A method for detecting antibacterial drug sensitivity based on physical morphological characteristics of single bacterial cells and related devices

Abstract

This application proposes a method and related apparatus for detecting antimicrobial susceptibility based on the physical morphological characteristics of single bacteria. The method includes: S1, establishing an independent detection unit; S2, drug processing; S3, single-bacterial imaging; S4, single-bacterial segmentation; S5, multidimensional feature extraction; S6, single-index performance evaluation; S7, joint modeling and performance evaluation; and S8, result output. This application can establish a rapid drug susceptibility discrimination process for common Gram-negative bacteria in the early single-bacterial morphological changes after antibiotic treatment, thereby improving the interpretability, standardization potential, and clinical translational applicability of the method while shortening the detection time.

Description

Technical Field

[0001] This application relates to the field of infectious disease diagnosis and phenotypic drug susceptibility testing technology, and in particular to a method and related device for detecting antimicrobial drug susceptibility based on the physical morphological characteristics of single bacteria. Background Technology

[0002] Bacterial infections are a major cause of the global mortality burden, and antibiotic resistance (AMR) further complicates the diagnosis and treatment of infection-related diseases. Clinically, antimicrobial susceptibility testing (AST) is a crucial step in guiding precise drug use and curbing the spread of resistance. Therefore, establishing rapid, accurate, and stable methods for drug susceptibility testing is of great significance.

[0003] Currently, the widely used AST methods in clinical practice mainly include agar diffusion, broth dilution, and their derivatives. These methods typically use the growth inhibition results of bacterial populations under antibiotic treatment as the basis for interpretation. Although relatively mature standard systems have been established, they still rely on a certain period of in vitro culture, usually 8–24 hours or even longer, making it difficult to meet the needs of precise medication in the early stages of infection. In addition, population-level detection results essentially reflect the average response of the bacterial community, making it difficult to reveal the heterogeneous changes of individual bacteria in the early stages of drug action, which may mask some individual differences and affect the sensitivity and stability of the results.

[0004] In recent years, with the development of microscopic imaging, image analysis, and artificial intelligence technologies, rapid drug susceptibility testing based on single-bacterial levels has gradually become a research hotspot. One existing single-bacterial drug susceptibility testing scheme mainly involves randomly selecting multiple bacteria at different antibiotic concentrations, acquiring continuous multi-frame image sequences of the same bacterium, reconstructing the images, and cropping the region of interest to form single-bacterial sequence data. This single-bacterial sequence data is then directly input into a pre-trained deep learning evaluation model to obtain drug resistance or susceptibility results. The core of this scheme lies in relying on continuous sequence image information and an end-to-end deep learning model to discriminate the survival status or subcellular dynamic changes of a single bacterium. For example, the technical solution provided by Chinese invention patent CN121249842A clearly defines the technical route as "multiple different antibiotic concentrations," "continuous multi-frame original imaging to form an original imaging sequence," and "inputting the region of interest sequence into a pre-trained evaluation model."

[0005] However, the above-mentioned schemes based on continuous sequence image acquisition and direct use of end-to-end deep learning models to determine the survival or susceptibility status of single bacteria still have the following shortcomings: First, the detection process relies on continuous time-series image acquisition, image reconstruction, and sequence-level modeling analysis, making the data acquisition and processing relatively complex; Second, this type of scheme mainly uses sequence images as input and directly outputs classification results through deep learning models, resulting in relatively limited interpretability of feature expressions, which is not conducive to analyzing drug action mechanisms and discrimination criteria from a morphological perspective; Third, this type of scheme usually constructs datasets based on multiple antibiotic concentration conditions, focusing more on sequence discrimination of single bacterial survival / non-survival status or susceptibility / resistance status, which still differs from the rapid drug sensitivity interpretation requirements in clinical AST based on specific bacterial species, specific antibiotics, and standard interpretation systems. Meanwhile, existing schemes based on continuous sequence image acquisition and direct end-to-end deep learning models to determine the survival or susceptibility status of single bacteria emphasize the random selection of multiple bacteria at different antibiotic concentrations, with each bacteria corresponding to multiple frames of sequence images, and adding them to the dataset for model training. The core of this approach is to directly output drug sensitivity results by combining continuous video data with an end-to-end deep learning model. However, such technologies usually lack independent interpretation logic for different bacterial species and different antibiotic combinations, and cannot clearly define the earliest time window for different bacterial-drug combinations to achieve stable drug sensitivity discrimination capabilities. Therefore, it is difficult to establish a standardized interpretation system suitable for rapid AST scenarios in clinical settings.

[0006] Therefore, it is necessary to develop a rapid drug susceptibility testing method that differs from deep learning schemes for continuous sequence imaging: under clinically relevant drug action conditions, based on interpretable morphological quantitative features of static images of single bacterial endpoints, a rapid drug susceptibility discrimination model for specific bacterial species-antibiotic combinations can be constructed, thereby shortening the detection time while improving the interpretability, standardization potential, and clinical applicability of the method. Summary of the Invention

[0007] This application provides a method and related apparatus for detecting antimicrobial susceptibility based on the physical morphological characteristics of single bacteria. Its core is not merely to achieve single-sample susceptibility classification, but rather to establish a rapid susceptibility interpretation framework based on independent detection units of "bacterial species-antibiotic-treatment duration." By analyzing the morphological response patterns of single bacteria after antibiotic treatment, it determines the earliest time window for different bacterial-antibiotic combinations to achieve stable discriminative ability, thereby addressing the problems existing in related technologies. The technical solution is as follows: In a first aspect, embodiments of this application provide a method for detecting antimicrobial susceptibility based on the physical morphological characteristics of a single bacterium, comprising: S1. Establish independent detection units: Select target Gram-negative bacteria, and establish different independent detection units for the target Gram-negative bacteria, with antibiotic type and treatment time as adjustable variables; S2. Drug treatment: For a single independent detection unit, the bacterial suspension containing the target Gram-negative bacteria is reacted with an antibiotic solution of a set concentration. After the set treatment time is reached, a bacterial sample corresponding to the single independent detection unit is obtained. S3. Single cell imaging: Fix, stain, and acquire microscopic images of the bacterial sample corresponding to a single independent detection unit to obtain several single cell images of a single independent detection unit; S4. Single bacterial cell segmentation: Using a specified image segmentation algorithm, identify and segment individual bacterial cells from each single bacterial cell image to obtain the segmentation result of each single bacterial cell image; S5. Multidimensional feature extraction: Based on the segmentation results of each single bacterial cell image, extract multiple physical morphological indicators of a single bacterial cell; S6. Single index performance evaluation: Within a single independent detection unit, the performance of multiple physical morphological indicators of the single bacterial cell is evaluated to assess the drug sensitivity discrimination ability of a single physical morphological indicator. S7. Joint modeling and performance evaluation: Multiple physical morphological indicators of different independent detection units are imported into the machine learning model as joint input variables to complete model training, cross-validation, performance evaluation and drug sensitivity discrimination, and obtain a rapid drug sensitivity prediction model for each independent detection unit. S8. Output Results: Output the evaluation results of the target Gram-negative bacteria under the conditions of antibiotic type and treatment duration, and the prediction results of the corresponding rapid drug sensitivity prediction model.

[0008] In one embodiment, selecting the target Gram-negative bacterium includes: selecting any one of Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa as the target Gram-negative bacterium; The antibiotics are selected from aztreonam, ceftazidime, meropenem, and ciprofloxacin; The processing time is set to 15 minutes, 30 minutes, 60 minutes or 90 minutes.

[0009] In one implementation, step S2 specifically includes: S21. For a single independent detection unit, fresh colonies of the target Gram-negative bacteria clinically isolated from a blood agar plate are picked, a bacterial suspension is prepared, and the concentration is adjusted to a predetermined treatment concentration. S22. Dilute the antibiotic solution of the set concentration and add it to Miller-Hinton broth medium and mix well. Then inoculate the prepared bacterial suspension into Miller-Hinton broth medium. S23. Place the Miller-Hinton broth culture medium in a constant temperature shaker and incubate for the set treatment time to obtain a bacterial sample corresponding to a single independent detection unit.

