KR102258181B1 - Non-discrete spectral analysis algorithms and methods for in vivo detection of tissue malignancy based on laser spectroscopy - Google Patents

Abstract

According to an embodiment of the present invention, a non-transitory computer-readable storage medium storing one or more programs for non-discrete spectrum analysis for laser spectroscopy-based lesion tissue detection is disclosed.
The one or more programs allow an electronic device having one or more processors to execute a method of detecting lesion tissue,
The lesion tissue detection method
Including; determining whether the lesion tissue is present by applying a learning model for detecting lesion tissue to the non-discrete spectrum measurement result,
The measurement result of the non-discrete spectrum is a result of measuring a spectrum of all generated light from the moment when the laser is irradiated to the sample and generated light is generated until the generated light is no longer generated.

Description

Non-discrete spectral analysis algorithms and methods for in vivo detection of tissue malignancy based on laser spectroscopy}

The present invention relates to a non-discrete spectrum analysis algorithm for detecting lesion tissue in vivo based on laser spectroscopy and a method therefor.

Conventionally, a technology for diagnosing a disease by analyzing a spectrum of light generated by irradiating a laser on an animal or human tissue has been disclosed. For example, US Patent No. 7,092,087 (2006. 8. 15) (hereinafter, '087' patent) discloses a technical concept for diagnosing a disease in an animal.

However, the '087' patent described above is a technology that detects lesion tissue based on a threshold value for specific components in a wavelength region, and it is not easy to ensure accuracy.

According to an embodiment of the present invention, a non-discrete spectrum analysis algorithm for detecting in vivo lesion tissue based on laser spectroscopy and a non-transitory computer-readable recording medium storing the algorithm are provided.

According to an embodiment of the present invention, a method for detecting lesion tissue in vivo based on laser spectroscopy and an apparatus or devices used in the method are provided.

According to an embodiment of the present invention, in a non-transitory computer-readable storage medium storing one or more programs for non-discrete spectrum analysis for detecting lesion tissue based on laser spectroscopy,

The one or more programs allow an electronic device having one or more processors to execute a method of detecting lesion tissue,

The lesion tissue detection method

Including; determining whether the lesion tissue is present by applying a learning model for detecting lesion tissue to the non-discrete spectrum measurement result,

The non-discrete spectrum measurement result is a result of measuring the spectrum of all generated light generated from the moment when the laser is irradiated to the sample and the generated light is generated until the generated light is no longer generated,

A non-transitory computer-readable storage medium storing one or more programs for non-discrete spectrum analysis for laser spectroscopy-based lesion tissue detection is provided.

According to one or more embodiments of the present invention, it is possible to more accurately detect a lesion through a method for detecting lesion tissue based on non-discrete spectrum measurement and machine learning.

In addition, the present embodiments may support both in-vivo lesion detection as well as ex-vivo lesion detection.

1 to 3 are diagrams for explaining a method of detecting lesion tissue based on machine learning according to an embodiment of the present invention.
4 is a view for explaining a laser spectroscopy-based independent device for detecting lesion tissue according to an embodiment of the present invention.
5 is a diagram for explaining measurement of a non-discrete spectrum according to an embodiment of the present invention.
FIG. 6 shows an example of a laser spectroscopy-based standalone device for in-vivo lesion tissue detection described with reference to FIG. 4.
7 to 9 are diagrams for explaining non-discrete spectrum measurement.

The above objects, other objects, features, and advantages of the present invention will be easily understood through the following preferred embodiments related to the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided so that the disclosed content may be thorough and complete, and the spirit of the present invention may be sufficiently conveyed to those skilled in the art.

In this specification, terms such as "... unit", "... device", and "module" mean a unit that processes at least one function or operation, which may be implemented by hardware or software or a combination of hardware and software. .

The terms used in the present specification are for describing exemplary embodiments and are not intended to limit the present invention. In this specification, the singular form also includes the plural form unless specifically stated in the phrase. As used in the specification, "comprises" and/or "comprising" does not exclude the presence or addition of one or more other components.

Definition of Terms

In the present specification, the term'program' or'algorithm' means'a set of instructions suitable for processing by a computer', and'program' and'algorithm' are used as the same meaning.

In the present specification, the expression “a program (or algorithm) performs (or executes) a certain operation (or step)” means, “a program (or algorithm) performs a certain operation (or step) by an electronic device equipped with a processor. To do or to make it happen.”

In the present specification, the term'laser' means a pulsed laser or a continuous light laser. In addition, the frequency band of the'laser' may have an arbitrary frequency band, and may have, for example, an ultra violet (UV) band, a visible light band, or an infra-red (IR) band.

In the present specification, the term'generated light' is meant to include all lights generated when a laser is irradiated onto the sample T. Accordingly,'generated light' may mean, for example, plasma light, reflected light, scattered light, and/or fluorescent light.

In the present specification, the term'sample' refers to a biological tissue, and may be, for example, a human tissue or an animal tissue. In the present specification, the term'measurement data' means'spectral data measured by a spectroscope for N0N-GATED light generated when a sample is irradiated with a laser', and'Non-discrete spectrum measurement. )'.

In the specification of the present application, all generated light measured over time from the moment when the laser is irradiated to the surface of the sample until the generated light is no longer generated is referred to as'N0N-GATED generated light'. The data measured by is referred to as non-discrete spectrum measurement data or non-discrete spectrum data. On the other hand, non-discrete spectrum measurement data or non-discrete spectrum data is a concept including Filtered Non-discrete spectrum data.

