KR102590387B1 - System for diagnosis of carpal tunnel syndrome using muscle ultrasound imaging based on artificial intelligence and method thereof - Google Patents

Abstract

A carpal tunnel syndrome diagnosis system according to an embodiment of the present invention includes a measurement unit configured to set a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image and measure muscle indices from the region of interest; and a diagnosis unit configured to input the muscle ultrasound image and the muscle index into a carpal tunnel syndrome diagnosis model and output a carpal tunnel syndrome diagnosis result for the muscle ultrasound image.

Description

Carpal tunnel syndrome diagnosis system and method using artificial intelligence-based muscle ultrasound

The present invention relates to a carpal tunnel syndrome diagnosis system using artificial intelligence-based muscle ultrasound and a method of using the same.

Currently, most cases of diagnosing carpal tunnel syndrome using ultrasound use the median nerve cross-sectional area, but the cut-off value varies greatly depending on the study, and sensitivity and specificity also vary greatly depending on the study.

In cases of neuropathy, the muscle fibers of the muscles controlled by the nerve are replaced by fat or fiber due to denervation, which results in increased echogenicity on ultrasound. By analyzing this quantitatively, it is possible to distinguish between normal muscles and muscles affected by neuropathy.

Until now, there have been several attempts to distinguish between muscles affected by carpal tunnel syndrome and normal muscles using muscle echo shadow changes that occur in neuropathy.

Most studies succeeded in showing statistically significant differences in muscle echo shade between patient groups and normal subjects, but because they attempted to distinguish using only simple variables such as the average value or standard deviation of echo shade, sensitivity and specificity were not high, and carpal tunnel No significant differences were found depending on the severity of the syndrome.

One embodiment of the present invention analyzes various input variables that can be derived from echo shadows by analyzing muscle ultrasound images based on artificial intelligence, and then weights input variables with high discrimination to improve the diagnostic accuracy of carpal tunnel syndrome. The purpose is to provide a carpal tunnel syndrome diagnosis system and method of use.

Meanwhile, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned will be clearly apparent to those skilled in the art from the description below. It will be understandable.

A carpal tunnel syndrome diagnosis system according to an embodiment of the present invention includes a measurement unit configured to set a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image and measure muscle indices from the region of interest; and a diagnosis unit configured to input the muscle ultrasound image and the muscle index into a carpal tunnel syndrome diagnosis model and output a carpal tunnel syndrome diagnosis result for the muscle ultrasound image.

The diagnostic unit may be configured to select a weighted muscle index, which is a muscle index of relatively high importance when diagnosing carpal tunnel syndrome, through at least one of a recursive feature elimination method or an F-test method for each of the muscle indexes measured by the measurement unit. there is.

It further includes a storage unit configured to store the muscle index measured by the measurement unit and the diagnostic result output by the carpal tunnel syndrome diagnostic model, wherein the storage unit: stores weights for each of the weighted muscle index and the weighted muscle index. It can be configured to store additional information.

The measuring unit: generates a mask image for a region of interest from the muscle ultrasound image; extracting a region of interest image from the muscle ultrasound image using the mask image; The feature of the muscle may be extracted from the region of interest image, but the muscle index may be extracted by excluding features with fixed values from among the extracted features.

The muscle indices are: First Order Statistics, Gray Level Cooccurrence, Gray Level Dependence, Gray Level Run Length, Gray Level Size Zone) and Neighboring Gray Tone Difference.

The primary statistic is: an index that describes the distribution of echo shadows in the region of interest image through a statistical method; Median Absolute Deviation, 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum , Mean AbsoluteDeviation, Mean, Median, Range, Mean Absolute Deviation between 10th and 90th percentiles, RootMeanSquared, Skewness, It may include at least one of Total Energy, Uniformity, and Variance.

The intensity co-occurrence is: a statistical index that examines echo shadowing taking into account the spatial relationship of pixels within the region of interest image; Autocorrelation, skewness and skewness (ClusterProminence), skewness and uniformity (ClusterShade), cluster tendency (ClusterTendency), local intensity change (Contrast), correlation (Correlation), intensity difference average (DifferenceAverage), Entropy difference (DifferenceEntropy), difference variance (DifferenceVariance), inverse difference (Id), inverse difference moment (Idm), inverse difference moment normalization (Idmn), normalized inverse difference (Idn), pixel-to-pixel distribution correlation 1 (Information measure of correlation 1) , Information measure of correlation 2, InverseVariance, JointAverage, JointEnergy, JointEntropy, MCC, Max. It may include at least one of MaximumProbability, SumAverage, SumEntropy, and SumSquares.

