KR102634008B1 - Method and apparatus for predicting early neurological deterioration of patients using multiple artificial intelligence models - Google Patents

Abstract

According to one embodiment of the present invention, a method of predicting a patient's END (Early Neurological Deterioration) using a plurality of artificial intelligence models includes obtaining the original brain image and clinical information of the patient from a server. step; Inputting the original brain image into a first artificial intelligence model that performs lesion analysis on the brain image and outputting a lesion image of the original brain image; Inputting the original brain image and the lesion image into a second artificial intelligence model to output a first probability value, which is the probability of occurrence of the END of the patient; Inputting the clinical information into a third artificial intelligence model to output a second probability value, which is the probability of occurrence of the END of the patient; And calculating a final probability value that is the probability of occurrence of the END of the patient based on the first probability value and the second probability value. A method for predicting END, including a step, may be provided.

Description

{Method and apparatus for predicting early neurological deterioration of patients using multiple artificial intelligence models}

The present invention relates to a method and device for predicting early neurological deterioration (END) in a patient using multiple artificial intelligence models. More specifically, by inputting different information into multiple artificial intelligence models, It relates to a method and device for predicting early neurological deterioration of a patient using the obtained outputs.

Early neurological deterioration (hereinafter referred to as 'END') occurs in approximately 15% of neurological patients, and approximately 75% of patients who develop END become patients with mobility difficulties who are unable to perform daily activities on their own in the future.

Although it is well known that predicting END in neurological patients is significantly helpful in improving the patient's prognosis, the specific numerical correlation between the causative factors causing END and the actual probability of occurrence has not been revealed, so clinical medical staff There were limitations that could only be judged empirically, and it was impossible to give a detailed explanation to the patient. Therefore, the act of predicting END was not an officially recognized medical practice.

In addition, even if existing clinical studies on END use only the patient's brain imaging information to predict, use only clinical information to predict, or use clinical and brain imaging information together, they provide limited information that is quantified by simply interpreting the image information. Only used.

Accordingly, the need for technology that can more accurately predict the occurrence of a patient's END using the patient's brain imaging information and clinical information is increasing compared to conventional technology.

Korean Patent Publication No. 10-2020157 (2019.09.03)

The present invention was created to solve the above problems, and the purpose of the present invention is to provide a method and device that provides a real-time predicted probability of END occurrence and can easily convey a dangerous situation to the patient using this. It is for.

In order to achieve the above object, a method of predicting a patient's END (Early Neurological Deterioration) using a plurality of artificial intelligence models according to an embodiment of the present invention includes the patient's original brain image and Obtaining clinical information; Inputting the original brain image into a first artificial intelligence model that performs lesion analysis on the brain image and outputting a lesion image of the original brain image; Inputting the original brain image and the lesion image into a second artificial intelligence model to output a first probability value, which is the probability of occurrence of the END of the patient; Inputting the clinical information into a third artificial intelligence model to output a second probability value, which is the probability of occurrence of the END of the patient; and calculating a final probability value, which is the probability of occurrence of the END of the patient, based on the first probability value and the second probability value.

In relation to this, the step of outputting the lesion image includes converting the original brain image into a two-dimensional original brain image; performing lesion analysis on the two-dimensional original brain image using the first artificial intelligence model to detect lesion location and lesion shape characteristics; and outputting a 3D lesion image displaying characteristics of the detected lesion location and shape of the lesion.

Meanwhile, the step of outputting the first probability value includes generating a synthetic input image by splicing the lesion image corresponding to the original brain image to the original brain image; Inputting the synthesized input image into the second artificial intelligence model and outputting the first probability value as a first END occurrence probability value and a first END non-occurrence probability value; and normalizing the sum of the first END occurrence probability value and the first END non-occurrence probability value to be 1.

Subsequently, calculating the second probability value includes inputting the clinical information into the third artificial intelligence model and outputting the second probability value as a second END occurrence probability value and a second END non-occurrence probability value; and normalizing the sum of the second END occurrence probability value and the second END non-occurrence probability value to be 1.

Meanwhile, calculating the final probability value includes calculating the final END occurrence probability value as a weighted average value of the first END occurrence probability value and the second END occurrence probability value; and calculating a final END non-occurrence probability value as a weighted average value of the first END non-occurrence probability value and the second END non-occurrence probability value.

A device for predicting a patient's END (Early Neurological Deterioration) using a plurality of artificial intelligence models according to another embodiment of the present invention includes a communication unit that obtains the patient's original brain image and clinical information from a server. ; a first analysis unit that inputs the original brain image into a first artificial intelligence model that performs lesion analysis on the brain image and outputs a lesion image of the original brain image; a second analysis unit that inputs the original brain image and the lesion image into a second artificial intelligence model and outputs a first probability value, which is the probability of occurrence of the END of the patient; a third analysis unit that inputs the clinical information into a third artificial intelligence model and outputs a second probability value, which is the probability of occurrence of the END of the patient; and a final probability value analysis unit that calculates a final probability value, which is the probability of occurrence of the END of the patient, based on the first probability value and the second probability value.

