Computer tomography image analysis device and analysis method using the same

Abstract

To provide a computer tomography image analysis device capable of simply and accurately predicting the presence or absence of invasiveness of a lesion based on the proportion of the area in which the Hounsfield value exceeds a reference value out of the total volume of the lesion in a computer tomography image, and an analysis method using the same. [Solution] A computed tomography image analysis device and an analysis method using the same are provided, which include a parameter calculation unit that calculates a reference value for the Hounsfield value based on a histogram showing the distribution of the volume of the lesion in the computed tomography image and the Hounsfield value, an excess rate calculation unit that calculates an excess rate corresponding to the proportion of the area where the Hounsfield value exceeds the reference value out of the total volume of the lesion in the computed tomography image, and a prediction value calculation unit that calculates a prediction value regarding the presence or absence of invasiveness of the lesion based on whether the excess rate exceeds a cutoff value on a standard ROC curve for the reference value.

Description

[Technical Field]

[0001] The present invention relates to a computed tomography image analyzing device and an analysis method using the same, and more particularly to a computed tomography image analyzing device and an analysis method using the same that can easily and accurately predict the presence or absence of invasiveness of a lesion based on the proportion of the area in which the Hounsfield value exceeds a reference value to the total volume of the lesion in a computed tomography image. [Background technology]

[0002] Recently, with the widespread use of computed tomography (CT) scans for the early detection of lung cancer, the rate at which lesions appearing as ground-glass opacities (GGO) are detected through medical imaging tests such as chest tomography has been continuously increasing.

[0003] Ground-glass nodules are round, nodular ground-glass opacities. Depending on whether or not solid components are present inside, they can be classified as part-solid ground-glass nodules (pGGNs) or pure ground-glass nodules (pGGNs).

[0004] Homogeneous ground-glass opacity nodules are divided into non-invasive adenocarcinoma and invasive adenocarcinoma, and treatment methods vary depending on whether the tumor is invasive or not.

[0005] Since homogeneous ground-glass nodules do not show solid components in the imaging window settings for observing the lungs and mediastinum, the Hounsfield (HU) value, which corresponds to the brightness value of the lesion in the computed tomography image, is relatively uniform. Therefore, it is difficult to predict whether a lesion is invasive or not based solely on the simple Hounsfield (HU) value in the computed tomography image.

[0006] Recently, it has been confirmed that the size, volume, and shape of ground-glass nodules are known to be the main characteristics of computed tomography images, but when the lesion size is small, there is a low correlation between these computed tomography image characteristics and the invasiveness of the lesion.Furthermore, when the size of ground-glass nodules is small, there is a low correlation between the mean, variance, and maximum Hounsfield value of the lesion and the invasiveness of the lesion. [Prior art documents] [Patent documents]

[0007] [Patent Document 1] Republic of Korea Patent Registration No. 10-2035381 (Publication Date: October 22, 2019) Summary of the Invention [Problem to be solved by the invention]

[0008] The problem to be solved by the present invention is to provide a computer tomography image analyzer that can easily and accurately predict the presence or absence of invasiveness of a lesion based on the proportion of the area in which the Hounsfield value exceeds the reference value out of the total volume of the lesion in a computer tomography image, and an analysis method using the same.

[0009] The problems to be solved by the present invention are not limited to the above problems, and other problems not mentioned will be clearly understood by a person having ordinary skill in the art to which the present invention pertains from the following description. [Means for solving the problem]

[0010] In order to solve the above problems, the present invention provides a computed tomography image analyzing device, which includes: a parameter calculation unit that calculates a reference value for the Hounsfield value based on a histogram showing the distribution of the volume of the lesion in the computed tomography image and the Hounsfield value; an excess rate calculation unit that calculates an excess rate corresponding to the proportion of the area where the Hounsfield value exceeds the reference value among the total volume of the lesion in the computed tomography image; and a predicted value calculation unit that calculates a predicted value regarding the presence or absence of invasiveness of the lesion based on whether the excess rate exceeds a cutoff value on a standard ROC curve for the reference value.

[0011] The excess rate calculation unit can calculate the excess rate using the following [Equation 1].