In the present specification, the term'Non-discrete spectrum measurement' refers to'all generated light measured over time from the moment when the laser is irradiated to the surface of the sample until the generated light is no longer generated, that is, It means to measure the spectrum of the light generated by non-gated light. That is, the result of'Non-discrete spectrum measurement' is not discrete in the wavelength band, that is, it has a continuous value. On the other hand,'Non-discrete spectrum measurement' is a concept including'Filtered Non-discrete spectrum Measurement'.

In the present specification, the term'Filtered Non-discrete spectrum Measurement' refers to measuring a spectrum of some light from the'NON-GATED generated light, or light belonging to a specific wavelength band in the NON-GATED generated light. It means to measure the spectrum for'.

In the present specification, the term'Filtered Non-discrete spectrum' means data obtained as a result of'Filtered Non-discrete spectrum Measurement'.

In the present specification, the term'parameter of the feature extractor' is used to refer to parameters constituting the feature extractor (eg, weights and main components of a feature function).

A method of detecting lesion tissue based on machine learning according to an embodiment of the present invention includes a pre-processing step (hereinafter, also referred to as a'first step') and a determining step (hereinafter, referred to as a'second step').

The first step is to normalize the non-discrete spectrum measurement (NSM) result of the NON-GATED generated light, the standardization step, and the normalized and standardized measurement results. It includes a Principle Component Analysis (PCA) step that extracts features of the main component.

Here, the steps of normalization and standardization are steps of removing deviation and noise between measurement results. For example, the steps of normalization and standardization may include an area-normalization and interpolation operation after removing background noise from a measurement result.

The method for detecting lesion tissue based on machine learning according to an embodiment of the present invention may further include a non-discrete spectrum measurement (NSM) step for non-GATED generated light. Here, in the non-discrete spectrum measurement (NSM) step, all generated light measured over time from the moment when the laser is irradiated to the sample surface until the generated light is no longer generated, that is, NON- This is the step of measuring the spectrum of the GATED generated light. According to an embodiment, in the non-discrete spectrum measurement (NSM) step, the spectrum is measured for some light from the light generated by NON-GATED, or the light belonging to a specific wavelength band from the light generated by NON-GATED. It may be a step of measuring a Filtered Non-discrete spectrum measuring a spectrum for the sample.

In the embodiments of the present invention, the'non-discrete spectrum measurement result' means'non-discrete spectrum measurement' data as it is, or with respect to such data. It refers to data after normalization and standardization.

A method for detecting lesion tissue based on machine learning according to an embodiment of the present invention includes: irradiating a laser of a specific band onto a sample surface; And a non-discrete spectrum measurement (NSM) step for the NON-GATED generated light. For example, the wavelength of the laser irradiated to the surface of the sample may be 1064 nm. In addition, in the non-discrete spectrum measurement (NSM) step, the spectrum is measured for some light from the light generated by NON-GATED, or the spectrum is measured for light belonging to a specific wavelength band from the light generated by NON-GATED. It may be a step of measuring a filtered non-discrete spectrum to be measured. Here, the wavelength of the generated light for which the spectrum is measured may have a band of 200 nm to 1000 nm, for example.

The principle component analysis (PCA) step is a step of extracting features for the main component from the results of non-discrete spectrum measurements. According to one embodiment, there are a plurality of main components (or, also referred to as'multidimensional'), and the main component analysis step extracts features (or also referred to as'feature values') for each of the plurality of such main components. Including the actions to be taken.

According to an embodiment, in the main component analysis step, each feature may be extracted for 16-dimensional main components. Here, since the 16th dimension is an exemplary numerical value, the present invention is not limited to such a numerical value, and the principal component analysis step may extract features for principal components of more dimensions than the 16th dimension.

In the embodiments of the present invention, for the purpose of description of the present invention,'extracting features for each of a plurality of main components' will be referred to as'multidimensional main component analysis'.

The first step described above is an electronic device including a memory (not shown), one or more processors (not shown), and one or more programs (not shown). ). Here, one or more programs (hereinafter, “pre-processing programs”) are stored in the memory and configured to be executed by the one or more processors. Here, the pre-processing programs include instructions for performing the above-described normalization and standardization steps and a principle component analysis (PCA) step.

According to an embodiment, the pre-processing programs may include a normalization and standardization program and a multidimensional principal component analysis program. Each of these programs includes instructions for performing a second operation.

According to an embodiment, the memory for the first step (not shown), one or more processors (not shown), and pre-processing programs (not shown), with reference to FIG. It may be located in the analysis device 10 to be described later.

According to another example, some of the memory for the first step (not shown), one or more processors (not shown), and pre-processing programs (not shown) are shown in FIG. 4. It is located in the analysis device 10 to be described later, the rest may be located in the handpiece 20.

According to another example, some of the memory for the first step (not shown), one or more processors (not shown), and pre-processing programs (not shown), see FIG. 4. Thus, it is located in the analysis device 10 to be described later, and some of the remainder may be located in an electronic device connected to the analysis device 10 and a communication network (a network capable of transmitting or receiving data, for example, the Internet).

In the second step, the machine learning algorithm determines whether or not there is a lesion tissue in the sample by applying a classifier, which is a learning model for detecting lesion tissue, to the result of analyzing the multidimensional principal component in the first step.

