A method for rapid detection of mixed bacterial resistance based on snapshot hyperspectral system

Abstract

The present invention relates to a method for rapidly detecting mixed bacterial resistance based on a snapshot hyperspectral system. In the first step, a snapshot hyperspectral microscopic imaging module is used to capture the atlas data of a mixed bacterial sample; in the second step, the atlas data is processed and reconstructed to obtain an atlas cube; in the third step, the atlas cube is input into a bacteria recognition and classification algorithm to identify the types of mixed bacteria; in the fourth step, a machine vision tracking module is used to record the movement trajectory of bacteria under antibiotic treatment; in the fifth step, the bacterial movement trajectory and morphological changes are analyzed to evaluate the bacterial resistance to different antibiotics. The detection method of the present invention has the advantages of simple operation, fast detection speed, high accuracy, good repeatability, etc., and has important application value for clinical diagnosis and treatment.

Description

Technical Field

[0001] The invention belongs to the technical field of hyperspectral technology and relates to a method for rapidly detecting mixed bacterial resistance based on a snapshot hyperspectral system. Background Art

[0002] The diagnosis of pathogenic microorganisms, especially the diagnosis of common clinical bacteria, is a major application area of ​​in vitro diagnostic technology. The current clinical diagnostic method for bacterial infection requires the sample to be enriched, cultured and purified first, and then the pathogens in the sample are identified and tested for species and drug sensitivity (or antibiotic sensitivity testing, AST). The process of clinical bacterial detection is slow and may take more than two days even under advanced laboratory conditions. Because the phenotypic antibiotic sensitivity test of blood samples involves three overnight culture steps, the turnaround time of the entire process usually takes at least 3-4 days. For infections with a high bacterial density in samples such as urinary tract infections, even if the in vitro diagnosis does not require amplification and culture of the original sample, it still takes up to 36 hours. It can be seen that the current clinical in vitro diagnosis process of pathogenic bacterial infections is relatively long, and the above methods cannot guide antibiotic treatment in the critical early stages of infection.

[0003] The main bottleneck of clinical pathogen diagnosis is that the diagnosis and drug sensitivity testing of pathogens often require long-term enrichment culture, or the detection speed needs to be increased at the expense of accuracy and cost. In recent years, polymerase chain reaction (PCR) has shown certain advantages in bacterial classification and drug resistance detection. By designing specific primers, PCR can amplify DNA fragments of target bacteria or drug resistance genes from complex biological samples within 8 hours, thereby achieving rapid classification of bacteria and accurate detection of drug resistance. However, PCR reactions need to be carried out under strict temperature cycling conditions and have certain requirements on the skill level of experimental operators. Although PCR can detect DNA fragments of bacteria or drug resistance genes, it cannot directly reflect the activity and number of bacteria in the sample. In clinical applications, PCR test results usually need to be combined with other detection methods to comprehensively evaluate bacterial infection and drug resistance. Therefore, the development of a new method for rapid, culture-free, and low-cost pathogen infection detection is crucial to guide the correct use of clinical drugs and delay the development of pathogen resistance problems. Summary of the invention

[0004] In order to overcome the shortcomings of the prior art, the present invention proposes a method for quickly detecting mixed bacterial resistance based on a snapshot hyperspectral system, which utilizes the dynamic detection, high resolution and high sensitivity characteristics of the snapshot hyperspectral imaging system to achieve rapid and accurate detection of mixed bacterial resistance.

[0005] The technical scheme for achieving the purpose of the present invention is as follows:

[0006] A method for rapidly detecting mixed bacterial resistance based on a snapshot hyperspectral system comprises the following steps:

[0007] a. Capture the spectral data of mixed bacterial samples using the snapshot hyperspectral microscopy module;

[0008] b. Calibrate, process and reconstruct the atlas data into single pixels to obtain an atlas cube;

[0009] c. Input the atlas cube into the bacteria recognition and classification algorithm for pixel-level recognition, thereby achieving single-cell level analysis in mixed bacterial samples;

[0010] d. Use machine vision tracking modules to record the movement trajectory of target bacteria under antibiotic treatment;

[0011] e. Analyze bacterial movement trajectories and morphological changes, and evaluate the resistance of target bacteria to different antibiotics.