[0010] In one implementation, step S3 specifically includes: S31. Fix and stain the cell nuclei of the bacterial sample corresponding to a single independent detection unit to obtain the processed bacterial sample. S32. Call the microscopic imaging system to acquire a full-field image of the processed bacterial sample, and obtain several single bacterial images of a single independent detection unit.

[0011] In one implementation, step S4 specifically includes: An instance segmentation algorithm is used to identify the contour of each bacterial cell in each bacterial cell image and generate a mask. Then, the mask of each bacterial cell is separated from the background and adjacent targets to obtain the segmentation result of each bacterial cell image.

[0012] In one implementation, step S6 specifically includes: S61. Statistical methods were used to compare the distribution differences of multiple physical morphological indicators between drug-sensitive and drug-insensitive bacteria and to perform statistical analysis to verify whether there are significant differences between different drug-sensitive phenotypes for a single physical morphological indicator. S62. Within a single independent detection unit, a subject operating characteristic curve is constructed for a single physical morphological index, and the drug sensitivity discrimination ability of the single physical morphological index is evaluated by the area under the curve, and the single physical morphological index is associated with a drug sensitivity label.

[0013] In one implementation, step S7 specifically includes: S71. Construct a machine learning model, wherein the input of the machine learning model is multiple physical morphological indicators, and the output is a binary label of drug sensitivity or drug insensitivity; S72. The multiple physical morphological indicators of different independent detection units are combined into a dataset, and then the dataset is divided into a training set and a test set. S73. The machine learning model is trained and its performance is evaluated using the cross-validation method based on the training set and the test set to obtain the trained machine learning model and its performance evaluation results. S74. Based on the performance evaluation results, determine the rapid drug sensitivity detection time for different combinations of bacterial species and antibiotics, and construct a rapid drug sensitivity prediction model for each of the independent detection units.

[0014] Secondly, embodiments of this application also provide an antimicrobial susceptibility detection device based on the physical morphological characteristics of a single bacterium, comprising: Establishment unit, used to establish independent detection unit: Select target Gram-negative bacteria, and establish different independent detection units for the target Gram-negative bacteria, with antibiotic type and treatment time as adjustable variables; The processing unit is used for drug processing: for a single independent detection unit, the bacterial suspension containing the target Gram-negative bacteria is reacted with an antibiotic solution of a set concentration, and after a set processing time is reached, a bacterial sample corresponding to the single independent detection unit is obtained. Imaging unit for single-cell imaging: fixing, staining and acquiring microscopic images of bacterial samples corresponding to a single independent detection unit to obtain several single-cell images of a single independent detection unit; A segmentation unit for single-cell segmentation: using a specified image segmentation algorithm to identify and segment individual cells from each single-cell image, thereby obtaining a segmentation result for each single-cell image; The extraction unit is used for multi-dimensional feature extraction: based on the segmentation results of each single bacterial cell image, multiple physical morphological indicators of a single bacterial cell are extracted; An evaluation unit is used for single-index performance evaluation: within a single independent detection unit, multiple physical morphological indices of the single bacterial cell are evaluated to assess the drug sensitivity discrimination capability of the single physical morphological index. Modeling unit, used for joint modeling and performance evaluation: multiple physical morphological indicators of different independent detection units are imported into the machine learning model as joint input variables to complete model training, cross-validation, performance evaluation and drug sensitivity discrimination, and obtain a rapid drug sensitivity prediction model for each independent detection unit. The output unit is used for result output: outputting the evaluation results of a single physical morphological index of the target Gram-negative bacteria under the antibiotic type-treatment time condition and the corresponding prediction results of the rapid drug sensitivity prediction model.

[0015] Thirdly, embodiments of this application also provide an electronic device, which includes: a memory and a processor, wherein the memory stores instructions, and the instructions are loaded and executed by the processor to implement the methods in any of the above embodiments, wherein the memory and the processor communicate with each other through an internal connection path.

[0016] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program that, when run on a computer, implements the methods in any of the above-described embodiments.

[0017] The advantages or beneficial effects of the above technical solutions include at least the following: This application constructs independent detection units based on different bacterial species, antibiotics, and drug treatment durations. Single-cell images are acquired within each independent detection unit, and multiple physical morphological indicators with drug sensitivity discrimination significance are extracted. Based on these, single-indicator discrimination ability evaluation and multi-indicator joint modeling analysis are conducted, ultimately forming a rapid drug sensitivity prediction model for different combinations of bacterial species, antibiotics, and treatment durations. This model can establish a rapid drug sensitivity discrimination process consisting of "drug treatment—single-cell imaging—single-cell segmentation—multi-dimensional feature extraction—single-indicator performance evaluation—joint modeling and performance evaluation—result output" for the early single-cell morphological changes of common Gram-negative bacteria after antibiotic treatment. This process can shorten detection time while improving the interpretability, standardization potential, and clinical translation applicability of the method.

[0018] Furthermore, unlike existing schemes that rely on continuous sequence image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria, this application does not rely on continuous dynamic video information for single-shot classification. Instead, it uses independent detection units of "bacterial species-antibiotic-treatment duration" under clinically sensitive breakpoint concentration conditions. Through static microscopic imaging at preset time points, single-bacterial segmentation, and explicit morphological feature extraction, combined with single-index statistical evaluation and multi-dimensional feature joint modeling, it achieves rapid determination of bacterial susceptibility / non-susceptibility status and further determines the earliest time window for different bacterial species and antibiotic combinations to achieve stable discrimination capability. Therefore, this application not only achieves rapid drug sensitivity classification but also establishes the time response patterns of different bacterial-antibiotic combinations. It differs substantially from existing schemes that rely on continuous sequence image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria in terms of technical approach, interpretation logic, feature construction method, and clinical application goals.

[0019] The above overview is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments, and features described above, further aspects, embodiments, and features of this application will become readily apparent from the accompanying drawings and the following detailed description. Attached Figure Description

[0020] In the accompanying drawings, unless otherwise specified, the same reference numerals throughout the various drawings denote the same or similar parts or elements. These drawings are not necessarily drawn to scale. It should be understood that these drawings depict only some embodiments disclosed in this application and should not be construed as limiting the scope of this application.

[0021] Figure 1 A flowchart illustrating an antimicrobial susceptibility detection method based on the physical morphological characteristics of a single bacterium, provided for an embodiment of this application; Figure 2This application provides a diagram showing the differential distribution of six morphological indicators (long axis length, short axis length, solidity, roundness, skeleton length, and nucleus / cell ratio) between susceptible and non-susceptible Escherichia coli strains after treatment with meropenem for 15–90 minutes. Figure 3 A comparative diagram of the morphological characteristics of sensitive and non-sensitive Klebsiella pneumoniae cells after different treatment times with ceftazidime, provided as an embodiment of this application; Figure 4 A comparative diagram of the differences in morphological indicators of sensitive and non-sensitive Pseudomonas aeruginosa cells after different treatment times with aztreonam, provided for an embodiment of this application; Figure 5 The area under the curve (AUC) of six indicators under different treatment durations of Escherichia coli-meropenem combination provided in the embodiments of this application reflects the ability of each indicator to individually determine sensitivity. Figure 6 This application provides a single-index area under curve (AUC) performance of the Klebsiella pneumoniae-ceftazidime combination. Figure 7 The area under the curve (AUC) of a single index under a combination of Pseudomonas aeruginosa and aztreonam is provided in the embodiments of this application. Figure 8 The performance comparison chart provided in this application embodiment shows the area under the curve (AUC) and accuracy of different models under each combination, with six indicators jointly input into six machine learning models. Figure 9 The performance comparison chart provided in this application embodiment shows the area under the curve (AUC) and accuracy of different models under each combination. Figure 10 The performance comparison chart provided in this application embodiment shows the area under the curve (AUC) and accuracy of different models under each combination of Pseudomonas aeruginosa-antibiotic-time. in, Figures 2-10 In Chinese, "ns" indicates no statistically significant difference, "*" indicates P < 0.05, "**" indicates P < 0.01, "***" indicates P < 0.001, and "****" indicates P < 0.0001. Figure 11 A structural block diagram of an antimicrobial susceptibility detection device based on the physical morphological characteristics of a single bacterium, provided in an embodiment of this application; Figure 12This is a structural block diagram of an electronic device provided in an embodiment of this application. Detailed Implementation

[0022] In the following description, only certain exemplary embodiments are briefly described. As those skilled in the art will recognize, the described embodiments can be modified in various ways without departing from the spirit or scope of this application. Therefore, the drawings and description are considered to be exemplary in nature and not restrictive.