The learning model for lesion tissue detection is a classifier generated by training (or learning) by labeled non-discrete spectral measurement data. According to an embodiment, the non-discrete spectrum measurement data may be Filtered Non-discrete spectrum data.

According to one embodiment, the machine learning algorithm is a deep learning algorithm configured to include an input layer, at least one hidden layer, and an output layer. Here, the hidden layer may reflect functions and coefficients constituting a learning model for detecting lesion tissue.

The input layer receives the result of the pre-processing step, the hidden layer applies a learning model for lesion tissue detection on the data received by the input layer, and the output layer outputs the result from the hidden layer. The output value of the output layer may be, for example, a value probabilistically indicating the presence or absence of a lesional tissue.

Machine learning algorithms that can be used in embodiments of the present invention may be, for example, Logistic regression, SVM (SVM = Space Vector Machine), Radom forest, DNN (Deep Neural Network), or CNN (Convolution Neural Network). have. On the other hand, those who are engaged in the technical field to which the present invention pertains (hereinafter, referred to as'person in charge') will readily understand that the present invention is not limited only to the above-described algorithms.

The second step described above is an electronic device including a memory (not shown), one or more processors (not shown), and one or more programs (not shown). ). Here, one or more programs (hereinafter, “machine learning programs”) are stored in the memory and configured to be executed by the one or more processors.

According to an embodiment, the machine learning program includes instructions for performing the step of determining the presence or absence of the lesion tissue described above. The machine learning program further includes instructions for performing a learning step for generating a learning model for lesion tissue detection.

According to an embodiment, the memory (not shown), one or more processors (not shown), and a machine learning program (not shown) for the second step will be described later with reference to FIG. 4. It may be located in the analysis device 10.

According to another example, some of the memory (not shown), one or more processors (not shown), and machine learning programs (not shown) for the second step will be described later with reference to FIG. 4. It is located in the analysis device 10 to be performed, and the remaining part may be located in the handpiece 20.

According to another example, some of the memory (not shown), one or more processors (not shown), and machine learning programs (not shown) for the second step will be described later with reference to FIG. 4. It is located in the analysis device 10 to be performed, and the remaining part may be located in an electronic device connected to the analysis device and a communication network (a network capable of transmitting or receiving data, for example, the Internet).

A method for detecting lesion tissue based on machine learning according to an embodiment of the present invention includes determining a parameter of a feature extractor from labeled non-discrete spectral measurement data and a classifier that is a learning model for detecting lesion tissue. It may further include the step of defining by learning. Here, the non-discrete spectrum measurement data may be, for example, filtered non-discrete spectrum data.

In the step of determining the parameter of the feature extractor described above, the feature extractor receives the labeled non-discrete spectrum measurement data and determines the parameter of the feature extractor.

The step of defining the above-described classifier by learning is a step in which a machine learning program is trained by labeled non-discrete spectrum measurement data.

Here, a classifier generated by learning from labeled non-discrete spectral measurement data may perform a step of determining whether there is a lesion tissue from unknown non-discrete spectral measurement data.

1 to 3 are diagrams for explaining a method of detecting lesion tissue based on machine learning according to an embodiment of the present invention.

The machine learning-based lesion tissue detection method described with reference to these drawings is an embodiment using a deep learning algorithm, and the numerical values and functions mentioned below are exemplary, and the scope of the present invention is limited only to those numerical values or functions. Those skilled in the art should know.

Referring to FIG. 1, the feature extractor 200 is a program for extracting features for main components from non-discrete spectrum measurement data ('multidimensional main component analysis program'), and the learning model 400 for detecting lesion tissue is It is a classifier to determine whether there is a lesion. Here, the non-discrete spectrum measurement data may be, for example, filtered non-discrete spectrum data.

According to an embodiment, the feature extractor 200 extracts each feature from a plurality of main components.

In the present embodiment, for the purpose of explanation, as shown in FIG. 1, it will be described as extracting each feature of the five main components.

That is, it is assumed that the feature extractor 200 receives the non-discrete spectrum measurement data as shown in FIG. 1A, and extracts features for each of the five main components from the measurement data.

In this embodiment, the feature extractor 200 determines the signal strengths (y1, y2, y3, y4, y5, ...) of the components (λ1, λ2, λ3, λ4, λ5, ...) in the wavelength band. Calculate the values of the input functions (f1, f2, f3, f4, f5).

In this embodiment, the functions f1, f2, f3, f4, and f5 calculated by the feature extractor 200 may have different coefficients (weights) for inputs.

For example, f1(y1, y2, y3, y4, y5, …) = (a1*y1) + (b1*y2) + (c1*y3) + (d1 * y4) + (e1*y5) +… F2(y1, y2, y3, y4, y5, …) = (a2*y1) + (b2*y2) + (c2*y3) + (d2 * y4) + (e2*y5) +… Assuming that it is defined as, a1 and a2 may be different, b1 and b2 may be different, c1 and c2 may be different, d1 and d2 may be different, and/or e1 and e2 may be different.

Referring to FIG. 1, the feature extractor 200 outputs features calculated by the feature extraction function.

In this embodiment, assuming that the feature extractor 200 has five (i.e., five-dimensional) feature extraction functions, the five feature extraction functions F1, F2, F3, F4, and F5 are each function. A value is calculated and output, and the outputs are input to the deep learning algorithm 300.