[0012] The mixed bacterial sample is pretreated or not, and comes from one or more of the following combinations, including directly or indirectly from the human body to be tested, animals, plants and isolates other than humans, and directly or indirectly from the environment to be tested.

[0013] The snapshot hyperspectral microscopy module in step a includes: a microscope objective unit, which is used to convert the information at the focus of the bacterial sample into infinite parallel light; a spectroscopic unit, which includes a filter, a beam splitter, a focusing lens, a square aperture, a collimating lens, a two-dimensional grating, and an imaging lens in sequence, and is used to spectrally split the infinite parallel light to form multiple spectral components; a detector unit, which is used to receive and record the light signal after spectral splitting; the wavelength of the light source selected by the snapshot hyperspectral microscopy module covers the visible light to near-infrared region to cover the key spectral characteristics of the bacterial community.

[0014] The spectroscopic unit uses a two-dimensional grating to generate a diffraction pattern: the center is zero-order diffraction, which locates the target spatial information; the periphery is high-order diffraction, which contains data at different projection angles and reconstructs the target spectral information, and finally integrates it into complete spatial and spectral data of the target.

[0015] Step b uses a mercury lamp for calibration.

[0016] In step b, the captured atlas data is processed by single pixel using the end-to-end UNET3 reconstruction algorithm and restored to an atlas cube, thereby achieving single-cell level recognition.

[0017] The bacterial recognition and classification algorithm in step c adopts a transformer-based deep learning model, which optimizes model parameters through training to improve recognition accuracy.

[0018] In step d, the machine vision tracking module combines image segmentation and target tracking algorithms to track the bacteria in the hyperspectral snapshot image, record the movement trajectory, and evaluate the bacterial resistance to different antibiotics by comparing and analyzing the changes in bacterial movement trajectory and morphology under different drug treatments.

[0019] The drug resistance assessment in step e includes using statistical analysis and pattern recognition methods to extract bacterial movement trajectories and morphological changes, learning the relationship between cell movement patterns and activity based on the long short-term memory network machine learning model, and realizing non-invasive real-time drug resistance assessment.

[0020] The drug resistance assessment in step e also includes training the model using the prepared dataset and adjusting the model parameters through an optimization algorithm to minimize the prediction error.

[0021] Beneficial effects of the present invention

[0022] The present invention proposes a method for rapidly detecting mixed bacterial resistance based on a snapshot hyperspectral system. By cleverly integrating snapshot hyperspectral imaging technology, it not only integrates high-resolution two-dimensional morphology and one-dimensional spectrum, realizes accurate analysis at the single-cell level, but also opens up a new way for rapid detection of bacteria. More importantly, the hyperspectral dynamic imaging detection method combined with the bacterial movement trajectory enables rapid detection of clinical pathogen drug sensitivity, significantly improving detection efficiency (which can be compressed to within half an hour). The implementation of this innovative method is of great significance for optimizing the current status of clinical reliance on empirical medication, guiding the formulation of precise treatment plans, and saving patients' lives in a timely manner. Through the application of the present invention, it is expected to bring revolutionary changes to the medical field, making the treatment of bacterial infections more scientific, efficient and personalized.

[0023] This method can integrate a hyperspectral snapshot imaging system, a data processing and analysis module, a machine vision processing software, and a drug resistance detection model to build a fast and convenient hybrid bacterial resistance detection platform that can realize automated operation and real-time data processing to improve detection efficiency and accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 Diagram of the steps for rapid detection of mixed bacterial resistance based on snapshot hyperspectral system.