[0023] Figure 1 A flowchart illustrating an antimicrobial susceptibility testing method based on the physical morphological characteristics of a single bacterium, according to an embodiment of this application, is shown. Figure 1 As shown, the method may include the following steps: S1. Establish independent detection units: Select target Gram-negative bacteria, and establish different independent detection units for the target Gram-negative bacteria, with antibiotic type and treatment duration as adjustable variables.

[0024] In step S1, different independent detection units are established for the target Gram-negative bacteria according to the combination of "bacterial species-antibiotics-treatment time".

[0025] In a real-time mode, any one of the Gram-negative bacteria, such as Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa, can be selected as the target Gram-negative bacterium.

[0026] In one embodiment, the antibiotic is selected from aztreonam, ceftazidime, meropenem, and ciprofloxacin. The treatment duration is set to 15 minutes, 30 minutes, 60 minutes, and 90 minutes.

[0027] For example, in conjunction with the above embodiments, when the target Gram-negative bacterium is Escherichia coli, its corresponding independent detection unit may include the 16 independent detection units shown in Table 1.

[0028] Table 1 In this embodiment of the application, by performing step S1, each combination of bacterial species, antibiotic type, and treatment duration can be used as an independent analysis object for subsequent image acquisition, feature extraction, statistical evaluation, and model construction.

[0029] S2. Drug treatment: For a single independent detection unit, the bacterial suspension containing the target Gram-negative bacteria is reacted with an antibiotic solution of a set concentration. After the set treatment time is reached, the bacterial sample corresponding to the single independent detection unit is obtained.

[0030] In one implementation, step S2 specifically includes: S21. For a single independent detection unit, pick fresh colonies of clinically isolated target Gram-negative bacteria from a blood plate, prepare a bacterial suspension, and adjust it to the predetermined treatment concentration.

[0031] In practice, the predetermined treatment concentration is preferably 0.01 OD. 600 .

[0032] In practice, the blood platelet can be provided by the relevant hospital.

[0033] For example, in step S21, a freshly isolated target bacterial strain (i.e., a strain of the target Gram-negative bacteria) can be picked from a blood agar plate provided by the hospital. A single colony is then picked and placed in 1 mL of sterile physiological saline, shaken and mixed. The OD value is then measured using a spectrophotometer (λ = 600 nm) and adjusted to 1.0 (subsequently, 20 μL of the bacterial suspension can be diluted to 0.01 OD by adding it to 1980 mL of broth). 600 (For use in antibiotic treatment), a prepared bacterial suspension can be obtained.

[0034] That is, step S21 is the strain isolation and pre-culture step.

[0035] S22. Dilute the antibiotic solution of the set concentration and add it to the Mueller-Hinton Broth (MHB) medium and mix well. Then inoculate the prepared bacterial suspension into the Mueller-Hinton Broth medium.

[0036] In practice, the optimal concentration of the antibiotic solution is the clinically sensitive breakpoint concentration.

[0037] For example, a 12-well plate can be prepared, with 1980 µL of Miller-Hinton broth added to each well. The antibiotic can then be selected and its concentration adjusted according to the target Gram-negative bacterial species classification, as follows: If the target Gram-negative bacteria are classified as Enterobacteriaceae (i.e., the target Gram-negative bacteria are Escherichia coli or Klebsiella pneumoniae), then the selected antibiotics and their solution concentrations can be: ceftazidime 4 µg / mL, aztreonam 4 µg / mL, meropenem 1 µg / mL, and ciprofloxacin 0.25 µg / mL. If the target Gram-negative bacteria is classified as a non-fermenting bacterium (i.e., the target Gram-negative bacteria is Pseudomonas aeruginosa), then the selected antibiotics and their solution concentrations can be: ceftazidime at 8 µg / mL, aztreonam at 8 µg / mL, meropenem at 2 µg / mL, and ciprofloxacin at 0.5 µg / mL.

[0038] Next, the antibiotic solution of the set concentration was diluted and added to Miller-Hinton broth medium and mixed. Then, the prepared bacterial suspension was diluted and 20 µL of the diluted bacterial suspension was added to each well, so that the final liquid volume of the 12-well plate was 2000 µL.

[0039] S23. Place the Miller-Hinton broth medium in a constant temperature shaker and incubate for the set treatment time to obtain the bacterial cell sample corresponding to a single independent detection unit.

[0040] For example, a 12-well plate can be placed in a thermostatic shaker, with the temperature set to 37°C and the shaking rate to 160 rpm. The bacterial sample corresponding to the single independent detection unit can then be obtained. For instance, if the set time for the single independent detection unit is 15 minutes, then incubate for 15 minutes; if the set time is 30 minutes, then incubate for 30 minutes; if the set time is 60 minutes, then incubate for 60 minutes; and if the set time is 90 minutes, then incubate for 90 minutes.

[0041] Steps S22 and S23 are the steps for establishing the drug sensitivity incubation system.

[0042] In the embodiments of this application, bacterial samples under different drug treatment conditions can be obtained by performing step S2.

[0043] S3. Single cell imaging: Fix, stain, and acquire microscopic images of the bacterial sample corresponding to a single independent detection unit to obtain several single cell images of the single independent detection unit.

[0044] In one implementation, step S3 specifically includes: S31. Fix and stain the cell nuclei of the bacterial sample corresponding to a single independent detection unit to obtain the processed bacterial sample.

[0045] For example, for a single independent detection unit, after incubation, the culture medium in each well of a 12-well plate is aspirated into a 2 mL centrifuge tube, centrifuged at 8000 rpm for 5 minutes, and the supernatant is discarded; then, 100 µL of 4% paraformaldehyde fixative is added to resuspend the bacterial cells, and the cells are fixed at room temperature for 5 minutes; the cells are centrifuged again and the supernatant is discarded, and 20 µL of 1×Hoechst 33342 staining solution is added, and the cells are stained in the dark for 5 minutes to obtain the processed bacterial sample of the single independent detection unit.

[0046] That is, step S31 is the cell fixation and staining step.

[0047] S32. Call the microscopic imaging system to acquire a full-field image of the processed bacterial sample, and obtain several single bacterial images of a single independent detection unit.

[0048] In practice, the microscopic imaging system preferably employs an inverted laser confocal microscope. Furthermore, the system preferably uses simultaneous acquisition of images from a 405nm excitation fluorescence channel and a brightfield channel to obtain high-resolution single-cell images under 100× oil immersion conditions for subsequent image analysis.

[0049] For example, 1.75 µL of the treated bacterial sample can be dropped onto the center of a pre-cleaned glass slide, covered with a 24 mm × 24 mm coverslip, and then an inverted confocal laser scanning microscope (Zeiss LSM880) can be used for image acquisition. The microscope is set to an excitation wavelength of 405 nm and a magnification of 100 × oil immersion. This allows for the acquisition of a high-resolution single-cell image (1024 × 1024 pixels) of a single independent detection unit. During acquisition, more than three fields of view can be recorded for the treated bacterial sample of a single independent detection unit, and the images can be saved in .tiff format to obtain several single-cell images of a single independent detection unit.

[0050] That is, step S32 is the slide preparation and microscopic imaging step.

[0051] As can be seen, single-cell imaging in step S3 involves taking a static image of the fixed and stained sample after the antibiotic treatment has reached the set treatment time.

[0052] In this embodiment of the application, by executing step S3, the processed bacterial samples of each independent detection unit can be subjected to endpoint static microscopic imaging at each preset drug treatment time. The resulting single-frame image is used for subsequent single-bacterial segmentation and morphological feature extraction, rather than performing continuous multi-frame sequence image acquisition, sequence reconstruction and region of interest sequence analysis on the same bacterium.

