Drug sensitivity test result recognition method, device, electronic device and readable storage medium

Abstract

The present application discloses a drug sensitivity result recognition method, device, electronic device and readable storage medium. Among them, the method includes obtaining the optical density data of the target strain at the target concentration of the drug within the first time period as the data to be tested. According to the optical density data of the target strain at different concentrations of the target drug and without adding the target drug, the growth information of the target strain in the future preset time period is determined to obtain the extended test data. According to the concentration information characteristics of the extracted extended test data and the similarity information characteristics between the optical density data of the target strain at the multiple reference concentrations of the target drug and without adding the target drug respectively, the growth characteristics to be recognized are generated. The growth characteristics to be recognized are input into the pre-trained growth prediction model to obtain the predicted value of the growth situation of the target strain in the target drug, so as to realize the efficient, low-cost and accurate recognition of the drug sensitivity result.

Description

Technical Field

[0001] The present application relates to the technical field of microbial drug sensitivity, and in particular to a method and device for identifying drug sensitivity results, an electronic device, and a readable storage medium. Background Art

[0002] Replacing traditional manual operations with automated systems in bacterial testing is one of the hallmarks of contemporary clinical testing progress. The application of fully automated microbial identification and drug sensitivity analysis instruments has created a material basis for rapid and accurate bacteriological reports, enabling a leap in the level of bacterial testing. Infectious diseases caused by a variety of pathogenic microorganisms are spread across all clinical departments, among which bacterial infections are the most common. Therefore, antibacterial drugs have become one of the most widely used drugs in clinical practice. While antibacterial drugs have cured and saved the lives of many patients, there have also been major impacts on patients' health and even lives due to unreasonable use of antibacterial drugs, such as incorrect selection of drug varieties and dosages, increased adverse reactions, and the growth of bacterial drug resistance.

[0003] Drug sensitivity tests are used to determine the sensitivity of detected drugs for accurate and effective treatment with drugs. Its basic principle is: by detecting the growth status of microorganisms under different concentrations of drugs, determining the concentration information of the drug that is most effective in killing cells. The drug sensitivity test process includes: placing the same concentration of bacterial solution in reaction wells containing different concentrations of drugs. The reaction system contains a color developer and a culture medium, and the color developer will produce color changes according to the growth of cells. The absorbance value (OD (optical density)) is collected by an enzyme-linked immunosorbent assay (ELISA) reader, and the absorbance value of the color change in the reaction wells is continuously measured every half hour. The growth status of bacteria is judged based on the change rule of the OD value.

[0004] In the process of identifying drug sensitivity results in related technologies, visual interpretation is usually adopted. For example, domestic manufacturers generally adopt the end-point interpretation method, which has poor timeliness and a long clinical reporting time. Even the participating algorithm models generally use methods such as the acceleration method, speed method, slope method, and threshold method to determine the positive and negative of whether there is microbial growth. It can be understood that visual interpretation of drug sensitivity requires trained laboratory technicians to observe with the naked eye, which not only depends on the professional skills of technical personnel and is affected by personal factors, but also has a high misjudgment rate and low efficiency when conducting a large number of experiments in scenarios with high labor intensity and fatigue. Further, the software and hardware systems for rapid drug sensitivity detection on the domestic and international markets are expensive and difficult to promote at the grass-roots level in China.

[0005] In view of this, how to efficiently, low-costly, and accurately identify drug sensitivity results is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0006] The present application provides a drug sensitivity result recognition method, device, electronic device and readable storage medium, which can efficiently, low-costly and accurately recognize drug sensitivity results.

[0007] To solve the above technical problems, the embodiments of the present invention provide the following technical solutions:

[0008] On the one hand, the embodiments of the present invention provide a drug sensitivity result recognition method, including:

[0009] Obtain the data to be tested, where the data to be tested is the optical density data of the target strain in the drug at the target concentration within the first time period;

[0010] According to the optical density data of the target strain without adding the target drug and the optical density data of the target strain at different concentrations of the target drug, determine the growth information of the target strain in the future preset time period to obtain extended test data;

[0011] Generate the growth characteristics to be recognized according to the concentration information characteristics of the extracted extended test data, the similarity information characteristics between the extended test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug;

[0012] Input the growth characteristics to be recognized into a pre-trained growth prediction model to obtain the predicted value of the growth situation of the target strain in the target drug.

[0013] Optionally, after inputting the growth characteristics to be recognized into a pre-trained growth prediction model to obtain the predicted value of the growth situation of the target strain in the target drug, it further includes:

[0014] Obtain the predicted values of the growth situation of the target strain at different concentrations of the target drug;

[0015] Determine the predicted value of the minimum inhibitory concentration of the target strain corresponding to the target drug according to the predicted values of the growth situation of the target strain at different concentrations of the target drug.

[0016] Optionally, the step of determining the growth information of the target strain in the future preset time period to obtain extended test data according to the optical density data of the target strain without adding the target drug and the optical density data of the target strain at different concentrations of the target drug includes:

[0017] Determine the original growth rate of the target strain according to the optical density data of the target strain at each known moment within the first time period;

[0018] Determine the general growth rate of the target strain according to the historical growth data of the target strain at different concentrations of the target drug and the original growth rate;

[0019] Determine the initial predicted optical density at multiple initial future times within the preset future time period according to the optical density value of the target strain at the target known time and the original growth rate;

[0020] Determine the later predicted optical density values at multiple later future times within the preset future time period according to the general growth rate and each initial predicted optical density value;

[0021] Based on the original maximum optical density value of the target strain without adding the target drug, each initial predicted optical density value, and each later predicted optical density value, determine the predicted optical density data of the data to be tested within the preset future time period;

[0022] Use the predicted optical density data and the data to be tested as augmented test data.

[0023] Optionally, the determining the general growth rate of the target strain according to the historical growth data of the target strain at different concentrations of the target drug and the original growth rate includes:

[0024] Determine the maximum optical density value of the target strain and the corresponding target time according to the historical growth data of the target strain at different concentrations of the target drug;

[0025] Determine the growth rate change rate according to the maximum optical density value and the optical density values at multiple times before the target time;

[0026] Determine the general growth rate of the target strain according to the original growth rate and the growth rate change rate.

[0027] Optionally, the generating the growth characteristics to be recognized according to the concentration information characteristics of the extracted augmented test data, the similarity information characteristics between the augmented test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug includes:

[0028] Calculate the optical density central tendency information, optical density dispersion and change information, optical density continuous change information, and correlation degree information between adjacent times according to the optical density values of the target strain at each time under the target concentration drug;

[0029] According to the geometric metric space distance, linear correlation degree information, and vector space distance between the optical density data of the target strain at the target concentration drug and the optical density data of the target strain without adding the target drug;

[0030] Based on the optical density data of the target bacterial strain in the drug at the target concentration, respectively, the geometric metric space distance, the linear correlation degree information, and the asymmetry metric information between the optical density data of the target bacterial strain in the target drug at each reference concentration.

[0031] Optionally, before inputting the to-be-identified growth characteristics into a pre-trained growth prediction model to obtain a predicted value of the growth situation of the target bacterial strain in the target drug, it further includes:

[0032] Pre-build a machine learning model framework;

[0033] Obtain the historical optical density data of the target bacterial strain in the absence of antibacterial drugs and different antibacterial drugs at different drug concentrations, and set corresponding labels for each historical optical density data, where the labels include growth labels and non-growth labels;

[0034] Divide the historical optical density data into a training sample set and a test sample set according to a preset ratio;

[0035] For each training sample in the training sample set, generate training growth characteristics according to the extracted concentration information characteristics of the current training sample, the similarity information characteristics between the current training sample and the optical density data of the target bacterial strain without adding antibacterial drugs, and the optical density data of the target bacterial strain at multiple reference concentrations of the target antibacterial drug;

[0036] For each test sample in the test sample set, determine the growth information of the target bacterial strain in the current test sample in a future preset time period according to the optical density data of the target bacterial strain in the absence of antibacterial drugs and different concentrations of the target antibacterial drug, so as to obtain the augmented test sample data of the current test sample;

[0037] Use each training sample in the training sample set and its corresponding training growth characteristics to continuously train the machine learning model until the accuracy of the trained machine learning model tested by the augmented test sample data of each test sample set is greater than a preset accuracy threshold, and obtain a growth prediction model.