The gray level dependency is: an indicator indicating a relationship with surrounding pixels having a specific intensity value within the region of interest image; DependenceEntropy, DependenceNonUniformity, DependenceNonUniformityNormalized, DependenceVariance, GrayLevelNonUniformity, GrayLevelVariance, High Gray Level Value Distribution (HighGrayLevelEmphasis), Distribution with large dependency (LargeDependenceEmphasis), High gray intensity distribution with large dependency (LargeDependenceHighGrayLevelEmphasis), Low gray intensity distribution with large dependency (LargeDependenceLowGrayLevelEmphasis), Low gray intensity distribution (LowGrayLevelEmphasis), Distribution with small dependency (SmallDependenceEmphasis) ) , It may include at least one of a high gray intensity distribution with few dependencies (SmallDependenceHighGrayLevelEmphasis), a low gray intensity distribution with few dependencies (SmallDependenceLowGrayLevelEmphasis), and normalized gray intensity non-uniformity (GrayLevelNonUniformityNormalized).

The intensity effect length is: an indicator related to the length of pixels in a specific direction with the same intensity value within the region of interest image; Gray intensity action distribution (HighGrayLevelRunEmphasis), long action length distribution (LongRunEmphasis), association distribution of long action length distribution with high gray intensity distribution (LongRunHighGrayLevelEmphasis), association distribution of long action length distribution with low gray intensity distribution (LongRunLowGrayLevelEmphasis), low gray intensity distribution. Strength action distribution (LowGrayLevelRunEmphasis), run length entropy (RunEntropy), run length non-uniformity (RunLengthNonUniformity), run length normalized non-uniformity (RunLengthNonUniformityNormalized), run length density (RunPercentage), run length variance (RunVariance), short It may include at least one of an action length distribution (ShortRunEmphasis), an association distribution of a short action length distribution and a high gray intensity distribution (ShortRunHighGrayLevelEmphasis), and an association distribution of a short action length distribution and a low gray intensity distribution (ShortRunLowGrayLevelEmphasis).

The gray level size area is: an indicator for quantifying the pixel area with the same intensity value within the region of interest image; HighGrayLevelZoneEmphasis, LargeAreaEmphasis, LargeAreaHighGrayLevelEmphasis, LargeAreaLowGrayLevelEmphasis, LowGrayLevelZoneEmphasis ), area non-uniformity (SizeZoneNonUniformity), area non-uniformity normalized (SizeZoneNonUniformityNormalized), small area distribution (SmallAreaEmphasis), correlation distribution of small area and high gray intensity area (SmallAreaHighGrayLevelEmphasis), small area and low gray intensity area It may include at least one of an association distribution (SmallAreaLowGrayLevelEmphasis), zone entropy (ZoneEntropy), zone density (ZonePercentage), and zone variance (ZoneVariance).

The ambient gray hue difference is: an indicator that quantifies the difference in pixels and brightness within a certain distance within the region of interest image; It may include at least one of the amount of intensity change to adjacent pixels (Busyness), the average difference between the center pixel and neighboring pixels (Coarseness), the complexity of the gray intensity change within the image (Complexity), and the size of the gray intensity change (Strength).

A method of using a carpal tunnel syndrome diagnosis system according to an embodiment of the present invention includes the steps of setting a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image through a measuring unit and measuring muscle indices from the region of interest; and inputting the muscle ultrasound image and the muscle index into a carpal tunnel syndrome diagnosis model through a diagnosis unit to output a carpal tunnel syndrome diagnosis result for the muscle ultrasound image.

A computer-readable recording medium according to an embodiment of the present invention is a recording medium on which a program for executing a method of using the carpal tunnel syndrome diagnosis system on a computer is recorded.

The carpal tunnel syndrome diagnosis system and method of using the same according to an embodiment of the present invention analyzes various input variables that can be derived from echo shadows by analyzing muscle ultrasound images based on artificial intelligence, and then weights the input variables with high discrimination power. can improve the diagnostic accuracy of carpal tunnel syndrome.

Meanwhile, the effects that can be obtained from the present invention are not limited to the effects mentioned above, and other effects not mentioned will be clearly understood by those skilled in the art from the description below. You will be able to.