[0053] S4. Single Bacterial Cell Segmentation: Using the specified image segmentation algorithm, identify and segment individual bacteria from each single bacterial cell image to obtain the segmentation result for each single bacterial cell image.

[0054] In one implementation, the specific implementation process of step S4 may include: using an instance segmentation algorithm to complete the single bacterial cell contour recognition and mask generation in each single bacterial cell image, and then separating the mask of the single bacterial cell from the background and adjacent targets to obtain the segmentation result of each single bacterial cell image.

[0055] In specific implementation, the preferred instance segmentation algorithm is the Cellpose segmentation algorithm. For example, each single bacterial cell image can be imported into Cellpose software, which will automatically execute step S4 to finally export the mask of the single bacterial cell. That is, the segmentation result of each single bacterial cell image includes the mask of its single bacterial cell.

[0056] In this embodiment of the application, by performing step S4, the accuracy and consistency of subsequent quantitative feature extraction can be guaranteed.

[0057] S5. Multidimensional feature extraction: Based on the segmentation results of each single bacterial cell image, extract multiple physical morphological indicators of a single bacterial cell.

[0058] In one embodiment, the multiple physical morphological indicators may include, but are not limited to: major length, minor length, solidity, roundness, skeleton length, and nucleus / cell ratio. The major length, minor length, and skeleton length characterize the length and spread of the bacterial cell; solidity and roundness characterize the regularity of the bacterial cell outline and morphological integrity; and the nucleus / cell ratio reflects the ratio of the fluorescent signal region within the bacterial cell to the overall cellular region. The units of these multiple physical morphological indicators may be micrometers or dimensionless parameters, and they are uniformly calibrated using an image scale.

[0059] For example, a Python script can be used to extract the following physical morphological indicators of a single bacterial cell based on the segmentation results of each single cell image: Major axis length (micrometers); Minor axis length (micrometers); Solidity (dimensionless); Roundness (dimensionless); Skeleton length (micrometers); Nucleus / cell ratio (dimensionless); These physical morphological parameters are all in uniform units (micrometers or dimensionless), and the images are calibrated using standard rulers.

[0060] In the embodiments of this application, by performing step S5, multidimensional morphological quantitative features can be extracted for each single bacterial cell.

[0061] It is evident that steps S4 and S5 are based on single-frame microscopic images for single-bacterial segmentation and extraction of six morphological features, rather than using a detection method that involves acquiring multiple consecutive frames of images of the same species, reconstructing the sequence, and directly inputting the sequence images into the evaluation model.

[0062] S6. Single indicator performance evaluation: Within a single independent detection unit, the performance of multiple physical morphological indicators of a single bacterial cell is evaluated separately to assess the drug sensitivity discrimination ability of a single physical morphological indicator.

[0063] In one implementation, step S6 specifically includes: S61. Statistical methods were used to compare the distribution differences of multiple physical morphological indicators between drug-sensitive and drug-insensitive bacteria and to perform statistical analysis to verify whether there are significant differences between different drug susceptibility phenotypes for a single physical morphological indicator.

[0064] In practice, for a single bacterial species, drug susceptibility tests can be performed separately for different antibiotics. For example, based on the minimum inhibitory concentration (MIC) and clinical interpretation criteria, the species can be classified as drug-sensitive or drug-insensitive. Furthermore, the drug susceptibility classification of a single bacterial species to antibiotics is determined based on the results of the gold standard antimicrobial susceptibility test. Therefore, the same bacterial species may have different susceptibility classification results under different antibiotic conditions.

[0065] In practice, the preferred statistical method is the Mann-Whitney U test to perform nonparametric difference testing, which is used to determine whether there are significant differences in various indicators between different drug sensitivity phenotypes.

[0066] S62. Within a single independent detection unit, construct a receiver operating characteristic curve for a single physical morphology indicator, evaluate the drug sensitivity discrimination ability of the single physical morphology indicator by the area under the curve, and associate the single physical morphology indicator with the drug sensitivity label.

[0067] In practice, the sensitivity (drug-sensitive / drug-insensitive) of a single physical morphological index of a single bacterial cell to any antibiotic can be labeled using the gold standard antimicrobial susceptibility testing method (such as the microbroth dilution method). In other words, the drug susceptibility label of a single physical morphological index of a single bacterial cell originates from the results of the gold standard antimicrobial susceptibility test.

[0068] In practice, within a single independent detection unit, a Receiver Operating Characteristic Curve (ROC) can be constructed for a single physical morphological indicator, and the Area Under the Curve (AUC) can be calculated to quantify the ability of a single physical morphological indicator to distinguish between drug-sensitive and drug-insensitive bacteria. Then, the single physical morphological indicator is associated with a drug susceptibility label. The closer the AUC value of a single physical morphological indicator is to 1, the stronger its discriminative ability.

[0069] For example, the formula for calculating AUC can be expressed by the following formula (1).

[0070] (1) In formula (1), TPR represents the True Positive Rate; FPR represents the False Positive Rate; TPR=TP / (TP+FN), FPR=FP / (FP+TN), where TP, FP, TN, and FN represent the number of true positives, false positives, true negatives, and false negatives, respectively.

[0071] In step S62, single-index AUC analysis can identify multiple physical morphological indicators with drug susceptibility discrimination capabilities, providing a basis for subsequent multi-index joint modeling. In other words, multiple physical morphological indicators of a single bacterial cell possess certain drug susceptibility discrimination efficacy, providing a basis for subsequent multi-index joint modeling.

[0072] In practice, under the same "strain-antibiotic" combination, images acquired at different treatment durations can all inherit the sensitivity label of the antibiotic corresponding to that strain. That is, images of sensitive strains acquired at 15, 30, 60, and 90 minutes are all labeled with the sensitivity label, and the same applies to drug-insensitive strains.

[0073] In this embodiment of the application, by executing step S6, multiple single physical morphological indicators can all have drug sensitivity discrimination efficacy through discrimination ability evaluation. Furthermore, these multiple physical morphological indicators can be used together as multi-dimensional feature inputs for joint modeling to improve drug sensitivity discrimination performance.

[0074] S7. Joint Modeling and Performance Evaluation: Multiple physical morphological indicators of different independent detection units are imported into the machine learning model as joint input variables to complete model training, cross-validation, performance evaluation and drug sensitivity discrimination, and obtain a rapid drug sensitivity prediction model for each independent detection unit.

[0075] In one implementation, step S7 specifically includes: S71. Construct a machine learning model.

[0076] In practice, the machine learning model takes multiple physical morphological indicators as input and outputs a binary label for drug sensitivity (S) or drug insensitivity (non-S). That is, in step S71, the machine learning model uses "bacterial species - antibiotic type" as the basic discrimination object, rather than using only bacterial species as a single classification object. In other words, the discrimination target of the machine learning model is the drug sensitivity classification corresponding to "bacterial species - antibiotic type," rather than performing end-to-end recognition directly based on the original image sequence or region of interest sequence.

[0077] In specific implementation, the machine learning model can be any one or more of the following: Logistic Regression L1 model, Logistic Regression L2 model, Random Forest model, Histogram Gradient Boosting (HistGB) model, Extreme Gradient Boosting (XGBoost) model, and Lightweight Gradient Boosting (LightGBM) model.

[0078] As an example, taking the random forest model as an example, the classification decision function used can be expressed by the following formula (2): (2) In formula (2), This represents the model output; N represents the number of decision trees. denoted as the classification result of the i-th decision tree; x represents the input feature vector (6 physical morphological indices).

[0079] S72. Combine multiple physical morphological indicators of different independent detection units into a dataset, and then divide the dataset into a training set and a test set.

[0080] For example, the dataset can be divided into a training set and a test set according to a certain ratio, such as 70%–90% for the training set and 10%–30% for the test set, preferably 8:2.

[0081] S73. Using cross-validation, the machine learning model is trained and its performance is evaluated based on the training set and the test set to obtain the trained machine learning model and its performance evaluation results.

[0082] In practice, the preferred method for cross-validation is the hierarchical K-fold cross-validation method, where the K value is preferably 5.

