KR20230173520A - Automatic bacteria identification method using cross-sectional images, system and recording medium for performing the method - Google Patents

Abstract

The present invention includes the steps of acquiring a tomography image of a non-destructive bacterium; Preprocessing the tomographic image; extracting feature vectors from the preprocessed image; and identifying the non-destructive bacteria based on the feature vector, wherein the feature vector is a morphological element of the non-destructible bacteria. As a result, this automatic bacterial identification technology allows identification at the same time as tomographic image acquisition without a general molecular biology-based identification process, so the results can be confirmed more quickly. Additionally, classification accuracy can be improved depending on the amount of data for each bacterial species learned in advance.

Description

Method for automatic identification of bacteria using tomography images and system and recording medium for performing the same {AUTOMATIC BACTERIA IDENTIFICATION METHOD USING CROSS-SECTIONAL IMAGES, SYSTEM AND RECORDING MEDIUM FOR PERFORMING THE METHOD}

The present invention relates to a method for automatic identification of bacteria using tomography images, and more specifically, tomography images that can improve classification accuracy according to the amount of data for each bacterial species learned quickly and in advance without a general molecular biology-based identification process. This relates to a method for automatic identification of bacteria using .

The DNA-based identification method used to identify existing bacteria had the disadvantage of requiring additional staining and time. To compensate for this, an automatic identification method based on artificial intelligence was proposed using images of the external shape of bacteria obtained through a microscope, but since only one data on the shape of the bacteria can be learned, additionally obtained DNA-based staining results are also learned.

This cannot solve the fundamental shortcomings of existing bacterial identification methods. Therefore, a new automatic bacterial identification technology is needed to minimize additional processes and time.

Republic of Korea Patent Publication No. 10-2246439

The present invention was created to solve the above problems, and the purpose of the present invention is to enable identification at the same time as tomographic image acquisition without a general molecular biology-based identification process, so that results can be confirmed more quickly and bacteria that have been learned in advance can be identified. The goal is to provide an automatic bacterial identification method using tomography images that can improve classification accuracy depending on the amount of data for each species.

To achieve the above object, a method for automatically identifying bacteria using tomographic images according to an embodiment of the present invention includes the steps of acquiring a tomographic image of non-destructive bacteria; Preprocessing the tomographic image; extracting feature vectors from the preprocessed image; and identifying the non-destructive bacteria based on the feature vector, wherein the feature vector is a morphological element of the non-destructible bacteria.

Additionally, the morphological elements may include the diameter, height, rise and fall angles, internal refractive index, and curvature of the non-destructible bacteria.

Additionally, in the step of extracting the feature vector, the feature vector may be extracted from an image preprocessed through a convolutional neural network (CNN).

It may also include providing the identification result probability using the extracted feature vector.

A computer-readable storage medium on which a computer program for realizing the other purposes of the present invention described above is recorded, and a computer program for performing an automatic bacterial identification method using tomography images is recorded thereon.

An automatic bacterial identification system using tomography images for realizing the other objects of the present invention described above includes a processor; Includes a memory connected to the processor, wherein the memory acquires a tomographic image of a non-destructive bacterium, preprocesses the tomographic image, extracts a feature vector from the pre-processed image, and extracts a feature vector from the pre-processed image, and the non-destructive bacterium based on the feature vector. wherein the feature vector stores program instructions executed by the processor to identify morphological elements of the non-destructible bacterium.

Additionally, the morphological elements may include the diameter, height, rise and fall angles, internal refractive index, and curvature of the non-destructible bacteria.

This automatic bacterial identification technology allows identification at the same time as tomographic image acquisition without a general molecular biology-based identification process, so the results can be confirmed more quickly.

Additionally, classification accuracy can be improved depending on the amount of data for each bacterial species learned in advance.

Figure 1 is a conceptual diagram of an automatic bacterial identification system using tomographic images according to the present invention.
Figure 2 is a flowchart of a method for automatic identification of bacteria using tomographic images according to the present invention.
Figure 3 is a flowchart of the image acquisition and image preprocessing steps in the automatic bacterial identification method using tomographic images according to the present invention.
Figure 4 is a flowchart of the step of determining characteristics for each bacterial species in the automatic bacterial identification method using tomographic images according to the present invention.
Figure 5 is a flowchart of the step of determining feature points for each bacterial species in the automatic bacterial identification method using tomographic images according to the present invention.
Figure 6 is a representative tomographic image for each bacterial species used in the automatic bacterial identification method according to the present invention.
Figure 7 is the result of the automatic bacterial identification method using tomographic images according to the present invention.
Figure 8 is another result of the automatic bacterial identification method using tomographic images according to the present invention.