[0012] [Number 1] TIFF0007754552000001.tif13166

[0013] γ means the exceedance rate, and V H means the volume of the area of ​​the lesion where the Hounsfield value is greater than the reference value, and V L means the volume of the area of ​​the lesion where the Hounsfield value is smaller than the reference value.

[0014] The parameter calculation unit can calculate, as the cutoff value, a candidate cutoff value that maximizes a Youden function among the candidates for cutoff values ​​on the reference ROC curve.

[0015] Furthermore, the parameter calculation unit can select a plurality of reference value candidates related to the reference value on the histogram, calculate a respective exceedance rate candidate value for each of the plurality of reference value candidates using the above [Equation 1], generate candidate ROC curves for each of the exceedance rate candidate values, and then select, from among the respective exceedance rate candidate values, the reference value candidate that maximizes the area under the curve for the candidate ROC curve as the reference value.

[0016] The predicted value calculation unit can calculate the accuracy of the predicted value using the following [Equation 2] and [Equation 3].

[0017] [Number 2] TIFF0007754552000002.tif13166

[0018] [Number 3] TIFF0007754552000003.tif13166

[0019] P I is the invasive accuracy, P nI means non-invasive accuracy, and P r [H i = 1, γ] means the probability value of invasiveness when the lesion is invasive for the excess rate (γ), and P r [H i = 0, γ] means the probability value of non-invasiveness that corresponds to the case where the lesion is non-invasive for the excess rate (γ).

[0020] In addition, the parameter calculation unit can calculate a modeling graph modeled based on the empirical joint probability distribution of the actual invasiveness of the lesion and the actual exceedance rate, and calculate, on the modeling graph, the probability value of invasiveness corresponding to when the lesion is invasive and the probability value of non-invasiveness corresponding to when the lesion is non-invasive for the exceedance rate, and transmit these to the prediction value calculation unit.

[0021] Furthermore, when a new computed tomography image is input from the outside, the parameter calculation unit can update the reference value and the cutoff value based on the new computed tomography image.

[0022] In addition, the parameter calculation unit can newly update the modeling graph based on the actual pathological value for the invasiveness of the lesion confirmed from the previous computed tomography image, and newly update the probability value of invasiveness and the probability value of non-invasiveness based on the newly updated modeling graph.

[0023] According to another embodiment of the present invention, there is provided a method for analyzing a computed tomography image using the above-mentioned computed tomography image analyzing apparatus, the method comprising: a reference value calculating step in which the parameter calculating unit calculates the reference value based on a histogram showing the volume of the lesion and the distribution of the Hounsfield value; an exceedance rate calculating step in which the exceedance rate calculating unit calculates an exceedance rate corresponding to the proportion of the area where the Hounsfield value exceeds the reference value to the total volume of the lesion; and a predicted value calculating step in which the predicted value calculating unit calculates a predicted value regarding the presence or absence of invasiveness of the lesion based on whether the exceedance rate exceeds a cutoff value on a standard ROC curve for the reference value.

[0024] The CT image analysis method may further include a cutoff value calculation step in which the parameter calculation unit calculates the cutoff value.

[0025] In the cutoff value calculation step, the parameter calculation unit can select, as the cutoff value, a candidate cutoff value that maximizes a Youden function from among the candidate cutoff values ​​on the reference ROC curve.

[0026] The reference value calculation step can include the following steps: a histogram calculation step in which the parameter calculation unit displays a distribution of the lesion volume and the Hounsfield value as a histogram; a reference value candidate selection step in which the parameter calculation unit selects a plurality of reference value candidates related to the reference value on the histogram; a step in which the parameter calculation unit calculates each exceedance rate candidate value for each of the plurality of reference value candidates using the above [Equation 1]; a curve generation step in which the parameter calculation unit generates a candidate ROC curve for each of the exceedance rate candidate values; and a reference value selection step in which the parameter calculation unit selects, from each of the exceedance rate candidate values, a reference value candidate that maximizes the area under the curve for the candidate ROC curve as the reference value.

[0027] The computed tomography image analysis method may further include a probability value calculation step in which the parameter calculation unit calculates a modeling graph modeled based on the empirical joint probability distribution of the actual invasiveness of the lesion and the actual excess rate, and calculates, on the modeling graph, a probability value of invasiveness corresponding to the case where the lesion is invasive and a probability value of non-invasiveness corresponding to the case where the lesion is non-invasive, for the excess rate.

