Method for judging sensitive type of escherichia coli to cefotaxime
By combining laser microconfocal Raman spectroscopy technology and improved convolutional neural network model, the high cost and complex operation problems in the identification of cefotaxime sensitive types in E. coli is solved, and a fast, accurate and low-cost detection effect is achieved.
Patent Information
- Application Number
- CN202510118608.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-24
- Publication Date
- 2025-06-20
AI Technical Summary
In the prior art, when judging the sensitive type of E. coli to cefotaxime, there are problems of high cost and complex operation, which is difficult to meet the needs of rapid clinical testing.
Laser microconfocal Raman spectroscopy technology combined with improved convolutional neural network (AlexNet) model is used to pre-process and classify Raman spectroscopy data to achieve rapid discrimination of E. coli sensitive types to cefotaxime.
This method simplifies the testing process, improves the testing speed and accuracy, reduces the operation difficulty and cost, is suitable for promotion in primary medical institutions, and meets the clinical fast and reliable drug sensitivity testing needs.
Smart Images
Figure CN120183684A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a method for discriminating the sensitive type of bacteria to cefotaxime, specifically a method for discriminating the sensitive type of Escherichia coli to cefotaxime. Background Art
[0002] Since the discovery of antibiotics, they have greatly promoted the treatment of infectious diseases and the progress of public health. However, the abuse and overuse of antibiotics have led to an increasingly serious problem of bacterial drug resistance. The World Health Organization (WHO) has listed bacterial drug resistance as one of the top ten global public health threats. The drug resistance problem has seriously affected the clinical treatment effect, significantly increased the treatment cost and the mortality rate of patients. Therefore, quickly and accurately detecting the sensitive type of bacteria to antibiotics has become an important research direction for solving the drug resistance problem, and the sensitive types include sensitivity and drug resistance. As the main pathogen causing urinary tract infections, abdominal infections and sepsis, the sensitivity detection of Escherichia coli to cefotaxime is particularly important. Cefotaxime is a third-generation cephalosporin widely used in clinical practice, but with the evolution of bacterial drug resistance mechanisms, such as the production of β-lactamase, its treatment effect has gradually declined. Accurately evaluating the sensitivity of bacteria to cefotaxime is of great significance for guiding rational clinical drug use and antibiotic management.
[0003] Currently, the methods for detecting bacterial drug resistance mainly include traditional detection methods, commercial systems and emerging technologies developed in recent years. Traditional methods such as the disk diffusion method and broth dilution method rely on bacterial culture and biochemical tests. These methods are technically mature and the results are accurate, but the detection cycle is relatively long (usually 18 hours to several days), and the operation is cumbersome, which is difficult to meet the needs of rapid clinical detection. In order to improve the detection efficiency, commercial systems such as 2System and Phoenix Automated Microbiology System shorten the detection time to about 18 hours through an automated process. These systems use technologies such as optical measurement and redox reaction, significantly improving the operational convenience. However, they still rely on bacterial culture, and have not fundamentally solved the time-consuming problem. At the same time, the high equipment cost also limits their promotion in primary medical institutions.
[0004] In recent years, emerging technologies (such as real-time imaging, biochemical detection and physical and chemical detection) have provided new ideas for bacterial sensitivity detection. Real-time imaging technology evaluates antibiotic sensitivity by monitoring the morphological changes of bacteria; biochemical detection determines drug sensitivity by detecting specific markers such as β-lactamase; physical and chemical detection (such as Raman spectroscopy and mass spectrometry) quickly evaluates sensitivity by analyzing the physical and chemical properties of bacterial cells. Although these technologies have made breakthroughs in detection speed and sensitivity classification, their high cost, operational complexity and the fact that some technologies still rely on the culture step limit their application in point-of-care testing (POCT). Summary of the Invention
[0005] The object of the present invention is to provide a method for discriminating the sensitive type of Escherichia coli to cefotaxime, so as to solve the problems of high cost and complex operation in classifying the sensitivity of Escherichia coli to cefotaxime.