The detailed description of the present invention described below refers to the accompanying drawings, which show by way of example specific embodiments in which the present invention may be practiced. These embodiments are described in sufficient detail to enable those skilled in the art to practice the invention. It should be understood that the various embodiments of the invention are different from one another but are not necessarily mutually exclusive. For example, specific shapes, structures and characteristics described herein may be implemented in one embodiment without departing from the spirit and scope of the invention. Additionally, it should be understood that the location or arrangement of individual components within each disclosed embodiment may be changed without departing from the spirit and scope of the present invention. Accordingly, the detailed description that follows is not intended to be taken in a limiting sense, and the scope of the present invention is limited only by the appended claims, together with all equivalents to what those claims assert, if properly described. Similar reference numbers in the drawings refer to identical or similar functions across various aspects.

Hereinafter, preferred embodiments of the present invention will be described in more detail with reference to the drawings.

Figure 1 is a conceptual diagram of an automatic bacterial identification system using tomographic images according to the present invention.

Referring to Figure 1, the automatic bacterial identification system using tomographic images according to the present invention includes a data input unit 100, a user DB 200, an image acquisition unit 300, an image preprocessor 400, and automatic identification model learning. It includes a unit 500, a feature point portion determination unit 600 for each type, and a judgment randomization unit 700.

Here, the automatic identification model learning unit 500 includes a learning model generating unit 510 and a model learning unit 530.

As will be described in more detail below, the data input unit 100 functions to input a tomographic image into the user DB 200. The data input unit 100 may be connected to an optical coherence tomography imaging device, which includes various devices for imaging tomography without causing additional damage to the sample.

In this way, the tomographic image obtained from the optical coherence tomography imaging device is stored in the user DB 200 from the data input unit 100. The user DB 200 is a storage that stores image data for training an AI model or input image data for determining characteristics of each species.

Afterwards, the image acquisition unit 300 receives and lists data from the DB 200. The image acquisition unit 300 reads image data from the user DB 200 to learn the AI model.

The image preprocessor 400 performs normalization, standardization, and transformation on image data. That is, the image preprocessor 400 reorganizes the image data to fit the input size of the AI model and performs image preprocessing tasks such as image normalization and noise removal.

The automatic identification model learning unit 500 and the species-specific feature point determination unit 600 perform deep learning to enable automatic identification through determination of feature points for each bacterial species.

The automatic identification model learning unit 500 receives transformed image data from the image preprocessor 400. Specifically, the automatic identification model learning unit 500 creates an AI model for determining characteristics of each bacterial species or loads a stored AI model and additionally learns new image data acquired through the image acquisition unit.

The learning model generator 510 extracts feature vectors from images preprocessed through a convolutional neural network (CNN), automatically identifies bacteria, and provides identification result probabilities.

Here, the feature vector is a morphological element that can be distinguished from the acquired bacterial tomography image, and includes image-based morphological features such as the diameter, height, and rise and fall angles of the bacteria, internal refractive index, and curvature calculation.

The learning model creation unit 510 creates a new AI model to be learned or loads a stored AI model depending on the existence and use of an existing stored AI model.

The model learning unit 530 performs a learning process on the learning model generated by the learning model generating unit 510. The model learning unit 530 sets a function for learning the model, extracts the feature vector of the image stored through CNN, reduces the error when an error occurs between the feature vector and the correct answer, and uses non-training data to improve the model. Judge learning progress. That is, the model learning unit 530 learns an AI model using images input through the user DB 200 and the image acquisition unit 300.

The species-specific feature point determination unit 600 loads the final trained model from the species-specific feature point model learning unit 500 for new input data to be evaluated and performs CNN calculation. Afterwards, the feature vector extracted through CNN from the new input data is used to determine bacterial identification. Then, the feature vector extracted in this way is calculated to calculate the probability of the identification result. In addition, it indicates in quantitative numbers whether the AI model learned through the input video data has progressed to the desired level.

The input image final decision randomization unit 700 displays the calculated identification result probability to the user. In other words, the basis for the judgment of the AI model, where the characteristic points of each species are determined through the AI model, is displayed on the image.

Such an automatic bacterial identification system using tomography images may include a processor connected to a memory, and this memory may include program instructions executed by the processor to implement the above-described functions.