[0028] The computed tomography image analysis method may further include an accuracy calculation step in which the prediction value calculation unit calculates the accuracy of the prediction value based on the invasive probability value and the non-invasive probability value transmitted from the parameter calculation unit.

[0029] The method for analyzing computed tomography images may further include a learning step in which the parameter calculation unit updates the reference value and the cutoff value based on a new computed tomography image input from the outside.

[0030] In the learning step, the parameter calculation unit can newly update the modeling graph based on the actual pathological value for the invasiveness of the lesion confirmed from previous computed tomography images, and newly update the probability value of invasiveness and the probability value of non-invasiveness based on the newly updated modeling graph. [Effects of the Invention]

[0031] The computer tomography image analyzing device and the analysis method using the same according to the present invention have the advantage that the presence or absence of invasiveness of a lesion can be easily and accurately predicted based on whether the excess rate exceeds the cutoff value on a standard ROC curve for the standard value, by calculating an excess rate corresponding to the proportion of the area in which the Hounsfield value exceeds the standard value out of the total volume of the lesion in the computer tomography image.

[0032] In addition, the computer tomography image analyzing device and the analysis method using the same according to the present invention have the advantage of being able to further improve the accuracy of determining whether a newly input computer tomography image is invasive or not through a learning process using the probability values ​​of invasiveness and non-invasiveness that are updated based on the actual pathological value of the invasiveness of the lesion.

[0033] The effects of the present invention are not limited to the above effects, but include all effects that can be inferred from the configuration of the invention described in the detailed description of the present invention or the claims. [Brief explanation of the drawings]

[0034] [Figure 1] 1 is a block diagram showing the structure of a computer tomography image analysis device according to the present invention; [Figure 2] FIG. 1 shows a histogram of Hounsfield values ​​for uniform ground-glass opacity nodules. [Figure 3] FIG. 1 is a diagram showing an example of an ROC curve with sensitivity and specificity based on excess rate as axes. [Figure 4] 10 is a graph comparing the areas under the ROC curves for candidate reference values. [Figure 5] FIG. 1 shows the value of Youden's function with sensitivity and specificity as variables for possible cutoff values ​​in the ROC curve. [Figure 6] FIG. 10 is a graph modeling the empirical joint probability distribution of the invasiveness or non-invasiveness of a lesion and the excess rate. DETAILED DESCRIPTION OF THE INVENTION

[0035] Hereinafter, preferred embodiments of the present invention that specifically achieve the above-mentioned object will be described with reference to the accompanying drawings. In describing the present embodiments, the same components are designated by the same names and reference numerals, and therefore, additional explanations will be omitted below.

[0036] Throughout this specification, when a part is said to be "connected (connected, contacted, or coupled)" to another part, this includes not only "directly connected" but also "indirectly connected" via another member therebetween. Furthermore, when a part is said to "include" a certain component, this does not mean that it excludes other components, but that it may further include other components, unless otherwise specified.

[0037] The terms used in this specification are merely used to describe specific embodiments and are not intended to limit the present invention. The singular expressions include the plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprise" or "have" specify the presence of features, numbers, steps, operations, components, parts, or combinations thereof described in the specification, and do not preclude the possibility of the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0038] The present invention can be used to predict the invasiveness of pathological lesions using features extracted from computed tomography images before surgery. The lesions referred to in this invention include homogeneous ground-glass nodules, and images of homogeneous ground-glass nodules can be obtained from three-dimensional computed tomography images of the chest.

[0039] A computer tomography image analyzing apparatus and an analyzing method using the same according to an embodiment of the present invention will be described below with reference to FIGS.

[0040] The computer tomography image analysis apparatus according to this embodiment may include an exceedance rate calculation unit 100, a predicted value calculation unit 200, and a parameter calculation unit 300.