[0006] The object of the present invention is achieved as follows:
[0007] A method for discriminating the sensitive type of Escherichia coli to cefotaxime, comprising the following steps:
[0008] S1. Obtain Escherichia coli from different patients. For one strain of Escherichia coli from each patient, measure the sensitive type of cefotaxime using the gold standard BMD method, and for the other strain, measure the Raman spectral data using a Raman spectrometer; until the sensitive types of Escherichia coli from at least two patients to cefotaxime are sensitivity, and the sensitive types of Escherichia coli from at least two patients to cefotaxime are drug resistance;
[0009] S2. Select the Raman spectral data of Escherichia coli with measurement results of drug resistance and sensitivity, and set corresponding drug sensitivity labels for the Raman spectral data; the number of selected Escherichia coli strains with drug resistance is the same as the number of Escherichia coli strains with sensitivity;
[0010] S3. Perform preprocessing on the Raman spectral data, including wavenumber selection, background removal, smoothing and noise reduction, and normalization;
[0011] S4. Remove the third convolutional layer to the fifth convolutional layer in the AlexNet model, and replace the ReLU activation function in the AlexNet model with the Leaky ReLU activation function to obtain a classification model;
[0012] S5. Divide the processed Raman spectral data with drug sensitivity labels into a training set and a validation set, and input them into the classification model for training to obtain a classification model for discriminating the sensitive type of Escherichia coli to cefotaxime;
[0013] S6. Preprocess the tested Escherichia coli, and collect the Raman spectral data of the preprocessed tested Escherichia coli; perform preprocessing on the Raman spectral data, including wavenumber selection, background removal, smoothing and noise reduction, and normalization; input the preprocessed Raman spectral data of the tested Escherichia coli into the classification model, and determine the sensitive type of the tested Escherichia coli according to the output result.
[0014] Furthermore, the Raman spectrometer uses a laser power of 50 mW, a grating of 300 l / mm to emit laser light to Escherichia coli, and 120 Raman spectral data are collected for each strain of Escherichia coli.
[0015] Furthermore, the specific method for preprocessing the Raman spectral data in step S3 is:
[0016] S3-1. Select the fingerprint region of the Raman spectral data from 400 to 1800 cm -1 ;
[0017] S3-2. Use the asymmetric least squares method to remove the background from the selected Raman spectral data;
[0018] S3-3. Smooth the high-frequency noise of the background-removed Raman spectral data using a Savitzky-Golay smoothing filter with a window size of 11 points and a third-degree polynomial fit;
[0019] S3-4. Perform maximum normalization on each smoothed spectral data.
[0020] Furthermore, the specific method for training the classification model in step S5 is as follows:
[0021] The initial learning rate is set to 0.001, and during the training process, it is set to 100 epochs. The batch size is adjusted to 20 and the Dropout rate is 0.1 through cross-validation.
[0022] The present invention innovatively combines the laser confocal Raman spectroscopy technology with an improved convolutional neural network (AlexNet) model. This method not only overcomes the limitations of the existing detection technology for the sensitivity of Escherichia coli to cefotaxime, such as insufficient accuracy, low time efficiency, and complex operation, but also provides a non-invasive, efficient, and reliable detection means. The present invention combines Raman spectroscopy technology with the AlexNet model, simplifies the detection process, improves the detection speed and accuracy, especially in the case of a small sample size, and meets the clinical requirements for rapid and reliable drug sensitivity testing.
[0023] The method for discriminating the sensitivity type of Escherichia coli to cefotaxime based on Raman spectroscopy technology and deep learning proposed by the present invention successfully solves the problems of long time and poor accuracy existing in traditional drug sensitivity tests. By innovatively combining the molecular information of Raman spectroscopy with the self-learning ability of deep learning, the present invention can efficiently and accurately identify the sensitivity of Escherichia coli to cefotaxime.
[0024] The present invention combines Raman spectroscopy with the AlexNet deep learning model to construct a spectral database containing 4 strains of Escherichia coli, and deeply explores the complex characteristics of bacterial drug resistance detection. This method does not require bacterial culture during classification, can complete the detection quickly, significantly shortens the diagnostic cycle, and meets the clinical point-of-care testing requirements. In addition, the technical process is fully automated, reducing the operation difficulty and technical threshold, while optimizing the cost, and has the advantages of fast, efficient, and low cost, and is especially suitable for promotion in primary medical institutions. Description of the Drawings
[0025] Figure 1 This is the flowchart of the present invention.
[0026] Figure 2 This is the Raman spectrum of Escherichia coli after pretreatment.