[0038] Optionally, the machine learning model is a support vector machine, and using each training sample in the training sample set and its corresponding training growth characteristics to continuously train the machine learning model includes:

[0039] Randomly construct an initial hyperplane according to the training growth characteristics of each training sample;

[0040] Obtain a corresponding decision function according to the initial hyperplane, and determine an optimization problem according to the decision function and a preset penalty parameter;

[0041] By calculating the optimization problem, parameter estimation is performed on the initial hyperplane to obtain the optimal parameters of the support vector machine;

[0042] Determine the support vector machine obtained in the current training cycle according to the optimal parameters.

[0043] Optionally, the machine learning model is an artificial neural network model. Using each training sample of the training sample set and its corresponding training growth characteristics to continuously train the machine learning model includes:

[0044] The artificial neural network model includes an input layer, a first neural network layer, a second neural network layer, and an output layer;

[0045] Input the training growth characteristics of each training sample in the training sample set into the artificial neural network model to obtain the predicted minimum inhibitory concentration value corresponding to each training sample;

[0046] According to the predicted minimum inhibitory concentration value and the true minimum inhibitory concentration value of each training sample, calculate the loss function of the artificial neural network model in the current training cycle;

[0047] Using the BP neural network algorithm, optimize the loss function according to the Adam algorithm to obtain the optimal parameters of the artificial neural network model in the current training cycle;

[0048] Determine the artificial neural network model obtained in the current training cycle according to the optimal parameters.

[0049] Another aspect of the embodiments of the present invention provides a drug sensitivity result recognition device, including:

[0050] A data acquisition module for acquiring data to be tested, where the data to be tested is the optical density data of the target strain in the target concentration drug within the first time period;

[0051] A data augmentation module for determining the growth information of the target strain in a future preset time period according to the optical density data of the target strain without adding the target drug and the optical density data of the target strain at different concentrations of the target drug, so as to obtain augmented test data;

[0052] A feature extraction module for generating growth characteristics to be recognized according to the concentration information characteristics of the augmented test data extracted, the similarity information characteristics between the augmented test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug;

[0053] A growth prediction module, configured to input the to-be-identified growth features into a pre-trained growth prediction model to obtain a predicted value of the growth condition of the target bacterial strain in the target drug.

[0054] An embodiment of the present invention further provides an electronic device, including a processor, where the processor is configured to implement the steps of the drug sensitivity result identification method as described in any one of the previous items when executing a computer program stored in a memory.

[0055] An embodiment of the present invention finally further provides a readable storage medium, on which a computer program is stored, and the computer program implements the steps of the drug sensitivity result identification method as described in any one of the previous items when being executed by a processor.

[0056] The advantages of the technical solution provided by this application are as follows: It can, according to the optical density values measured by the instrument in sequence, and in combination with test data information and drug information, automatically judge the growth condition and growth probability of microorganisms under different drugs and different concentrations of the same drug by analyzing the variation law of the optical density values. Furthermore, it can automatically judge the positive and negative results of each growth curve, replacing the existing method of visual judgment based on experimental experience, and at the same time solving all the drawbacks of this existing method. At the same time, it can also solve the problems of inaccurate identification of drug sensitivity results caused by the trailing growth prediction and interpretation in drug sensitivity tests and the problem of color change without bacteria in drug sensitivity tests. The entire testing process does not require adding new hardware costs, and can efficiently, low-costly and accurately identify drug sensitivity results. Further, according to the continuously interpreted data, real-time dynamic calculation is performed. By using the future growth data predicted from the to-be-tested data at the current moment and the current test data as the overall test data, the drug sensitivity result can be reported in advance, which is beneficial to obtaining the minimum inhibitory concentration with the highest probability in the shortest possible time, thereby guiding drug use.

[0057] In addition, an embodiment of the present invention also provides a corresponding implementation device, electronic device and readable storage medium for the drug sensitivity result identification method, further making the method more practical, and the device, electronic device and readable storage medium have corresponding advantages.

[0058] It should be understood that the above general description and the following detailed description are only exemplary and do not limit the present disclosure. Description of the Drawings

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention or related technologies, the following will briefly introduce the drawings required for use in the description of the embodiments or related technologies. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0060] Figure 1 Schematic flowchart of a drug sensitivity result recognition method provided by an embodiment of the present invention;

[0061] Figure 2 Schematic diagram of the growth curve before data augmentation in an exemplary application scenario provided by an embodiment of the present invention;

[0062] Figure 3 Schematic diagram of the growth curve after data augmentation in an exemplary application scenario provided by an embodiment of the present invention;

[0063] Figure 4 Schematic flowchart of a training growth feature extraction method provided by an embodiment of the present invention;

[0064] Figure 5 Schematic flowchart of a test growth feature extraction method provided by an embodiment of the present invention;

[0065] Figure 6 Schematic diagram of a support vector machine algorithm model provided by an embodiment of the present invention;

[0066] Figure 7 Schematic diagram of an artificial neural network model provided by an embodiment of the present invention;

[0067] Figure 8 Schematic flowchart of the training process of an artificial neural network model provided by an embodiment of the present invention;

[0068] Figure 9 Schematic flowchart of the MIC prediction process provided by an embodiment of the present invention;

[0069] Figure 10 Structural diagram of a specific implementation manner of a drug sensitivity result recognition device provided by an embodiment of the present invention;

[0070] Figure 11 Structural diagram of a specific implementation manner of an electronic device provided by an embodiment of the present invention. Specific implementation manner

[0071] In order to enable those skilled in the art to better understand the solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation manners. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0072] In the description and claims of this application and the above-mentioned drawings, the terms "first", "second", "third", "fourth", etc. are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but may include steps or units not listed.

[0073] After introducing the technical solutions of the embodiments of the present invention, various non-limiting embodiments of this application will be described in detail below.

[0074] First, refer to Figure 1 , Figure 1 which is a schematic flow chart of a drug sensitivity result identification method provided by an embodiment of the present invention. The embodiments of the present invention may include the following content:

[0075] S101: Obtain the data to be tested.

[0076] In this step, the data to be tested is the optical density data of the target bacterial species in the drug at the target concentration during the first time period. The first time period may be the time from the first moment to the current moment. As for the first moment, it may be the moment when the target drug is added, or it may be a certain moment after the target drug is added. Those skilled in the art can make a flexible choice according to actual needs. For the sake of distinction, in this embodiment, the bacterial species to be tested is called the target bacterial species, the antibacterial drug used to test the required drug resistance or growth condition of this bacterial species is called the target drug, and the target concentration refers to the concentration of the antibacterial drug when testing the drug resistance or growth condition at present in the entire culture dish or reaction well.

[0077] S102: Determine the growth information of the target bacterial species in a preset time period in the future according to the optical density data of the target bacterial species without adding the target drug and at different concentrations of the target drug, so as to obtain the extended test data.

[0078] In this step, the optical density data refers to the set of optical density values of the target strain at multiple moments. The optical density value, i.e., the OD value, can be used to describe the optical density absorbed by the detected substance, and the detected unit is represented by the OD value. Optical density has no dimension unit and is the logarithm of the ratio of incident light to transmitted light. The optical density data of the target strain without adding the target drug is used to represent the growth of the target strain itself, and the optical density data of the target strain at different concentrations of the target drug is used to represent the influence of different concentrations of the target drug on the growth of the target strain. The growth information in the future preset time period refers to after obtaining the data to be tested in S101, in order to predict the growth of the strain under the condition of only having OD values in a very short time. Based on the growth of the target strain itself and the influence of different concentrations of the target drug on the growth of the target strain, jointly predict the growth of the target strain under the target concentration drug in a future period of time. The growth situation data in the preset future period of time and the data to be tested in S101 are jointly used as the growth data of the target strain under the target concentration drug, that is, to expand the test data. By comparing Figure 2 and Figure 3 it can be seen that in this step, by expanding the original data to be tested, under the condition of only having OD values in a very short time, the prediction accuracy can be effectively improved.

[0079] S103: Generate the growth characteristics to be identified according to the concentration information characteristics of the extracted expanded test data, and the similarity information characteristics between the expanded test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug.

