taking several front-end detection devices as nodes, a detection system for daily screening of blood banks is established, covering sample flow, data flow, quality control flow and model maintenance.

[0254] The system includes:

[0255] Sampling and receiving:

[0256] Implementation content: the front-end nurse / technician collects the plasma tube sample according to the process and pastes a unique barcode (sample ID), puts the sample into the front-end detection device at the receiving station and scans the code for registration, obtains the sample to be tested and metadata (sampling time, operator, temperature label, etc.).

[0257] Key points: the barcode is connected with the LIS (laboratory information system) to realize sample traceability.

[0258] Multi-wavelength irradiation and local collection:

[0259] Implementation: The front-end device performs 577 / 600 / 850nm alternate irradiation and collects N frames of signals according to the configuration, generates S_raw and caches locally; meanwhile, it uploads the original timestamp and device ID to the system server for auditing.

[0260] Key points: Sampling parameters (N, exposure) can be issued by the server according to batch / environment.

[0261] Local preprocessing and preliminary upload:

[0262] Implementation: The device performs dark current subtraction, blank normalization and denoising locally to obtain Aλ and uploads the compressed feature vector V to the central server; if the network is unavailable, it will be queued for upload.

[0263] To protect privacy, data is encrypted before transmission and only necessary features (or de-identified aggregate data) are uploaded.

[0264] Central compensation and mapping service:

[0265] Implementation: The central server receives feature vectors V from multiple devices, performs batch inference using the centralized model f_calib (which can be deployed on the server or cloud), obtains C_est and returns it together with the model confidence σ_pred; the server is responsible for unified version management and can remotely issue new models.

[0266] Centralized inference facilitates unified model update, performance statistics and reduces the computational burden of the front end (edge computing can also be selected to reduce latency).

[0267] Quality control, retest strategy and manual review:

[0268] Implementation: The server automatically decides whether to trigger retesting (indicating the front-end device to collect again) or mark the sample as "manual review" according to the Q indicator; quality control personnel view the sample history, original absorbance spectrum and model output in the system and make the final decision and register in the LIS.

[0269] The system can set automatic disposal rules for different risk levels (e.g. critical zone automatic retest, high risk direct manual review).

[0270] Data management, model update and regulatory reporting (extension):

[0271] Implementation: The system periodically aggregates artificial review samples with reference concentrations into a training set, performs offline retraining and pushes the new model version to the front end after regression testing with the validation suite; it also automatically generates calibration / control reports for regulatory audit.

[0272] The model update process must include change control, rollback mechanism and verification report to meet regulatory requirements.

[0273] Deployment and operation:

[0274] Interface: LIS, barcode scanner, printer, network (wired / Wi-Fi), user management and log system.

[0275] Security: transmission encryption, access control, audit record, model version management and signature.

[0276] Maintenance: regular calibration of equipment, monthly review of model performance by the center, establishment of emergency rollback and abnormal alarm mechanism.

[0277] As can be seen from the above, the implementation of centralized model management and unified quality control facilitates the realization of high consistency in the blood bank / hospital level; the automatic retest and manual review process reduces the risk of misjudgment; and the centralized management of data is conducive to long-term model optimization and regulatory compliance;

[0278] An electronic device, the electronic device comprising:

[0279] at least one processor; and a memory connected to the at least one processor in communication; wherein:

[0280] The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described above.

[0281] A computer-readable storage medium, the medium being a computer-readable storage medium, the medium storing a computer program, the computer program being executed by a processor to implement the method described above.