[0025] Figure 2a A schematic diagram of a snapshot hyperspectral system;

[0026] In the figure: 1. Microscope objective; 2. 4500-700nm filter; 3. Beam splitter; 4. Focusing lens; 5. Liquid crystal tunable filter; 6. CMOS camera; 7. Square aperture; 8. Collimating lens; 9. Two-dimensional grating; 10. Imaging lens.

[0027] Figure 2b A physical picture of a snapshot hyperspectral system.

[0028] Figure 3 Snapshot hyperspectral system imaging results for different grating diffraction orders.

[0029] Figure 4 For calibration of the snapshot hyperspectral system using a standard mercury lamp.

[0030] Figure 5 This is a snapshot hyperspectral imaging image of bacteria. The center is the zero-order image, and the eight surrounding areas are the first-order diffraction spectral images caused by two-dimensional grating diffraction.

[0031] Figure 6 This is a zero-order snapshot hyperspectral image of bacteria.

[0032] Figure 7 Comparison of the reconstructed snapshot hyperspectral and the corresponding LCTF spectrum in the background light source region.

[0033] Figure 8 Comparison of snapshot hyperspectral reconstructions for Escherichia coli and Staphylococcus aureus. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0035] Example 1

[0036] like Figure 1 As shown, a method for rapidly detecting mixed bacterial resistance based on a snapshot hyperspectral system utilizes the dynamic detection, high resolution, and high sensitivity characteristics of a snapshot hyperspectral imaging system to achieve rapid and accurate detection of mixed bacterial resistance.

[0037] The bacteria used in the experiment are derived from common clinical samples, such as body fluids (including blood, cerebrospinal fluid, gastric juice, digestive juice, semen, saliva, tears, sweat, urine, vaginal secretions, etc.) and processed samples thereof. For those skilled in the art, according to the needs of the test, the mixed bacterial sample described in the present invention may be pre-treated or not, and may come from one or more of the following combinations, including directly or indirectly from the human body to be tested, animals other than humans, plants and isolates, and directly or indirectly from the environment to be tested.

[0038] Use a pipette to draw 0.5mL of a sample containing multiple bacteria and drop it into a centrifuge tube; then use a pipette to draw 2.5mL of ultrapure water and add it to the centrifuge tube. Then, after sufficient oscillation using a vortex oscillator, filter the diluted sample using a 3μm mixed cellulose filter membrane. Combine the syringe and filter, and slowly push the syringe to make the diluted sample flow through the filter membrane to obtain the filtrate containing mixed bacteria. Drop the filtrate sample on a glass slide, and use the snapshot hyperspectral microscopy module to capture the spectral data of the mixed bacterial sample.

[0039] The atlas data is calibrated, processed and single-pixel reconstructed to obtain the atlas cube.

[0040] Inputting the atlas cube into the bacterial recognition and classification algorithm can achieve pixel-level recognition, thereby realizing accurate analysis at the single-cell level in mixed bacterial samples.

[0041] A machine vision tracking module is used to record the movement trajectory of specific types of bacteria under antibiotic treatment, including re-identification and tracking after collision with other bacteria.

[0042] Analyze the bacterial movement trajectory and morphological changes, and evaluate the resistance of specific types of bacteria to different antibiotics. The worse the resistance, the slower the movement or even stops under antibiotic treatment.