[0027] Figure 3 This is the changing trend of accuracy and loss function value during the training process of the classification model.
[0028] Figure 4 This is the confusion matrix of the classification results of the cefotaxime sensitivity type of Escherichia coli using the classification model.
[0029] Figure 5 This is the ROC curve of the classification results of the cefotaxime sensitivity type of Escherichia coli using the classification model. Detailed implementation mode
[0030] The present invention will be further described in detail below.
[0031] As Figure 1 shown, the present invention provides a method for discriminating the cefotaxime sensitivity type of Escherichia coli, including the following steps:
[0032] S1. Obtain Escherichia coli from different patients. For one strain of Escherichia coli from each patient, measure the cefotaxime sensitivity type using the gold standard BMD method, and for the other strain, measure the Raman spectrum data of Escherichia coli using a Raman spectrometer, until at least two patients' Escherichia coli are sensitive and at least two patients' Escherichia coli are resistant.
[0033] The source of Escherichia coli is patients. Obtain Escherichia coli from at least four patients. Divide the Escherichia coli of each patient into two strains. Measure the sensitivity of one strain of Escherichia coli using the gold standard BMD method, and measure the Raman spectrum data of the other strain of Escherichia coli using a Raman spectrometer. The Escherichia coli measured by the gold standard BMD method and the Raman spectrometer are both obtained from one bacterial culture.
[0034] Take one strain of Escherichia coli from each patient for culture, add cefotaxime during the culture, and measure the minimum inhibitory concentration (Minimum Inhibitory Concentration, MIC) of cefotaxime against these Escherichia coli. The MIC values measured in the present invention range from 0.03 μg / mL -1 to 256 μg / mL -1 , including Escherichia coli sensitive to cefotaxime to Escherichia coli resistant to cefotaxime. Escherichia coli with an MIC value less than or equal to 1 is a sensitive bacterium, Escherichia coli with an MIC value greater than or equal to 4 is a resistant bacterium, and Escherichia coli with 1 < MIC < 4 is a neutral bacterium.
[0035] Until the Escherichia coli with measured sensitivity comes from at least two sources, namely two patients, and the Escherichia coli with drug resistance comes from at least two patients.
[0036] Before measuring the Raman spectral data of Escherichia coli, it is pretreated by the method of washing with deionized water and centrifugal precipitation. The pretreated Escherichia coli is transferred to a centrifuge tube and centrifuged at a speed of 2000 rpm for 3 minutes to precipitate the Escherichia coli cells to the bottom of the centrifuge tube. Discard the supernatant; add an appropriate amount of deionized water to the centrifuge tube, gently pipette or vortex to resuspend the precipitated Escherichia coli cells, and then centrifuge again at a speed of 2000 rpm for 3 minutes and discard the supernatant. Repeat this washing process 2 - 3 times to remove the residual culture medium components and impurities on the cell surface.
[0037] Set the parameters of the Raman spectrometer. Select a laser wavelength of 707 nm, which effectively excites the Raman scattering signal in the bacterial sample; a laser power of 50 mW to ensure that the signal intensity and signal - to - noise ratio reach an ideal level; a grating of 300 l / mm to further optimize the resolution of the spectral data. Collect 120 Raman spectral data for each strain of Escherichia coli.
[0038] According to the type of sample stage of the Raman spectrometer, select a suitable method to fix the sample. A common method is to use a glass slide or a quartz slide, drop the bacterial solution on its surface, and then let it dry naturally or dry it at a low temperature to fix the Escherichia coli on the glass slide. It is also possible to use a special sample cell or fixing device to fix the liquid sample containing Escherichia coli.
[0039] Place the glass slide or sample cell with the fixed Escherichia coli sample on the sample stage of the Raman spectrometer, and adjust the position of the sample so that it is perpendicular to the laser beam and focused on the sample surface to obtain the best measurement effect. After setting the relevant parameters in the instrument control software, click the "Start" button to start the acquisition process of Raman spectral data. During the acquisition process, closely observe the status of the spectrometer and the reaction of the sample to ensure the smooth progress of the test. After the acquisition is completed, the software will display the Raman spectrogram of Escherichia coli and generate the corresponding data file. Record the position, intensity and other data of the characteristic peaks in the spectrogram, as well as the instrument parameters and sample information used during the acquisition process for subsequent data processing and analysis.
