A wound bleeding multi-mode FBG spectrum feature fusion detection method

Abstract

The application discloses a kind of wound bleeding multimode FBG spectrum feature fusion detection methods, it is related to wound bleeding detection technical field, including the following steps: based on the medical dressing of integrated FBG sensor carries out bleeding simulation, corresponding spectrum data is collected under different bleeding amount, and corresponding spectrum feature is extracted, and original spectrum feature data is obtained;Data screening is carried out, and error data is removed, and feature analysis is carried out, and bleeding related features are screened out;Wound bleeding detection model is established;The spectrum data of wound is collected using the medical dressing of integrated FBG sensor, and wound bleeding is detected based on wound bleeding detection model;The application is used to solve the problem that existing wound bleeding detection technology cannot continuously and real-timely detect the bleeding amount of wound by integrated FBG sensor to collect spectrum features without uncovering the dressing when detecting the bleeding of wound with medical dressing.

Description

Technical Field

[0001] This invention relates to the field of wound bleeding detection technology, specifically a detection method for wound bleeding by fusing multimodal FBG spectral features. Background Technology

[0002] Wound bleeding detection technology is an interdisciplinary technology that integrates medicine, electronic engineering, computer science and materials science. It refers to a series of technical methods that use physical, chemical or biological means to non-invasively or minimally invasively sense, identify, quantify and dynamically monitor the bleeding status of a wound.

[0003] Current wound bleeding detection technologies often require removing the dressing and visually inspecting or using machine vision to detect bleeding in wounds covered with medical dressings. However, this method can easily cause secondary damage to the wound during dressing removal, prolonging the healing period, increasing patient pain, and posing a risk of wound infection, thus affecting the patient's medical experience. Furthermore, this method can only achieve intermittent detection, not continuous real-time monitoring. For example, medical staff cannot frequently remove the dressing for examination, resulting in a significant time blind spot in bleeding detection. By the time bleeding symptoms are observed, substantial bleeding may have already occurred, missing the optimal clinical intervention window and increasing treatment risks. Furthermore, manual visual observation is highly subjective and prone to error, making it impossible to obtain specific bleeding amounts and thus failing to provide objective and quantitative evidence for clinical decision-making. While machine vision can detect specific bleeding amounts to some extent, it also requires removing the dressing to obtain wound images, and it cannot provide continuous real-time monitoring. Moreover, the accuracy of machine vision is easily affected by factors such as lighting conditions, wound secretions, and shooting angle, making it difficult to guarantee the stability and reliability of the detection results. Therefore, existing wound bleeding detection technologies cannot continuously and in real-time detect the amount of bleeding in wounds with medical dressings without removing the dressing and by collecting spectral features through an integrated FBG sensor. Summary of the Invention

[0004] This invention aims to at least partially address one of the technical problems in the prior art. It simulates bleeding using a medical dressing with an integrated FBG sensor, collects corresponding spectral data under different bleeding amounts, extracts corresponding spectral features to obtain raw spectral feature data, performs data filtering to remove erroneous data, and conducts feature analysis to filter bleeding-related features. A wound bleeding detection model is established. The invention utilizes the medical dressing with the integrated FBG sensor to collect spectral data of the wound and detects wound bleeding based on the wound bleeding detection model. This addresses the problem that existing wound bleeding detection technologies cannot continuously and in real-time detect the amount of bleeding without removing the dressing, using the integrated FBG sensor to collect spectral features.

[0005] To achieve the above objectives, this application provides a method for detecting multimodal FBG spectral feature fusion of wound bleeding, comprising the following steps: Bleeding simulation was performed on medical dressings with integrated FBG sensors. Spectral data were collected under different bleeding volumes, and corresponding spectral features were extracted to obtain raw spectral feature data. The original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features and obtain relevant feature information. A wound bleeding detection model was established based on relevant feature information and corresponding standard spectral feature data. Spectral data of the wound were collected using a medical dressing with an integrated FBG sensor, and wound bleeding was detected based on a wound bleeding detection model.

[0006] Furthermore, bleeding simulation was performed using a medical dressing with an integrated FBG sensor. Spectral data was collected at different bleeding volumes, and corresponding spectral features were extracted to obtain the raw spectral feature data. This process included the following sub-steps: Medical dressings with integrated FBG sensors are referred to as FBG dressings, and any type of FBG dressing is referred to as Type I dressing. Set up a simulated wound to simulate bleeding, and make the first type of dressing fit tightly to the simulated wound; set the bleeding range for the bleeding simulation, and select multiple points evenly from the bleeding range, and record them as simulated quantity 1 to simulated quantity n in ascending order.

[0007] Furthermore, bleeding simulation was performed using a medical dressing with an integrated FBG sensor. Spectral data was collected at different bleeding volumes, and corresponding spectral features were extracted to obtain the raw spectral feature data. This process included the following sub-steps: For analog quantity 1, the simulated bleeding from the simulated wound reaches analog quantity 1 and is absorbed by the first type of dressing, and the corresponding spectral data is collected by the integrated FBG sensor. Multiple spectral features are set, and two or more spectral features are selected and fused to obtain multiple fused features. The set spectral features and fused features are collectively referred to as original features, and are sequentially referred to as original feature 1 to original feature m, where m is the total number of original features. Based on the acquired spectral data, the corresponding original features 1 to m are extracted and denoted as the corresponding original feature group; multiple simulations are performed based on analog quantity 1, and the corresponding spectral data is acquired in each simulation, and the corresponding original feature group is extracted; the resulting multiple original feature groups are denoted as the original feature data of analog quantity 1. Repeated simulations are used to obtain the original characteristic data of all simulated quantities, thus obtaining the original spectral characteristic data.