[0080] In this step, feature extraction refers to the information extraction and compression of the original data, including feature processing and feature selection. The concentration information characteristics refer to the growth characteristic information of the target strain under the target concentration drug extracted from the expanded test data. The similarity information characteristics include two parts. One part is the similarity information between the growth of the target strain under the target concentration drug and its own growth, and the other part is the similarity information between the growth of the target strain under the target concentration drug and its growth at other concentrations of the target drug. The concentration information characteristics and the similarity information characteristics jointly serve as the characteristic information that can reflect the growth of the target strain under the target concentration drug. In this step, by reducing the dimension of the large-scale OD value data, the overfitting phenomenon is effectively solved, and the prediction accuracy of the final growth situation is improved.

[0081] S104: Input the growth characteristics to be identified into the pre-trained growth prediction model to obtain the predicted value of the growth of the target strain in the target drug.

[0082] The growth prediction model of this embodiment is a binary classification model, which is used to identify the input OD value data and determine whether the bacterial strain grows or does not grow in the corresponding environment. The growth prediction model is a model pre-trained based on machine learning algorithms. Any machine learning model in the prior art can be used to build the framework, and the OD value data of different antibacterial drugs of the bacterial strain at different concentrations are used to train the framework to obtain the growth prediction model in the case of whether it has grown or not.

[0083] In the technical solution provided by the embodiment of the present invention, it is possible to automatically judge the growth situation and growth probability of microorganisms under different drugs and different concentrations of the same drug by analyzing the change law of the optical density value according to the optical density value measured by the instrument in time sequence and combining the test data information and drug information, and then automatically judge the positive and negative results of each growth curve, which can replace the existing method of visual judgment based on experimental experience and solve all the drawbacks of this existing method at the same time; it can also solve the problem of inaccurate identification of drug sensitivity results caused by the trailing growth prediction interpretation in drug sensitivity tests and the problem of color change in sterile conditions in drug sensitivity tests. No new hardware costs need to be added during the whole test process, and the drug sensitivity results can be identified efficiently, at low cost and accurately. Further, according to the continuously interpreted data, real-time dynamic calculation is performed. By using the future growth data predicted from the test data to be tested at the current moment and the current test data together as the overall test data, the drug sensitivity results can be reported in advance, which is beneficial to obtaining the minimum inhibitory concentration with the highest probability in the shortest possible time, so as to guide drug use.

[0084] It should be noted that there is no strict order of execution among the steps in this application. As long as it conforms to the logical order, these steps can be executed simultaneously or in a certain preset order. Figure 1 It is only a schematic way and does not mean that it can only be executed in this order.

[0085] It can be understood that due to the differences in bacterial strains and antibiotics, the interpretation criteria may be inconsistent in visual interpretation. For Staphylococcus, Enterococcus, Streptococcus pneumoniae, β-hemolytic Streptococcus, and Viridans Streptococcus, when testing the MIC (Minimum Inhibitory Concentration) of chloramphenicol, clindamycin, erythromycin, linezolid, tedizolid, and tetracycline, trailing growth will make it difficult to determine the end point value. In such a case, read the MIC at the lowest concentration where trailing growth begins, and weak growth should be ignored; for Staphylococcus and Streptococcus pneumoniae, when measuring the MIC of trimethoprim and sulfonamides, some slight growth is allowed. Therefore, when reading the concentration end point, compared with the control growth well, the reduction of the growth in the well should be ≥80%. Based on the above embodiments, this application may further include:

[0086] Obtain the predicted growth values of the target bacterial strain at different concentrations of the target drug;

[0087] Determine the predicted minimum inhibitory concentration value of the target bacterial strain corresponding to the target drug according to the predicted growth values of the target bacterial strain at different concentrations of the target drug.

[0088] Among them, MIC is the minimum antibiotic concentration that can effectively kill bacteria. By obtaining the predicted growth values of the target bacterial strain at each concentration of the target drug according to the above embodiments, the predicted minimum inhibitory concentration value of the target bacterial strain corresponding to the target drug can be quickly and accurately determined based on these predicted growth values, solving the problem of inaccurate MIC judgment caused by the trailing growth prediction and reading in the drug sensitivity test and the problem of no-color change in the drug sensitivity test.

[0089] In the above embodiments, there is no limitation on how to execute step S102. In this embodiment, an optional generation method for expanding test data is given, which may include the following content:

[0090] Determine the original growth rate of the target bacterial strain according to the optical density data of the target bacterial strain at each known moment in the first time period;

[0091] Determine the general growth rate of the target bacterial strain according to the historical growth data and the original growth rate of the target bacterial strain at different concentrations of the target drug;

[0092] Determine the initial predicted optical density at multiple initial future moments within a preset future time period according to the optical density value of the target bacterial strain at a target known moment and the original growth rate;

[0093] Determine the later predicted optical density values at multiple later future moments within a preset future time period according to the general growth rate and each initial predicted optical density value;

[0094] Based on the original maximum optical density value of the target bacterial strain without adding the target drug, each initial predicted optical density value and each later predicted optical density value, determine the predicted optical density data of the data to be tested within a preset future time period;

[0095] Use the predicted optical density data and the data to be tested as the expanded test data.

[0096] In this embodiment, the target known time is any time within the first time period. Multiple initial future times within the preset future time period refer to several times within the prediction time period, and the number of initial future times can be selected according to the actual situation. Multiple later future times within the preset time period refer to several times within the prediction time period after the initial future times. For the sake of distinction, they are called later future times. Similarly, the number of later future times can be selected according to the actual situation. For example, if the preset future time period is 3 hours in the future, then the initial future times can be several times within the first two hours, and the later future times can be several times within the last hour. Similarly, for the sake of distinction and to avoid ambiguity, the OD value of the target strain at the initial future time is called the initial predicted optical density, and the OD value of the target strain at the later future time is called the later predicted optical density value. Among them, the process of determining the general growth rate of the target strain based on the historical growth data and the original growth rate of the target strain at different concentrations of the target drug may include:

[0097] Based on the historical growth data of the target strain at different concentrations of the target drug, determine the maximum optical density value of the target strain and the corresponding target time; based on the maximum optical density value and the optical density values at multiple times before the target time, determine the growth rate change rate; based on the original growth rate and the growth rate change rate, determine the general growth rate of the target strain.

[0098] To make those skilled in the art more clearly understand the data augmentation method of the present application, this embodiment also takes the example of artificially supplementing 2 hours of data for a specific 8-hour data to be predicted to illustrate the implementation manner of the above technical solution:

[0099] Based on the analysis of the OD value data and the research on pharmacological knowledge, artificially supplement 2 hours of data for a specific 8-hour data to be predicted. The OD value data of a specific concentration in the data to be predicted is represented by x1, x2,..., x 16 , that is, obtain the OD value of the target strain at the target concentration drug every half hour. The specific supplementation rules are as follows:

[0100] (1) First, grow the 8-hour data by the original growth rate r = x 16 -x 15 for 1 hour, that is, the OD values of the 8.5th hour and the 9th hour can be obtained, which can be expressed as:

[0101] x 17 = x 16 + r, x 18 = x 17 + r (1)

[0102] (2) By analyzing a large amount of OD value data, it can be seen that after the OD value data reaches the peak for the first time, it will show a short-term decline and gradually maintain balance. For the concentration data of the growth of the target drug - target strain in the historical data, find its maximum OD value data x t , and use the data at this moment and the data of the previous two moments to calculate the growth rate change rate l and round it up. Then, take the mode of all the rounded-up growth rate change rates l as the growth rate change rate l' of the target drug - target strain. That is, the growth rate change rate can be calculated through the following relational formula (2):

[0103]

[0104] Among them, for the convenience of subsequent use, the growth rate change rates under different corresponding relationships of drug - strain can be pre-stored in the database and directly called when in use.

[0105] (3) According to the growth rate change rate l' of the historical data of the target drug - target strain, calculate a general growth rate r' = r × l'. Then, grow for another 1h according to the general growth rate r'. That is, the OD values at the 9.5th hour and the 10th hour can be obtained:

[0106] x 19 = x 18 + r', x 20 = x 19 + r' (3)

[0107] (4) Note that for the concentration curve that has already grown, making the same supplement will cause a sharp increase in the OD value, which will have an adverse impact on the subsequent feature selection and MIC prediction. In actual supplementation, the supplemented data can be judged first. If the supplemented OD value data has exceeded the maximum OD value data corresponding to the growth relationship of the target strain when no antibacterial drug is added, that is, the 0 concentration curve set it to That is, after obtaining the OD values in the future preset time period, further adjustment can be made based on the following relational formula (4), and the adjusted OD value is used as the final predicted OD value:

[0108]

[0109] Among them, i = 17, 18, 19, 20. After the above four steps of supplementation, the data x1, x2,..., x 16 to be predicted for 8h becomes the data x1, x2,..., x 19 , x 20 .