[0040] S2. Select the Raman spectral data of Escherichia coli with measured drug resistance and sensitivity, set the corresponding drug sensitivity labels for the Raman spectral data, and select the same number of strains of drug - resistant Escherichia coli as the number of strains of sensitive Escherichia coli.
[0041] The present invention selects the Raman spectral data of drug - resistant and sensitive Escherichia coli as the samples for model training.
[0042] The selected sensitive and drug-resistant Escherichia coli in the present invention are from the same number of patients, that is, the number of strains of the two types of Escherichia coli is the same.
[0043] According to the drug sensitivity and Raman spectrum data of Escherichia coli tested in step S1, drug sensitivity labels are set for the Raman spectrum data of Escherichia coli. For one strain of Escherichia coli from patient A, when the MIC of cefotaxime measured by the gold standard BMD method is drug-resistant, the label of the Raman spectrum data of another strain of Escherichia coli from the same source as this Escherichia coli is also drug-resistant.
[0044] S3. Perform wavenumber selection, background removal, smoothing and noise reduction, and normalization on the Raman spectrum data.
[0045] Select the fingerprint region of the Raman spectrum data at 400 - 1800 cm -1 This region contains signals closely related to bacterial molecular characteristics (such as proteins, lipids, nucleic acids, etc.); use the asymmetric least squares (ALS) method to remove the background of the spectrum, accurately fit the low-frequency background noise, remove the baseline drift caused by instrument and environmental factors, and enhance the signal contrast. Use a Savitzky-Golay smoothing filter with a window size of 11 points and a third-order polynomial fit to smooth the high-frequency noise, effectively retaining the peak characteristics of the signal. Perform maximum normalization on each spectrum data to eliminate the influence of experimental environment changes and equipment fluctuations, and ensure the comparability of data under different experimental conditions.
[0046] Such as Figure 2 shown, after wavenumber selection, background removal, smoothing and noise reduction, and normalization, the noise in the Raman spectrum data is effectively removed, the quality of the Raman spectrum data is improved, and reliable input is provided for the subsequent deep learning model. Figure 2 Represents Escherichia coli after preprocessing.
[0047] S4. Remove the third convolutional layer to the fifth convolutional layer in the AlexNet model, and replace the ReLU activation function in the AlexNet model with the Leaky ReLU activation function to obtain a classification model.
[0048] The present invention makes some improvements on the basis of the classical AlexNet model to adapt to the characteristics of Raman spectrum data and achieve the classification task of the sensitive type of Escherichia coli to cefotaxime. The classical AlexNet model includes five convolutional units and three fully connected layers.
[0049] The present invention removes the last three convolutional units and the last fully connected layer of the AlexNet model, and replaces the ReLU activation function of the AlexNet model with the Leaky ReLU activation function. Further increases the non-linear representation ability of the network.
[0050] The classification model includes a first convolutional unit, a second convolutional unit, a Flatten layer, a first fully connected layer, a second fully connected layer, and an output layer that are connected in sequence.
[0051] The input data of the first convolutional unit is output after passing through eight 8×1 convolutional kernels, a batch normalization layer, a LeakyReLU activation function, and a max pooling layer with a scale of 2×1 in sequence, and then enters the second convolutional unit. The input data of the second convolutional unit is output after passing through sixteen 16×1 convolutional kernels, a batch normalization layer, a LeakyReLU activation function, and a max pooling layer with a scale of 2×1 in sequence..
[0052] The present invention uses eight 8×1 convolutional kernels to extract low-order features in Raman spectroscopy data, such as spectral peak positions and intensities; sixteen 16×1 convolutional kernels further capture higher-order feature information of Raman spectroscopy data. A max pooling layer is added after the convolutional layer, which effectively reduces the data dimension through pooling, while retaining key information, reducing the computational complexity, and improving the classification performance. A batch normalization layer is added after each pooling layer, which avoids gradient disappearance by normalizing the output of each layer, improves the training stability and convergence speed of the model, and accelerates the network training process.
[0053] The first layer of the fully connected layer contains 64 neurons, and the second layer contains 32 neurons. After passing through the first fully connected layer and the second fully connected layer, they are both activated by the Leaky ReLU activation function.