[0041] The parameter calculation unit 300 calculates a reference value (θ) for the Hounsfield value based on a histogram showing the distribution of the volume of the lesion and the Hounsfield value in the computed tomography image. HU ) is calculated. The parameter calculation unit 300 calculates the reference value (θ HU The process of calculating ) is as follows:

[0042] First, when a DICOM (Digital Imaging and Communication in Medicine) file for a lesion is transmitted from the outside, the parameter calculation unit 300 can acquire the volume of the lesion and the distribution of Hounsfield values ​​of a computed tomography (CT) image in the form of a histogram, as shown in Figure 2. That is, the parameter calculation unit 300 can display the distribution of the volume of the lesion and the Hounsfield values ​​as a histogram.

[0043] Here, DICOM refers to a standard used when expressing digital image data in medical equipment or communicating using digital image data. Figure 2 shows the histogram of the reference value (θHU ) is -300HU.

[0044] Then, the parameter calculation unit 300 calculates the reference value (θ HU 2, in this embodiment, the reference value candidates can be selected from the range of Hounsfield values ​​that appear on the histogram, that is, from -1000 HU to 100 HU.

[0045] Preferably, the reference value candidates can be selected within a range in which high Hounsfield values ​​within the lesion are distributed on average. For example, the parameter calculation unit 300 can select the reference value candidates based on a range in which previous reference values ​​selected from previous pre-stored computed tomography images are distributed. In this embodiment, based on Figure 2, the reference value candidates are selected from a range of -500 HU to 100 HU, and the interval between the reference value candidates is set to 50.

[0046] Next, the parameter calculation unit 300 calculates each exceedance rate candidate value for each reference value candidate using [Equation 1].

[0047] The parameter computation unit 300 then generates a candidate ROC curve for each candidate rate of exceedance.

[0048] Figure 3 shows an example of a candidate ROC curve with axes of sensitivity and specificity (1-specificity) for candidate excess rates when classifying lesions into invasive and non-invasive. Referring to Figure 3, it can be seen that the candidate ROC curve is in the upper region of the reference line.

[0049] Next, the parameter calculation unit 300 selects, as the reference value, the reference value candidate that maximizes the area under the curve (AUC) for the candidate ROC curve from among the exceedance rate candidate values.

[0050] 4 is an example of a graph comparing the area under the curve (AUC) of candidate ROC curves for candidate exceedance rates. Referring to FIG. 4, in this embodiment, −300 HU, which maximizes the area under the curve (AUC), is selected as the reference value.

[0051] On the other hand, the excess rate calculation unit 100 calculates the Hounsfield value of the total volume of the lesion in the computed tomography image of the input lesion by the reference value (θ HU ) can be calculated as the excess rate (γ).

[0052] The excess rate (γ) can be calculated by the following [Equation 1].

[0053] [Number 1] TIFF0007754552000004.tif13166

[0054] where γ is the excess rate and V H is the Hounsfield value of the lesions compared to the reference value (θ HU ), and V L is the Hounsfield value of the lesions compared to the reference value (θ HU ) means the volume of an area smaller than

[0055] Also, V H and V L is obtained from a graph showing the distribution of the volume and Housefield value for the lesion in the form of a histogram. For example, in the histogram shown in Figure 2, the reference value (θ HU ) as a reference, V H is the Hounsfield value relative to the reference value (θ HU ) corresponds to the area under the graph in the area larger than V L is the Hounsfield value relative to the reference value (θ HU ) corresponds to the area under the graph in an area smaller than

[0056] The excess rate calculation unit 100 calculates the reference value (θ HU ) is transmitted from the parameter calculation unit 300, and the exceedance rate (γ) can be calculated by [Equation 1].

[0057] TIFF0007754552000005.tif32166

[0058] The predicted value calculation unit 200 calculates the excess rate (γ) based on the cutoff value (θ γ ) or more, the predicted value is output as 1, and the excess rate (γ) is the cutoff value (θ γ ), the predicted value can be output as 0.

[0059] TIFF0007754552000006.tif20166

[0060] The predicted value calculation unit 200 calculates the cutoff value (θ γ ) is transmitted from the parameter calculation unit 300.

[0061] The parameter calculation unit 300 calculates the cutoff value (θ γ The process of calculating ) is as follows:

[0062] The parameter calculation unit 300 calculates Youden function values ​​for all cutoff value candidates on the reference ROC curve for the selected reference value. The Youden function values ​​are calculated using Youden's Index, so detailed description thereof will be omitted.