Figure 1 is a diagram schematically showing each configuration of the carpal tunnel syndrome diagnosis system 10 according to an embodiment of the present invention.
Figure 2 is an example diagram showing the structure of a tree-based classifier included in the carpal tunnel syndrome diagnosis model.
Figure 3 is an example diagram showing the structure of a linear classifier included in the carpal tunnel syndrome diagnostic model.
Figure 4 is a flowchart showing a method of using the carpal tunnel syndrome diagnosis system (S10).
Figure 5 is a flowchart showing step S200 of Figure 4 in more detail.
Figure 6 is an example diagram showing a muscle ultrasound image.
Figure 7 is an example diagram showing a mask image.
Figure 8 is a flowchart showing step S240 of Figure 5 in more detail.
Figure 9 is a flowchart showing step S300 of Figure 4 in more detail.

Other advantages and features of the present invention and methods for achieving them will become clear by referring to the embodiments described in detail below along with the accompanying drawings. However, the present invention is not limited to the embodiments disclosed below and may be implemented in various different forms. The present embodiments are only provided to ensure that the disclosure of the present invention is complete and to provide common knowledge in the technical field to which the present invention pertains. It is provided to fully inform those who have the scope of the invention, and the present invention is only defined by the scope of the claims.

Even if not defined, all terms (including technical or scientific terms) used herein have the same meaning as generally accepted by the general art in the prior art to which this invention belongs.

Terms defined by general dictionaries may be interpreted as having the same meaning as they have in the related art and/or text of the present application, and should not be conceptualized or interpreted in an overly formal manner even if expressions are not clearly defined herein. won't

The terms used in this specification are for describing embodiments and are not intended to limit the invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context.

As used in the specification, 'comprises' and/or various conjugations of this verb, such as 'comprises', 'comprising', 'includes', 'comprises', etc., refer to the composition, ingredient, or component mentioned. A step, operation and/or element does not exclude the presence or addition of one or more other compositions, ingredients, components, steps, operations and/or elements. As used herein, the term 'and/or' refers to each of the listed components or various combinations thereof.

Meanwhile, terms such as '~unit', '~unit', '~block', and '~module' used throughout this specification may mean a unit that processes at least one function or operation. For example, it can refer to software, hardware components such as FPGAs or ASICs.

However, '~part', '~unit', '~block', '~module', etc. are not limited to software or hardware. '~ part', '~ base', '~ block', and '~ module' may be configured to reside in an addressable storage medium and may be configured to reproduce one or more processors.

Therefore, as an example, '~part', '~gi', '~block', and '~module' refer to components such as software components, object-oriented software components, class components, and task components. , processes, functions, properties, procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and Contains variables.

The functions provided within components and '~part', '~unit', '~block', and '~module' are divided into smaller numbers of components and '~unit', '~unit', '~block'. ', '~module' or can be further separated into additional components and '~sub', '~base', '~block', and '~module'.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings attached to this specification.

Figure 1 is a diagram schematically showing each configuration of the carpal tunnel syndrome diagnosis system 10 according to an embodiment of the present invention.

Referring to FIG. 1, the carpal tunnel syndrome diagnosis system 10 may include an input unit 100, a measurement unit 200, a diagnosis unit 300, a storage unit 400, and an output unit 500, and may include muscle ultrasound. After measuring muscle-related indicators within the area of interest in the image, the data can be input into an artificial intelligence model included in the diagnostic unit 300 to output whether carpal tunnel syndrome is present.

The input unit 100 may receive a muscle ultrasound image from an external device, for example, an ultrasound imaging device, and transmit it to the measurement unit 200.

The measurement unit 200 may set a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image and measure muscle indicators from the region of interest.

More specifically, the measurement unit 200 generates a mask image for the region of interest of the muscle from the muscle ultrasound image, extracts the region of interest image from the muscle ultrasound image using the mask image, and then measures various characteristics of the muscle. You can. At this time, among the measured characteristics, except for characteristics that always have a fixed value, the remaining characteristics can be used as muscle indicators.

The measured muscle index can be transmitted to the diagnostic unit 300 and used as an input variable for diagnosing carpal tunnel syndrome. The measured muscle index is the energy, entropy, mean, median, standard deviation, and interquartile range of the echo shade through at least one calculation method among the first-order statistical method, image gray level continuity, and image gray level intensity for the region of interest. A characteristic (muscle index) that includes at least one of the following can be measured.