[0008] Furthermore, the original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features, obtaining relevant feature information. This process includes the following sub-steps: The original feature data of analog quantity 1 is denoted as the first feature data; a reasonable range and a reasonable sign are set for each original feature; data in the first feature data that are not within the corresponding reasonable range or do not meet the corresponding reasonable sign are filtered out and denoted as absolute error data; Mark the original feature groups containing absolute error data in the first feature data as absolute error groups; denote the remaining original feature groups as the second feature data; Extract all original features 1 from the second feature data and calculate the corresponding median M1 and median absolute deviation MD1. Record [M1-3×MD1, M1+3×MD1] as the reasonable range of the original feature 1 values. Based on the second feature data, repeatedly obtain the reasonable range of the original features values. Data in the first feature data that does not fall within the corresponding reasonable range of values ​​are recorded as numerical error data; the original feature groups containing numerical error data in the second feature data are marked as numerical error groups; and the remaining original feature groups are recorded as the third feature data.

[0009] Furthermore, the original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features and obtain relevant feature information. This process also includes the following sub-steps: Extract all original features 1 from the third feature data and calculate the corresponding mean AP1 and variance R1; denote 1 / R1 as the original weight of original feature 1; Based on the third feature data, the average value of all original features is repeatedly obtained and recorded as AP1 to APn in sequence according to the original feature number; and the original weights of all original features are repeatedly obtained and normalized to obtain the feature weights of all original features. Combine AP1 to APn into an n-dimensional vector, denoted as the cluster center C, C = (AP1, AP2, ..., APn); denote any original feature group in the third feature data as the first feature group; The first feature group is also combined into a corresponding n-dimensional vector, and the weighted Mahalanobis distance to the cluster center C is calculated based on the feature weights of the original features, which is denoted as the weighted distance of the first feature group. Repeatedly calculate the weighted distance of all original feature groups in the third feature data; and select a confidence level α, look up the chi-square distribution table with n degrees of freedom, and obtain the critical value corresponding to the confidence level α, denoted as AX; The original feature groups with a weighted distance greater than √AX are marked as distance outliers and removed from the third feature data. The remaining third feature data are recorded as the standard feature data of analog quantity 1. Standard spectral characteristic data are obtained by repeatedly acquiring standard characteristic data of all analog quantities.

[0010] Furthermore, the original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features and obtain relevant feature information. This process also includes the following sub-steps: Arrange all analog quantities in ascending order and denote them as the blood volume sequence; for original feature 1, extract all original feature 1 from the standard spectral feature data and divide them according to the corresponding analog quantities to obtain n feature subsets; Randomly select one data point from each feature subset and form it as a combination, denoted as the feature value combination; repeat this process to obtain all insufficient feature value combinations; and denote any feature combination as the first combination. The first combination is arranged in ascending order of the corresponding analog quantities and is denoted as the first feature sequence. The Pearson correlation coefficient, Spearman correlation coefficient and maximum information coefficient between the first feature sequence and the bleeding volume sequence are denoted as XM1, XM2 and XM3 respectively. Calculate q1×|XM1|+q2×|XM2|+q3×XM3, and denote it as the comprehensive coefficient of the first combination; repeat the calculation of the comprehensive coefficients of all feature value combinations to obtain the set of comprehensive coefficients of the original feature 1, where q1, q2 and q3 are the set weights.

[0011] Furthermore, the original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features and obtain relevant feature information. This process also includes the following sub-steps: Arrange the set of comprehensive coefficients of the original feature 1 in ascending order and denote it as the comprehensive correlation sequence; calculate the 10th percentile, 50th percentile and 90th percentile of the comprehensive correlation sequence and denote them as Q10, Q50 and Q90 respectively; calculate 0.2×Q10+0.3×Q50+0.5×Q90 and denote it as the weighted basic coefficient HF; Calculate the 5th percentile Q5, 25th percentile Q25, 75th percentile Q75, and 95th percentile Q95 of the composite correlation sequence, and calculate the coefficients of dispersion CV1 and CV2, where CV1 = (Q75 - Q25) / Q50 and CV2 = (Q95 - Q5) / Q50; calculate 0.6 × |CV1| + 0.4 × |CV2|, denoted as BV; and calculate the stability correction coefficient EA, where EA = 0.2 + 0.8 × e -3×VD VD = tanh(BV), where tanh() represents the tanh function.