[0063] The parameter calculation unit 300 selects a candidate cutoff value that maximizes the Youden function value as the cutoff value (θ γ ) Referring to FIG. 5, in this embodiment, 5.41% is selected as the cutoff value. On the other hand, the predicted value calculation unit 200 calculates the excess rate (γ) and the cutoff value (θ γ ) and the accuracy of the predicted value corresponding to the difference value can be calculated.

[0064] The accuracy of the predicted value is determined by the excess rate (γ) and the cutoff value (θ γ ) or more for the accuracy of invasiveness (P I ) and the excess rate (γ) is the cutoff value (θ γ ) for non-invasive accuracy (P nI ) is included.

[0065] Invasive accuracy (P I ) is calculated by [Equation 2], and the non-invasive precision (P nI ) is calculated using [Equation 3].

[0066] [Number 2] TIFF0007754552000007.tif13166

[0067] [Number 3] TIFF0007754552000008.tif13166

[0068] where P r [H i = 1, γ] means the probability value of invasiveness when the lesion is invasive for the excess rate (γ), and P r [H i = 0, γ] means the probability value of non-invasiveness that corresponds to the case where the lesion is non-invasive for the excess rate (γ).

[0069] TIFF0007754552000009.tif20166

[0070] Invasive accuracy (P I ) is the excess rate (γ) when the cutoff value (θ γ ), that is, the greater the exceedance rate (γ) and the cutoff value (θ γ ) the greater the difference, the closer it is to 100, and the more accurate the invasiveness (P I ) increases. Also, the non-invasive accuracy (P nI ) is the excess rate (γ) when the cutoff value (θ γ ), that is, the greater the excess rate (γ) and the cutoff value (θγ ) the greater the difference, the closer it is to 100, and the greater the accuracy of non-invasiveness (P nI ) increases.

[0071] The prediction value calculation unit 200 calculates the probability value of invasiveness (P r [H i = 1, γ]) and the probability value of non-invasiveness (P r [H i =0, γ]) is transmitted from the parameter calculation unit 300.

[0072] The process by which the parameter calculation unit 300 calculates the probability value of invasiveness and the probability value of non-invasiveness will be described as follows.

[0073] The probability value of invasiveness (P r [H i = 1, γ]) and the probability value of non-invasiveness (P r [H i = 0, γ) can be obtained from a modeling graph based on the empirical joint probability distribution of the actual presence or absence of invasiveness for the lesion and the actual excess rate.

[0074] First, the parameter calculation unit 300 calculates a modeling graph based on the empirical joint probability distribution of the actual presence or absence of invasiveness and the actual excess rate of the lesion.

[0075] As shown in Figure 6, the modeling graph is a graph obtained by curve fitting based on the actual values ​​for the presence or absence of invasiveness of the lesion obtained from the actual pathological results and the excess rate values ​​at that time.

[0076] The empirical joint probability distribution satisfies the following [Equation 4], [Equation 5], and [Equation 6].

[0077] [Number 4] TIFF0007754552000010.tif13166

[0078] [Number 5] TIFF0007754552000011.tif13166

[0079] [Number 6] TIFF0007754552000012.tif13166

[0080] where N t is the number of collected homogeneous ground-glass nodules, i.e., lesions, and N I and N nI are the numbers of lesions classified as invasive adenocarcinoma and non-invasive adenocarcinoma, respectively, among the collected lesions.

[0081] In FIG. 6, G1 is a curve showing the joint probability distribution according to the excess rate when the lesion is invasive, and G0 is a curve showing the joint probability distribution according to the excess rate when the lesion is non-invasive.

[0082] Then, the parameter calculation unit 300 calculates the probability value of invasiveness (P) when the lesion is invasive against the exceedance rate (γ) on the modeling graph. r [H i = 1, γ]) and the probability of non-invasiveness when the lesion is non-invasive (P r [H i =0,γ]).

[0083] As shown in Figure 6, when the value of exceedance rate (γ) is α, the probability value of invasiveness (P r [H i =1,γ]) is the value corresponding to α in the G1 graph, and the probability value of non-invasiveness (P r [H i = 0, γ]) is calculated as the value corresponding to α in the G0 graph. Here, the probability value of invasiveness (P r [H i = 1, γ]) and the probability value of non-invasiveness (P r [H i =0, γ]) can have a value between 0 and 1.