[0110] As can be seen from the above, this embodiment achieves the effect of improving the overall prediction accuracy on the premise of effectively shortening the prediction time, and can guide drug use more accurately.

[0111] In the above embodiments, there is no limitation on how to execute step S103. In this embodiment, an optional generation method for the growth characteristics to be recognized is given, which may include the following content:

[0112] According to the optical density values of the target strain at each moment under the drug at the target concentration, calculate the central tendency information of the optical density, the dispersion and change information of the optical density, the continuous change information of the optical density, and the correlation degree information between adjacent moments under the drug at the target concentration;

[0113] According to the geometric metric space distance, linear correlation degree information, and vector space distance between the optical density data of the target strain under the drug at the target concentration and the optical density data of the target strain without adding the target drug;

[0114] According to the optical density data of the target strain under the drug at the target concentration, respectively, the geometric metric space distance, linear correlation degree information, and asymmetry metric information between the optical density data of the target strain and the optical density data of the target strain at each reference concentration of the target drug.

[0115] In this embodiment, the central tendency information of the optical density can be represented by the average value of the optical density values at different moments of the target strain under the drug at the target concentration of the target strain. The dispersion and change information of the optical density can be represented by the standard deviation and range of the optical density values at different moments of the target strain under the drug at the target concentration of the target drug. The continuous change information of the optical density can be represented by the difference information of the optical density values at different moments of the target strain in the drug at the target concentration, and the correlation degree information between adjacent moments can be represented by the autocorrelation coefficient of the optical density values at adjacent moments of the target strain under the drug at the target concentration. The geometric metric space distance includes but is not limited to the Manhattan distance and the Euclidean distance, the linear correlation degree information can be but is not limited to the correlation coefficient, the vector space distance can be but is not limited to the cosine similarity, and the asymmetry metric information can be but is not limited to the relative entropy. To make the technical solution of the present application clearer to those skilled in the art, the present application also provides an optional implementation manner of the above embodiments, which may include:

[0116] The optical density data of the target drug - target strain can be used to represent, where i represents the 0.5×i moment, j represents the jth concentration in ascending order of the corresponding concentration sequence, and the value ranges of i and j can be i = 1, 2,... n; j = 1, 2,... m. n and m are the total number of moments and the total number of concentrations of the OD value data in the target drug - target strain combination.

[0117] (1) The characteristics of the concentration curve itself data are extracted from the data of each concentration itself, reflecting the situation and change law of the OD value data of this concentration. That is, the concentration information characteristics for expanding the test data may include the following several characteristics:

[0118] 1) Average value

[0119]

[0120] The average value is a measure representing the central tendency of a set of data. It refers to the sum of all data in a set of data divided by the number of data in this set. The average value of the OD value data of the concentration curve at all times describes the central tendency of the OD value of this concentration and is closely related to the growth of bacteria at this concentration. Generally speaking, the higher the average value of the OD value data, the higher the probability of bacterial growth.

[0121] 2) Standard deviation:

[0122]

[0123] The standard deviation is the arithmetic square root of the variance. Since the variance is generated by taking the average of the square terms of the data, the magnitude of the variance is quite different from the data itself and is not easy to compare with the data. The standard deviation can reflect the degree of dispersion of the data, and it works better as a feature when combined with the mean. Specifically in this embodiment, the average value of the OD value data of the concentration curve itself will cause great interference to the average value result in the case of extreme values in the data. The standard deviation of the OD value data of the concentration curve itself can reasonably describe this situation, and combined with the average value, it can more accurately reflect the growth of bacteria corresponding to the concentration.

[0124] 3) Range:

[0125]

[0126] The range is used to describe the maximum value range of the data, which is obtained by subtracting the minimum value from the maximum value, and has advantages such as convenient calculation and intuitive meaning. Similar to the standard deviation, the range can also reflect the degree of dispersion of the data and is a very effective measurement feature. The range of the OD value data of the concentration curve itself is the difference between the maximum value and the minimum value of all data at all times. It can reflect the change of the OD value better than the maximum value. The larger the range of the OD value data, the higher the corresponding probability of bacterial growth.

[0127] 4) Maximum value of the first-order difference:

[0128]

[0129] The difference, that is, the difference function, whose result reflects the change of continuous data. The difference is divided into three types: forward difference, backward difference, and central difference. The forward difference is adopted in this embodiment. The maximum difference value is the case where the adjacent data changes the most, which can reflect the information inside the continuous data. The maximum value of the first-order difference of the OD value data of the concentration curve itself can reflect the maximum value of the change of the OD value data every 0.5h, and this value has a close relationship with the growth of bacteria. The larger the maximum value of the first-order difference of the OD value data, the higher the corresponding growth probability of bacteria.

[0130] 5) Maximum value of the second-order difference:

[0131]

[0132] Similar to the maximum value of the first-order difference, the second-order difference is to take the maximum value by repeating the difference operation on the first-order difference values again. Similar to the relationship between speed and acceleration, it can represent the deep information of data change. The maximum value of the second-order difference of the OD value data of the concentration curve itself can reflect the change of the first-order difference of the OD value data, and can be combined with the maximum value of the first-order difference to jointly reflect the growth of bacteria.

[0133] 6) Autocorrelation coefficient:

[0134]

[0135] where μ j , σ j represent the mean and standard deviation of the OD value at concentration j. The autocorrelation coefficient describes the correlation degree between the same random variable at two different times. Generally speaking, it is the influence of the past data of the random variable on the current data. According to the lag number, the specific time period can be selected. In this embodiment, the autocorrelation coefficient with a lag number of 1 can be selected. The autocorrelation coefficient of the OD value data of the concentration curve itself can describe the common information of the change of the OD value data at different times in this concentration curve.

[0136] The above 6 features are all features extracted from the concentration curve itself, and each concentration curve can extract these 6 features.

[0137] (2) Features jointly extracted with the 0-concentration curve. The growth curve with a drug concentration of 0, that is, the growth curve without adding drugs, must have growth in terms of the corresponding bacterial growth situation, which has good reference significance. The features jointly extracted by the concentration curve to be extracted and the 0-concentration curve can more accurately reflect the growth situation of the concentration to be extracted. In this embodiment, corresponds to the growth data of the 0-concentration at the 0.5×i moment. That is, the features of the similarity information extracted from the extended test data and the optical density data of the target bacterial species without adding the target drug can include the following features:

[0138] 1) Manhattan distance from the 0-concentration curve

[0139]

[0140] 2) Euclidean distance from the 0-concentration curve

[0141]

[0142] 3) Correlation coefficient with the 0-concentration curve

[0143]

[0144] where μ j , σ j represent the mean and standard deviation of the OD values at concentration j.

[0145] 4) Cosine similarity with the 0-concentration curve

[0146]

[0147] The second type of features are all features co-extracted with the 0-concentration curve. These 4 features can reflect the similarity degree between the concentration curve of the feature to be extracted and the 0-concentration curve from different perspectives. For each concentration curve of the feature to be extracted, these 4 features are calculated and extracted through the above relational expressions.

[0148] (3) Features co-extracted from all adjacent concentration curves. This type of features is the features co-extracted from all concentration curves and can accurately describe the overall information of the OD value data of all concentration curves under the target strain - target drug combination. Considering the higher dependence between adjacent concentrations, the features extracted here are all features of adjacent concentration curves. As the features of the specific numbered bacteria under the specific target strain - target drug combination, they are added to the features of each concentration curve to jointly judge the growth situation of the corresponding bacteria. That is, the similarity information features between the expanded test data and the optical density data of the target strain at multiple reference concentrations of the target drug can include the following multiple features:

[0149] 1) Manhattan distance between adjacent concentration curves

[0150]

[0151] 2) Euclidean distance between adjacent concentration curves

[0152]

[0153] 3) Correlation coefficient between adjacent concentration curves

[0154]

[0155] where μ j , σj Represents the mean and standard deviation of the OD value at concentration j.