[0012] Furthermore, the original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features and obtain relevant feature information. This process also includes the following sub-steps: Calculate the skewness SK and kurtosis KU of the aggregate coefficient set, and calculate 1 + 0.2 × tanh (SK) + 0.1 × tanh (KU-3), denoted as the distribution correction coefficient EB; Calculate HF×EA×EB, and record it as the final correlation coefficient of the original feature 1; repeatedly obtain the final correlation coefficients of all original features; and select the k1 original features with the largest final correlation coefficients, mark them as bleeding-related features, and record all bleeding-related features as related feature information, where k1 is the set number.

[0013] Furthermore, establishing a wound bleeding detection model based on relevant feature information and corresponding standard spectral feature data includes the following sub-steps: Remove the data corresponding to non-hemorrhagic features from the standard spectral feature data, and record the remaining data as hemorrhage feature simulation data; A multilayer perceptron was used as the initial detection model. The input of the initial detection model was set as bleeding-related features, and the output was the amount of bleeding. The initial detection model was trained using simulated data of bleeding features, and the wound bleeding detection model was obtained after the training was completed.

[0014] Furthermore, the use of medical dressings with integrated FBG sensors to collect spectral data of the wound, and the detection of wound bleeding based on a wound bleeding detection model, includes the following sub-steps: The first type of dressing is applied to the wound to be tested, and the corresponding spectral data is collected periodically. The corresponding bleeding-related features are extracted and input into the wound bleeding detection model to obtain the corresponding bleeding amount, which is recorded as the bleeding detection amount. If the obtained bleeding detection value is less than e0, the bleeding detection value is adjusted to 0 and output; if it is not less than e0, it is not adjusted and output, where e0 is the set threshold.

[0015] The beneficial effects of this invention are as follows: This invention simulates bleeding using a medical dressing with an integrated FBG sensor. It collects corresponding spectral data under different bleeding volumes and extracts corresponding spectral features to obtain raw spectral feature data. The raw spectral feature data is then filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is performed to filter bleeding-related features, obtaining relevant feature information. A wound bleeding detection model is established based on the relevant feature information and the corresponding standard spectral feature data. The medical dressing with an integrated FBG sensor collects spectral data of the wound and detects bleeding based on the wound bleeding detection model. When detecting bleeding on a wound with a medical dressing applied, the integrated FBG sensor can collect spectral features without removing the dressing, enabling continuous real-time detection of the bleeding volume, thus improving the real-time performance and reliability of wound bleeding detection. This invention first filters the original spectral feature data, successively removing absolute error data, numerical error data, and distance anomaly data, thereby obtaining standard spectral feature data. This effectively eliminates abnormal fluctuations, random noise, and outliers during the acquisition process, reducing the interference of invalid data on subsequent analysis, improving the reliability of the original data, avoiding the influence of erroneous samples on subsequent modeling and feature distribution, and improving the accuracy and robustness of subsequent detection results. By extracting data from feature subsets to form feature numerical combinations and calculating a comprehensive coefficient set, and then comprehensively evaluating the correlation between the original features and hemorrhage through stability correction coefficients and distribution correction coefficients, this invention can perform joint screening from multiple dimensions such as correlation, stability, and distribution characteristics. This can more accurately retain features that are highly correlated with hemorrhage and more sensitive to hemorrhage response, eliminate redundant and weakly correlated features, reduce feature dimensionality, reduce the burden of subsequent model training, and improve the accuracy and reliability of the model in hemorrhage detection. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the method of the present invention; Figure 2 This is a flowchart of the standard feature data acquisition process of the present invention; Figure 3 This is a flowchart of the bleeding-related feature screening process of the present invention; Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 As shown, this application provides a method for detecting multimodal FBG spectral feature fusion of wound bleeding, including the following steps: Step S1 involves simulating bleeding using a medical dressing with an integrated FBG sensor, collecting corresponding spectral data under different bleeding volumes, and extracting the corresponding spectral features to obtain raw spectral feature data. Step S1 includes the following sub-steps: Step S101: The medical dressing with integrated FBG sensor is denoted as FBG dressing, and any type of FBG dressing is denoted as type I dressing; generally, a medical dressing will integrate one FBG sensor; if multiple FBG sensors are integrated, each FBG sensor will be analyzed and processed separately. Step S102: Set up a simulated wound to mimic bleeding, and ensure the first type of dressing adheres tightly to the simulated wound; set the bleeding volume range for the bleeding simulation, and uniformly select multiple points from the bleeding volume range, denoting them sequentially as simulated quantity 1 to simulated quantity n in ascending order; different bleeding volumes will cause the medical dressing to expand in different volumes, resulting in inconsistent spectral characteristics. The bleeding volume range and simulated quantities can be set according to the actual application scenario. To ensure data separation, the step size of the simulated quantities should not be less than 1 ml.

[0019] Step S103: For the simulated amount 1, the simulated bleeding from the simulated wound reaches the simulated amount 1 and is absorbed by the first type of dressing. Then, the corresponding spectral data is collected by the integrated FBG sensor. Step S104: Set multiple spectral features, and select two or more spectral features to fuse them to obtain multiple fused features. The set spectral features and fused features are collectively referred to as original features, and are sequentially referred to as original feature 1 to original feature m, where m is the total number of original features. Spectral features can be set according to the actual use scenario, such as common center wavelength shift, reflection peak intensity change rate, etc.; fusion features can be selected from two or more spectral features for calculation and fusion, such as intensity wavelength ratio, peak value to valley value ratio of reflection peak, etc., or can be designed by yourself or by using AI.