[0084] The parameter calculation unit 300 calculates the calculated probability value of invasiveness (P r [Hi = 1, γ]) and the probability value of non-invasiveness (P r [H i =0, γ]) to the prediction calculation unit 200.

[0085] As a result, the value of the exceedance rate (γ) is γ ) is greater than the invasive precision (P I ) is output as a value close to 100, and the excess rate (γ) is the cutoff value (θ γ ), the accuracy of invasiveness (P I ) is output as a value of approximately 50. In this way, the accuracy of the predicted value is output as a percentile, improving the reliability of the predicted value for the invasiveness of the lesion.

[0086] Meanwhile, when a new CT image is input from the outside, the parameter calculation unit 300 can update the reference value and the cutoff value based on the new CT image.

[0087] The parameter calculation unit 300 transmits the newly updated reference value to the exceedance rate calculation unit 100 to which the new computed tomography image has been input, and also transmits the newly updated cutoff value to the prediction value calculation unit 200.

[0088] In addition, the parameter calculation unit 300 can newly update the modeling graph based on the actual pathological value for the invasiveness of the lesion confirmed in the previous step. For example, the G1 curve and the G0 curve in Fig. 6 can be newly set by adding the pathological result for the presence or absence of invasiveness of the new lesion.

[0089] The parameter calculation unit 300 can newly update the probability value of invasiveness and the probability value of non-invasiveness based on the newly updated modeling graph and transmit them to the prediction value calculation unit 200.

[0090] The process of analyzing a computer tomography image using the above-described computer tomography image analysis apparatus will be described below.

[0091] First, the parameter calculation unit 300 performs a reference value calculation step in which the parameter calculation unit 300 calculates a reference value based on a histogram showing the distribution of the volume and Hounsfield value of a lesion in a computed tomography image.

[0092] When the computed tomography image of the lesion is input to the parameter calculation unit 300, the parameter calculation unit 300 calculates the reference value (θ HU ) and transmits the calculated reference value to the exceedance rate calculation unit 100.

[0093] In the reference value calculation step, the parameter calculation unit 300 first performs a histogram calculation step in which the distribution of the lesion volume and Hounsfield value is displayed as a histogram.

[0094] Next, the parameter calculation unit 300 performs a reference value candidate selection step of selecting a plurality of reference value candidates related to the reference value on the histogram.

[0095] Next, the parameter calculation unit 300 calculates each exceedance rate candidate value for each of the plurality of reference value candidates using the above [Equation 1].

[0096] Next, the parameter calculation unit 300 performs a curve generation step to generate candidate ROC curves for each of the candidate exceedance rates.

[0097] Next, the parameter calculation unit 300 performs a reference value selection step of selecting, as the reference value, a reference value candidate that maximizes the area under the curve for the candidate ROC curve from among the exceedance rate candidate values.

[0098] In addition, the parameter calculation unit 300 calculates the reference value (θ HU) based on the reference ROC curve for the cutoff value (θ γ ) and transmits the calculated cutoff value to the predicted value calculation unit 200. That is, the parameter calculation unit 300 performs a cutoff value calculation step in which the cutoff value is calculated.

[0099] In the cutoff value calculation step, a candidate cutoff value that maximizes the Youden function is selected as the cutoff value from among the candidates for cutoff values ​​on the reference ROC curve.

[0100] After the calculation process for the reference value and the cutoff value is completed, the excess rate calculation unit 100 calculates the excess rate (γ), which corresponds to the proportion of the area in which the Hounsfield value exceeds the reference value among the total volume of the lesion in the computed tomography image of the lesion, using [Equation 1].

[0101] Next, the prediction value calculation unit 200 calculates the cutoff value (θ) on the reference ROC curve relative to the reference value by calculating the overshoot rate (γ). γ A predicted value calculation step is performed in which a predicted value regarding the presence or absence of invasiveness of the lesion is calculated based on whether or not the predicted value exceeds the threshold value.

[0102] As a result, the present invention has the advantage of being able to easily and accurately predict the presence or absence of invasiveness of ground-glass opacity nodules with little solid component based on the distribution map of Hounsfield values ​​obtained by image analysis of three-dimensional chest computed tomography images.