The deep learning algorithm 300 determines whether or not there is a lesion tissue by applying the learning model 400 for detecting lesion tissue to the received features.

2 is a diagram illustrating a deep learning algorithm 300.

According to an embodiment, the deep learning algorithm 300 receives features from the feature extractor 200, applies the learning model 400 for detecting lesion tissue to the features, and determines whether there is a lesion tissue.

In this embodiment, for illustrative purposes, it is configured to include an input layer 302, a hidden layer 304 (first hidden layer 303, second hidden layer 305), and an output layer 306. The deep learning algorithm 300 will be described as an example.

The input layer 302, the hidden layer 304, and the output layer 306 are each composed of one or more nodes, each of which receives a plurality of inputs, and the number of coefficients equal to the number of inputs (' Also known as'weights'). That is, the node performs an operation of calculating predetermined coefficients for each of the inputs it receives. In addition, weights calculated for inputs in a node are defined differently for each node.

Referring to FIG. 2, the input layer 302 is composed of five nodes, and these five nodes (hereinafter,'input nodes') receive each of the five features extracted from the feature extractor 200, and are hidden. Output to layer 304.

Referring to FIG. 2, for the purpose of explanation, five input nodes are represented by in1, in2, in3, in4, and in5, and each output of in1, in2, in3, in4, and in5 is IN1, IN2, Indicated by IN3, IN4, and IN5. Here, IN1 may be the received feature F1 as it is, or may be a value of any function that is input to F1.

The first hidden layer 303 is composed of a plurality of nodes (hereinafter, “first hidden node”), and each of the first hidden nodes calculates and outputs a function that takes output values of the input node as inputs. Referring to FIG. 2, a plurality of first hidden nodes are m (where m is a positive integer), and h11, h12, h13, h14, ... , it can be seen that it is marked as h1m.

FIG. 2 exemplarily shows functions and their operations calculated in m first hidden nodes.

For the purpose of explanation, the function applied in the h11 node of the first hidden node is defined as h11, and the result (or'output') of the function h11 is defined as H11. In addition, if the function h11 that takes IN1, IN2, IN3, IN4, and IN5 as inputs is defined as an equation, it is the same as H11 = h11 ( IN1, IN2, IN3, IN4, IN5 ). That is, the node h11 of the first hidden node calculates a function h11 that takes IN1, IN2, IN3, IN4, and IN5 as inputs, and outputs the calculation result H11.

The remaining nodes h12, h13, h14, …, h1m of the first hidden node are also displayed in a similar manner and may be defined by an equation. That is, h12 of the first hidden node calculates a function h12 that takes IN1, IN2, IN3, IN4, and IN5 as inputs, and outputs the calculation result H12.

In the same way, the remaining first hidden nodes are also H13, H14,… , Output H1M.

According to an embodiment, each coefficient included in the functions h11, h12, h13, h14, …, h1m calculated in the first hidden nodes is at least one or more functions (h11, h12, h13, h14). ,…, H1m).

For example, H11 = h11( IN1, IN2, IN3, IN4, IN5 ) = (f11 * IN1 ) + (g11 * IN2 ) + (j11 * IN3 ) + (k11 * IN4 ) + (p11 * IN5 ) And H12 = h12( IN1, IN2, IN3, IN4, IN5 ) = (f12 * IN1 ) + (g12 * IN2 ) + (j12 * IN3 ) + (k12 * IN4 ) + (p12 * IN5 ) Then, f11 and f12 may be different, g11 and g12 may be different, j11 and j12 may be different, k11 and k12 may be different, and/or p11 and p12 may be different.

That is, the function of node h11 of the first hidden layer 303 receives IN1, IN2, IN3, IN4, and IN5 , and is defined to calculate a weight corresponding to each of the inputs, and the h12 node of the first hidden node The function is also defined to receive inputs IN1, IN2, IN3, IN4, and IN5 , and a weight is calculated for each of those inputs.The weights used in the function of the h11 node and the weights used in the function of the h12 node are at least one. They are defined differently. In the same way, at least one or more weights used for each of the functions included in the first hidden node are defined differently.

The second hidden layer 305 is composed of a plurality of nodes (hereinafter,'second hidden node'), and each of the second hidden nodes calculates and outputs a function using the output values of the first hidden node as variables. do. Referring to FIG. 2, a plurality of second hidden nodes are n (where n is a positive integer) and h21, h22, h23, h24, ... , it can be seen that it is denoted by h2n.

In FIG. 2, functions calculated in n second hidden nodes and their operations are exemplarily shown.

For the purpose of explanation, the function applied to the h21 node of the second hidden node is defined as h21, and the result of the function h21 is defined as H21. And, H11, H12, H13, H14, ... , H1M If the function h21 as an input variable is defined by an equation, it is the same as H21 = h21( H11, H12, H13, H14, …, H1M ). That is, the h21 node of the second hidden node is H11, H12, H13, H14, ... Calculates a function h21 with H1M as an input variable, and outputs H21 as a result of the calculation.

The remaining nodes h22, h23, h24, …, h2n of the second hidden node are also displayed in a similar manner and may be defined by an equation. That is, h22 of the second hidden node is H11, H12, H13, H14, ... Calculates a function h22 with H1M as an input variable, and outputs H22 as a result of the calculation.

In the same way, the remaining second hidden nodes are also H23, H24,… , Output H2N.