The diagnostic unit 300 may receive muscle indicators based on an artificial intelligence algorithm performed by an artificial intelligence model and then output a determination result as to whether the image is an image of a patient with carpal tunnel syndrome.

More specifically, the diagnosis unit 300 inputs the muscle ultrasound image for diagnosing carpal tunnel syndrome and the muscle index measured by the measurement unit 200 into a carpal tunnel syndrome diagnosis model based on an artificial intelligence algorithm, and then inputs the input muscle ultrasound image. You can receive a classification result (diagnosis result) regarding whether this is a muscle ultrasound image of a patient with carpal tunnel syndrome.

The carpal tunnel syndrome diagnosis model may include a tree-based classifier and a linear classifier.

A tree-based classifier can use a method of generating multiple tree classifiers, predicting classification results for a subset of input variables, and predicting the final result by assigning weights to the predicted results.

Figure 2 is an example diagram showing the structure of a tree-based classifier included in the carpal tunnel syndrome diagnosis model.

Referring to Figure 2, muscle indicators corresponding to input variables can be divided into several subsets by allowing overlap. At this time, the number of divided subsets allows the classifier to produce optimal performance, and this value (number of subsets) may be stored in the storage unit 400.

For each subset, the classification result can be predicted through a different decision tree classifier. Afterwards, the final classification result can be predicted by weighting and combining the results predicted by each decision tree classifier. This method is called ensemble.

A linear classifier can be a baseline or plane that determines which of the predetermined classes an input variable belongs to.

Figure 3 is an example diagram showing the structure of a linear classifier included in the carpal tunnel syndrome diagnostic model.

Referring to FIG. 3, variables used for learning a linear classifier may have respective variable values and predetermined classes. A linear classifier can distinguish which classification an input variable belongs to depending on where the input variable appears in the classifier.

The learning method of a linear classifier is a method of maximizing the distance between the variables of each classification that are closest to the linear classifier, and a method of properly separating the given data as much as possible, and a method of mixing the two methods to maximize the distance between variables that are properly separated. Methods such as separating given variables as properly as possible can be used.

Referring again to FIG. 1, the diagnosis unit 300 selects a weighted muscle index, which is a muscle index with relatively high discrimination power, for diagnosing carpal tunnel syndrome by measuring the performance of the classifier for each muscle index measured by the measurement unit 200. This can be done by applying weights to the weighted muscle index and providing it to the carpal tunnel syndrome diagnosis model.

More specifically, the diagnosis unit 300 measures the diagnostic performance of each muscle index for each classifier and then selects a weighted muscle index, which is a muscle index with relatively high discriminatory power in diagnosing carpal tunnel syndrome. . At this time, the diagnostic performance can be measured as high when the muscle ultrasound image diagnosed as carpal tunnel syndrome according to the carpal tunnel syndrome diagnostic model is an actual muscle ultrasound image of a carpal tunnel syndrome patient.

The recursive feature elimination method trains a machine learning-based carpal tunnel syndrome classifier for all muscle indicators, then calculates the priority of indicators deemed important for classification performance, and recursively removes unimportant features while recursively removing the necessary features for training. You can find the optimal muscle index.

The F-test method obtains a statistical test quantity for each muscle index, and the calculated value can be used as a weight for the muscle index.

The storage unit 400 contains the characteristics (muscle index) measured in the measurement unit 200, the classification result by the carpal tunnel syndrome diagnostic model included in the diagnosis unit 300, and the weighted muscle index selected in the diagnosis unit 300. Weights for weighted muscle indices can be stored.

The output unit 500 may output the carpal tunnel syndrome diagnosis result generated by the diagnosis unit 300 and the discrimination evaluation result for each muscle index.

Figure 4 is a flowchart showing a method of using the carpal tunnel syndrome diagnosis system (S10).

Referring to FIG. 4, the method of using the carpal tunnel syndrome diagnosis system (S10) may include steps S100 to S500.

In step S100, a muscle ultrasound image can be input from an external device, for example, an ultrasound imaging device, through the input unit 100, and the input muscle ultrasound image can be transmitted to the measurement unit 200.

In step S200, a region of interest to be used for diagnosing carpal tunnel syndrome can be set in the muscle ultrasound image through the measuring unit 200, and muscle indices can be measured from the region of interest.

In step S300, muscle indicators may be input based on an artificial intelligence algorithm performed by an artificial intelligence model through the diagnostic unit 300, and then a determination result as to whether the image is of a patient with carpal tunnel syndrome may be output.

