Method for determining minimum inhibitory concentration based on coaxial holographic images

Abstract

The present invention discloses a method for determining the minimum inhibitory concentration based on a coaxial holographic image. First, the coaxial holographic diffraction image data of pathogens after being incubated for the same time at different antibiotic concentrations is collected, and the angular spectrum reconstruction algorithm is combined to realize the recovery of pathogens in the diffraction image and then establish a sample data set; secondly, the pathogen detection model is trained by a deep learning target detection algorithm, and the target detection of pathogens under different concentrations of antibiotics is realized, and the number of pathogens under different concentrations of antibiotics is counted again by a feature point counting method; finally, the minimum antibiotic concentration value that inhibits the growth of pathogens is determined according to a threshold. The advantage of the present invention is that it combines the advantages of lensless digital holographic imaging technology, such as large field of view, no need for labeling, and low cost, with a deep learning method, and proposes a method that combines a coaxial digital holographic image with a deep learning algorithm and realizes the counting of pathogens, which can automatically detect the minimum inhibitory concentration and solve the limitations of professional operation.

Description

Technical Field

[0001] The invention relates to the field of antimicrobial drug sensitivity test, and in particular to a method for determining minimum inhibitory concentration based on coaxial holographic images. Background Art

[0002] Antimicrobial drugs play an important role in ensuring the health of humans and livestock, reducing pathogen infection, and ensuring the safety of animal food. Before using antimicrobial drugs, drug sensitivity tests need to be performed to determine the drug sensitivity so that the drugs can be used for treatment accurately and effectively. At present, the main methods for drug sensitivity testing in clinical microbiology laboratories include paper diffusion method, dilution method (including agar and broth dilution method), antibiotic concentration gradient method (E-test method), paper diffusion method, etc. These methods require specialized technicians to operate, the test process is relatively cumbersome, and it is not suitable for large-scale promotion and use.

[0003] With the vigorous development of artificial intelligence technology and the maturity of lensless digital holographic imaging technology, deep learning based on artificial intelligence has also been well applied in the field of biology. For example, scholar DaeHan Ahn realized human-level blood cell counting based on lensless digital holographic diffraction images based on deep neural networks, and scholar Gao Junhui used machine learning models to predict minimum inhibitory concentrations based on gene sequence data.

[0004] This method collects holographic images of drug-resistant bacteria after incubation for the same time at different antibiotic concentrations, reproduces the images through a digital image restoration algorithm, and then combines deep learning methods to monitor the growth status of drug-resistant bacteria. Finally, the MIC (minimum inhibitory concentration) value is determined based on the threshold. Summary of the invention

[0005] The present invention aims to provide a method for determining the minimum inhibitory concentration based on a coaxial holographic image, so as to realize the detection of the minimum inhibitory concentration.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] The method for determining the minimum inhibitory concentration based on the coaxial holographic image of the present invention comprises the following steps:

[0008] S1, collects the on-axis holographic diffraction image data of drug-resistant bacteria after incubation at different antibiotic concentrations for the same time;

[0009] S2, obtaining the reconstruction image data of the coaxial holographic diffraction image;

[0010] S3, removing the conjugate image and the zero-order image in the reconstructed image to establish a sample data set;

[0011] S4, target labeling of sample data sets;

[0012] S5, iteratively train, verify, and test sample data sets through computer target detection deep learning algorithms to obtain a pathogen detection model;

[0013] S6, obtaining coaxial holographic diffraction image data of the test bacterial suspension with the same concentration after being incubated for the same time under different concentrations of antibiotics;

[0014] S7, using the methods of S2 and S3 to obtain the reconstructed image data of the coaxial holographic diffraction image data of the bacterial suspension to be tested;

[0015] S8, using the pathogen detection model to detect the pathogen to be tested;

[0016] S9, counting the number of pathogens at each antibiotic concentration by the characteristic point counting method;

[0017] S10, finding the minimum antibiotic concentration value that inhibits bacterial growth according to the threshold.

[0018] Furthermore, the coaxial holographic diffraction image data in step S1 is collected using a coaxial lens-free digital holographic imaging device.

[0019] Furthermore, the reproduced image data in step S2 is realized by an angular spectrum reconstruction algorithm.

[0020] Furthermore, in step S3, a four-step phase shift technique is used to remove the conjugate image and the zero-order image in the reconstructed image.

[0021] Furthermore, the computer target detection deep learning algorithm described in step S5 includes any deep learning algorithm that can achieve target detection.

[0022] Furthermore, the feature point counting method in step S9 is specifically as follows: for each target detected by the pathogen detection model, count the pixels in each target whose grayscale value is greater than the average grayscale value of the target and whose grayscale value is locally the largest, and then eliminate the pixels whose length is less than a fixed length k from the pixels and then determine the position of the single pathogen, thereby determining the number of pathogens.