According to an embodiment, each coefficient included in the functions h21, h22, h23, h24, …, h2n calculated in the second hidden nodes is at least one or more functions (h21, h22, h23, h24). ,…, H2n) is different.

For example, H21 = h21( H11, H12, H13, H14,…, H1M ) = (f21 * H11 ) + (g21 * H12 ) + (j21 * H13 ) + (k21 * H14 ) +… + (p21 * H1M ), H22 = h22 ( H11, H12, H13, H14,…, H1M ) = (f22 * H11 ) + (g22 * H12 ) + (j22 * H13 ) + (k22 * H14 ) +… Assuming that it is defined as + (p22 * H1M ), f21 and f22 are different, g21 and g22 are different, j21 and j22 are different, k21 and k22 are different,… , And/or p21 and p22 may be different.

That is, the function of the node h21 of the second hidden node is H11, H12, H13, H14, ... , H1M are inputted as variables, and the weights are calculated for each of those variables, and the functions of the h22 node of the second hidden node are also H11, H12, H13, H14,… , H1M is inputted as variables, and weights are calculated for each of those variables. However, the weights used in the function of the h21 node and the weights used in the function of the h22 node are defined differently from at least one. .

In a similar manner, at least one or more weights used for functions included in the second hidden node are defined differently.

The output layer 306 is composed of one node (hereinafter, “output node”), and the output node calculates and outputs a predefined function that takes output values of the second hidden node as inputs. In FIG. 2, functions and operations calculated in the output node are illustrated as an example.

For the purpose of explanation, the function of the output node is defined as out, and the result of the function out is defined as OUT. And, H21, H22, H23, H24, ... If the output function using, H2N as an input variable is defined as an equation, it is the same as OUT = out( H21, H22, H23, H24,…, H2N ). That is, the output nodes are H21, H22, H23, H24, ... Calculates a function out with H2N as an input variable, and outputs OUT as a result of the calculation.

In this embodiment, OUT may be a value capable of determining whether it is a lesion tissue. E.g., ≤1 0≤ OUT it may be a value defined by or defined by (OUT is a real number), or 0≤ ≤100 OUT (OUT is a percentage value).

Meanwhile, the deep learning algorithm 300 generates a learning model 400 for detecting lesion tissue by performing learning on labeled measurement data.

Each of the hidden nodes has one function (hereinafter referred to as'hidden node function'), and each hidden node outputs a value of the hidden node function corresponding to itself when each of the hidden nodes receives an input.

Hidden node functions mathematically compute (eg multiply) inputs and coefficients. The learning of the learning model 400 for detecting lesion tissue will be described in detail later with reference to FIG. 3.

The learning model 400 for detecting lesion tissue is optimized by learning (or training). Learning the lesion tissue detection learning model 400 means a process of optimizing each coefficient included in the function of each node of the hidden layers. The learning will be described later in detail with reference to FIG. 3.

According to an alternative embodiment, the deep learning algorithm 300 directly inputs measurement data without passing through the feature extractor 200 and applies the learning model 400 for detecting lesion tissue to determine whether there is a lesion tissue. That is, without using the feature extractor 200, it is determined whether or not there is a lesion tissue by applying the learning model 400 for detecting lesion tissue to all values of the wavelength band.

Referring to FIG. 3, a learning operation according to an embodiment of the present invention will be described.

Referring to FIG. 3, the feature extractor 200 can optimally determine the parameters of the feature extractor 200 using a plurality of labeled measurement data, and the deep learning algorithm 300 is a plurality of labeled measurement data. The learning model 400 for detecting lesion tissue is trained by using them. Here, the plurality of labeled measurement data may be Filtered Non-discrete spectrum data, for example.

In the present specification, a learning model for detecting lesion tissue after learning is specifically referred to as a classifier.

The label indicates whether the lesion tissue is present. For example, measurement data labeled “cancer” refers to non-discrete spectrum measurement data collected from tissues of a patient determined to be cancer as a result of a doctor's diagnosis.

A method of determining feature extractor parameters will be described first by way of example.

The feature extractor 200 receives all of the measurement data (all the collected measurement data) labeled (e.g., a label meaning cancer), and the feature extractor ( 200 determines the parameters of the feature extractor 200.

According to an embodiment, the feature extractor 200 may determine weights included in a function for extracting features as optimized values.

According to another embodiment, the feature extractor 200 may determine not only the weights included in the feature extracting function, but also the main component (the number of main components and/or the position of the main component) as an optimized value.

According to another embodiment, the feature extractor 200 determines the weights included in the feature extracting function as optimized values, and the main component can be defined by a person (for example, a person implementing the present invention). have. According to this embodiment, based on the number of main components and/or the location of the main components defined by the embodiment of the present invention, the feature extractor 200 may determine weights included in the function for extracting features as optimized values. .

A method of optimizing the learning model 400 for detecting lesion tissue by learning will be exemplarily described.

The deep learning algorithm 300 sequentially receives measurement data (all collected measurement data) labeled with a label (e.g., a label indicating cancer), while the deep learning algorithm is a learning model for detecting lesion tissue ( 400).

Since the method described with reference to FIG. 3 is exemplary, the present invention is not limited to such a method. In the above-described embodiments, specific numerical values are illustrative, and it should be understood by those skilled in the art that the present invention is not limited thereto.

4 is a view for explaining a laser spectroscopy-based independent device (hereinafter,'independent device') for detecting lesion tissue according to an embodiment of the present invention.