Figure 2 is a flowchart of a method for automatic identification of bacteria using tomographic images according to the present invention.

Referring to Figure 2, the method for automatic identification of bacteria using tomographic images according to the present invention includes the steps of entering data into the user DB (S1), acquiring the image (S2), and preprocessing the image (S3). , It includes the step of determining the characteristics of each bacterial species (S4), the step of determining the feature point area of each bacterial species (S5), and the step of probabilizing the final decision of the input image (S6).

More specifically, the step S1 of inputting data into the user DB is a step in which the data input unit 100 inputs a tomography image into the user DB 200. The data input unit 100 connected to the optical coherence tomography imaging device stores the tomographic image acquired from the optical coherence tomography imaging device from the data input unit 100 to the user DB 200.

Thereafter, the step of acquiring the image (S2) and the step of pre-processing the image (S3) will be described with reference to FIG. 3.

Figure 3 is a flowchart of the image acquisition and image preprocessing steps in the automatic bacterial identification method using tomographic images according to the present invention.

Referring to FIG. 3, the step of acquiring an image (S2) includes a step of storing data (S21), a step of receiving an input data storage path (S22), and a step of listing the input data (S23).

More specifically, the data storage step (S21) stores data through a data collection server. In the step S22 of receiving the input data storage path, the image data is loaded into the computer main memory through a path designated by the user (for example, a folder location designated by the user) where the input data is stored. The step of listing input data (S23) lists the input data existing in the path. This is to learn the image data in the user DB 200 without duplication.

The image preprocessing step (S3) unifies the different extensions of the input image data, transforms or cuts the image to an image size suitable for learning the AI model, and additionally performs image processing such as noise removal and normalization.

Specifically, it includes the step of data normalizing the entire image (S31), standardizing the image size (S32), and transforming the image for learning (S33).

More specifically, the step of data normalization of the entire image (S31) normalizes the entire tomographic image data, and normalization is intended to keep the value range of the tomographic image data to be used during learning constant between 0 and 1. am. The way to make data into a range between 0 and 1 is (value to be normalized - minimum value among data values)/ (maximum value among data values - minimum value among data values).

Thereafter, the image size standardization step (S32) standardizes the input tomographic image size, and standardization is to change the range (scale) of the + value so that the mean is 0 and the variance is 1. The standardization method is (value to be standardized - average)/standard deviation. Afterwards, transformation for learning of the input image is performed (S33)

The step (S4) of determining the characteristics of each bacterial species will be described with reference to FIG. 4.

Figure 4 is a flowchart of the step of determining characteristics for each bacterial species in the automatic bacterial identification method using tomographic images according to the present invention.

More specifically, in the step of determining the characteristics of each bacterial species (S4), parameters for creating a learning model are first received (S42). Here, the parameters for creating a learning model are weights. Parameters include the input image size. Afterwards, it passes through CNN to automatically identify the final bacteria using the feature vector extracted from the input data (S42). Then, to determine the characteristics of each species, the extracted feature vector is calculated and the identification result probability is provided (S43).

In steps S41 to S43, an AI model is used to load an existing stored AI model or apply parameters input from the user, implement a CNN inside the AI model, and finally determine and output feature points from the input image data. This is the process of creating a .

Afterwards, the loss function and optimizer function for training the model are set and initialized (S44). The activation function used in CNN may be an activation function that outputs positive values among preprocessed images as is and outputs negative values as reduced values by a predetermined size.

Here, the activation function refers to a function that assigns weights to multiple input information, combines them, and outputs a complete result.

The loss function refers to the difference between the value output by the artificial neural network and the actual correct answer. Therefore, the smaller the loss function, the better. Through deep learning, the output value of the artificial neural network is compared with the actual value to find a combination of weight (W) and bias that minimizes the difference. Deep learning learning has been defined as the process of finding the weights and biases of an artificial neural network that minimizes the loss function. The algorithm that updates weights in the direction of minimizing the loss function is called an optimizer.

Afterwards, feature vectors of the stored input data are extracted through the generated CNN model (S45). Then, learning progresses in the direction of reducing the error of the feature vector by comparing the feature vector value with the correct answer entered by the user (S46).

Afterwards, after a certain number of training sessions, the learning progress of the model is determined using training data that was not used for training (S47).

Steps S44 to S47 are a process of learning the generated AI model by setting a loss function and an optimization function. The input image data passes through the CNN inside the AI model to determine the feature point area and output it to classify it according to the type specified by the user. Learning is conducted so that the species classifications determined by the and AI models are ultimately the same.