Additionally, in step S300, a weighted muscle index, which is a muscle index with relatively high discriminatory power in diagnosing carpal tunnel syndrome, can be selected from among the input muscle indexes, and carpal tunnel syndrome can be diagnosed using the weighted muscle index.

Figure 5 is a flowchart showing step S200 of Figure 4 in more detail.

Referring to Figure 5, step S200 may include steps S210 to S240.

In step S210, a muscle ultrasound image may be received from the input unit 100.

Figure 6 is an example diagram showing a muscle ultrasound image.

Referring to FIG. 6, the measuring unit 200 may receive a muscle ultrasound image as shown in FIG. 4 from the input unit 100 (S210).

Referring again to FIG. 5, in step S220, a mask image for the region of interest may be generated from the muscle ultrasound image.

Figure 7 is an example diagram showing a mask image.

Referring to FIG. 7 , the measurement unit 200 may generate a mask image for the region of interest of the muscle ultrasound image as shown in FIG. 5 (S220).

Referring again to FIG. 5, in step S230, a region of interest image can be extracted from the muscle ultrasound image of FIG. 6 using the mask image of FIG. 7.

In step S240, various characteristics of the muscle can be extracted from the image of the region of interest extracted in step S230, and among the extracted features, except for features that always have a fixed value, the remaining features can be extracted as muscle indicators.

More specifically, in step S240, at least one calculation method among the first-order statistical method, image gray level continuity, and image gray level intensity is applied to the region of interest image to calculate the energy, entropy, mean, median, standard deviation, and distribution of echo shading. It can be extracted by measuring a characteristic (muscle index) that includes at least one of the quantile ranges.

At this time, the extracted muscle indicators are First Order Statistics, Gray Level Cooccurrence, Gray Level Dependence, Gray Level Run Length, and Gray Level Size Area. Size Zone) and Neighboring Gray Tone Difference.

Figure 8 is a flowchart showing step S240 of Figure 5 in more detail.

Referring to FIG. 8, step S240 may include steps S241 to S246.

In step S241, the muscle index corresponding to the first statistic among the muscle indexes to be extracted can be measured.

Primary statistics refer to indices that describe the distribution of echo shading within an area of interest through commonly used basic statistical methods. Includes 18 muscle indicators, and the corresponding muscle indicators are Median Absolute Deviation, 10 Percentile, 90 Percentile, Energy, Entropy, and Interquartile InterquartileRange, Kurtosis, Maximum, Mean AbsoluteDeviation, Mean, Median, Range, Mean Absolute Deviation between 10th and 90th percentiles , RootMeanSquared, Skewness, TotalEnergy, Uniformity, and Variance.

In step S242, among the muscle indicators to be extracted, the muscle indicator corresponding to the simultaneous occurrence of light and dark can be measured.

Contrast co-occurrence is a statistical method that examines echo shading by considering the spatial relationship of pixels within a region of interest and includes 24 muscle indices. The corresponding muscle indices are autocorrelation, skewness and skewness (ClusterProminence), skewness and uniformity (ClusterShade), cluster tendency (ClusterTendency), local intensity change (Contrast), correlation (Correlation), and intensity difference. Average (DifferenceAverage), entropy difference (DifferenceEntropy), difference variance (DifferenceVariance), inverse difference (Id), inverse difference moment (Idm), inverse difference moment normalization (Idmn), normalized inverse difference (Idn), pixel-to-pixel distribution correlation 1 (Information) measure of correlation 1), information measure of correlation 2, inverse variance, pixel joint average, pixel joint energy, pixel joint energy, pixel joint entropy, maximum correlation coefficient May include (MCC), MaximumProbability, SumAverage, SumEntropy, and SumSquares.

In step S243, among the muscle indicators to be extracted, the muscle indicator corresponding to gray level dependency can be measured.

Gray level dependency includes 15 muscle indicators as indicators of relationships with surrounding pixels with specific intensity values within the region of interest. The corresponding muscle indices are DependenceEntropy, DependenceNonUniformity, DependenceNonUniformityNormalized, DependenceVariance, GrayLevelNonUniformity, and Gray Intensity Variance. (Graylevelvariance), high -colored value distribution (Highgraylevelemphasis), large -dependent distribution, and high -dependency high -colored strength distribution (LARGEDEPENCEHINCEIHGIGIDEINCEHIHIHIHIHIHGRAHILEMPHA SIS), low -colored strength distribution (LARGEDEPENCELOWGRAYLEMPHASIS) , a distribution with few dependencies (SmallDependenceEmphasis), a high gray intensity distribution with few dependencies (SmallDependenceHighGrayLevelEmphasis), a low gray intensity distribution with few dependencies (SmallDependenceLowGrayLevelEmphasis), and normalized gray intensity non-uniformity (GrayLevelNonUniformityNormalized).