[0083] In practice, this performance evaluation uses, but is not limited to, the area under the curve (AUC) and classification accuracy to comprehensively evaluate the performance of the trained machine learning model. The AUC and classification accuracy can also be used to compare and analyze the performance of different models and models with different processing times.

[0084] S74. Based on the performance evaluation results, determine the rapid drug susceptibility detection time for different bacterial species and antibiotic combinations, and construct a rapid drug susceptibility prediction model for each independent detection unit.

[0085] In practice, based on the performance evaluation results, the performance of the trained machine learning model can be compared under different combinations of bacterial species, antibiotic types, and treatment durations to determine the optimal discrimination model and its drug sensitivity prediction effect in each independent detection unit, and a rapid drug sensitivity prediction model for each independent detection unit can be constructed.

[0086] For example, based on the performance evaluation results, the area under the curve and classification accuracy of the trained machine learning model under different drug action durations can be compared to determine the shortest discrimination time that meets the preset performance threshold. This time can be used to establish the rapid drug sensitivity detection time for different bacterial species-antibiotic combinations. Then, based on the rapid drug sensitivity detection time for different bacterial species-antibiotic combinations, a rapid drug sensitivity prediction model for each independent detection unit can be constructed.

[0087] In other words, to determine the drug susceptibility discrimination time for different bacterial species and antibiotic combinations, this application establishes rapid drug susceptibility prediction models for different drug action durations, and calculates the area under the curve (AUC) and classification accuracy of the rapid drug susceptibility prediction models as performance evaluation indicators. A preset performance threshold is set; for a given treatment duration, if both the AUC and classification accuracy of the rapid drug susceptibility prediction model are not lower than the preset performance threshold, the treatment duration is considered to have drug susceptibility discrimination capability. Among multiple treatment durations that meet the performance threshold, the shortest treatment duration is selected as the optimal discrimination treatment duration, thereby achieving rapid drug susceptibility detection while ensuring discrimination accuracy. The preset performance threshold is an AUC or classification accuracy of not less than 0.70, preferably not less than 0.80, and more preferably not less than 0.90.

[0088] The analytical objective of step S74 is to obtain the earliest discriminable time for different independent detection units, unlike the approach of making a one-time judgment on the survival status of a single strain based solely on continuous sequence images. Specifically, the drug susceptibility discrimination performance of the trained machine learning model is compared under different treatment durations to determine the shortest treatment duration that can achieve drug susceptibility discrimination while ensuring discrimination performance. Since the morphological response time of different bacterial species and different antibiotic combinations is different, their optimal discrimination time may be different. Therefore, step S74 can use this analysis to determine the rapid drug susceptibility detection time for different bacterial species-antibiotic combinations.

[0089] In the embodiments of this application, by executing step S7, not only can drug sensitivity classification be achieved, but also the earliest discriminable time of different bacterial species and antibiotic combinations can be determined, thereby achieving faster drug sensitivity detection.

[0090] In this embodiment, by executing steps S4-S7, interpretable features can be extracted and statistically evaluated based on six explicit morphological indicators obtained after single-cell segmentation: long axis length, short axis length, solidity, roundness, skeleton length, and nucleocell ratio. Joint modeling is then performed on this basis, rather than directly inputting the region of interest sequence of a single bacterial cell into an end-to-end deep learning model for discrimination. Specifically, in steps S4-S7, this embodiment employs an explicit feature analysis path based on the endpoint static image: first, a single bacterial outline is obtained through single-cell segmentation; then, the six morphological indicators are quantitatively extracted; subsequently, statistical difference analysis of single indicators and area under the curve evaluation are performed; and based on this, multi-dimensional feature joint machine learning modeling is conducted to obtain sensitive / non-sensitive discrimination results, rather than directly inputting multiple consecutive frames of the same bacterial cell into the deep learning model.

[0091] S8. Results Output: Output the evaluation results of a single physical morphological index of the target Gram-negative bacteria under the conditions of antibiotic type and treatment duration, as well as the prediction results of the corresponding rapid drug susceptibility prediction model.

[0092] In one implementation, the output of step S8 may include, but is not limited to: statistical differences of various physical morphological indicators, area under the curve, model classification performance indicators, and corresponding drug sensitivity discrimination conclusions.

[0093] In this embodiment of the application, by executing step S8, a rapid drug susceptibility determination basis for the target Gram-negative bacteria can be formed based on the output of step S8.

[0094] To further facilitate those skilled in the art to understand the antimicrobial susceptibility detection method based on the physical morphological characteristics of single bacteria provided in the embodiments of this application, several examples are described below.

[0095] Example 1: Escherichia coli-meropenem combination The antibiotic meropenem (1 µg / mL) was used to analyze parameters such as long axis length, solidity, and skeleton length. The most significant differences were observed at a treatment time of 60 minutes, as shown in the figure. Figure 2 (Differences in morphological characteristics) and Figure 5 (Area under curve (AUC) for a single index).

[0096] Example 2: Klebsiella pneumoniae-ceftazidime combination The treatment method was the same as in Example 1, and cefotaxime (4 µg / mL) was used as the antibiotic. Analysis revealed a clear distinction between sensitive and non-sensitive roundness and solidity indices at 30 minutes, as illustrated in the figure. Figure 3 and Figure 6 .

[0097] Example 3: Pseudomonas aeruginosa-aztreonam Treatment with aztreonam at a concentration of 8 µg / mL resulted in significant differences in roundness and nucleus / cell ratio within 15–30 minutes. (See figure). Figure 4 and Figure 7 .

[0098] Example 4: Multi-index joint machine learning modeling of E. coli with different antibiotics and treatment durations In this embodiment, Escherichia coli was used as the research object. Under different antibiotic treatment and different treatment time conditions, a multi-dimensional feature combined machine learning model was established based on the morphological characteristics of a single bacterium for drug sensitivity discrimination, and the discrimination performance of different models was compared.

[0099] First, single bacterial microscopic images were acquired at different treatment durations after antibiotic treatment. Image segmentation and morphological feature extraction were performed on individual bacteria to obtain six morphological indicators: long axis length, short axis length, solidity, roundness, skeleton length, and nucleus / cell ratio. These six indicators were then used as joint input features.

[0100] Then, the six morphological indicators are input into different machine learning models for drug sensitivity discrimination modeling. These different machine learning models include, but are not limited to: • Logistic Regression Model (L1) • Logistic Regression Model (L2) • Random Forest Model • Histogram Gradient Boosting Model •Extreme gradient boosting model • Lightweight gradient boosting model During model training, the dataset is divided into training and testing sets, preferably in an 8:2 ratio. In the training set, a hierarchical five-fold cross-validation method is used for model training and validation to improve model stability and generalization ability.

[0101] Model performance evaluation metrics include area under the curve (AUC) and classification accuracy. The effectiveness of the multidimensional morphological feature-based machine learning model in E. coli drug sensitivity assessment was evaluated by comparing the performance of different machine learning models under different combinations of antibiotics and treatment durations.

[0102] Figure 8 This is a performance comparison graph showing the combined input of six indicators into six machine learning models under the E. coli-antibiotic-time combination in this embodiment. The graph shows the area under the curve (AUC) and accuracy of different models under each combination.

[0103] pass Figure 8It can be seen that under different combinations of antibiotics and treatment times, all machine learning models can achieve drug sensitivity discrimination of Escherichia coli based on the morphological characteristics of single bacteria, and the model performance reaches a high level for some treatment times, indicating that the method has good drug sensitivity discrimination ability.

[0104] Example 5: Multi-index joint machine learning modeling of Klebsiella pneumoniae with different antibiotics and treatment durations In this embodiment, Klebsiella pneumoniae was used as the research object. Under different antibiotic treatment and different treatment time conditions, a multi-dimensional feature combined machine learning model was established based on the morphological characteristics of a single bacterium for drug sensitivity discrimination, and the discrimination performance of different models was compared.

[0105] Microscopic images of single bacteria were acquired at different treatment durations after antibiotic treatment. Image segmentation and morphological feature extraction were performed on individual bacteria to obtain six morphological indicators: long axis length, short axis length, solidity, roundness, skeleton length, and nucleus / cell ratio. These six indicators were then used as joint input features.