Referring to FIG. 4, the independent device according to an embodiment of the present invention irradiates a laser to a sample T, collects generated light generated from the sample T, and describes the spectrum of the collected generated light. It is possible to diagnose in real time whether there is lesional tissue in the sample by analyzing it by'based lesion tissue detection method'.

On the other hand, the stand apparatus in-vivo (in- vivo) as well as the lesion diagnostic X-VIVO (ex-vivo) is also possible lesion diagnosis.

The independent device according to an embodiment of the present invention can diagnose a disease such as cancer. For example, the independent device can diagnose diseases such as skin cancer, and can diagnose other types of cancer as well as skin cancer.

Meanwhile, the skin cancer may be, for example, squamous cell carcinoma, basal cell carcinoma, or melanoma.

4, the independent device according to an embodiment of the present invention guides the laser generated by the analysis device 10, the laser generation device 11, and the laser generation device 11 to be irradiated to the sample T. The first optical elements 13 to be performed, the second optical elements 15 for collecting light generated when the laser is irradiated to the sample T, and the generated light collected by the second optical elements 15 It may include a cable 31 providing a path through which the analysis device can be moved.

According to an embodiment, the analysis device 10 irradiates a laser to the sample T and performs a non-discrete spectrum measurement on the NON-GATED generated light collected therefrom, and the measurement result It is an electronic device that determines whether a lesional tissue is present by performing the first and second steps.

According to another embodiment, the analysis device 10 determines parameters of a feature extractor for extracting a feature from the labeled non-discrete spectrum measurement data, and from the labeled non-discrete spectrum measurement data, The step of defining a learning model for detecting lesion tissue may be additionally performed.

According to an embodiment, the analysis device 10 includes a plurality of electronic devices. For example, the analysis device 10 includes a spectroscope 21, a disease analysis module 23, a power supply 25, and a display unit 27.

The spectrometer 21 irradiates a laser to the sample T and performs a non-discrete spectrum measurement on the NON-GATED generated light collected therefrom.

5 is a diagram illustrating a non-discrete spectrum according to an embodiment of the present invention. In the present invention, it is very important to use non-discrete spectrum measurement data. Hereinafter, the non-discrete spectrum will be described in detail with reference to FIG. 5.

In general, a pulsed laser having a pulse length of several nanoseconds or less (fs~ns) is condensed and irradiated on the surface of the sample.If energy is generally 1GW/cm2 or more at the sample surface, a very small amount of the sample surface It is ablated and becomes plasma.

At this time, "plasma light" (including an electron emission spectrum of an atom, a molecular emission spectrum, and continuous emission) is generated on the surface of the sample. As shown, if the gating of the spectrometer is performed after a certain period of time (for example, 1 us) (it is to measure the light generated after a certain period of time has elapsed from the moment when the laser is irradiated on the sample surface), the initial continuous emission spectrum Is excluded, and mainly electron emission spectra of atoms can be obtained.

In contrast, the spectrum of the NON-GATED generated light collected from the laser irradiated to the sample is non-discrete spectrum data including all of the electron emission spectrum of the atom, the molecular emission spectrum, and the continuous emission spectrum.

According to an embodiment of the present invention, the spectrometer 21 may measure a spectrum of light generated by NON-GATED.

According to another embodiment of the present invention, the spectrometer 21 is configured to measure a filtered non-discrete spectrum in the light generated by NON-GATED.

For example, for measuring a filtered non-discrete spectrum, the spectroscope 21 measures spectrums for the rest of the light excluding reflected light, scattered light, and fluorescent light in the NON-GATED generated light collected from the sample irradiated by a laser and collected therefrom. That is, the spectroscope 21 is configured to perform a filtering operation of excluding reflected light, scattered light, and fluorescent light from the NON-GATED generated light and measuring a spectrum for the remaining light, thereby Light is excluded and a non-discrete spectrum is obtained for the rest of the generated light.

As another example of measuring the Filtered Non-discrete spectrum, the spectrometer 21 is configured to perform a filtering operation of measuring the spectrum for the plasma light in the non-GATED generated light that is irradiated to the sample and collected therefrom by a laser. With this configuration, a non-discrete spectrum for plasma light is obtained.

As another example of measuring a filtered non-discrete spectrum, in the NON-GATED generated light collected from the laser irradiated to the sample and collected therefrom, the spectrometer 21 is used only for the generated light in a specific wavelength band (200 nm to 1000 nm). It is configured to measure the spectrum. With this configuration, a non-discrete spectrum for light generated in a specific wavelength band (200 nm to 1000 nm) is obtained.

The disease analysis module 23 sequentially performs a pre-treatment step (first step) and a step (second step) of determining whether there is a lesioned tissue with respect to the result of the non-discrete spectrum measurement. By doing so, it is an electronic device that determines whether the lesion tissue is present in the sample T.

According to another embodiment, the disease analysis module 23 includes the steps of determining parameters of the feature extractor from the labeled non-discrete spectral measurement data, and from the labeled non-discrete spectral measurement data, detecting lesion tissue. You can additionally perform the steps of defining a learning model for use.

According to an embodiment, the disease analysis module 23 includes a memory (not shown), one or more processors (not shown), and one or more programs. Here, the one or more programs include pre-processing programs and/or deep learning programs.

Here, the pre-processing programs are for performing the pre-processing step and include a multidimensional principal component analysis program, and the deep learning program performs an operation of determining whether or not there is a lesion tissue. For more details on these programs, please refer to the above section.