In step S244, among the muscle indicators to be extracted, the muscle indicator corresponding to the light/dark action length can be measured.

The intensity effect length is an index related to the length of pixels in a specific direction with the same intensity value within the region of interest, and includes 13 muscle indices. The corresponding muscle indices are: high gray intensity action distribution (HighGrayLevelRunEmphasis), long action length distribution (LongRunEmphasis), correlation distribution of long action length distribution and high gray intensity distribution (LongRunHighGrayLevelEmphasis), and long action length distribution and low gray intensity distribution. correlation distribution (LongRunLowGrayLevelEmphasis), low gray intensity action distribution (LowGrayLevelRunEmphasis), runEntropy, runLengthNonUniformity, runLengthNonUniformityNormalized, runPercentage , runVariance, short runlength distribution (ShortRunEmphasis), association distribution of short run length distribution with high gray intensity distribution (ShortRunHighGrayLevelEmphasis), and association distribution of short run length distribution with low gray intensity distribution (ShortRunLowGrayLevelEmphasis). You can.

In step S245, among the muscle indicators to be extracted, the muscle indicator corresponding to the gray level size area can be measured.

The gray level size area is an indicator that quantifies the pixel area with the same intensity value within the region of interest and includes 13 muscle indicators. The corresponding muscle indicators are as follows: high gray intensity area distribution (HighGrayLevelZoneEmphasis), large area distribution (LargeAreaEmphasis), correlation distribution between large areas and high gray intensity areas (LargeAreaHighGrayLevelEmphasis), and correlation distribution between large areas and low gray intensity areas (LargeAreaLowGrayLevelEmphasis). ), low gray intensity area distribution (LowGrayLevelZoneEmphasis), area non-uniformity (SizeZoneNonUniformity), normalized area non-uniformity (SizeZoneNonUniformityNormalized), small area distribution (SmallAreaEmphasis), association distribution of small area and high gray intensity area (SmallAreaHighGrayLevelEmphasis) , may include the association distribution of small areas and low gray intensity areas (SmallAreaLowGrayLevelEmphasis), area entropy (ZoneEntropy), area density (ZonePercentage), and area variance (ZoneVariance).

In step S246, among the muscle indicators to be extracted, the muscle indicator corresponding to the difference in surrounding gray color tone can be measured.

Ambient gray hue difference is an index that quantifies the contrast between pixels and brightness within a certain distance within the region of interest and includes four muscle indices. Corresponding muscle indicators may include the amount of intensity change to adjacent pixels (Busyness), the average difference between the center pixel and neighboring pixels (Coarseness), the complexity of the gray intensity change within the image (Complexity), and the size of the gray intensity change (Strength), as follows. You can.

Figure 9 is a flowchart showing step S300 of Figure 4 in more detail.

Referring to FIG. 9, step S300 may include steps S310 to S330.

In step S310, the muscle ultrasound image for diagnosing carpal tunnel syndrome and muscle indicators that can be used as input variables are input into the carpal tunnel syndrome diagnosis model, and then the carpal tunnel syndrome diagnosis result can be output.

In step S320, after measuring the diagnostic performance of each muscle index for each classifier, a weighted muscle index, which is a muscle index with relatively high discriminatory power in diagnosing carpal tunnel syndrome, can be selected. At this time, the diagnostic performance can be measured as high when the muscle ultrasound image diagnosed as carpal tunnel syndrome according to the carpal tunnel syndrome diagnostic model is an actual muscle ultrasound image of a carpal tunnel syndrome patient.

More specifically, in step S320, a weighted muscle index, which is a muscle index with relatively high discriminatory power in diagnosing carpal tunnel syndrome, can be selected through recursive feature elimination or the F-test method. A detailed explanation of this can be found in the diagnosis. Since it was described above along with the part 300, it will be omitted.

The recursive feature elimination method trains a machine learning-based carpal tunnel syndrome classifier for all muscle indicators, then calculates the priority of indicators deemed important for classification performance, and recursively removes unimportant features while recursively removing the necessary features for training. You can find the optimal muscle index.