[0156] 4) Relative entropy of adjacent concentration curves

[0157] First, perform normalization

[0158]

[0159] Then calculate

[0160]

[0161] These 4 types of features are all features jointly extracted from adjacent concentration curves. Similar to the second type of features, 4 angles are selected to calculate the similarity degree between adjacent concentration curves. Specifically, each angle will generate m - 1 features, that is, the third type of features can generate a total of 4×(m - 1) features.

[0162] In summary, for each concentration curve from which features are to be extracted, the above three types of features are extracted through the calculation formula, arranged in order to form the final feature vector, that is, the total features extracted from each concentration curve are 6 + 4 + 4×(m - 1) = 4m + 6 features.

[0163] In order to overcome the defects and problems of the existing technology, in this embodiment, a growth prediction model trained based on machine learning algorithms such as support vector machine algorithms or artificial neural network algorithms is used to perform binary classification on the growth curves of each reaction well to judge the positive and negative results of each reaction well. The above embodiment does not make any limitations on the training process of the model. This embodiment also provides an optional training method for the growth prediction model, which may include the following contents:

[0164] Pre - build a machine learning model framework;

[0165] Obtain the historical optical density data of the target strain at different drug concentrations in the absence of antibacterial drugs and different antibacterial drugs, and set corresponding labels for each historical optical density data. The labels include growth labels and non - growth labels;

[0166] Divide the historical optical density data into a training sample set and a test sample set according to a preset ratio;

[0167] For each training sample in the training sample set, generate training growth features according to the concentration information features of the current training sample extracted, the similarity information features between the current training sample and the optical density data of the target strain without adding antibacterial drugs, and the optical density data of the target strain at multiple reference concentrations of the target antibacterial drug;

[0168] For each test sample in the test sample set, based on the optical density data of the target bacterial species under different concentrations of the target antibacterial drug and without the addition of antibacterial drugs, determine the growth information of the target bacterial species in the current test sample in a preset future time period, so as to obtain the augmented test sample data of the current test sample;

[0169] Use each training sample in the training sample set and its corresponding training growth characteristics to continuously train the machine learning model until the accuracy of the trained machine learning model tested with the augmented test sample data of each test sample set is greater than the preset accuracy threshold, and obtain the growth prediction model.

[0170] In this embodiment, first, through operations such as artificial culture and historical data extraction, collect the OD values generated by the microplate reader to obtain a large amount of OD value data and its growth conditions under the same model (specific antibacterial drug - specific bacterial species), and label the growth label or non-growth label. For example, 0 represents the non-growth label, and 1 represents the growth label. The obtained data can be divided into a training set and a test set according to a ratio of 7:3 or 8:2. For the training set, the extraction of training growth characteristics can be carried out based on the Figure 4 method shown. Specifically, starting from the original OD value data of the training set, divide it into data of multiple concentrations according to different drug concentrations. For each concentration of data, extract the three types of characteristics as shown in Figure 4 , that is, the growth curve characteristics, the characteristics between the growth curve and the positive control curve, and the characteristics between the growth curve and the curves of the same antibiotic and the same bacteria at different concentrations of 0.5×i. Combine the three types of characteristics to form a feature vector corresponding to each concentration. For the test set, the extraction of test growth characteristics can be carried out based on the Figure 5 method shown. The method of extracting characteristics from the test set is the same as that of the training set. When the test set is actually tested, since the dynamic test method is used and in order to achieve the ultimate goal of early prediction, all OD value data cannot be obtained. For the OD value data within several hours, according to the statistical growth law of the OD value data of the model, a predictive supplement can be made to the data within several hours. The specific supplement time and supplement method are determined by statistics, and the supplement methods for different models are slightly different.

[0171] Based on the above embodiments, the present application also takes the machine learning models as the support vector machine and the artificial neural network model as examples to elaborate on various optional implementation manners of the above embodiments, which may include the following content:

[0172] For the machine learning model being the support vector machine, use each training sample in the training sample set and its corresponding training growth characteristics to continuously train the machine learning model, including:

[0173] Randomly construct an initial hyperplane according to the training growth characteristics of each training sample; obtain the corresponding decision function based on the initial hyperplane, and determine the optimization problem according to the decision function and the preset penalty parameter; estimate the parameters of the initial hyperplane by calculating the optimization problem to obtain the optimal parameters of the support vector machine; determine the support vector machine obtained in the current training cycle according to the optimal parameters.

[0174] In this embodiment, the support vector machine, also known as Support Vector Machine, abbreviated as SVM, is a class of generalized linear classifiers that perform binary classification on data in a supervised learning manner. The support vector machine is a widely used binary classification algorithm. Different from the perceptron algorithm, the support vector machine also requires the geometric margin from the hyperplane to be maximized on the basis of correct classification, and it is a linear classification algorithm defined in the feature space. In fact, after using the kernel method, the support vector machine algorithm can become a non-linear classification algorithm, and the linear support vector machine algorithm used here. The essence of the support vector machine is to set a hyperplane (such as Figure Four the yin and yang parts in the figure), and the data on both sides of the hyperplane belong to different categories. In fact, Figure 6 it can be seen that the support vector machine has natural processing advantages for binary classification problems, and the problem of whether the bacteria grow or not in this application is exactly a binary classification problem, which is very suitable for this model. The specific training process can include the following steps:

[0175] A1: Purchase or cooperate with units such as hospitals to obtain the strains to be tested, and the strains to be tested can be any pathogenic strains.

[0176] A2: Record the OD value data every half hour through the instrument optical module such as an enzyme-linked immunosorbent assay (ELISA) reader.

[0177] A3: Use the traditional broth dilution method for 18 - 24 hours of culture to obtain the accurate growth condition of the strain to be tested;

[0178] A4: Obtain all the OD value data and their corresponding growth conditions, where 0 represents that the bacteria have not grown, and 1 represents that the bacteria have grown.

[0179] A5: For the support vector machine model, divide all the data into a training set and a test set. Among them, the training set is used as training data to train the model, and the test set is used to verify the model.

[0180] A6: For the training set data, use the feature extraction method provided in S103 above to extract and compress the OD values.

[0181] A7: For the test set data, for the OD value data within a short time (such as within 8 hours), use the method provided in S102 above to supplement the growth, and use the statistical law of historical data to supplement the data to a certain extent.

[0182] A8: The feature extraction method provided in S103 extracts features from the supplemented test set.

[0183] A9: Use the training set to iterate the parameters of the support vector machine model.

[0184] For a specific concentration data x to be predicted of the target drug - target strain, the specific process is as follows:

[0185] 1) Extract the historical data of the drug - strain model where the concentration data x to be predicted is located, and extract features from the historical data according to the feature extraction rules in S103. All concentration data obtain 4m + 6 features. And centralize these features as the training set. Change the data with label 0 to -1 for convenient use, and obtain the training set T as follows:

[0186] T = {(x1, y1), (x2, y2), …, (x N , y N )}

[0187] Among them, x i corresponds to the feature vector y i is its growth situation, that is, the label is -1 or 1.

[0188] 2) Randomly construct the initial hyperplane w·x i +b = 0 according to the extracted feature dimension, and obtain the corresponding decision function f.

[0189] f(x i ) = sign(w·x i +b)

[0190] 3) According to the theoretical knowledge of the support vector machine, transform the optimized problem after transformation:

[0191]

[0192] s.t.y i (w·x i +b)≥1 + ξ i , i = 1, 2, …, N

[0193] ξ i ≥0, i = 1, 2, …, N

[0194] Among them, C is the penalty parameter (usually can be set to 1 or 2), and its role is to adjust the interval size and the importance of misclassification.

[0195] 4) Using the solution of the above optimized problem, the parameters of the hyperplane can be estimated, and finally the optimal parameters w * , b* , the corresponding decision function is:

[0196] f * (x i ) = sign(w * ·x i +b * )

[0197] 5) First, perform 2h of manual supplementation on the data to be predicted according to the above S102. Similarly, perform feature extraction and centering processing on the supplemented data to be predicted x to obtain the vector x = (x 1 , x 2 , …, x 4m+6 ). Then, substitute it into the optimal decision function f * . Judge the result f * (x). If f * (x) >= 0, then classify this concentration as growth, and the label is 1; if f * (x) < 0, then classify this concentration as non - growth, and the label is 0. The label is changed to - 1 before the model operation only to facilitate the prediction of the support vector machine, and it can be adjusted back to the normal value after the prediction is completed.

[0198] A10: Use the trained support vector machine model to predict the test set data to obtain the predicted values of the growth conditions of each concentration.