[0282] Although embodiments of the present application have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions and alterations can be made thereto without departing from the principles and spirit of the present application, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A spectroscopic method for detecting the free hemoglobin content of a plasma tube braid, comprising: The method comprises the following steps: A section of the sample to be measured is taken out from the plasma tube braid to obtain an original optical measurement object, and the sample to be measured is obtained. The sample to be measured is alternately irradiated with multiple wavelengths, and continuous time-series photoelectric responses are collected to obtain an original time-series signal set. Dark current deduction and blank reference normalization are performed on the original time-series signal set, and time-series signal denoising and steady-state determination are performed to obtain steady-state transmission values Tλ and convert them into absorbance Aλ to obtain an absorbance set. Based on the absorbance set, differential absorbance is calculated, and scattering and temperature compensation are performed to obtain a normalized absorption feature that is double compensated for scattering and temperature. The normalized absorption feature is combined with an auxiliary measurement parameter to form a feature vector, which is input into a pre-calibrated correction model to be mapped into an estimated value of free hemoglobin concentration to obtain an initial estimated value of free hemoglobin concentration. Quality evaluation is performed on the initial estimated value of free hemoglobin concentration, and based on the quality evaluation result, direct output or adaptive re-measurement and backup regression are selected to obtain a final free hemoglobin concentration and a determination result.

2. The method according to claim 1, wherein the method is characterized by, The calculation formula of the initial estimated value of free hemoglobin concentration is as follows: Wherein, Cest: estimated free hemoglobin concentration, unit: g / L; β0: constant term, unit: g / L; β1: linear sensitivity coefficient of normalized absorption feature X, unit: g / L; X: normalized absorption feature, defined as the absorbance after differential, scattering and temperature compensation divided by the equivalent optical path Leff; Sct: scattering, turbidity reference; β2: linear coefficient of scattering, turbidity term Sct; β3: temperature linear coefficient, unit: g.L -1 • °C -1 , for first order temperature correction; Tenv: sample / environment temperature, unit: ℃; ψj(V) and γj: nonlinear correction term and its weight, ψj is a basis function with the feature vector V=[X, A577, A600, Sct, Tenv, Leff] as input, and γj scales it to g / L; J: number of nonlinear basis functions.

3. The method according to claim 1, wherein the method is characterized by, The multiple-wavelength alternately irradiation and collection of continuous time-series photoelectric responses comprise: The sample to be measured is alternately irradiated with a first wavelength of 577 nm, a second wavelength of 600 nm and a third wavelength of 850 nm, and continuous N frames of time-series data of each wavelength are collected simultaneously to obtain hemoglobin absorption channel signals and scattering, turbidity reference signals respectively, thereby obtaining three-wavelength original signals S577, S600 and S850.

4. The method according to claim 1, wherein the method is characterized by, The preprocessing steps of the original time-series signal set are as follows: The time-series signal of each wavelength is first deducted by the corresponding dark current Dλ and normalized by the blank reference I0λ to calculate the normalized transmission intensity Iλ(t)=(Sλ(-t)-Dλ) / I0λ, thereby obtaining Iλ(t); Then, Iλ(t) is denoised by using sliding median filtering or Savitzky-Golay filtering, and the average value in the steady-state section is taken to obtain the steady-state transmission value Tλ and calculate the absorbance Aλ=-log10(Tλ), thereby obtaining the absorbance Aλ, λ∈{577, 600, 850}.

5. The spectroscopic detection method for detecting the content of free hemoglobin in plasma according to claim 4, characterized in that, The differential, scattering and temperature compensation steps comprise: The differential absorbance ΔA_basic=A577-A600 is calculated to preliminarily suppress the background pigment, thereby obtaining the differential absorbance ΔA_basic; With A850 or normalized A850 as scattering reference Sct for quantifying turbidity effect, scattering reference Sct is obtained; With linear correction ΔA_corr1 = ΔA_basic - α1·Sct for scattering compensation, linear correction ΔA_corr1 is obtained; With measured sample ambient temperature T_env for temperature compensation, temperature correction ΔA_corr2 is obtained: ΔA_corr2 = ΔA_corr1 - α2·(T_env - T_ref) Finally, sample equivalent optical path L_eff is determined or calculated and ΔA_corr2 is normalized to absorption feature X = ΔA_corr2 / L_eff for subsequent quantitative mapping, absorption feature X is obtained.