[0054] The present invention adds a Dropout layer between the fully connected layers, randomly discarding the connections of some neurons, thereby improving the generalization ability of the model and reducing the over-dependence on the training data. The output layer uses the Softmax activation function for binary classification to determine whether the bacteria are resistant to cefotaxime. The Softmax function converts the various predicted values output by the network into probability values, thereby ensuring the accuracy of classification.
[0055] S5. Divide the processed Raman spectroscopy data with drug sensitivity labels into a training set and a validation set, and input them into the classification model for training to obtain a classification model for discriminating the sensitive type of Escherichia coli to cefotaxime.
[0056] The present invention uses the Adam optimizer to train the classification model, which has a good convergence speed and low hyperparameter sensitivity. The Raman spectroscopy data is divided into a training set and a validation set, and the ratio of the training set to the validation set is 7:3 to ensure that the training and evaluation data of the model are sufficiently diverse. Hyperparameter settings: The initial learning rate is 0.001, and a learning rate decay strategy is used to dynamically adjust the learning rate to optimize the training effect. During the training process, it is set to 100 epochs, and the batch size (20) and Dropout rate (0.1) are adjusted through cross-validation. These optimization measures ensure the stability and efficiency of the model and significantly improve the classification performance.
[0057] The output results of the classification model are the probabilities corresponding to sensitivity and drug resistance respectively. Compare their probabilities, and take the drug sensitivity type with the larger probability as the type of the measured Escherichia coli.
[0058] Due to its characteristics of being label-free, fast detection speed, and high sensitivity, Raman spectroscopy technology has become a research hotspot in bacterial sensitivity detection. Existing studies have attempted to combine Raman spectroscopy with machine learning algorithms, such as principal component analysis (PCA) and support vector machine (SVM). However, traditional algorithms have limitations in processing complex spectral features and small sample data, and the classification accuracy and robustness still need to be improved.
[0059] The present invention only requires the Raman spectroscopy data of four strains of Escherichia coli to train the classification model, which is suitable for use with small sample data.
[0060] S6. Pretreat the measured Escherichia coli and collect the Raman spectroscopy data of the pretreated measured Escherichia coli; perform pretreatment on the Raman spectroscopy data, including wavenumber selection, background removal, smoothing and noise reduction, and normalization; input the Raman spectroscopy data of the pretreated measured Escherichia coli into the classification model, and determine the sensitive type of the measured Escherichia coli according to the output results.
[0061] After obtaining the classification model for discriminating the sensitive type of Escherichia coli to cefotaxime, when classifying and identifying the measured Escherichia coli with unknown drug sensitivity, steps S2 - S3 also need to be passed through. Wash the measured Escherichia coli with deionized water and perform centrifugal precipitation treatment, collect the Raman spectroscopy data of the treated measured Escherichia coli using a Raman spectrometer, perform wavenumber selection, background removal, smoothing and noise reduction, and normalization on the Raman spectroscopy data of the measured Escherichia coli, and input the processed Raman spectroscopy data of the measured Escherichia coli into the classification model to obtain the sensitivity of the measured Escherichia coli to cefotaxime.
[0062] In the present invention, Escherichia coli only needs to be cultured during the training of the classification model. In subsequent detections, there is no need to culture Escherichia coli again, and the trained classification model can be directly used, saving time.
[0063] S7. Method evaluation.
[0064] As Figure 3 shown, by training on the Raman spectral data of Escherichia coli, as the number of training rounds increases, the accuracy of the training set and the validation set gradually increases, and the value of the loss function gradually decreases and stabilizes. The model training process converges well and there is no overfitting. The loss function values of both the training set and the validation set gradually decrease, and the accuracy gradually increases, indicating that the model effectively learns the data features during the training process. The loss function value and accuracy of the validation set are similar to those of the training set, and there is no obvious overfitting phenomenon (i.e., the validation set loss suddenly increases or the accuracy decreases), indicating that the model has good robustness.