[0199] A11: Obtain the MIC predicted value according to the predicted values of the growth conditions of each concentration.

[0200] After all the concentration data of the target drug - target strain are predicted and classified, according to the calculation rule of MIC, the minimum inhibitory concentration MIC of the target drug - target strain is obtained finally.

[0201] A12: Compare the MIC predicted value with the true MIC to obtain the model accuracy.

[0202] If the model accuracy meets the requirements, stop training, and the parameters obtained in A9 are the parameters of the finally trained model. If the model accuracy does not meet the requirements, train again, that is, repeat A9 until the requirements are met. After obtaining the support vector machine model through training, according to the finally obtained growth prediction accuracy of each concentration and the finally desired MIC prediction accuracy.

[0203] As can be seen from the above, the model obtained by training with the support vector machine algorithm in this embodiment is very appropriate for predicting the growth situation, and can effectively improve the prediction accuracy of the growth situation of the strain.

[0204] If the machine learning model is an artificial neural network model, the process of continuously training the machine learning model using each training sample in the training sample set and its corresponding training growth characteristics may include:

[0205] The artificial neural network model includes an input layer, a first neural network layer, a second neural network layer, and an output layer; the training growth characteristics of each training sample in the training sample set are input into the artificial neural network model to obtain the minimum inhibitory concentration prediction value corresponding to each training sample; according to the minimum inhibitory concentration prediction value and the true value of the minimum inhibitory concentration of each training sample, the loss function of the artificial neural network model in the current training cycle is calculated; using the BP neural network algorithm, the loss function is optimized according to the Adam algorithm to obtain the optimal parameters of the artificial neural network model in the current training cycle; the artificial neural network model obtained in the current training cycle is determined according to the optimal parameters.

[0206] In this embodiment, the artificial neural network, also known as the connection model, generally has a very complex structure, and the relationship between the nodes between each layer of the network is different. According to these complex relationships, the purpose of processing complex information can be achieved. Before training the model, the artificial neural network can be constructed first. The artificial neural network is mainly composed of an input layer, a hidden layer, and an output layer, and each layer of the neural network is composed of several nodes. Different from traditional machine learning algorithms, the neural network is not a fixed dead algorithm. It does not perform operations step by step according to a written program, but can summarize rules through the self-adaptability of the network to achieve a predetermined purpose. By introducing a non-linear activation function into the neural network, the neural network can theoretically approximate any non-linear function, greatly increasing the application range of the neural network. The artificial neural network of this embodiment is as Figure 7 shown as a three-layer neural network: the first layer of the neural network, the number of its nodes can be m128, and the ReLU function is taken as the activation function of this layer of the network. The second layer of the neural network, the number of its nodes is m64, and the Sigmoid function is taken as the activation function of this layer of the network. The third layer of the neural network, that is, the output layer, is constructed, the number of its nodes is 1, and the Sigmoid function is selected as the activation function of this layer of the network. The input layer is a 4m + 6-dimensional feature vector; the first layer has 8m + 12 neurons; the second layer has 4m + 6 neurons; the third layer is the output layer, with only one neuron. For this third layer of the neural network, the input data is This data is the total feature spliced by three types of features, and the output after passing through the third layer of the neural network is

[0207] Record the OD value data of the target bacterial strain in the target drug every half hour through an instrument optical module such as an enzyme-labeling instrument. Traverse these data one by one in order. When traversing to the data of each concentration, use it as the test set, and use the data of each concentration of the remaining data as the training set. The specific traversal method is as follows: For a certain antibiotic-bacterial strain combination, when traversing to the data of a specific antibiotic-bacterial strain of a certain patient, we select all other data of this antibiotic-bacterial strain as the training set. The data of the specific antibiotic-bacterial strain unique to the traversed patient is used as the test set. Traverse all patients under this combination in turn. Each time a traversal is performed, the following operations are carried out:

[0208] 1) Input the training set into the neural network model for training.

[0209] As Figure 8 shown, input all the training set data i = 1, 2, …, N into the initialized three-layer neural network as Figure 7 shown, and the obtained output value and the true value y i make cross-entropy LOSS, and accumulate and sum according to the following formula to obtain the loss function:

[0210]

[0211] Use the BP (Back Propagation) neural network algorithm to optimize the loss function according to the Adam (Adaptive Momentum Estimation) algorithm to obtain the optimal parameters of the neural network.

[0212] Among them, the common methods for judging the termination conditions are as follows:

[0213] 1. Specified number of times method, specify the number of iterations of the neural network, such as 50, 100, etc.

[0214] 2. Specified error method, such as setting an error threshold, and stop the iteration when the error is less than this threshold.

[0215] 3. Iterative determination method, when the degree of reduction of the error in each iteration is not obvious enough, the iteration can be stopped.

[0216] Specifically, the first specified number of times method is used in the model of this embodiment. The specific parameters are related to each drug model and can be flexibly determined according to experiments.

[0217] 2) Use the trained model to predict the test set. Specifically: Predict the growth situation corresponding to the concentration according to the OD value data of each concentration of the drug of this patient in the test set, and save the prediction results of each concentration.

[0218] In this step, when using the trained neural network for testing, only extract features by using the method of extracting features from the test set, and input the extracted features into the neural network to obtain the growth prediction accuracy rates for each concentration.

[0219] 3) Obtain the growth prediction accuracy rates of each model concentration according to all the prediction results

[0220] As Figure 9 described, calculate the MIC according to the positive and negative classification results of the micro-wells on the drug susceptibility plate and the drug layout information of the micro-wells; judge the drug resistance result according to the MIC result and the breakpoint value standard.

[0221] The embodiment of the present invention also provides a corresponding device for the drug susceptibility result recognition method, which further makes the method more practical. Among them, the device can be described from the perspective of functional modules and the perspective of hardware respectively. The drug susceptibility result recognition device provided by the embodiment of the present invention will be introduced below, and the drug susceptibility result recognition device described below can be mutually referred to the drug susceptibility result recognition method described above.

[0222] From the perspective of functional modules, refer to Figure 10 , Figure 10 which is the structural diagram of the drug susceptibility result recognition device provided by the embodiment of the present invention under a specific embodiment. The device may include:

[0223] A data acquisition module 101, configured to acquire data to be tested, and the data to be tested is the optical density data of the target strain in the target concentration drug within the first time period.

[0224] A data expansion module 102, configured to determine the growth information of the target strain in a future preset time period according to the optical density data of the target strain without adding the target drug and the optical density data of the target strain at different concentrations of the target drug, so as to obtain expanded test data.

[0225] A feature extraction module 103, configured to generate growth features to be recognized according to the concentration information features of the extracted expanded test data, the similarity information features between the expanded test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug;

[0226] A growth prediction module 104, configured to input the growth features to be recognized into a pre-trained growth prediction model to obtain the growth situation prediction value of the target strain in the target drug.

[0227] Optionally, in some embodiments of this embodiment, the above device may further include a MIC prediction module, configured to obtain the predicted growth values of the target strain at different concentrations of the target drug; and determine the minimum inhibitory concentration prediction value of the target strain corresponding to the target drug according to the predicted growth values of the target strain at different concentrations of the target drug.

[0228] In some other embodiments of this embodiment, the above data augmentation module 102 may further be configured to: determine the original growth rate of the target strain according to the optical density data of the target strain at each known moment in the first time period; determine the general growth rate of the target strain according to the historical growth data of the target strain at different concentrations of the target drug and the original growth rate; determine the initial predicted optical density at multiple initial future moments within a preset future time period according to the optical density value of the target strain at the target known moment and the original growth rate; determine the later predicted optical density values at multiple later future moments within the preset future time period according to the general growth rate and each initial predicted optical density value; determine the predicted optical density data of the data to be tested within the preset future time period based on the original maximum optical density value of the target strain without adding the target drug, each initial predicted optical density value, and each later predicted optical density value; and use the predicted optical density data and the data to be tested as augmented test data.

[0229] As an optional implementation manner of the above embodiment, the above data augmentation module 102 may further be configured to: determine the maximum optical density value of the target strain and the corresponding target moment according to the historical growth data of the target strain at different concentrations of the target drug; determine the growth rate change rate according to the maximum optical density value and the optical density values at multiple moments before the target moment; and determine the general growth rate of the target strain according to the original growth rate and the growth rate change rate.