6. The method according to claim 1, wherein the method is characterized by, The correction and quality evaluation steps include: The feature vector V = [X, A577, A600, Sct, T_env, L_eff] is input into the fused correction model f_calib for mapping to the estimated free hemoglobin concentration C_est, the correction model f_calib is a combination of linear terms and machine learning nonlinear correction for consideration of interpretability and nonlinear compensation, the estimated free hemoglobin concentration C_est is obtained; Then, the signal-to-noise ratio SNR and the model prediction uncertainty σ_pred are calculated, and the quality index Q = g(SNR, σ_pred) is formed for judging the measurement reliability, if Q ≥ Q_min, C_final = C_est is output and the determination result R is given; if Q < Q_min, adaptive retest is performed: Increase the sampling number M on the original sample and repeat the above steps to obtain the free hemoglobin concentration retest value C_est2, if the retest still does not meet Q_min, linear least squares multi-wavelength regression is performed with A577, A600 and A850 as independent variables to obtain the backup concentration C_backup, and the final concentration C_final and the determination result are selected according to the pre-set rule from the free hemoglobin concentration retest value C_est2 and the backup concentration C_backup.

7. A spectroscopic detection system for detecting free hemoglobin content in a braid of plasma tubes, comprising: Including: Sampling module: taking out a section of the sample to be measured from the plasma tube braid for obtaining the sample to be measured; Multi-wavelength irradiation and collection module: alternately irradiating the sample to be measured with multiple wavelengths and collecting N frames of time series data for each wavelength to obtain the original time series signal; Preprocessing module: dark current subtraction, blank reference normalization, denoising and steady state judgment are performed on the original time series signal to obtain the steady state transmission value Tλ and calculate the absorbance Aλ, and the absorbance set is obtained; Compensation module: based on the absorbance set, the differential absorbance is calculated, and the scattering and temperature compensation are performed with the 850nm scattering reference and the measured temperature to obtain the normalized absorption feature and normalize it to the absorption feature; Quantitative mapping module: the absorption feature and the auxiliary measurement parameters are combined into a feature vector and input into the pre-calibrated correction model for mapping to obtain the estimated concentration; Quality determination and output module: the signal-to-noise ratio and the prediction uncertainty of the obtained estimated concentration are evaluated, and the direct output or the adaptive retest and the backup regression are determined accordingly to obtain the final concentration and the determination result.

8. A spectroscopic detection device for detecting the free hemoglobin content of a plasma tube braid, characterized by Including: A multi-wavelength light source is used to alternately irradiate the plasma tube braid, and the irradiation wavelengths are 577 nm, 600 nm and 850 nm, so as to obtain a multi-wavelength irradiation light beam; An optical acquisition component is used to acquire N continuous frames of photoelectric responses at each wavelength, so as to obtain a set of original time sequence signals; A signal preprocessing unit is used to perform dark current deduction and blank reference normalization on the set of original time sequence signals, and to perform time sequence signal denoising and steady state determination, so as to obtain steady state absorbance Aλ; A compensation processor is used to calculate differential absorbance based on the steady state absorbance Aλ, and to perform scattering and temperature compensation, so as to obtain normalized absorption characteristics; A correction estimation unit is used to combine the normalized absorption characteristics and auxiliary parameters into a feature vector, and to input the feature vector into a pre-labeled mixed correction model to map to a free hemoglobin concentration estimation value; A quality evaluation and control unit is used to evaluate quality based on a signal-to-noise ratio and a model prediction uncertainty, and to determine direct output or trigger adaptive retesting, backup regression to obtain a final concentration and a determination result.

9. An electronic device, comprising: The electronic device comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein: the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, The medium is a computer readable storage medium, and the medium stores a computer program, and the computer program is executed by a processor to implement the method of any one of claims 1-6.

Images