The F-test method obtains a statistical test quantity for each muscle index, and the calculated value can be used as a weight for the muscle index.

In step S330, the carpal tunnel syndrome diagnosis result generated in step S310 and the discrimination evaluation result for each muscle index generated in step S320 may be transmitted to the storage unit 400 and the output unit 500.

Referring again to FIG. 4, in step S400, the characteristics (muscle index) measured in the measurement unit 200 through the storage unit 400, the classification result by the carpal tunnel syndrome diagnostic model included in the diagnosis unit 300, and the diagnosis unit Information about the weighted muscle index selected at 300 and the weight applied to each weighted muscle index can be stored.

In step S500, the carpal tunnel syndrome diagnosis result and the discrimination evaluation result for each muscle index generated by the diagnosis unit 300 can be output through the output unit 500.

Although the present invention has been described above through examples, the above examples are merely for illustrating the spirit of the present invention and are not limited thereto. Those skilled in the art will understand that various modifications may be made to the above-described embodiments. The scope of the present invention is determined only through interpretation of the appended claims.

10 Carpal Tunnel Syndrome Diagnosis System
100 input section
200 measuring unit
300 Diagnostic Department
400 storage unit
500 output

Claims (13)

a measurement unit configured to set a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image and measure muscle indices from the region of interest; and
A diagnosis unit configured to input the muscle ultrasound image and the muscle index into a carpal tunnel syndrome diagnosis model and output a carpal tunnel syndrome diagnosis result for the muscle ultrasound image,
The diagnostic department:
A carpal tunnel syndrome diagnostic system configured to select a weighted muscle index, which is a muscle index of relatively high importance when diagnosing carpal tunnel syndrome, through at least one of the recursive feature elimination method or the F-test method for each of the muscle indexes measured by the measurement unit. .
delete According to paragraph 1,
It further includes a storage unit configured to store the muscle index measured by the measurement unit and the diagnostic result output by the carpal tunnel syndrome diagnostic model,
The storage unit:
A carpal tunnel syndrome diagnosis system configured to additionally store the weighted muscle index and weights for each of the weighted muscle index.
a measurement unit configured to set a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image and measure muscle indices from the region of interest; and
A diagnosis unit configured to input the muscle ultrasound image and the muscle index into a carpal tunnel syndrome diagnosis model and output a carpal tunnel syndrome diagnosis result for the muscle ultrasound image,
The measuring unit:
generating a mask image for a region of interest from the muscle ultrasound image;
extracting a region of interest image from the muscle ultrasound image using the mask image;
A carpal tunnel syndrome diagnostic system configured to extract muscle indicators from the region of interest image by extracting muscle characteristics and excluding characteristics with fixed values from among the extracted characteristics.
According to paragraph 4,
The muscle indicators are:
First Order Statistics, Gray Level Cooccurrence, Gray Level Dependence, Gray Level Run Length, Gray Level Size Zone, and Ambient Gray A carpal tunnel syndrome diagnostic system that is configured to fall into one of the categories of Neighboring Gray Tone Difference.
According to clause 5,
The above primary statistics are:
As an index describing the distribution of echo shades within the region of interest image through statistical methods;
Median Absolute Deviation, 10Percentile, 90Percentile, Energy, Entropy, InterquartileRange, Kurtosis, Maximum , Mean AbsoluteDeviation, Mean, Median, Range, Mean Absolute Deviation between 10th and 90th percentiles, RootMeanSquared, Skewness, A carpal tunnel syndrome diagnosis system comprising at least one of total energy, uniformity, and variance.
According to clause 5,
The above contrast co-occurrence is:
As a statistical indicator for examining echo shading considering the spatial relationship of pixels within the region of interest image;
Autocorrelation, skewness and skewness (ClusterProminence), skewness and uniformity (ClusterShade), cluster tendency (ClusterTendency), local intensity change (Contrast), correlation (Correlation), intensity difference average (DifferenceAverage), Entropy difference (DifferenceEntropy), difference variance (DifferenceVariance), inverse difference (Id), inverse difference moment (Idm), inverse difference moment normalization (Idmn), normalized inverse difference (Idn), pixel-to-pixel distribution correlation 1 (Information measure of correlation 1) , Information measure of correlation 2, InverseVariance, JointAverage, JointEnergy, JointEntropy, MCC, Max. A carpal tunnel syndrome diagnosis system comprising at least one of MaximumProbability, SumAverage, SumEntropy, and SumSquares.