[0230] In some other embodiments of this embodiment, the above feature extraction module 103 may further be configured to: calculate the optical density central tendency information, optical density dispersion and change information, optical density continuous change information, and the correlation degree information between adjacent moments under the target concentration drug according to the optical density values of the target strain at each moment under the target concentration drug; calculate the geometric metric space distance, linear correlation degree information, and vector space distance between the optical density data of the target strain under the target concentration drug and the optical density data of the target strain without adding the target drug; and calculate the geometric metric space distance, linear correlation degree information, and asymmetry metric information between the optical density data of the target strain under the target concentration drug and the optical density data of the target strain of the target drug at each reference concentration.

[0231] Optionally, in some other embodiments of this embodiment, the above device may further include a model training module, which is used to pre-build a machine learning model framework; obtain the historical optical density data of the target bacterial strain at different drug concentrations in the absence of antibacterial drugs and different antibacterial drugs, and set corresponding labels for each piece of historical optical density data, where the labels include growth labels and non-growth labels; divide the historical optical density data into a training sample set and a test sample set according to a preset ratio; for each training sample in the training sample set, generate training growth features according to the extracted concentration information features of the current training sample, the similarity information features between the current training sample and the optical density data of the target bacterial strain without added antibacterial drugs, and the optical density data of the target bacterial strain at multiple reference concentrations of the target antibacterial drug; for each test sample in the test sample set, determine the growth information of the target bacterial strain in the current test sample in a preset future time period according to the optical density data of the target bacterial strain at different concentrations of the absence of antibacterial drugs and the target antibacterial drug, so as to obtain the extended test sample data of the current test sample; use each training sample in the training sample set and its corresponding training growth features to continuously train the machine learning model until the accuracy of the trained machine learning model tested by the extended test sample data of each test sample set is greater than a preset accuracy threshold, and obtain a growth prediction model.

[0232] As an alternative implementation of the above embodiment, the above model training module may further be used for: when the machine learning model is a support vector machine, randomly construct an initial hyperplane according to the training growth features of each training sample; obtain the corresponding decision function according to the initial hyperplane, and determine the optimization problem according to the decision function and a preset penalty parameter; estimate the parameters of the initial hyperplane by calculating the optimization problem to obtain the optimal parameters of the support vector machine; determine the support vector machine obtained in the current training cycle according to the optimal parameters.

[0233] As another alternative implementation of the above embodiment, the above model training module may further be used for: when the machine learning model is an artificial neural network model, the artificial neural network model includes an input layer, a first neural network layer, a second neural network layer, and an output layer; input the training growth features of each training sample in the training sample set into the artificial neural network model to obtain the minimum inhibitory concentration prediction value corresponding to each training sample; calculate the loss function of the artificial neural network model in the current training cycle according to the minimum inhibitory concentration prediction value and the true value of the minimum inhibitory concentration of each training sample; use the BP neural network algorithm to optimize the loss function according to the Adam algorithm to obtain the optimal parameters of the artificial neural network model in the current training cycle; determine the artificial neural network model obtained in the current training cycle according to the optimal parameters.

[0234] The functions of the functional modules of the drug sensitivity result recognition device according to the embodiments of the present invention can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can refer to the relevant descriptions in the above method embodiments and will not be elaborated here.

[0235] As can be seen from the above, the embodiments of the present invention can efficiently, accurately, and at low cost identify drug sensitivity results.

[0236] The drug sensitivity result recognition device mentioned above is described from the perspective of functional modules. Further, the present application also provides an electronic device, which is described from the perspective of hardware. Figure 11 It is a schematic structural diagram of the electronic device provided by the embodiments of the present application in an implementation manner. As Figure 11 shown, the electronic device includes a memory 110 for storing computer programs; a processor 111 for implementing the steps of the drug sensitivity result recognition method mentioned in any of the above embodiments when executing the computer programs.

[0237] Among them, the processor 111 may include one or more processing cores, such as a 4-core processor or an 8-core processor. The processor 111 may also be a controller, a microcontroller, a microprocessor, or other data processing chips. The processor 111 may be implemented in at least one hardware form of DSP (Digital Signal Processing), FPGA (Field-Programmable Gate Array), or PLA (Programmable Logic Array). The processor 111 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the wake state, also known as the CPU (Central Processing Unit); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 111 may be integrated with a GPU (Graphics Processing Unit) for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 111 may also include an AI (Artificial Intelligence) processor for processing computational operations related to machine learning.

[0238] The memory 110 may include one or more computer-readable storage media, which may be non-transitory. The memory 110 may also include high-speed random access memory and non-volatile memory, such as one or more disk storage devices and flash storage devices. In some embodiments, the memory 110 may be an internal storage unit of an electronic device, such as the hard disk of a server. In other embodiments, the memory 110 may also be an external storage device of an electronic device, such as a plug-in hard disk equipped on a server, a Smart Media Card (SMC), a Secure Digital (SD) card, a Flash Card, etc. Further, the memory 110 may also include both an internal storage unit and an external storage device of the electronic device. The memory 110 can be used not only to store application software installed in the electronic device and various types of data, such as the code of a program for executing the vulnerability handling method, etc., but also to temporarily store data that has been output or will be output. In this embodiment, the memory 110 is at least used to store the following computer program 1101, wherein, after being loaded and executed by the processor 111, the computer program can implement the relevant steps of the drug sensitivity result identification method disclosed in any of the foregoing embodiments. Additionally, the resources stored in the memory 110 may also include an operating system 1102 and data 1103, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 1102 may include Windows, Unix, Linux, etc. The data 1103 may include, but is not limited to, data corresponding to the drug sensitivity result identification results, etc.

[0239] In some embodiments, the above-mentioned electronic device may further include a display screen 112, an input / output interface 113, a communication interface 114 or a network interface, a power supply 115, and a communication bus 116. Among them, the display screen 112 and the input / output interface 113 such as a keyboard belong to user interfaces. Optional user interfaces may also include standard wired interfaces, wireless interfaces, etc. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, and an OLED (Organic Light-Emitting Diode) toucher, etc. The display may also be appropriately referred to as a display screen or a display unit, which is used to display the information processed in the electronic device and to display a visual user interface. The communication interface 114 may optionally include a wired interface and / or a wireless interface, such as a WI-FI interface, a Bluetooth interface, etc., and is generally used to establish a communication connection between the electronic device and other electronic devices. The communication bus 116 may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. This bus may be divided into an address bus, a data bus, a control bus, etc. For the sake of convenience of representation, Figure 11 it is only represented by a thick line in the figure, but it does not mean that there is only one bus or one type of bus.

[0240] Those skilled in the art can understand that Figure 11 the structure shown in the figure does not constitute a limitation on the electronic device, and it may include more or fewer components than those shown in the figure. For example, it may further include sensors 117 for implementing various functions.

[0241] The functions of the functional modules of the electronic device described in the embodiments of the present invention can be specifically implemented according to the methods in the above method embodiments. The specific implementation process can refer to the relevant descriptions of the above method embodiments and will not be elaborated here.

[0242] As can be seen from the above, the embodiments of the present invention can efficiently, low-costly, and accurately identify the drug sensitivity results.

[0243] It can be understood that if the drug sensitivity result recognition method in the above embodiments is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and executes all or part of the steps of the methods in the various embodiments of the present application. The aforementioned storage medium includes: USB flash drives, mobile hard disks, read-only memories (ROM), random access memories (RAM), electrically erasable programmable ROMs, registers, hard disks, multimedia cards, card-type memories (such as SD or DX memories, etc.), magnetic memories, removable disks, CD-ROMs, magnetic disks, or optical discs, etc., all kinds of media that can store program codes.

[0244] Based on this, the embodiments of the present invention further provide a readable storage medium storing a computer program, and when the computer program is executed by a processor, it performs the steps of the drug sensitivity result recognition method as described in any one of the above embodiments.

[0245] In this specification, the various embodiments are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts among the various embodiments, reference can be made to each other. For the hardware, including devices and electronic devices, disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For the relevant parts, reference can be made to the description in the method part.

[0246] Those skilled in the art can further realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be implemented by electronic hardware, computer software, or a combination of the two. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described according to their functions in the above description. Whether these functions are executed in a hardware or software manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered to exceed the scope of the present invention.