[0065] Table 1: Evaluation of the classification results of drug-resistant bacteria using Alexnet
[0066] Sensitivity (%) Specificity (%) Accuracy (%) Drug-resistant bacteria classification 97.3 93.3 95.3
[0067] As shown in Table 1 and Figure 4 shown, according to the confusion matrix, the accuracy of the method of the present invention in discriminating the sensitivity of Escherichia coli to cefotaxime reaches 95.3%, fully verifying that the classification model of the present invention can effectively distinguish the sensitive type and resistant type of Escherichia coli to cefotaxime. The sensitivity also reaches 97.32% and the specificity is 93.3%. Among them, sensitivity = true positive / (true positive + false negative) × 100%, specificity = true negative / (true negative + false positive) × 100%, where drug resistance is the positive class and sensitivity is the negative class. The low error rates (such as 2.68% and 6.65%) in the confusion matrix indicate that the model of the present invention has fewer classification errors for different classes, indicating that the model has good robustness when facing different classes.
[0068] Among them, true positive (TP): the number of samples correctly classified as the positive class; true negative (TN): the number of samples correctly classified as the negative class; false positive (FP): the number of samples misclassified as the positive class (actually the negative class); false negative (FN): the number of samples misclassified as the negative class (actually the positive class); sensitivity = true positive / (true positive + false negative) × 100%, specificity = true negative / (true negative + false positive) × 100%, where drug resistance is the sample of the positive class and sensitivity is the sample of the negative class.
[0069] As Figure 5As shown, the AUC value of the ROC curve is 0.993. The higher the AUC value, the better the performance of the model, indicating that the classification model performs excellently in the classification of Escherichia coli sensitive to cefotaxime. The high AUC value indicates that the model can maintain a high true positive rate and a low false positive rate at different thresholds, indicating that the model has good robustness under different conditions.
Claims
1. A method for distinguishing the type of Escherichia coli sensitive to cefotaxime, characterized in that: The following steps are involved: S1. Obtain Escherichia coli from different patients, measure the cefotaxime sensitive type of one strain of Escherichia coli from each patient by using the gold standard broth dilution method, and measure the Raman spectrum data of the other strain by using a Raman spectrometer; until the Escherichia coli of at least two patients are sensitive to cefotaxime, and the Escherichia coli of at least two patients are resistant to cefotaxime; S2. Select Raman spectral data of Escherichia coli with drug resistance and sensitivity as the measurement results, and set corresponding drug sensitivity labels for the Raman spectral data; the number of selected drug-resistant Escherichia coli strains is the same as the number of selected sensitive Escherichia coli strains; S3. Preprocessing of Raman spectroscopy data using wave number selection, background removal, smoothing and noise reduction, and normalization; S4. Remove the third to fifth convolutional layers in the AlexNet model, replace the ReLU activation function in the AlexNet model with the Leaky ReLU activation function, and obtain a classification model; S5. dividing the processed Raman spectral data with drug sensitivity labels into a training set and a validation set, and inputting the data into a classification model for training to obtain a classification model for distinguishing the sensitivity type of Escherichia coli to cefotaxime; S6. Pre-process the Escherichia coli to be tested, and collect Raman spectrum data of the pre-processed Escherichia coli to be tested; pre-process the Raman spectrum data by wavenumber selection, background removal, smoothing noise reduction and normalization; input the pre-processed Raman spectrum data of the Escherichia coli to be tested into the classification model, and determine the sensitive type of the Escherichia coli to be tested according to the output result.
2. The method for distinguishing the type of Escherichia coli sensitive to cefotaxime according to claim 1, characterized in that: The Raman spectrometer uses a laser power of 50 mW and a grating of 300 l / mm to emit laser light to E. coli, and collects 120 Raman spectral data for each strain of E. coli.
3. The method for distinguishing the type of Escherichia coli sensitive to cefotaxime according to claim 1, characterized in that: The specific method of preprocessing the Raman spectrum data in step S3 is: S3-1. Select Raman spectrum data 400-1800cm -1 Fingerprint area; S3-2. Background removal of selected Raman spectral data using asymmetric least squares method; S3-3. The high-frequency noise of the Raman spectral data after background removal was smoothed using a Savitzky-Golay smoothing filter with a window size of 11 points and a third-order polynomial fitting; S3-4. Perform maximum normalization on each smoothed spectral data.
4. The method for distinguishing the type of Escherichia coli sensitive to cefotaxime according to claim 1, characterized in that: The specific method of training the classification model in step S5 is: The initial learning rate is set to 0.001, the training process is set to 100 epochs, the batch size is adjusted to 20 through cross-validation, and the dropout rate is 0.1.