[0103] Next, the parameter calculation unit 300 performs a probability value calculation step to calculate a probability value of invasiveness and a probability value of non-invasiveness, which are used to calculate the accuracy of the predicted value.

[0104] In the probability value calculation step, the parameter calculation unit 300 first performs a graph calculation step of calculating a modeling graph modeled based on the empirical joint probability distribution of the actual presence or absence of invasiveness of the lesion and the actual excess rate.

[0105] Next, the parameter calculation unit 300 performs a probability value calculation step to calculate, on the modeling graph, the probability value of invasiveness when the lesion is invasive and the probability value of non-invasiveness when the lesion is non-invasive, relative to the exceedance rate.

[0106] The parameter calculation unit 300 then transmits the probability value of invasiveness and the probability value of non-invasiveness to the prediction value calculation unit 200 .

[0107] Thereafter, the prediction value calculation unit 200 performs an accuracy calculation step in which it calculates the accuracy of the prediction value based on the invasive probability value and non-invasive probability value transmitted from the parameter calculation unit 300 using [Equation 2] and [Equation 3].

[0108] The accuracy of the predicted value may include accuracy of invasiveness when the excess rate is equal to or greater than a cutoff value, and accuracy of non-invasiveness when the excess rate is less than the cutoff value. Of course, the present invention is not limited to this, and the predicted value calculation step and the accuracy calculation step may be performed simultaneously.

[0109] Meanwhile, the CT image analysis method may further include a learning step in which the parameter calculation unit 300 updates the reference value and the cutoff value based on a new CT image input from the outside.

[0110] In the learning step, the parameter calculation unit can newly update the modeling graph based on the actual pathological value for the invasiveness of the lesion confirmed from previous computed tomography images, and newly update the probability value of invasiveness and the probability value of non-invasiveness based on the newly updated modeling graph.

[0111] As a result, through a learning process using updated probability values ​​of invasiveness and non-invasiveness based on the actual pathological values ​​of the lesions confirmed from previous computed tomography images, the accuracy of determining whether or not a lesion is invasive for newly input computed tomography images can be further improved.

[0112] As described above, the present invention is not limited to the specific preferred embodiments described above, and various modifications can be made by a person having ordinary skill in the art to which the invention pertains without departing from the gist of the present invention as claimed in the claims, and such modifications fall within the scope of the present invention. [Explanation of symbols]

[0113] 100: Excess rate calculation unit 200: Prediction calculation unit 300: Parameter calculation unit

Claims

1. a parameter calculation unit for calculating a reference value for the Hounsfield value based on a histogram showing the distribution of the volume of the lesion and the Hounsfield value in the computed tomography image; an excess rate calculation unit for calculating an excess rate corresponding to a rate of an area in which the Hounsfield value exceeds the reference value in the total volume of the lesion in the computed tomography image; a prediction value calculation unit that calculates a prediction value regarding the presence or absence of invasiveness of the lesion based on whether the excess rate exceeds a cutoff value on a reference ROC curve for the reference value; The exceedance rate calculation unit is the following [Equation 1]: [Equation 1] (γ means the excess rate, V Ｈ means the volume of the area in the lesion where the Hounsfield value is greater than the reference value, and V Ｌ indicates the volume of the area of ​​the lesion where the Hounsfield value is below the reference value. Calculate the excess rate using A computer tomography image analysis device characterized by:

2. The parameter calculation unit calculates, as the cutoff value, a candidate cutoff value at which a Youden function is maximized for the candidates of cutoff values ​​on the reference ROC curve.

2. A computer tomography image analysis system according to claim 1.

3. the parameter calculation unit selects a plurality of reference value candidates related to the reference value on the histogram; Calculating each excess rate candidate value for each of the plurality of reference value candidates using the above [Equation 1]; After generating candidate ROC curves for each of the exceedance rate candidate values, a reference value candidate that maximizes the area under the candidate ROC curve is selected as the reference value from among the exceedance rate candidate values.