[0106] The above six features were input into various machine learning models for drug sensitivity discrimination modeling, including logistic regression, random forest, histogram gradient boosting, extreme gradient boosting, and lightweight gradient boosting models. During model training, training and test sets were partitioned, and hierarchical five-fold cross-validation was used for model training and performance evaluation.

[0107] The model performance evaluation metrics include area under the curve (AUC) and classification accuracy. By comparing the model performance of different models under different antibiotic and time combinations, the application effect of multidimensional morphological features in the drug sensitivity discrimination of Klebsiella pneumoniae is analyzed.

[0108] Figure 9 This is a performance comparison graph showing the combined input of six indicators into six machine learning models under the Klebsiella pneumoniae-antibiotic-time combination in this embodiment. The graph shows the area under the curve (AUC) and accuracy of different models under each combination.

[0109] pass Figure 9 It can be seen that under different antibiotic effects and different time conditions, multiple machine learning models can all achieve drug sensitivity discrimination of Klebsiella pneumoniae, and the model performance reaches the preset performance threshold at some processing times. This indicates that the antimicrobial drug sensitivity detection method based on the physical morphological characteristics of single bacteria provided in this application embodiment can be used for rapid drug sensitivity detection of Klebsiella pneumoniae.

[0110] Example 6: Multi-index joint machine learning modeling of Pseudomonas aeruginosa with different antibiotics and treatment durations In this embodiment, Pseudomonas aeruginosa is used as the research object. Under different antibiotic treatment and different treatment time conditions, a multi-dimensional feature combined machine learning model is established based on the morphological characteristics of a single bacterium for drug sensitivity discrimination, and the discrimination performance of different models is compared.

[0111] Microscopic images of single bacteria were acquired at different treatment durations after antibiotic treatment. Image segmentation and morphological feature extraction were performed on individual bacteria to obtain six morphological indicators: long axis length, short axis length, solidity, roundness, skeleton length, and nucleus / cell ratio. These six indicators were then used as joint input features.

[0112] Six morphological features were input into various machine learning models for drug sensitivity discrimination modeling. Model training and performance evaluation were conducted using a training and test set split and a hierarchical five-fold cross-validation method. Model performance evaluation metrics included the area under the curve (AUC) and classification accuracy.

[0113] Figure 10 This is a performance comparison graph showing the combined input of six indicators into six machine learning models under the Pseudomonas aeruginosa-antibiotic-time combination in this embodiment. The graph shows the area under the curve (AUC) and accuracy of different models under each combination.

[0114] pass Figure 10 It can be seen that Pseudomonas aeruginosa can show obvious morphological changes in a short period of time after the action of some antibiotics. The multidimensional morphological features combined with the machine learning model can achieve drug sensitivity discrimination in a short time, indicating that the antimicrobial drug sensitivity detection method based on the physical morphological characteristics of single bacteria provided in this application is suitable for rapid drug sensitivity detection of Pseudomonas aeruginosa.

[0115] The above-described embodiments extract the morphological characteristics of single bacterial cells from different bacterial species under different antibiotic treatments and for different durations. By employing a multi-dimensional feature-based machine learning model for drug susceptibility assessment, a rapid drug susceptibility detection time for different bacterial species-antibiotic combinations can be established, achieving rapid drug susceptibility detection based on the morphological characteristics of individual bacteria. Furthermore, it demonstrates that the antimicrobial susceptibility detection method based on the physical morphological characteristics of single bacteria provided in this application does not rely on bacterial population growth results, but rather completes drug susceptibility assessment based on single-cell morphological changes, significantly reducing the drug susceptibility detection time from 8-24 hours to within 90 minutes.

[0116] In summary, by implementing the antimicrobial susceptibility detection method based on the physical morphological characteristics of single bacteria provided in this application, short-term antibiotic treatment (15–90 minutes) can be applied to Gram-negative bacteria (such as Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa). Subsequently, images of individual bacteria are extracted using inverted confocal microscopy imaging and segmentation algorithms. The main physical morphological characteristics, including major axis length, minor axis length, solidity, roundness, skeleton length, and nucleocellular ratio, are calculated using Python scripts. By performing Mann-Whitney U tests and area under the curve (AUC) analysis on single indicators, and further introducing multiple machine learning models for joint feature modeling, rapid prediction of the susceptibility of bacterial strains to antimicrobial drugs can be achieved in a short time. For example, different bacterial species and different antibiotic combinations can exhibit different morphological response speeds and discrimination time windows. For instance, Pseudomonas aeruginosa can form stable discrimination characteristics within 15–30 minutes after treatment with aztreonam, while some Enterobacteriaceae require a longer time to reach a stable discrimination state after treatment with β-lactam drugs.

[0117] Furthermore, compared to existing schemes that rely on continuous sequence image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria, the antimicrobial susceptibility detection method based on the physical morphological characteristics of single bacteria provided in this application (hereinafter referred to as "this method") does not identify the status of single bacteria based on multiple antibiotic concentrations and continuous sequence imaging. Instead, it establishes an independent detection unit based on "bacterial species-antibiotic-treatment duration," performs short-term drug treatment at clinically sensitive breakpoint concentrations, and performs endpoint static microscopic imaging at the set treatment duration. Subsequently, it segments the individual bacteria in the image and extracts six interpretable quantitative morphological features: major axis length, minor axis length, solidity, roundness, skeleton length, and nucleocellular ratio. Based on this, this method first clarifies the independent discriminative ability of each physical morphological indicator through single-index statistical analysis and area under the curve evaluation, and then further combines various traditional machine learning models to complete the sensitive / non-sensitive drug susceptibility discrimination. Therefore, this method differs substantially from existing schemes that use continuous sequence image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria in terms of technical approach, input data format, feature construction method, label definition method, and final interpretation target. In particular, the independent detection unit established by this method is not only used for drug sensitivity classification, but also for establishing the time response patterns of different bacterial-drug combinations.

[0118] In summary, compared with existing technologies, this method has the following characteristics and beneficial effects: (I) Characteristics (1.1) A rapid drug sensitivity analysis framework with independent detection units is established based on the basic organizational logic of "bacterial species-antibiotic-treatment time". Static microscopic imaging of the endpoint is performed based on the set treatment time, rather than continuous multi-frame sequence imaging of the same bacterial species. (1.2) Technical route for drug sensitivity determination based on the morphological changes of single bacteria after early antibiotic treatment rather than the end-stage proliferation results of the bacterial community; (1.3) Extract multidimensional morphological quantitative features based on single-strain results, especially the six physical morphological indicators: long axis length, short axis length, solidity, roundness, skeleton length and nucleus / cell ratio. (1.4) Analytical methods that use statistical difference analysis and area under curve evaluation to characterize the independent discriminant ability of a single physical morphological index; (1.5) Multiple single-cell morphological indicators (i.e. physical morphological indicators) with discriminative ability are jointly input to construct a machine learning model to achieve rapid drug sensitivity prediction. The joint modeling is based on explicit and interpretable morphological features, rather than directly inputting the region of interest sequence of a single cell into an end-to-end deep learning model. (1.6) The gold standard antimicrobial susceptibility test results corresponding to “strain-antibiotic” are used as model labels, and the analysis target of drug susceptibility / drug insensitivity judgment and the earliest discriminant time is output. In addition, the drug susceptibility label comes from the gold standard antimicrobial susceptibility test results corresponding to “strain-antibiotic”, and the model output is the sensitive or insensitive classification result. At the same time, by comparing the performance of the model under different treatment durations, the earliest discriminant time of different strain-antibiotic combinations can be further determined, thereby establishing the corresponding rapid drug susceptibility detection time window.

[0119] (1.7) The processes involved, such as drug processing, experimental setup, antibiotic concentration setting, image acquisition, segmentation, feature extraction and analysis, are all reproducible.

[0120] (ii) Beneficial effects (2.1) Fast detection speed This method uses "bacterial species-antibiotic-treatment time" as the basic organizational unit. It achieves rapid identification of bacterial drug susceptibility by performing microscopic imaging, single-cell segmentation, feature extraction, and joint modeling analysis on the early morphological response of bacteria after antibiotic treatment. For example, drug susceptibility determination can be achieved based on the early morphological response of single bacteria within 15–90 minutes after antibiotic treatment, eliminating the need to wait for the traditional culture endpoint and significantly shortening the drug susceptibility detection cycle from 8–24 hours to within 90 minutes.