[0043] Example 2

[0044] In this embodiment, mixed bacteria are detected and imaged by a snapshot hyperspectral microscopy system. The system is installed on a conventional microscope, and a liquid crystal tunable filter (LCTF) spectral imaging module is placed in a conjugate position to obtain the true value of the training. Before using the system, the LCTF module should be used to obtain the true spectral image cube of the microscopic target. Subsequently, the snapshot hyperspectral module is trained using a machine learning algorithm to reconstruct the spectral image cube. After the training process, the LCTF module is no longer required, and only the snapshot hyperspectral system is used for single-exposure microscopy imaging. The structural schematic diagram and the actual picture of the snapshot hyperspectral system are shown in Figure 1 and Figure 2, respectively. Figure 2a , Figure 2b As shown. The microscope objective 1 converts the target information at the focus into parallel light through the 450nm-700nm filter 2 and the beam splitter 3, generating transmitted light and reflected light. The transmitted light is detected by the CMOS camera 6 through the focusing lens 4 and the liquid crystal tunable filter 5. The 26 wavelength channel ranges can be scanned by the C++ control algorithm to achieve LCTF spectral imaging to obtain accurate spectral imaging data cubes from 450-700nm. The reflected light is focused on a square aperture 7 through the focusing lens 4, and then collimated by the collimating lens 8 and incident on the two-dimensional grating 9 for spectral splitting. Through the imaging lens 10, light of different diffraction orders and wavelengths is focused at different positions in the CMOS camera 6 to achieve snapshot hyperspectral microscopy imaging.

[0045] The snapshot hyperspectral microscopy system uses a two-dimensional grating to realize diffraction spectroscopy. The middle part of the diffraction image is the zero-order diffraction of the grating, which is used to determine the spatial information of the target image; the rest of the diffraction image is the high-order diffraction of the grating, which corresponds to the projection data of the target object data cube at different projection angles, and is used to reconstruct the spectral information of the target object, thereby obtaining the complete spatial and spectral information of the target object. The hyperspectral diffraction image collected by the system is shown in Figure 1. Figure 3 shown.

[0046] A standard mercury lamp that emits multiple known and stable characteristic wavelengths is used as the calibration light source to achieve accurate calibration of the snapshot hyperspectral imaging system, aiming to ensure that the system can accurately map the spectral information to the corresponding pixel position. The slit device is used to replace the traditional aperture to help reduce the diffusion of the light spot, making the spectral image clearer and facilitating subsequent calibration analysis.

[0047] Figure 4 The image shows the detection image captured by the system after being irradiated by a standard mercury lamp light source. Here, a row of pixels in the image is selected (marked with a dotted line), and the spectral distribution of the light source is extracted along this line to clearly identify the characteristic wavelength peaks. Combined with the pixel index differences corresponding to these peaks (i.e., the position difference of different wavelength peaks on the pixel row), a second-order polynomial model is used to fit the relationship between wavelength and pixel index.

[0048] The snapshot hyperspectral system was used to perform spectral imaging of Escherichia coli and Staphylococcus aureus with a 40x objective. A dataset containing 46 samples was created from 46 sets of real data (3361×3361) collected by the snapshot hyperspectral system and 46 sets of imaging spectral cube truth values ​​(541×541×26) collected by the LCTF module. This dataset was used to train the reconstruction of single-channel snapshot hyperspectral image data into a 26-channel atlas cube. The wavelength range is from 450nm to 700nm, and the spectral resolution is 10nm. The dataset was randomly divided into two parts: one part contains 40 samples as a training set, and the other part contains 6 samples as a test set.

[0049] The U-Net3 network structure is used for snapshot atlas cube reconstruction, which is significantly enhanced compared to the traditional U-Net architecture, thus significantly improving the accuracy of tasks such as image segmentation. Through the encoder-decoder structure, cross-layer connections, and optimized long-short connection design, efficient and accurate atlas cube pixel-level reconstruction of high-dimensional data such as hyperspectral images is achieved. Snapshot hyperspectral images of mixed bacteria are shown in the attached figure. Figure 5As shown, the exposure time is 10 ms, and the zero-order diffraction occupies the central position, achieving high spatial resolution imaging; while the adjacent area is the first-order diffraction spectral imaging, which is produced by superimposing the spectral cube of the detection target along eight different diffraction directions to reconstruct the spectral information, thereby realizing the acquisition of the spectral data of the video frame.