[0020] Step S105: Based on the acquired spectral data, extract the corresponding original features 1 to m, and denot them as the corresponding original feature groups; perform multiple simulations based on analog quantity 1, and acquire the corresponding spectral data for each simulation, and extract the corresponding original feature groups; denot the resulting multiple original feature groups as the original feature data of analog quantity 1; to ensure the accuracy and reliability of subsequent modeling, the number of original feature groups for each analog quantity should be as large as possible, at least 50 groups; Step S106: Repeat the simulation to obtain the original feature data of all simulated quantities, and obtain the original spectral feature data; In practice, FBG is a periodic refractive index modulation structure inscribed inside the core of a single-mode quartz fiber using ultraviolet laser interference or femtosecond laser direct writing technology. When blood soaks into a medical dressing, it causes the dressing to expand in volume, generating axial tensile strain on the FBG embedded within it. This increases the grating period and leads to changes in a series of spectral characteristics. By establishing the correspondence between spectral characteristics and bleeding volume, quantitative detection of bleeding volume can be achieved.

[0021] Step S2 involves filtering the original spectral feature data to remove erroneous data, obtaining standard spectral feature data, and performing feature analysis to filter out hemorrhage-related features and obtain relevant feature information. Step S2 includes the following sub-steps: For step S201, please refer to... Figure 2 As shown, the original feature data of analog quantity 1 is denoted as the first feature data; the reasonable range and reasonable sign of each original feature are set; that is, the positive or negative sign of the original feature. For example, some features can only be positive, so the reasonable sign is positive; the data in the first feature data that are not in the corresponding reasonable range and do not meet the corresponding reasonable sign are filtered out and denoted as absolute error data. Absolutely erroneous data refers to erroneous data that does not conform to the laws of physics; these data are not experimental errors, but invalid data caused by hardware failure, operational errors or extreme interference, and have no analytical value in themselves.

[0022] Step S202: Mark the original feature groups containing absolute error data in the first feature data as absolute error groups; and denote the remaining original feature groups as the second feature data. Step S203: Extract all original features 1 from the second feature data and calculate the corresponding median M1 and median absolute deviation MD1. Record [M1-3×MD1, M1+3×MD1] as the reasonable range of values ​​for the original feature 1. Based on the second feature data, repeatedly obtain the reasonable range of values ​​for all original features. Using the median and median absolute deviation to construct the reasonable range of values ​​can more robustly identify outliers and reduce the impact of extreme outliers on the overall distribution. Step S204: Filter the data in the first feature data that are not in the corresponding reasonable range of values ​​and record them as numerical error data; mark the original feature groups containing numerical error data in the second feature data as numerical error groups; and record the remaining original feature groups as the third feature data. Although numerical error data conforms to physical laws in terms of numerical sign and magnitude, it differs significantly from other normal data under the same blood loss. It is usually caused by random external interference during the simulation process and has no analytical value. Although obvious errors have been removed from the third feature data, there may still be hidden errors due to abnormal feature combinations. Each feature in these data may be within the normal range when viewed individually, but the combination relationship of multiple features does not conform to the statistical regularity of normal bleeding. This is usually caused by systematic deviations such as batch differences in dressings, abnormal FBG pre-stretching, micro-local compression, and demodulator system drift, and further removal is required.

[0023] Step S205: Extract all original features 1 from the third feature data and calculate the corresponding mean AP1 and variance R1; denote 1 / R1 as the original weight of the original feature 1. Variance can measure the degree of fluctuation of the original features. Using the inverse of the variance as the initial weight can make subsequent analysis focus more on features with high stability, because the essence of weight is to quantify the reliability of each feature. The smaller the variance, the more stable the feature is under the same experimental conditions, and the greater its contribution to distinguishing between normal and abnormal features. Step S206: Based on the third feature data, repeatedly obtain the average value of all original features and record them as AP1 to APn in sequence according to the original feature number; and repeatedly obtain the original weights of all original features and normalize all original weights, that is, scale all weights to [0, 1] and make the sum of all feature weights equal to 1, so as to obtain the feature weights of all original features.

[0024] Step S207: Combine AP1 to APn into an n-dimensional vector, denoted as cluster center C, C = (AP1, AP2, ..., APn); denote any original feature group in the third feature data as the first feature group; Step S208: Combine the first feature group into a corresponding n-dimensional vector, and calculate the weighted Mahalanobis distance to the cluster center C based on the feature weights of the original features, denoted as the weighted distance of the first feature group; Weighted Mahalanobis distance can be used to determine whether each set of features deviates from the center from the perspective of multidimensional joint distribution, rather than judging anomalies based on a single feature. It can identify abnormal samples in combination relationships. At the same time, the introduction of feature weights can strengthen the role of key features in anomaly judgment and reduce misjudgments caused by irrelevant or weakly correlated features. Step S209: Repeatedly calculate the weighted distance of all original feature groups in the third feature data; and select the confidence level α, look up the chi-square distribution table with n degrees of freedom, and obtain the critical value corresponding to the confidence level α, denoted as AX; the normal feature group follows a multivariate normal distribution in the high-dimensional feature space, and the square of its weighted Mahalanobis distance follows a chi-square distribution with n degrees of freedom; in this embodiment, the confidence level α=0.99, which is generally selected, but can be adjusted according to the actual application scenario; for example, if the total number of original features n=15 and α=0.99, the corresponding critical value is 5.23; Step S210: Mark the original feature groups with weighted distance greater than √AX as distance outlier groups and remove them from the third feature data. Record the remaining third feature data as the standard feature data of analog quantity 1. Step S211: Repeatedly acquire the standard characteristic data of all analog quantities to obtain standard spectral characteristic data.