Using the learning progress determined in this way, the process proceeds to the feature point determination step (S5) for each bacterial species.

Figure 5 is a flowchart of the step of determining feature points for each bacterial species in the automatic bacterial identification method using tomographic images according to the present invention.

Referring to Figure 5, in the step (S5) of determining the feature point area for each bacterial species, the AI model that has completed learning is loaded, new image data to be tested is input, and the extracted feature vectors are calculated as the model operation progresses to reach the target value. This is the process of checking with quantitative numbers whether the learning of the AI model has been completed, and in the case of the AI model whose learning has been completed, using it in an actual situation to make a final judgment on the input feature points for each species.

In the step of determining the feature point area for each bacterial species (S5), a deep neural network operation is performed on the new input data to be evaluated using the final learned model (S51).

Afterwards, the bacterial identification result is determined using the feature vector extracted from the input data through CNN (S52). Then, the feature vector extracted for bacterial species classification is calculated and the identification result probability is provided (S53).

And finally, the final judgment of the input image is probabilized based on the probability of the identification result (S6).

Figure 6 is a representative tomographic image for each bacterial species used in the automatic bacterial identification method according to the present invention.

Referring to Figure 6, this is a tomographic image obtained using an optical coherence tomography device for a total of 16 bacterial single colonies. It can be confirmed with the naked eye that the morphological characteristics of the monolayer are different for each bacterial species. In this drawing, 16 tomography images are used as an example, but it is not limited to this.

In addition, all tomographic images of bacteria obtained using various tomographic imaging devices can be used, and based on this, the results of automatic bacterial identification can be checked after database accumulation and learning.

Figure 7 is the result of the automatic bacterial identification method using tomographic images according to the present invention. It shows that a total of 7 types of bacteria can be automatically identified at once with a probability of 93%. In this drawing, seven types of bacteria are used as examples, but it is not limited to these.

Figure 8 is another result of the automatic bacterial identification method using tomographic images according to the present invention.

Referring to Figure 8, this is the result of an additional deep learning model that indicates which region of the bacterial tomography image the deep learning model extracted features based on. The closer it is to red, the more it has characteristics that distinguish it from other species. The fact that red areas are concentrated in bacterial areas proves that it is possible to automatically identify bacteria using tomographic images based on deep learning.

Although various embodiments of the present invention have been shown and described above, the present invention is not limited to the specific embodiments described above, and may be used in the technical field to which the invention pertains without departing from the gist of the invention as claimed in the claims. Of course, various modifications can be made by those skilled in the art, and these modifications should not be understood individually from the technical idea or perspective of the present invention.

100: data input unit
200: User DB
300: Image acquisition unit
400: Image preprocessing unit
500: Automatic identification model learning unit
600: Feature point judgment unit for each type
700: Decision probability unit

Claims (7)

Obtaining a tomography image of non-destructive bacteria;
Preprocessing the tomographic image;
extracting feature vectors from the preprocessed image; and
Including: identifying the non-destructive bacteria based on the feature vector,
A method for automatically identifying bacteria using tomography images, wherein the feature vector is a morphological element of the non-destructive bacteria.
According to paragraph 1,
The morphological elements include the diameter, height, elevation and descent angles, internal refractive index, and curvature of the non-destructive bacteria. Method for automatically identifying bacteria using tomography images.
According to paragraph 1,
The step of extracting the feature vector is a method of automatically identifying bacteria using a tomography image, in which the feature vector is extracted from an image preprocessed through a convolutional neural network (CNN).

According to paragraph 1,
A method for automatically identifying bacteria using a tomography image, comprising providing the probability of the identification result using the extracted feature vector.
A computer-readable storage medium recording a computer program for performing the method for automatically identifying bacteria using the tomography image according to any one of claims 1 to 4.
An automatic bacterial identification system using tomographic images,
processor;
Including a memory connected to the processor,
The memory is,
Acquire tomography images of non-destructive bacteria,
Preprocess the tomographic image,
Extracting feature vectors from the preprocessed image,
An automatic bacterial identification system using tomography images, wherein the non-destructive bacteria are identified based on the feature vector, and program instructions executed by the processor are stored so that the feature vector becomes a morphological element of the non-destructible bacteria.
According to clause 6,
The morphological elements include the diameter, height, rise and fall angles, internal refractive index and curvature of the non-destructive bacteria. An automatic bacterial identification system using tomography images.

Images