[0121] (2.2) Strong early discrimination ability This method directly utilizes morphological information at the single-cell level for analysis, enabling earlier capture of bacterial response characteristics to antibiotics and providing a basis for early anti-infection decision-making. Specifically, this method does not rely on the interpretation method based on bacterial colony proliferation endpoints in traditional antimicrobial susceptibility testing. Instead, it utilizes quantifiable changes in the morphology of single cells at the early stage of antibiotic action to rapidly classify bacterial susceptibility, thereby shortening the detection cycle, improving the timeliness of early and precise drug administration in anti-infection treatment, and providing a new implementation path for rapid drug susceptibility testing technology.

[0122] (2.3) The technical routes are clearly distinguished and the features are more interpretable. Compared with existing methods that use continuous sequence image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria, this method adopts a technical approach that involves short-term processing at clinically sensitive breakpoint concentrations, static image acquisition at a preset processing time endpoint, extraction of six interpretable morphological features after single-bacterial segmentation, and joint evaluation by multiple traditional machine learning models. This approach results in simpler input data, a clearer feature construction process, and more interpretable discrimination criteria.

[0123] (2.4) The label correspondence is clearer, and the results are closer to the logic of clinical drug sensitivity interpretation. This method uses the gold standard antimicrobial susceptibility test results corresponding to "species-antibiotic" as the label source, and outputs drug susceptibility / drug insensitivity classification conclusions. It can further determine the earliest discriminative time window for different species-antibiotic combinations. Therefore, the correspondence with the clinical drug susceptibility interpretation system is clearer.

[0124] (2.5) Rich multidimensional feature information, joint modeling improves discrimination performance The multidimensional morphological indicators extracted by this method, such as major axis length, minor axis length, solidity, roundness, skeleton length, and nucleus / cell ratio, can characterize the structural changes of bacteria under drug action from multiple perspectives, which is more comprehensive than a single indicator. At the same time, by combining multidimensional features with machine learning modeling, information complementarity is achieved, thereby improving the accuracy and stability of drug sensitivity assessment.

[0125] (2.6) It is suitable for combinations of various bacterial strains and antibiotics, and has good scalability. This method employs an independent detection unit design for "bacterial species-antibiotic-treatment time," making it suitable for rapid drug sensitivity analysis under different bacterial species, drugs, and treatment time conditions, and exhibiting good scalability.

[0126] (2.7) Easy to standardize and systematize The process of this method consists of modules such as sample processing, endpoint static image acquisition, single-bacterial segmentation, feature extraction, statistical analysis, machine learning modeling, and result output. The technical path is clear, and the equipment and computing process are easier to standardize and promote. It can also be further developed into a database system, automated analysis software, or embedded intelligent detection platform.

[0127] In summary, this method constructs independent detection units based on different bacterial species, antibiotics, and drug treatment durations. Within each independent detection unit, single-cell images are acquired, and multiple physical morphological indicators with drug sensitivity discrimination significance are extracted. Based on these, single-indicator discrimination ability evaluation and multi-indicator joint modeling analysis are conducted, ultimately forming a rapid drug sensitivity prediction model for different combinations of bacterial species, antibiotics, and treatment durations. This model can identify early single-cell morphological changes in common Gram-negative bacteria after antibiotic treatment. A rapid drug sensitivity discrimination process consisting of "drug treatment—single-cell imaging—single-cell segmentation—multi-dimensional feature extraction—single-indicator performance evaluation—joint modeling and performance evaluation—result output" is established, thereby shortening detection time while improving the method's interpretability, standardization potential, and clinical translation applicability.

[0128] Furthermore, unlike existing methods that rely on continuous image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria, this method does not rely on continuous dynamic video information for single-shot classification. Instead, it is based on independent detection units of "bacterial species-antibiotic-treatment duration." Under clinically sensitive breakpoint concentration conditions, it achieves rapid discrimination of bacterial susceptibility / non-susceptibility status through static microscopic imaging at preset time points, single-bacterial segmentation, and explicit morphological feature extraction, combined with single-index statistical evaluation and multi-dimensional feature joint modeling. It further determines the earliest time window for different bacterial species and antibiotic combinations to achieve stable discrimination capability. Therefore, this method not only enables rapid drug sensitivity classification but also establishes the temporal response patterns of different bacterial-antibiotic combinations. It differs substantially from existing methods that rely on continuous image acquisition and directly employ end-to-end deep learning models to determine the survival or susceptibility status of single bacteria in terms of technical approach, interpretation logic, feature construction method, and clinical application goals.

[0129] Figure 11 This diagram illustrates a structural block diagram of an antimicrobial susceptibility detection device based on the physical morphological characteristics of a single bacterium, according to an embodiment of this application. Figure 11 As shown, the device may include: Unit 10 is established to create independent detection units: target Gram-negative bacteria are selected, and different independent detection units are established with antibiotic type and treatment time as adjustable variables for the target Gram-negative bacteria. Processing unit 20 is used for drug processing: for a single independent detection unit, the bacterial suspension containing the target Gram-negative bacteria is reacted with an antibiotic solution of a set concentration, and after the set processing time is reached, the bacterial sample corresponding to the single independent detection unit is obtained. Imaging unit 30 is used for single-cell imaging: the bacterial sample corresponding to a single independent detection unit is fixed, stained and microscopic images are acquired to obtain several single-cell images of a single independent detection unit; Segmentation unit 40 is used for single bacterial cell segmentation: using a specified image segmentation algorithm to identify and segment individual bacterial cells from each single bacterial cell image, and obtain the segmentation result of each single bacterial cell image; Extraction unit 50 is used for multi-dimensional feature extraction: based on the segmentation results of each single bacterial cell image, multiple physical morphological indicators of a single bacterial cell are extracted; Evaluation unit 60 is used for single indicator performance evaluation: within a single independent detection unit, the performance of multiple physical morphological indicators of a single bacterial cell is evaluated separately to assess the drug sensitivity discrimination ability of a single physical morphological indicator. Modeling unit 70 is used for joint modeling and performance evaluation: multiple physical morphological indicators of different independent detection units are imported into the machine learning model as joint input variables to complete model training, cross-validation, performance evaluation and drug sensitivity discrimination, and obtain a rapid drug sensitivity prediction model for each independent detection unit. Output unit 80 is used for result output: outputting the evaluation results of a single physical morphological index of the target Gram-negative bacteria under the conditions of antibiotic type and treatment time, and the prediction results of the corresponding rapid drug susceptibility prediction model.

[0130] The functions of each unit in the antimicrobial susceptibility detection device based on the physical morphological characteristics of single bacteria in this application embodiment can be found in the corresponding description in the above method, and will not be repeated here.

[0131] Figure 12 A structural block diagram of an electronic device according to an embodiment of this application is shown. Figure 12 As shown, the electronic device includes a memory 100 and a processor 200. The memory 100 stores instructions, which are loaded and executed by the processor 200 to implement the antimicrobial susceptibility detection method based on the physical morphological characteristics of a single bacterium in the above embodiments. The number of memories 100 and processors 200 can be one or more.

[0132] The electronic device also includes: The communication interface 300 is used to communicate with external devices and perform data exchange and transmission.

[0133] If the memory 100, processor 200, and communication interface 300 are implemented independently, they can be interconnected via a bus to communicate with each other. This bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 12 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0134] Optionally, in a specific implementation, if the memory 100, processor 200, and communication interface 300 are integrated on a single chip, then the memory 100, processor 200, and communication interface 300 can communicate with each other through an internal interface.

[0135] This application provides a computer-readable storage medium storing a computer program. When the computer program is run on a computer, it implements the method provided in this application.

[0136] This application also provides a chip, which includes a processor for calling and executing instructions stored in a memory, causing a communication device with the chip installed to perform the method provided in this application.