According to clause 5,
The above gray level dependencies are:
As an indicator indicating a relationship with surrounding pixels having a specific brightness value within the region of interest image;
DependenceEntropy, DependenceNonUniformity, DependenceNonUniformityNormalized, DependenceVariance, GrayLevelNonUniformity, GrayLevelVariance, High Gray Level Value Distribution (HighGrayLevelEmphasis), Distribution with large dependency (LargeDependenceEmphasis), High gray intensity distribution with large dependency (LargeDependenceHighGrayLevelEmphasis), Low gray intensity distribution with large dependency (LargeDependenceLowGrayLevelEmphasis), Low gray intensity distribution (LowGrayLevelEmphasis), Distribution with small dependency (SmallDependenceEmphasis) ) , A carpal tunnel syndrome diagnostic system comprising at least one of a high gray intensity distribution with small dependency (SmallDependenceHighGrayLevelEmphasis), a low gray intensity distribution with small dependency (SmallDependenceLowGrayLevelEmphasis), and normalized gray intensity non-uniformity (GrayLevelNonUniformityNormalized).
According to clause 5,
The contrast action length is:
As an indicator related to the length of pixels in a specific direction with the same intensity value within the region of interest image;
Gray intensity action distribution (HighGrayLevelRunEmphasis), long action length distribution (LongRunEmphasis), association distribution of long action length distribution with high gray intensity distribution (LongRunHighGrayLevelEmphasis), association distribution of long action length distribution with low gray intensity distribution (LongRunLowGrayLevelEmphasis), low gray intensity distribution. Strength action distribution (LowGrayLevelRunEmphasis), run length entropy (RunEntropy), run length non-uniformity (RunLengthNonUniformity), run length normalized non-uniformity (RunLengthNonUniformityNormalized), run length density (RunPercentage), run length variance (RunVariance), short A carpal tunnel syndrome diagnostic system comprising at least one of an action length distribution (ShortRunEmphasis), an association distribution of a short action length distribution and a high gray intensity distribution (ShortRunHighGrayLevelEmphasis), and an association distribution of a short action length distribution and a low gray intensity distribution (ShortRunLowGrayLevelEmphasis). .
According to clause 5,
The gray level size areas are:
As an indicator for quantifying a pixel area with the same brightness value within the region of interest image;
HighGrayLevelZoneEmphasis, LargeAreaEmphasis, LargeAreaHighGrayLevelEmphasis, LargeAreaLowGrayLevelEmphasis, LowGrayLevelZoneEmphasis ), area non-uniformity (SizeZoneNonUniformity), area non-uniformity normalized (SizeZoneNonUniformityNormalized), small area distribution (SmallAreaEmphasis), correlation distribution of small area and high gray intensity area (SmallAreaHighGrayLevelEmphasis), small area and low gray intensity area A carpal tunnel syndrome diagnosis system comprising at least one of an association distribution (SmallAreaLowGrayLevelEmphasis), area entropy (ZoneEntropy), area density (ZonePercentage), and area variance (ZoneVariance).
According to clause 5,
The ambient gray tint difference is:
As an index for quantifying the difference between pixels and brightness within a certain distance within the image of the region of interest;
Diagnosis of carpal tunnel syndrome, including at least one of the intensity change to adjacent pixels (Busyness), the average difference between the center pixel and neighboring pixels (Coarseness), the complexity of the gray intensity change in the image, and the size of the gray intensity change (Strength). system.
A method of using a carpal tunnel syndrome diagnosis system including a measurement unit and a diagnosis unit,
setting a region of interest to be used for diagnosing carpal tunnel syndrome in a muscle ultrasound image through the measurement unit and measuring muscle indices from the region of interest; and
Inputting the muscle ultrasound image and the muscle index into a carpal tunnel syndrome diagnosis model through the diagnosis unit and outputting a carpal tunnel syndrome diagnosis result for the muscle ultrasound image,
The steps for printing the carpal tunnel syndrome diagnosis result are:
Carpal tunnel comprising the step of selecting a weighted muscle index, which is a muscle index of relatively high importance when diagnosing carpal tunnel syndrome, through at least one of a recursive feature elimination method or an F-test method for each of the muscle indexes measured by the measurement unit. How to use the syndrome diagnosis system.
In a computer-readable recording medium,
A recording medium on which a program for executing the method of using the carpal tunnel syndrome diagnosis system according to paragraph 12 on a computer is recorded.