[0050] Figure 6 The zero-order snapshot hyperspectral image of mixed bacteria is shown, and the distribution of bacteria at the μm scale can be clearly observed, proving the image resolution at the single-cell level; as shown in the attached figure Figure 7 As shown in the figure, the snapshot hyperspectrum reconstructed in the background light source area and the corresponding spectrum restored by LCTF, it can be seen that the hyperspectrum reconstructed by the snapshot system is basically consistent with the real hyperspectral information obtained by LCTF; as shown in the attached figure, Figure 8 As shown in the figure, the reconstructed snapshot hyperspectral image can well distinguish Escherichia coli and Staphylococcus aureus, proving that the atlas cube acquired by the snapshot microscopy system can realize the recognition and classification of mixed bacterial samples at the cellular level. The snapshot hyperspectral system can greatly improve the speed of atlas cube acquisition and realize the acquisition of video frame atlas data, thereby realizing the recognition and tracking of dynamic bacteria.

[0051] Combined with machine vision technology, the bacteria in the hyperspectral snapshot images are tracked and their movement trajectories are recorded. By comparing the changes in bacterial movement trajectories and morphology under different drug treatments, the bacterial resistance to different antibiotics is evaluated.

[0052] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0053] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for rapidly detecting mixed bacterial resistance based on a snapshot hyperspectral system, characterized in that: The following steps are involved: a. Capture the spectral data of mixed bacterial samples using the snapshot hyperspectral microscopy module; b. Calibrate, process and reconstruct the atlas data into single pixels to obtain an atlas cube; c. Input the atlas cube into the bacteria recognition and classification algorithm for pixel-level recognition, thereby achieving single-cell level analysis in mixed bacterial samples; d. Use machine vision tracking modules to record the movement trajectory of target bacteria under antibiotic treatment; e. Analyze bacterial movement trajectories and morphological changes, and evaluate the resistance of target bacteria to different antibiotics; The snapshot hyperspectral microscopy imaging module in step a includes: a microscope objective unit, which is used to convert the information at the focus of the bacterial sample into infinite parallel light; a spectroscopic unit, which includes a filter, a beam splitter, a focusing lens, a square aperture, a collimating lens, a two-dimensional grating, and an imaging lens in sequence, and is used to spectrally split the infinite parallel light to form multiple spectral components; a detector unit, which is used to receive and record the light signal after spectral splitting; the wavelength of the light source selected by the snapshot hyperspectral microscopy imaging module covers the visible light to near-infrared region to cover the key spectral characteristics of the bacterial community; the spectroscopic unit uses a two-dimensional grating to generate a diffraction pattern: the center is zero-order diffraction, which locates the target spatial information; the periphery is high-order diffraction, which contains different projection angle data, reconstructs the target spectral information, and finally integrates it into the complete spatial and spectral data of the target; Step b: using the end-to-end UNET3 reconstruction algorithm to process the captured atlas data into single pixels and restore them into an atlas cube, thereby achieving single-cell level recognition; The bacteria recognition and classification algorithm in step c adopts a transformer-based deep learning model, and optimizes the model parameters through training to improve the recognition accuracy; In step d, the machine vision tracking module combines image segmentation and target tracking algorithms to track the bacteria in the hyperspectral snapshot image, record the movement trajectory, and evaluate the bacterial resistance to different antibiotics by comparing and analyzing the changes in the movement trajectory and morphology of bacteria under different drug treatments; The drug resistance assessment in step e includes using statistical analysis and pattern recognition methods to extract bacterial movement trajectories and morphological changes, learning the relationship between cell movement patterns and activity based on the long short-term memory network machine learning model, and realizing non-invasive real-time drug resistance assessment.

2. The method according to claim 1, characterized in that The mixed bacterial sample is pretreated or not, and comes from one or more of the following combinations, including directly or indirectly from the human body to be tested, animals, plants and isolates other than humans, and directly or indirectly from the environment to be tested.

3. The method according to claim 1, characterized in that: Step b uses a mercury lamp for calibration.

4. The method according to claim 1, characterized in that: The drug resistance assessment in step e involves training the model using the prepared dataset and adjusting the model parameters through an optimization algorithm to minimize the prediction error.

Images