[0025] For step S212, please refer to... Figure 3 As shown, all analog quantities are arranged in ascending order and denoted as the blood volume sequence; for original feature 1, all original features 1 are extracted from the standard spectral feature data and divided according to the corresponding analog quantities to obtain n feature subsets; Step S213: Randomly select one data point from each feature subset as a combination, denoted as a feature value combination; repeat the process of obtaining all insufficient feature value combinations; denote any feature combination as the first combination; if the total number of feature value combinations is too large, a fixed number of feature value combinations can be generated using the Monte Carlo random sampling method, which can be adjusted appropriately according to computing power. Step S214: Arrange the first combination according to the corresponding analog quantities from smallest to largest, and denote it as the first feature sequence; denote the Pearson correlation coefficient, Spearman correlation coefficient, and maximum information coefficient between the first feature sequence and the bleeding volume sequence as XM1, XM2, and XM3 respectively; XM1 is used to measure the linear correlation between the original feature and the bleeding volume; XM2 is used to measure the monotonic correlation between the original feature and the bleeding volume; XM3 is used to measure the degree of any type of correlation between the original feature and the bleeding volume, such as linear, nonlinear, etc.

[0026] Step S215: Calculate q1×|XM1|+q2×|XM2|+q3×XM3, denoted as the comprehensive coefficient of the first combination; repeat the calculation of the comprehensive coefficients of all feature value combinations to obtain the set of comprehensive coefficients of the original feature 1, where q1, q2, and q3 are set weights; in this embodiment, q1=0.4, q2=0.3, q3=0.3, which can be flexibly adjusted according to the actual application scenario; for example, if XM1=0.92, XM2=0.90, XM3=0.93, then the comprehensive coefficient is 0.4×0.92+0.3×0.90+0.3×0.93=0.917; The comprehensive coefficient can avoid the bias caused by selecting features based on a single correlation coefficient. At the same time, by repeatedly calculating the comprehensive coefficient for different combinations, the instability caused by random sampling can be reduced, making the judgment of the association between features and bleeding volume more robust.

[0027] Step S216: Arrange the set of comprehensive coefficients of the original feature 1 in ascending order, denoted as the comprehensive correlation sequence; calculate the 10th percentile, 50th percentile, and 90th percentile of the comprehensive correlation sequence, denoted as Q10, Q50, and Q90 respectively; calculate 0.2×Q10+0.3×Q50+0.5×Q90, denoted as the weighted basic coefficient HF; where 0.2, 0.3, and 0.5 are set weights that can be flexibly adjusted. Under normal circumstances, the weights of Q10, Q50, and Q90 increase sequentially, reflecting the principle that excellent performance is more important than average performance; the weighted basic coefficient considers the central tendency, dispersion, and lower limit performance of the scores simultaneously, avoiding the one-sidedness of a single mean or median; For example, if Q10 = 0.930, Q50 = 0.970, and Q90 = 0.990, then the weighted basic coefficient HF = 0.2 × 0.93 + 0.3 × 0.97 + 0.5 × 0.99 = 0.972; Step S217: Calculate the 5th percentile Q5, 25th percentile Q25, 75th percentile Q75, and 95th percentile Q95 of the composite correlation sequence, and calculate the coefficients of dispersion CV1 and CV2, where CV1 = (Q75 - Q25) / Q50 and CV2 = (Q95 - Q5) / Q50; calculate 0.6 × |CV1| + 0.4 × |CV2|, denoted as BV; and calculate the stability correction coefficient EA, where EA = 0.2 + 0.8 × e -3×VD VD = tanh(BV), where tanh() represents the tanh function, and 0.4 and 0.6 are the weights that can be adjusted flexibly; 0.2 in EA is the lower limit of EA to avoid over-correction later; By using a stability correction coefficient, we can avoid the problem that although a certain original feature may have strong local correlation, its overall distribution may be unstable and fluctuate too much. This allows for a more comprehensive assessment of the stability and repeatability of the original features, ensuring that the features selected later are not only correlated but also stable. For example, given Q5=0.920, Q25=0.950, Q50=0.970, Q75=0.980, and Q95=0.995, then CV1=(Q75-Q25) / Q50=0.0309; CV2=(Q95-Q5) / Q50=0.0773; therefore, BV=0.6×|CV1|+0.4×|CV2|=0.0494, VD=tanh(BV)=0.0494, and EA=0.2+0.8×e -3×VD =0.89.