The power supply 25 supplies power to the laser generator 11 and the disease analysis module 23.

The display unit 27 outputs the analysis result by the disease analysis module 23 in a form that can be recognized by the user visually and/or auditoryly.

The laser generating device 11 may be one generally used for skin treatment, for example.

When the pulsed laser by the laser generator 11 is condensed and irradiated on the surface of the sample, not only plasma light but also reflected light, scattered light, or fluorescence emission is generated.

According to an embodiment of the present invention, plasma light except for reflected light, scattered light, and fluorescence emission among light generated when a pulsed laser is condensed and irradiated on a sample surface. It is desirable to obtain a non-discrete spectrum only for.

For example, the laser generator 10 irradiates a laser having a wavelength of a specific band on the surface of a sample, and the spectroscope 21 measures a spectrum only for the generated light belonging to a wavelength of a specific band among the generated light. do.

For a more specific example, a laser generator called a Q-switched Nd:YAG laser generates lasers with wavelengths of 1064 nm and 532 nm. Light with a wavelength band of 1000 nm is generated when the tissue on the surface of the sample breaks down. That is, the spectra measured for the generated light having a wavelength band of 200 nm to 1000 nm are all atomic electron emission spectra and molecular emission spectra. Therefore, a laser called a Q-switched Nd:YAG laser is irradiated on the surface of the sample, and the spectroscope 21 measures the spectrum only for light in the wavelength band of 200 nm to 1000 nm (1 μm) among the generated light. Non-discrete spectrum can be obtained.

When a source laser with a wavelength of 1064 nm is irradiated on the surface of the sample, the wavelengths of reflected light, scattered light, and fluorescent light are almost unchanged from the wavelength of the source laser and remain at 1064 nm, and inelastic scattering, which corresponds to only a small part of the actual scattered light. The wavelength change occurs at, but the signal strength is very weak compared to the elastic scattering and is negligible. Accordingly, the wavelengths (1064 nm) of reflected light, scattered light, and fluorescent light among the generated light can be completely separated from the wavelength band (200 to 1000 nm) in which the electron emission spectrum and the molecular emission spectrum of the atom to be measured are actually emitted.

According to this embodiment, when the laser generator 11 generates a laser having a wavelength of 1064 nm and irradiates it on the surface of a sample, and the spectroscope 21 no longer generates plasma light from the moment the laser is irradiated on the surface of the sample. It is configured to measure the light over time as non-gated until, but to measure the non-discrete spectrum only for the wavelength band between 200 nm and 1000 nm. In this case, the spectroscope 21 can remove reflected light, scattered light, and fluorescent light, and effectively obtain an electron emission spectrum and a molecular emission spectrum of significant atoms.

The first optical elements 13 may include a medium and/or optical elements (eg, lenses) that guide light and control a focus of light so that a laser is irradiated onto the sample T.

The second optical elements 15 collect and guide the generated light and/or optical elements (e.g., a medium for collecting and guiding the generated light and adjusting the focus of the light) in order to collect the generated light and guide the collected generated light to the cable 31. For example, a lens) may be included.

The cable 31 may include a light transmission medium that provides a path through which the generated light collected by the handpiece can be moved to the analysis instrument. The optical transmission medium may be composed of, for example, an optical fiber.

6 shows an exemplary embodiment of a laser spectroscopy-based standalone device for in vivo lesion tissue detection described with reference to FIG. 4.

Referring to FIG. 6, an independent device based on laser spectroscopy for detecting lesion tissue in vivo according to an embodiment of the present invention includes an analyzer 10, a handpiece 20, and a cable 30.

The analysis device 10 has a substantially cylindrical case, and the inside of the analysis device 10 has a spectroscope 21, a disease analysis module 23, a power supply 25, and a display unit ( 27) is provided.

The handpiece 20 has a shape for gripping by hand, and a laser generated by the laser generating device 11 and the laser generating device 11 is irradiated to the body tissue T in the interior of the hand piece 20. First optical elements 13 for collecting the light generated when the laser is irradiated to the body tissue T are provided with second optical elements 15.

The cable 30 may include an electric wire and an optical transmission medium for providing electric power.

According to an embodiment, the spectroscope 20 performs a non-discrete spectrum measurement on the generated light collected by the second optical elements 15, and the disease analysis module 23 The lesion tissue may be detected by the above-described machine learning-based lesion tissue detection method.

According to another embodiment, the spectroscope 20 measures a filtered non-discrete spectrum of the generated light collected by the second optical elements 15, and the disease analysis module 23 is based on machine learning as described above. The lesion tissue can be detected by a method of detecting lesion tissue.

For details on the non-discrete spectrum measurement and the method for detecting lesion tissue based on machine learning, please refer to the above.

7 to 9 are diagrams for explaining non-discrete spectrum measurement.

That is, FIG. 7 shows a result of a discrete spectrum measurement according to the prior art, and FIG. 8 shows a result of a non-discrete spectrum measurement according to an embodiment of the present invention, 9 is a comparison of the results of FIGS. 7 and 8.

Referring to these drawings, the result of discrete spectrum measurement according to the prior art is that only threshold values of wavelength values corresponding to a specific component can be measured, whereas non-discrete according to an embodiment of the present invention It can be seen that the result of non-discrete spectrum measurement is to measure values for all wavelength values.