[0023] Furthermore, the fixed length k is the average size of the initial state of the pathogen to be detected in the reproduction graph.

[0024] Furthermore, the threshold value in step S10 refers to the critical value of pathogens when the number of pathogens under the action of antibiotics changes significantly compared with the control bacterial suspension without adding antibiotics.

[0025] The advantage of the present invention is that it combines the advantages of lens-free digital holographic imaging technology such as large field of view, no need for marking, and low cost with computer deep learning methods, and proposes a method for counting bacteria according to the local feature point method, which can automatically detect the minimum inhibitory concentration, solves the limitations of professional operators, and can realize batch program and intelligent operation, improve the work efficiency of doctors, and reduce labor costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 is a flow chart of the method of the present invention.

[0027] Figure 2 It is a schematic diagram of a grayscale holographic image of the method of the present invention.

[0028] Figure 3 It is a schematic diagram of the reproduction graph data of the method described in the present invention.

[0029] Figure 4 It is a schematic diagram of the number of pathogens in the method of the present invention.

[0030] Figure 5 It is a schematic diagram of the 4x inverted microscope image of the present invention. DETAILED DESCRIPTION

[0031] The technical solutions in the embodiments of the present invention are described clearly and completely below. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] like Figure 1 As shown, the method for determining the minimum inhibitory concentration based on the coaxial holographic image of the present invention comprises the following steps:

[0033] S1, using a coaxial lensless digital holographic imaging device to collect coaxial holographic diffraction image data of drug-resistant bacteria after incubation at different antibiotic concentrations for the same time;

[0034] Coaxial lensless digital holographic imaging technology is an interference imaging method that can achieve label-free quantitative phase imaging of transparent objects such as biological tissues and cells without the need for operations such as staining the cells. Therefore, it can observe the life activities of cells for a long time, and has the advantages of a large field of view, high resolution, no labeling, low cost, good portability, and the ability to achieve three-dimensional (3D) imaging. The coaxial lensless digital holographic imaging device can expand the boundaries of traditional microscopic imaging technology and form a new type of fast and portable on-site inspection tool. In this embodiment, the coaxial lensless digital holographic imaging device uses a red light source, and the CMOS detector is a black and white camera, which can directly obtain grayscale holographic images of drug-resistant bacteria after incubation for the same time at different antibiotic concentrations, such as Figure 2 shown.

[0035] S2, obtaining the reconstruction image data of the coaxial holographic diffraction image through an angular spectrum reconstruction algorithm, such as Figure 3 As shown;

[0036] There are many reconstruction algorithms for digital holographic images, which can be roughly divided into Fresnel diffraction integral reconstruction algorithm, convolution reconstruction algorithm, and angular spectrum reconstruction algorithm. The angular spectrum reconstruction algorithm used in the present invention is based on the Helmtz equation of scalar diffraction and is an accurate representation of the diffraction of light in the frequency domain based on the angular spectrum theory. Its expression is formula (1):

[0037] (1)

[0038] in, is the wavelength of light, is the distance from the sample to the photoelectric sensor, , for and The Fourier transform of is the complex amplitude distribution of the phenomenon, is the complex amplitude distribution of the holographic image, , is the spatial frequency, which is the number of changes in the complex amplitude of the light field per unit spatial distance. Therefore, the complex amplitude distribution of the reconstructed object light field is given by formula (2), where is the inverse Fourier transform.

[0039] (2)

[0040] S3, using a four-step phase shifting technique to remove the conjugate image and the zero-order image in the reconstructed image, and establishing a sample data set;

[0041] S4, a large number of reproduction images in the sample data set are labeled with objects and randomly divided into training set, validation set, and test set in a ratio of 7:2:1.

[0042] S5, iteratively train, verify, and test sample data sets through computer target detection deep learning algorithms to obtain a pathogen detection model; the present invention can adopt any deep learning algorithm that can achieve target detection, such as the YOLOv5 target detection algorithm, and use the Adam optimizer with an initial learning rate of 0.001 to train the training set for 300 rounds to obtain the pathogen detection model, so that the pathogen detection model achieves 0.96map@0.5 on the test set.

[0043] S6, obtaining coaxial holographic diffraction image data of the test bacterial suspension of the same concentration after being incubated for the same time under different concentrations of antibiotics; adding antibiotics of different concentration gradients to the bacterial suspension of the same concentration and incubating for several hours, and then using a coaxial lensless digital holographic imaging device to collect coaxial holographic diffraction image data.