[0028] Step S218: Calculate the skewness SK and kurtosis KU of the composite coefficient set, and calculate 1 + 0.2 × tanh(SK) + 0.1 × tanh(KU-3), denoted as the distribution correction coefficient EB. Skewness is used to evaluate the symmetry of the composite coefficient distribution and determine whether the composite coefficients are concentrated in the high coefficient range or the low coefficient range. Kurtosis is used to evaluate the degree of concentration of the composite coefficient distribution and determine the consistency of the original feature performance. The larger the kurtosis, the more concentrated the composite coefficients are, the more consistent the feature performance is, and the higher the reliability. Among them, 0.2, 0.1 and the tanh function are used to limit the influence of skewness and kurtosis to (-0.2, 0.2) and (-0.1, 0.1) to avoid excessive correction in the future. The distribution correction coefficient can comprehensively measure the adaptability of the original features to changes in bleeding volume from multiple perspectives, so that the bleeding-related features retained in the end usually have strong correlation, good stability and better distribution characteristics, thereby improving the accuracy and generalization ability of the model for bleeding detection. For example, if skewness SK = 1.2 and kurtosis KU = 4.5, then the distribution correction coefficient EB = 1 + 0.2 × tanh(SK) + 0.1 × tanh(KU-3) = 1.26; Step S219: Calculate HF×EA×EB, and record it as the final correlation coefficient of the original feature 1; repeatedly obtain the final correlation coefficients of all original features; and select the k1 original features with the largest final correlation coefficients, mark them as bleeding-related features, and record all bleeding-related features as related feature information; where k1 is the set number, in this embodiment k1=4, which can be set according to the actual application scenario; For example, if the original feature 1 has HF=0.972, EA=0.89, and EB=1.26, then the final correlation coefficient of the original feature 1 is HF×EA×EB=1.09; In the specific implementation process, by using weighted basic coefficients, stability correction coefficients, and distribution correction coefficients, we ensure that only features that simultaneously possess high correlation, high stability, and a good distribution pattern can obtain high scores; this guarantees the effectiveness and reliability of the screened bleeding-related features, thereby ensuring the accuracy and reliability of subsequent bleeding detection.

[0029] Step S3: Establish a wound bleeding detection model based on relevant feature information and corresponding standard spectral feature data; Step S3 includes the following sub-steps: Step S301: Remove the data corresponding to non-hemorrhagic features from the standard spectral feature data, and record the remaining data as hemorrhage feature simulation data; Step S302: Use a multilayer perceptron as the initial detection model; set the input of the initial detection model to bleeding-related features and the output to bleeding volume; use bleeding feature simulation data to train the initial detection model, and obtain the wound bleeding detection model after completion; the multilayer perceptron has a strong nonlinear fitting ability and is suitable for handling the complex correspondence between bleeding-related features and bleeding volume, enabling the model to achieve quantitative prediction of wound bleeding. In the actual implementation process, other machine learning models or other algorithms can be selected to build a wound bleeding detection model according to the actual application scenario.

[0030] Step S4 involves acquiring spectral data of the wound using a medical dressing with an integrated FBG sensor, and detecting wound bleeding based on a wound bleeding detection model. Step S4 includes the following sub-steps: Step S401: Apply the first type of dressing to the wound to be tested, periodically collect the corresponding spectral data, extract the corresponding bleeding-related features, input the extracted bleeding-related features into the wound bleeding detection model, obtain the corresponding bleeding amount, and record the bleeding detection amount. Step S402: For the obtained bleeding detection amount, if it is less than e0, adjust the bleeding detection amount to 0 and output it; if it is not less than e0, do not adjust it and output it, where e0 is the set threshold; in this embodiment, e0=1ml; it can be flexibly set according to the actual application scenario. In practical applications, even if there is no bleeding from the wound, the influence of the external environment or the patient's actions can cause changes in bleeding-related characteristics to a certain extent, resulting in a bleeding amount that is extremely small but not zero. In order to avoid misjudging slight fluctuations with no clinical significance as bleeding and to make the final output more in line with the actual application scenario, e0 is set to treat tiny detection values ​​below the threshold as no obvious bleeding and output them as zero, while retaining the output of results that reach the threshold and above. In the specific implementation process, by establishing a wound bleeding detection model, medical dressings with integrated FBG sensors can achieve continuous, real-time, and quantitative detection of wound bleeding without removing the dressing. This not only avoids secondary damage to the wound and monitoring blind spots caused by traditional detection methods, but also improves the objectivity, accuracy, and stability of the detection results.

[0031] Example 2, please refer to Figure 4 As shown, Figure 4 A schematic diagram of an electronic device is provided, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer-readable instructions, which the processor can call. When the processor executes a computer-readable instruction, it performs steps such as those in a method for detecting multimodal FBG spectral features of wound bleeding, to achieve the following functions: simulating bleeding using a medical dressing with an integrated FBG sensor; collecting corresponding spectral data under different bleeding amounts and extracting corresponding spectral features to obtain raw spectral feature data; filtering the raw spectral feature data to remove erroneous data, obtaining standard spectral feature data, and performing feature analysis to filter bleeding-related features to obtain relevant feature information; establishing a wound bleeding detection model based on the relevant feature information and the corresponding standard spectral feature data; collecting spectral data of the wound using a medical dressing with an integrated FBG sensor, and detecting wound bleeding based on the wound bleeding detection model.