3. A computer tomography image analysis system as claimed in claim 2.

4. The predicted value calculation unit is configured to calculate the predicted value using the following [Equation 2] and [Equation 3]: [Equation 2] [Equation 3] (P Ｉ is the accuracy of invasiveness, P ｎＩ means non-invasive accuracy, and P ｒ [H ｉ = 1, γ] means the probability value of invasiveness that corresponds to the case where the lesion is invasive for the excess rate (γ), and P ｒ [H ｉ = 0, γ] means the probability value of non-invasiveness that corresponds to the case where the lesion is non-invasive for the excess rate (γ)) Calculate the accuracy of the predicted value using 2. A computer tomography image analysis system according to claim 1.

5. The parameter calculation unit calculates a modeling graph based on an empirical joint probability distribution of the actual presence or absence of invasiveness and the actual excess rate of the lesion; On the modeling graph, a probability value of invasiveness corresponding to when the lesion is invasive and a probability value of non-invasiveness corresponding to when the lesion is non-invasive are calculated for the excess rate, and transmitted to the prediction value calculation unit.

5. A computer tomography image analysis system according to claim 4.

6. When a new computed tomography image is input from the outside, the parameter calculation unit updates the reference value and the cutoff value based on the new computed tomography image.

4. A computer tomography image analysis system according to claim 3.

7. The parameter calculation unit updates the modeling graph based on the actual pathological values ​​for the invasiveness of the lesion confirmed from the previous computed tomography images; The probability value of invasiveness and the probability value of non-invasiveness are updated based on the newly updated modeling graph.

6. A computer tomography image analysis system according to claim 5.

8. 2. A method for analyzing a computed tomography image using the computed tomography image analysis apparatus of claim 1, comprising: a reference value calculation step in which the parameter calculation unit calculates the reference value based on a histogram showing the distribution of the volume of the lesion and the Hounsfield value; an excess rate calculation step in which the excess rate calculation unit calculates an excess rate corresponding to a ratio of an area in which the Hounsfield value exceeds the reference value to the total volume of the lesion; a prediction value calculation step in which the prediction value calculation unit calculates a prediction value regarding the presence or absence of invasiveness of the lesion based on whether the excess rate exceeds a cutoff value on a reference ROC curve for the reference value. A method for analyzing computed tomography images, comprising:

9. a cutoff value calculation step in which the parameter calculation unit calculates the cutoff value; In the cutoff value calculation step, the parameter calculation unit selects, as the cutoff value, a candidate cutoff value that maximizes a Youden function from among the candidates for cutoff values ​​on the reference ROC curve.

9. The method of claim 8, wherein the method comprises:

10. The reference value calculation step a histogram calculation step in which the parameter calculation unit displays a distribution of the volume of the lesion and the Hounsfield value as a histogram; a reference value candidate selection step in which the parameter calculation unit selects a plurality of reference value candidates related to the reference value on the histogram; the parameter calculation unit calculating each exceedance rate candidate value for each of the plurality of reference value candidates using the above [Equation 1]; a curve generation step in which the parameter calculation unit generates a candidate ROC curve for each of the exceedance rate candidate values; a reference value selection step in which the parameter calculation unit selects, as the reference value, a reference value candidate that maximizes the area under the curve for the candidate ROC curve from among the exceedance rate candidate values.

9. The method of claim 8, wherein the method comprises:

11. The parameter calculation unit calculates a modeling graph modeled based on an empirical joint probability distribution of the actual presence or absence of invasiveness and the actual excess rate of the lesion; The method further includes a probability value calculation step of calculating, on the modeling graph, a probability value of invasiveness corresponding to when the lesion is invasive and a probability value of non-invasiveness corresponding to when the lesion is non-invasive, for the exceedance rate.

9. The method of claim 8, wherein the method comprises:

12. The prediction value calculation unit further includes an accuracy calculation step of calculating accuracy of the prediction value based on the invasive probability value and the non-invasive probability value transmitted from the parameter calculation unit.

12. The method of claim 11, wherein the method comprises:

13. The parameter calculation unit further includes a learning step of updating the reference value and the cutoff value based on a new computed tomography image input from the outside.

12. The method of claim 11, wherein the method comprises:

14. In the learning step, the parameter calculation unit newly updates the modeling graph based on the actual pathological value for the invasiveness of the lesion confirmed from the previous computed tomography image, and newly updates the probability value of invasiveness and the probability value of non-invasiveness based on the newly updated modeling graph.

14. The method of claim 13, wherein the method comprises:

Images