[0044] S7, adopting the methods of S2 and S3, obtains the reconstruction image data of the coaxial holographic diffraction image data of the bacterial suspension to be tested through the angular spectrum reconstruction algorithm; the four-step phase shift technology removes the conjugate image and the zero-order image in the reconstruction image.

[0045] S8, using the pathogen detection model to detect the pathogen to be tested;

[0046] S9, counting the number of pathogens at each antibiotic concentration by the characteristic point counting method;

[0047] Since each type of bacteria has a specific size range, such as the size of Klebsiella , E. coli is about The present application uses the fixed size characteristic of pathogens to calculate the number of adherent bacteria.

[0048] For each target detected by the pathogen detection model, a statistical window is set with the size of the pathogen to be detected in the reproduction image as a fixed length K. The pixel points whose grayscale value in the current statistical window is greater than the average grayscale value of the current statistical window and whose grayscale value is the maximum value in the statistical window area are counted by sliding and translating. Then, the pixel points whose Euclidean distance between the maximum value pixel points is less than the fixed length K are eliminated. The pixel points finally selected are used as the position of a single bacterium, and then the number of pathogens is counted (such as Figure 4 The bacterial count method of the present invention was applied to the test set, and the statistical results were consistent with the 4x inverted microscope image (as shown in Figure 5 The consistency of human statistics results is as high as 0.92.

[0049] S10, find the minimum antibiotic concentration value that inhibits bacterial growth according to the threshold value. The threshold value refers to the critical value of pathogens at which the number of pathogens under the action of antibiotics changes significantly compared with the test bacterial suspension. In actual operation, the coaxial holographic diffraction image data of the test bacterial suspension of the same concentration is incubated for the same time under different concentrations of antibiotics for many times; the reconstructed image data of the coaxial holographic diffraction image data of the test bacterial suspension is obtained by the angular spectrum reconstruction algorithm; the conjugate image and the zero-order image in the reconstructed image are removed by the four-step phase shift technology, and the pathogen detection model is used to detect the test pathogen; the number of pathogens under each concentration of antibiotics is counted by the feature point counting method. Compare the number of pathogens under each concentration of antibiotics obtained each time, find the critical value of pathogens when there is a significant change compared with the control bacterial suspension without added antibiotics, and the corresponding antibiotic concentration is determined as the minimum antibiotic concentration value that inhibits bacterial growth.

Claims

1. A method for determining minimum inhibitory concentration based on coaxial holographic images, characterized in that: The following steps are involved: S1, collects the on-axis holographic diffraction image data of drug-resistant bacteria after incubation at different antibiotic concentrations for the same time; S2, obtaining the reconstruction image data of the coaxial holographic diffraction image by an angular spectrum reconstruction algorithm; S3, removing the conjugate image and the zero-order image in the reconstructed image to establish a sample data set; S4, target labeling of sample data sets; S5, iteratively train, verify, and test sample data sets through computer target detection deep learning algorithms to obtain a pathogen detection model; S6, obtaining on-axis holographic diffraction image data of the test bacterial suspension with the same concentration after being incubated with different concentrations of antibiotics for the same time; S7, using the methods of S2 and S3 to obtain the reconstructed image data of the coaxial holographic diffraction image data of the bacterial suspension to be tested; S8, using the pathogen detection model to detect the pathogen to be tested; S9, counting the number of pathogens under each concentration of antibiotics by a feature point counting method; specifically, for each target detected by the pathogen detection model, counting the pixel points in each target whose grayscale value is greater than the average grayscale value of the target and whose grayscale value is locally the largest, and then removing the pixel points whose length is less than a fixed length k from the pixel points, and then determining the position of the single pathogen; the fixed length k is the average size of the initial state of the pathogen to be detected in the reproduction image; S10, finding the minimum antibiotic concentration value that inhibits bacterial growth according to the threshold.

2. The method for determining minimum inhibitory concentration based on coaxial holographic images according to claim 1, characterized in that: The coaxial holographic diffraction image data described in step S1 is collected using a coaxial lensless digital holographic imaging device.

3. The method for determining minimum inhibitory concentration based on coaxial holographic images according to claim 1, characterized in that: In step S3, a four-step phase shift technique is used to remove the conjugate image and the zero-order image in the reconstructed image.

4. The method for determining minimum inhibitory concentration based on coaxial holographic images according to claim 1, characterized in that: The computer target detection deep learning algorithm described in step S5 includes any deep learning algorithm that can achieve target detection.

5. The method for determining minimum inhibitory concentration based on coaxial holographic images according to claim 1, characterized in that: The threshold value in step S10 refers to the critical value of pathogens when the number of pathogens under the action of antibiotics changes significantly compared with the control bacterial suspension without adding antibiotics.

Images