[0032] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0033] Example 3: This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs steps as described in a method for detecting multimodal FBG spectral feature fusion of wound bleeding, to achieve the following functions: simulating bleeding using a medical dressing with an integrated FBG sensor; collecting corresponding spectral data under different bleeding amounts and extracting corresponding spectral features to obtain raw spectral feature data; filtering the raw spectral feature data to remove erroneous data, obtaining standard spectral feature data, and performing feature analysis to filter bleeding-related features to obtain relevant feature information; establishing a wound bleeding detection model based on the relevant feature information and the corresponding standard spectral feature data; collecting spectral data of the wound using a medical dressing with an integrated FBG sensor, and detecting wound bleeding based on the wound bleeding detection model.

[0034] Based on the above description of the embodiments, the embodiments of the present invention can be provided as methods, systems, or computer program products. Based on this understanding, the above technical solutions, in essence or in terms of their contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments.

[0035] In the embodiments provided in this application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces. The indirect coupling or communication connection between systems, modules, and units may be electrical, mechanical, or other forms.

[0036] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for detecting wound bleeding by fusing multimodal FBG spectral features, characterized in that, Includes the following steps: Bleeding simulation was performed on medical dressings with integrated FBG sensors. Spectral data were collected under different bleeding volumes, and corresponding spectral features were extracted to obtain raw spectral feature data. The original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features and obtain relevant feature information. A wound bleeding detection model was established based on relevant feature information and corresponding standard spectral feature data. Spectral data of the wound were collected using a medical dressing with an integrated FBG sensor, and wound bleeding was detected based on a wound bleeding detection model.

2. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 1, characterized in that, Bleeding simulation was performed using a medical dressing with an integrated FBG sensor. Spectral data was collected at different bleeding volumes, and corresponding spectral features were extracted to obtain the raw spectral feature data. The process included the following sub-steps: Medical dressings with integrated FBG sensors are referred to as FBG dressings, and any type of FBG dressing is referred to as Type I dressing. Set up a simulated wound to simulate bleeding, and make the first type of dressing fit tightly to the simulated wound; set the bleeding range for the bleeding simulation, and select multiple points evenly from the bleeding range, and record them as simulated quantity 1 to simulated quantity n in ascending order.

3. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 2, characterized in that, Bleeding simulation was performed using a medical dressing with an integrated FBG sensor. Spectral data was collected at different bleeding volumes, and corresponding spectral features were extracted to obtain the raw spectral feature data. The process included the following sub-steps: For analog quantity 1, the simulated bleeding from the simulated wound reaches analog quantity 1 and is absorbed by the first type of dressing, and the corresponding spectral data is collected by the integrated FBG sensor. Multiple spectral features are set, and two or more spectral features are selected and fused to obtain multiple fused features. The set spectral features and fused features are collectively referred to as original features, and are sequentially referred to as original feature 1 to original feature m, where m is the total number of original features. Based on the acquired spectral data, the corresponding original features 1 to m are extracted and denoted as the corresponding original feature group. Multiple simulations were performed based on analog quantity 1, and corresponding spectral data were collected for each simulation, and the corresponding original feature groups were extracted. The resulting sets of original features are denoted as the original feature data of analog quantity 1; Repeated simulations are used to obtain the original characteristic data of all simulated quantities, thus obtaining the original spectral characteristic data.

4. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 3, characterized in that, The original spectral feature data is filtered to remove erroneous data, resulting in standard spectral feature data. Feature analysis is then performed to filter out hemorrhage-related features, obtaining relevant feature information. This process includes the following sub-steps: The original feature data of analog quantity 1 is denoted as the first feature data; a reasonable range and a reasonable sign are set for each original feature; data in the first feature data that are not within the corresponding reasonable range or do not meet the corresponding reasonable sign are filtered out and denoted as absolute error data; Mark the original feature groups containing absolute error data in the first feature data as absolute error groups; denote the remaining original feature groups as the second feature data; Extract all original features 1 from the second feature data and calculate the corresponding median M1 and median absolute deviation MD1. Record [M1-3×MD1, M1+3×MD1] as the reasonable range of the original feature 1 values. Based on the second feature data, repeatedly obtain the reasonable range of the original features values. Data in the first feature data that does not fall within the corresponding reasonable range of values ​​are selected and recorded as numerical error data; the original feature group containing numerical error data in the second feature data is marked as the numerical error group. The remaining original feature groups are then recorded as the third feature data.

5. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 4, characterized in that, The process of filtering the raw spectral feature data to remove erroneous data and obtain standard spectral feature data, followed by feature analysis to filter out hemorrhage-related features and obtain relevant feature information, also includes the following sub-steps: Extract all original features 1 from the third feature data and calculate the corresponding mean AP1 and variance R1; denote 1 / R1 as the original weight of original feature 1; Based on the third feature data, the average value of all original features is repeatedly obtained and recorded as AP1 to APn in sequence according to the original feature number; and the original weights of all original features are repeatedly obtained and normalized to obtain the feature weights of all original features. Combine AP1 to APn into an n-dimensional vector, denoted as the cluster center C, C = (AP1, AP2, ..., APn); denote any original feature group in the third feature data as the first feature group; The first feature group is also combined into a corresponding n-dimensional vector, and the weighted Mahalanobis distance to the cluster center C is calculated based on the feature weights of the original features, which is denoted as the weighted distance of the first feature group. Repeatedly calculate the weighted distance of all original feature groups in the third feature data; and select a confidence level α, look up the chi-square distribution table with n degrees of freedom, and obtain the critical value corresponding to the confidence level α, denoted as AX; The original feature groups with a weighted distance greater than √AX are marked as distance outliers and removed from the third feature data. The remaining third feature data are recorded as the standard feature data of analog quantity 1. Standard spectral characteristic data are obtained by repeatedly acquiring standard characteristic data of all analog quantities.

6. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 5, characterized in that, The process of filtering the raw spectral feature data to remove erroneous data and obtain standard spectral feature data, followed by feature analysis to filter out hemorrhage-related features and obtain relevant feature information, also includes the following sub-steps: Arrange all analog quantities in ascending order and denote them as the blood volume sequence; for original feature 1, extract all original feature 1 from the standard spectral feature data and divide them according to the corresponding analog quantities to obtain n feature subsets; Randomly select one data point from each feature subset and form a combination, denoted as the feature value combination; Repeatedly obtain all insufficient feature value combinations; Any combination of features is denoted as the first combination; The first combination is arranged in ascending order of the corresponding analog quantities and is denoted as the first feature sequence. The Pearson correlation coefficient, Spearman correlation coefficient and maximum information coefficient between the first feature sequence and the bleeding volume sequence are denoted as XM1, XM2 and XM3 respectively. Calculate q1×|XM1|+q2×|XM2|+q3×XM3, and denote it as the comprehensive coefficient of the first combination; repeat the calculation of the comprehensive coefficients of all feature value combinations to obtain the set of comprehensive coefficients of the original feature 1, where q1, q2 and q3 are the set weights.

7. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 6, characterized in that, The process of filtering the raw spectral feature data to remove erroneous data and obtain standard spectral feature data, followed by feature analysis to filter out hemorrhage-related features and obtain relevant feature information, also includes the following sub-steps: Arrange the set of comprehensive coefficients of the original feature 1 in ascending order and denote it as the comprehensive correlation sequence; calculate the 10th percentile, 50th percentile and 90th percentile of the comprehensive correlation sequence and denote them as Q10, Q50 and Q90 respectively; calculate 0.2×Q10+0.3×Q50+0.5×Q90 and denote it as the weighted basic coefficient HF; Calculate the 5th percentile Q5, 25th percentile Q25, 75th percentile Q75, and 95th percentile Q95 of the composite correlation sequence, and calculate the coefficients of dispersion CV1 and CV2, where CV1 = (Q75 - Q25) / Q50 and CV2 = (Q95 - Q5) / Q50; calculate 0.6 × |CV1| + 0.4 × |CV2|, denoted as BV; and calculate the stability correction coefficient EA, where EA = 0.2 + 0.8 × e -3×VD VD = tanh(BV), where tanh() represents the tanh function.

8. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 7, characterized in that, The process of filtering the raw spectral feature data to remove erroneous data and obtain standard spectral feature data, followed by feature analysis to filter out hemorrhage-related features and obtain relevant feature information, also includes the following sub-steps: Calculate the skewness SK and kurtosis KU of the aggregate coefficient set, and calculate 1 + 0.2 × tanh (SK) + 0.1 × tanh (KU-3), denoted as the distribution correction coefficient EB; Calculate HF×EA×EB, and denote it as the final correlation coefficient of original feature 1; repeat to obtain the final correlation coefficients of all original features; The k1 original features with the highest final correlation coefficients are selected and marked as bleeding-related features. All bleeding-related features are recorded as related feature information, where k1 is the set number.

9. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 8, characterized in that, Establishing a wound bleeding detection model based on relevant feature information and corresponding standard spectral feature data includes the following sub-steps: Remove the data corresponding to non-hemorrhagic features from the standard spectral feature data, and record the remaining data as hemorrhage feature simulation data; A multilayer perceptron was used as the initial detection model. The input of the initial detection model was set as bleeding-related features, and the output was the amount of bleeding. The initial detection model was trained using simulated data of bleeding features, and the wound bleeding detection model was obtained after the training was completed.

10. The detection method for multimodal FBG spectral feature fusion of wound bleeding according to claim 9, characterized in that, The process of acquiring spectral data of the wound using a medical dressing with an integrated FBG sensor and detecting wound bleeding based on a wound bleeding detection model includes the following sub-steps: The first type of dressing is applied to the wound to be tested, and the corresponding spectral data is collected periodically. The corresponding bleeding-related features are extracted and input into the wound bleeding detection model to obtain the corresponding bleeding amount, which is recorded as the bleeding detection amount. If the obtained bleeding detection value is less than e0, the bleeding detection value is adjusted to 0 and output; if it is not less than e0, it is not adjusted and output, where e0 is the set threshold.

Images