[0137] This application also provides a chip, including: an input interface, an output interface, a processor, and a memory. The input interface, output interface, processor, and memory are connected through an internal connection path. The processor is used to execute code in the memory. When the code is executed, the processor is used to execute the method provided in the application embodiment.

[0138] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. General-purpose processors can be microprocessors or any conventional processor. It is worth noting that the processor can be a processor supporting Advanced Reduced Instruction Set Machines (ARM) architecture.

[0139] Further, optionally, the aforementioned memory may include read-only memory and random access memory, and may also include non-volatile random access memory. The memory may be volatile or non-volatile, or may include both. Non-volatile memory may include read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory. Volatile memory may include random access memory (RAM), which serves as an external cache. Many forms of RAM are available by way of example, but not limitation. Examples include static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (SLDRAM), and direct rambus RAM (DR RAM).

[0140] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the flow or function according to this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another.

[0141] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of those different embodiments or examples.

[0142] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "a plurality of" means two or more, unless otherwise explicitly specified.

[0143] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process. Furthermore, the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functionality involved.

[0144] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus or device (such as a computer-based system, a processor-included system or other system that can fetch and execute instructions from, an instruction execution system, apparatus or device).

[0145] It should be understood that various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. All or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware, the program being stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiments.

[0146] Furthermore, the functional units in the various embodiments of this application can be integrated into a processing module, or each unit can exist physically separately, or two or more units can be integrated into a module. The integrated module can be implemented in hardware or as a software functional module. If the integrated module is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. This storage medium can be a read-only memory, a disk, or an optical disk, etc.

[0147] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various variations or substitutions within the technical scope disclosed in this application, and these should all be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for detecting antimicrobial susceptibility based on the physical morphological characteristics of single bacteria, characterized in that, include: S1. Establish independent detection units: Select target Gram-negative bacteria, and establish different independent detection units for the target Gram-negative bacteria, with antibiotic type and treatment time as adjustable variables; S2. Drug treatment: For a single independent detection unit, the bacterial suspension containing the target Gram-negative bacteria is reacted with an antibiotic solution of a set concentration. After the set treatment time is reached, a bacterial sample corresponding to the single independent detection unit is obtained. S3. Single cell imaging: Fix, stain, and acquire microscopic images of the bacterial sample corresponding to a single independent detection unit to obtain several single cell images of a single independent detection unit; S4. Single bacterial cell segmentation: Using a specified image segmentation algorithm, identify and segment individual bacterial cells from each single bacterial cell image to obtain the segmentation result of each single bacterial cell image; S5. Multidimensional feature extraction: Based on the segmentation results of each single bacterial cell image, extract multiple physical morphological indicators of a single bacterial cell; S6. Single index performance evaluation: Within a single independent detection unit, the performance of multiple physical morphological indicators of the single bacterial cell is evaluated to assess the drug sensitivity discrimination ability of a single physical morphological indicator. S7. Joint modeling and performance evaluation: Multiple physical morphological indicators of different independent detection units are imported into the machine learning model as joint input variables to complete model training, cross-validation, performance evaluation and drug sensitivity discrimination, and obtain a rapid drug sensitivity prediction model for each independent detection unit. S8. Output Results: Output the evaluation results of the target Gram-negative bacteria under the conditions of antibiotic type and treatment duration, and the prediction results of the corresponding rapid drug sensitivity prediction model.

2. The method according to claim 1, characterized in that, The target Gram-negative bacteria are selected from Escherichia coli, Klebsiella pneumoniae, and Pseudomonas aeruginosa as the target Gram-negative bacteria. The antibiotics are selected from aztreonam, ceftazidime, meropenem, and ciprofloxacin; The processing time is set to 15 minutes, 30 minutes, 60 minutes or 90 minutes.

3. The method according to claim 1, characterized in that, Step S2 specifically includes: S21. For a single independent detection unit, fresh colonies of the target Gram-negative bacteria clinically isolated from a blood agar plate are picked, a bacterial suspension is prepared, and the concentration is adjusted to a predetermined treatment concentration. S22. Dilute the antibiotic solution of the set concentration and add it to Miller-Hinton broth medium and mix well. Then inoculate the prepared bacterial suspension into Miller-Hinton broth medium. S23. Place the Miller-Hinton broth culture medium in a constant temperature shaker and incubate for the set treatment time to obtain a bacterial sample corresponding to a single independent detection unit.

4. The method according to claim 1, characterized in that, Step S3 specifically includes: S31. Fix and stain the cell nuclei of the bacterial sample corresponding to a single independent detection unit to obtain the processed bacterial sample. S32. Call the microscopic imaging system to acquire a full-field image of the processed bacterial sample, and obtain several single bacterial images of a single independent detection unit.

5. The method according to claim 1, characterized in that, Step S4 specifically includes: An instance segmentation algorithm is used to identify the contour of each bacterial cell in each bacterial cell image and generate a mask. Then, the mask of each bacterial cell is separated from the background and adjacent targets to obtain the segmentation result of each bacterial cell image.

6. The method according to claim 1, characterized in that, Step S6 specifically includes: S61. Statistical methods were used to compare the distribution differences of multiple physical morphological indicators between drug-sensitive and drug-insensitive bacteria and to perform statistical analysis to verify whether there are significant differences between different drug-sensitive phenotypes for a single physical morphological indicator. S62. Within a single independent detection unit, a subject operating characteristic curve is constructed for a single physical morphological index, and the drug sensitivity discrimination ability of the single physical morphological index is evaluated by the area under the curve, and the single physical morphological index is associated with a drug sensitivity label.

7. The method according to any one of claims 1-6, characterized in that, Step S7 specifically includes: S71. Construct a machine learning model, wherein the input of the machine learning model is multiple physical morphological indicators, and the output is a binary label of drug sensitivity or drug insensitivity; S72. The multiple physical morphological indicators of different independent detection units are combined into a dataset, and then the dataset is divided into a training set and a test set. S73. The machine learning model is trained and its performance is evaluated using the cross-validation method based on the training set and the test set to obtain the trained machine learning model and its performance evaluation results. S74. Based on the performance evaluation results, determine the rapid drug sensitivity detection time for different combinations of bacterial species and antibiotics, and construct a rapid drug sensitivity prediction model for each of the independent detection units.

8. A device for detecting antimicrobial susceptibility based on the physical morphological characteristics of a single bacterium, characterized in that, include: Establishment unit, used to establish independent detection unit: Select target Gram-negative bacteria, and establish different independent detection units for the target Gram-negative bacteria, with antibiotic type and treatment time as adjustable variables; The processing unit is used for drug processing: for a single independent detection unit, the bacterial suspension containing the target Gram-negative bacteria is reacted with an antibiotic solution of a set concentration, and after a set processing time is reached, a bacterial sample corresponding to the single independent detection unit is obtained. Imaging unit for single-cell imaging: fixing, staining and acquiring microscopic images of bacterial samples corresponding to a single independent detection unit to obtain several single-cell images of a single independent detection unit; A segmentation unit for single-cell segmentation: using a specified image segmentation algorithm to identify and segment individual cells from each single-cell image, thereby obtaining a segmentation result for each single-cell image; The extraction unit is used for multi-dimensional feature extraction: based on the segmentation results of each single bacterial cell image, multiple physical morphological indicators of a single bacterial cell are extracted; An evaluation unit is used for single-index performance evaluation: within a single independent detection unit, multiple physical morphological indices of the single bacterial cell are evaluated to assess the drug sensitivity discrimination capability of the single physical morphological index. Modeling unit, used for joint modeling and performance evaluation: multiple physical morphological indicators of different independent detection units are imported into the machine learning model as joint input variables to complete model training, cross-validation, performance evaluation and drug sensitivity discrimination, and obtain a rapid drug sensitivity prediction model for each independent detection unit. The output unit is used for result output: outputting the evaluation results of a single physical morphological index of the target Gram-negative bacteria under the antibiotic type-treatment time condition and the corresponding prediction results of the rapid drug sensitivity prediction model.

9. An electronic device, characterized in that, include: A memory and a processor, wherein the memory stores instructions which are loaded and executed by the processor to implement the method as described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a computer, implements the method as described in any one of claims 1-7.

Images