As described above, although the present invention has been described by the limited embodiments and drawings, the present invention is not limited to the above embodiments and drawings, and those of ordinary skill in the art to which the present invention pertains, such description From, various modifications and variations are possible. Therefore, the scope of the present invention is limited to the described embodiments and should not be defined, but should be defined by the claims to be described later as well as equivalents to the claims.

10: analysis unit
11: laser
13, 15: optical elements
20: handpiece
21: spectrograph
23: disease analysis module
25: power
27: display
30: cable

Claims (18)

A laser irradiation module for inducing plasma ablation on the surface of the specimen by irradiating a pulsed laser on the specimen to be diagnosed with or without a disease;
Atoms operating in a non-gated manner and having a specific wavelength according to the intrinsic energy level of the material generated in the process of transitioning at least a part of the specimen from the plasma state to the molecular state by the plasma ablation/ A spectrometer for acquiring first spectral data including a discontinuous spectrum due to atomic-molecular emission and a continuous spectrum generated before the atomic/molecular emission light and emitted from free electrons in a plasma state with an unspecified wavelength; And
The first spectral data obtained by inducing plasma ablation on a learning specimen in which the diagnosis result regarding the presence or absence of a disease is known in advance, and the machine learning model learned using the learning data labeled with the diagnosis result regarding the presence or absence of the disease, are used. Including; a processor for detecting information on the presence or absence of disease in the specimen from the spectral data
Device for providing disease detection information.
The method of claim 1,
The wavelength of the pulsed laser is 1064 nm,
Device for providing disease detection information.
The method of claim 2,
The spectroscope measures light in a predetermined wavelength band so that spectral data related to reflected light, scattered light, and fluorescent light by the pulsed laser are not included in the first spectral data.
Device for providing disease detection information
The method of claim 3,
The predetermined wavelength band is 200 ~ 1000nm.
Device for providing disease detection information
The method of claim 1,
The disease includes skin cancer, wherein the skin cancer is squamous cell carcinoma, basal cell carcinoma, or melanoma,
Device for providing disease detection information
The method of claim 1,
The spectroscope receives plasma light by plasma ablation,
Device for providing disease detection information
A laser irradiation module for inducing plasma ablation on the surface of the specimen by irradiating a pulsed laser on the specimen to be diagnosed with or without a disease;
The generated light is measured from the moment the pulsed laser is irradiated to the specimen until the generated light generated by the plasma ablation is no longer generated, and at least a part of the specimen is removed from the plasma state by the plasma ablation. Discontinuous spectrum by atomic-molecular emission having a specific wavelength according to the intrinsic energy level of a substance generated in the process of transitioning to a molecular state, and free electrons in the plasma state that occur before the atomic/molecular emission light A spectrometer for acquiring first spectral data including a continuous spectrum emitted with an unspecified wavelength from from; And
The first spectral data obtained by inducing plasma ablation on a learning specimen in which the diagnosis result regarding the presence or absence of a disease is known in advance, and the machine learning model learned using the learning data labeled with the diagnosis result regarding the presence or absence of the disease, are used. Including; a processor for detecting information on the presence or absence of disease in the specimen from the spectral data
Device for providing disease detection information.
The method of claim 7,
The wavelength of the pulsed laser is 1064 nm,
Device for providing disease detection information.
The method of claim 8,
The spectroscope measures light in a predetermined wavelength band so that spectral data related to reflected light, scattered light, and fluorescent light by the pulsed laser are not included in the first spectral data.
Device for providing disease detection information
The method of claim 9,
The predetermined wavelength band is 200 ~ 1000nm.
Device for providing disease detection information
The method of claim 8,
The disease includes skin cancer, wherein the skin cancer is squamous cell carcinoma, basal cell carcinoma, or melanoma,
Device for providing disease detection information
The method of claim 8,
The generated light is plasma light,
Device for providing disease detection information
A laser irradiation module for inducing plasma ablation on the surface of the specimen by irradiating a pulsed laser on the specimen to be diagnosed with or without a disease;
Discontinuous spectrum by atomic-molecular emission having a specific wavelength according to the material intrinsic energy level generated in the process of transitioning at least a part of the specimen from the plasma state to the molecular state by the plasma ablation, and A spectrometer with a gating time set to obtain first spectral data including a continuous spectrum generated before the atomic/molecular emission light and emitted from free electrons in a plasma state with an unspecified wavelength; And
The first spectral data obtained by inducing plasma ablation on a learning specimen for which the diagnosis result regarding the presence or absence of a disease is known in advance is used, using a machine learning model learned using the learning data labeled with the diagnosis result regarding the presence or absence of the disease. Including; a processor for detecting information on the presence or absence of disease in the specimen from the spectral data
Device for providing disease detection information.
The method of claim 13,
The wavelength of the pulsed laser is 1064 nm,
Device for providing disease detection information.
The method of claim 14,
The spectroscope measures a wavelength band of 200 to 1000 nm so that spectral data about reflected light, scattered light, and fluorescent light by the pulsed laser are not included in the first spectral data.
Device for providing disease detection information
The method of claim 13,
The disease includes skin cancer, wherein the skin cancer is squamous cell carcinoma, basal cell carcinoma, or melanoma,
Device for providing disease detection information
The method of claim 13,
The spectroscope acquires plasma light by plasma ablation,
Device for providing disease detection information
The method of claim 13,
The gating time is 1 ns ~ 100us,
Device for providing disease detection information

Images