[0247] The above has introduced in detail a drug sensitivity result recognition method, device, electronic device and readable storage medium provided by the present application. Specific examples are used in this article to elaborate on the principles and implementation manners of the present invention. The description of the above embodiments is only used to help understand the method and its core idea of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and modifications can still be made to the present application, and these improvements and modifications also fall within the protection scope of the claims of the present application.

Claims

1. A drug sensitivity result recognition method, characterized in that, Including: Obtain the data to be tested, where the data to be tested is the optical density data of the target strain in the target concentration of the drug within the first time period; According to the optical density data of the target strain when no target drug is added and the optical density data of the target strain at different concentrations of the target drug, determine the growth information of the target strain in a preset future time period to obtain extended test data; Generate the growth characteristics to be identified based on the concentration information characteristics of the extracted extended test data, the similarity information characteristics between the extended test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug; Input the growth characteristics to be identified into a pre-trained growth prediction model to obtain the predicted growth value of the target strain in the target drug; Wherein, the process of obtaining the extended test data includes: Determine the original growth rate of the target strain according to the optical density data of the target strain at each known moment in the first time period; determine the general growth rate of the target strain according to the historical growth data of the target strain at different concentrations of the target drug and the original growth rate; determine the initial predicted optical density at multiple initial future moments in the preset future time period according to the optical density value of the target strain at a target known moment and the original growth rate; determine the later predicted optical density values at multiple later future moments in the preset future time period according to the general growth rate and each initial predicted optical density value; based on the original maximum optical density value of the target strain without adding the target drug, each initial predicted optical density value, and each later predicted optical density value, determine the predicted optical density data of the data to be tested in the preset future time period; use the predicted optical density data and the data to be tested as the extended test data; Wherein, the process of generating the growth characteristics to be identified includes: Calculate the optical density central tendency information, optical density dispersion and change information, optical density continuous change information, and correlation degree information between adjacent moments under the target concentration of the drug according to the optical density values of the target strain at each moment under the target concentration of the drug; calculate the geometric metric space distance, linear correlation degree information, and vector space distance between the optical density data of the target strain under the target concentration of the drug and the optical density data of the target strain without adding the target drug; calculate the geometric metric space distance, linear correlation degree information, and asymmetry metric information between the optical density data of the target strain under the target concentration of the drug and the optical density data of the target strain at each reference concentration of the target drug.

2. The drug sensitivity result identification method according to claim 1, characterized in that After inputting the growth characteristics to be identified into a pre-trained growth prediction model to obtain the predicted growth value of the target strain in the target drug, it further includes: Obtain the predicted growth values of the target strain at different concentrations of the target drug; Determine the minimum inhibitory concentration prediction value of the target strain corresponding to the target drug according to the growth condition prediction values of the target strain at different concentrations of the target drug.

3. The drug sensitivity result identification method according to claim 1, wherein The determining of the general growth rate of the target strain according to the historical growth data of the target strain at different concentrations of the target drug and the original growth rate includes: Determine the maximum optical density value of the target strain and the corresponding target time according to the historical growth data of the target strain at different concentrations of the target drug. Determine the growth rate change rate according to the maximum optical density value and the optical density values at multiple times before the target time. Determine the general growth rate of the target strain according to the original growth rate and the growth rate change rate.

4. The drug sensitivity result identification method according to any one of claims 1 to 3, characterized in that Before inputting the growth characteristics to be recognized into a pre-trained growth prediction model to obtain the growth condition prediction value of the target strain in the target drug, it further includes: Pre-build a machine learning model framework. Obtain the historical optical density data of the target strain at different drug concentrations in the absence of antibacterial drugs and different antibacterial drugs respectively, and set corresponding labels for each historical optical density data, where the labels include growth labels and non-growth labels. Divide the historical optical density data into a training sample set and a test sample set according to a preset ratio. For each training sample in the training sample set, generate training growth characteristics according to the concentration information characteristics of the current training sample extracted, the similarity information characteristics between the current training sample and the optical density data of the target strain without adding antibacterial drugs, and the optical density data of the target strain at multiple reference concentrations of the target antibacterial drug. For each test sample in the test sample set, determine the growth information of the target strain in the current test sample in a preset future time period according to the optical density data of the target strain at different concentrations of the target antibacterial drug without adding antibacterial drugs, so as to obtain the extended test sample data of the current test sample. Use each training sample in the training sample set and its corresponding training growth characteristics to continuously train the machine learning model until the accuracy of the trained machine learning model tested by the extended test sample data of each test sample set is greater than a preset accuracy threshold to obtain a growth prediction model.

5. The drug sensitivity result identification method according to claim 4, wherein The machine learning model is a support vector machine. The continuously training the machine learning model using each training sample in the training sample set and its corresponding training growth characteristics includes: Randomly construct an initial hyperplane according to the training growth characteristics of each training sample. Obtain the corresponding decision function according to the initial hyperplane, and determine the optimization problem according to the decision function and a preset penalty parameter. Estimate the parameters of the initial hyperplane by calculating the optimization problem to obtain the optimal parameters of the support vector machine. Determine the support vector machine obtained in the current training cycle according to the optimal parameters.

6. The drug sensitivity result identification method according to claim 4, characterized in that, The machine learning model is an artificial neural network model. The continuously training the machine learning model using each training sample in the training sample set and its corresponding training growth characteristics includes: The artificial neural network model includes an input layer, a first neural network layer, a second neural network layer, and an output layer; Input the training growth characteristics of each training sample in the training sample set into the artificial neural network model to obtain the minimum inhibitory concentration prediction value corresponding to each training sample; According to the minimum inhibitory concentration prediction value and the true value of the minimum inhibitory concentration of each training sample, calculate the loss function of the artificial neural network model in the current training cycle; Use the BP neural network algorithm to optimize the loss function according to the Adam algorithm to obtain the optimal parameters of the artificial neural network model in the current training cycle; Determine the artificial neural network model obtained in the current training cycle according to the optimal parameters.

7. A drug sensitivity result recognition device, characterized in that, Including: A data acquisition module for acquiring the data to be tested, where the data to be tested is the optical density data of the target strain in the target concentration drug within the first time period; A data augmentation module for determining the growth information of the target strain in a future preset time period according to the optical density data of the target strain when no target drug is added and the optical density data of the target strain at different concentrations of the target drug, so as to obtain augmented test data; A feature extraction module for generating the growth characteristics to be recognized according to the concentration information characteristics of the augmented test data extracted, the similarity information characteristics between the augmented test data and the optical density data of the target strain without adding the target drug, and the optical density data of the target strain at multiple reference concentrations of the target drug; A growth prediction module for inputting the growth characteristics to be recognized into a pre-trained growth prediction model to obtain the predicted value of the growth situation of the target strain in the target drug; Wherein, the data augmentation module is further used for: Determine the original growth rate of the target strain according to the optical density data of the target strain at each known moment in the first time period; determine the general growth rate of the target strain according to the historical growth data of the target strain at different concentrations of the target drug and the original growth rate; determine the initial predicted optical density at multiple initial future moments in the future preset time period according to the optical density value of the target strain at the target known moment and the original growth rate; determine the later predicted optical density values at multiple later future moments in the future preset time period according to the general growth rate and each initial predicted optical density value; based on the original maximum optical density value of the target strain without adding the target drug, each initial predicted optical density value, and each later predicted optical density value, determine the predicted optical density data of the data to be tested in the future preset time period; use the predicted optical density data and the data to be tested as augmented test data; Wherein, the feature extraction module is further used for: Calculate the central tendency information of the optical density, the discrete and variation information of the optical density, the continuous variation information of the optical density, and the correlation degree information between adjacent time points based on the optical density values of the target strain at each time point under the drug at the target concentration; calculate the geometric metric space distance, the linear correlation degree information, and the vector space distance between the optical density data of the target strain under the drug at the target concentration and the optical density data of the target strain without adding the target drug; calculate the geometric metric space distance, the linear correlation degree information, and the asymmetry metric information between the optical density data of the target strain under the drug at the target concentration and the optical density data of the target strain at each reference concentration of the target drug.

8. An electronic device, characterized in that, It includes a processor and a memory. When the processor executes the computer program stored in the memory, it implements the steps of the drug sensitivity result identification method according to any one of claims 1 to 6.

9. A readable storage medium, characterized in that, A computer program is stored on the readable storage medium. When the computer program is executed by the processor, it implements the steps of the drug sensitivity result identification method according to any one of claims 1 to 6.

Images