In relation to this, the first analysis unit converts the original brain image into a two-dimensional original brain image and performs lesion analysis on the two-dimensional original brain image through the first artificial intelligence model to determine the location of the lesion and the characteristics of the lesion shape. , and output a 3D lesion image displaying the detected location of the lesion and the characteristics of the lesion shape.

Meanwhile, the second analysis unit generates a synthetic input image by splicing the lesion image corresponding to the original brain image to the original brain image, and inputs the synthetic input image into the second artificial intelligence model to determine the first probability value. The 1st END occurrence probability value and the 1st END non-occurrence probability value can be output, and the sum of the 1st END occurrence probability value and the 1st END non-occurrence probability value can be normalized to be 1.

Subsequently, the third analysis unit inputs the clinical information into the third artificial intelligence model and outputs the second probability value as a second END occurrence probability value and a second END non-occurrence probability value, and the second END occurrence probability value and the first END occurrence probability value. 2 END The sum of the probability values of non-occurrence can be normalized to be 1.

Meanwhile, the final probability value analysis unit calculates the final END occurrence probability value as a weighted average value of the first END occurrence probability value and the second END occurrence probability value, and calculates the final END occurrence probability value and the first END non-occurrence probability value and the second END non-occurrence probability value. The final END non-occurrence probability value can be calculated using the weighted average value.

According to one aspect of the present invention described above, the predicted probability of occurrence of END can be provided in real time and the risk situation can be easily communicated to the patient using this.

In addition, using the predicted probability, risk situations can be explained to patients, consent to active intervention treatment can be obtained, and ultimately the prognosis of more patients can be improved compared to when the present invention is not used.

Figure 1 is a block diagram showing the schematic configuration of an END prediction device according to an embodiment of the present invention.
Figure 2 is a diagram schematically showing the structure of a first artificial intelligence model according to an embodiment of the present invention.
Figure 3 is a diagram schematically showing the structure of a first artificial intelligence model and a second artificial intelligence model according to an embodiment of the present invention.
Figure 4 is a diagram schematically showing the structure of a third artificial intelligence model according to an embodiment of the present invention.
Figure 5 is a diagram schematically showing the entire process of deriving the final probability value using the first to third artificial intelligence models according to an embodiment of the present invention.
Figure 6 is a flowchart showing each step in which the END prediction device operates according to an embodiment of the present invention.

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

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

Figure 1 is a block diagram showing the schematic configuration of an END prediction device 100 according to an embodiment of the present invention.

In the present invention, the description is focused on embodiments in which the END prediction device 100 independently predicts the probability of occurrence of END, but the embodiments can also be performed through linkage with a related server that operates to predict the probability of occurrence of END. there is.

That is, the END prediction device 100 and the related server according to one embodiment may be integrated and implemented in terms of their functions, and the related server may be omitted, and it can be seen that the END prediction device 100 and the related server according to one embodiment are not limited to any one embodiment.

The components of the END prediction device 100 shown in FIG. 1 may be connected wired or wirelessly to operate organically with each other, and may be provided inside or individually outside the END prediction device 100.

In addition, the END prediction device 100 may be equipped with a wireless communication device and a UAV (Unmanned Aerial Vehicle) that ensures portability and mobility, and such wireless communication devices include Personal Communication System (PCS) and Global System for GSM (GSM). Mobile communications), PDC (Personal Digital Celluar), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (W- All types of handhelds (Code Division Multiple Access), Wibro (Wireless Broadband Internet) terminals, smartphones, smartpads, tablet PCs, VR (Virtual Reality) devices, HMDs (Head Mounted Displays), etc. Handheld)-based wireless communication devices may be included, but are not limited to this.

Referring again to FIG. 1, the END prediction device 100 includes a communication unit 110, a first analysis unit 120, a second analysis unit 130, a third analysis unit 140, and a final probability value analysis unit 150. may include.

In one embodiment, the communication unit 110 may obtain the patient's original brain image and clinical information from the server. Here, the server may include a server that can provide the patient's medical information, such as a hospital server. Additionally, the original brain image may be an original, unprocessed image (e.g., Diffusion Weighted Image (DWI)) related to the patient's brain, and may display brain lesions. Clinical information may include the patient's age, BMI, fasting blood sugar, etc.

The first analysis unit 120 according to an embodiment includes a first artificial intelligence model (e.g., brain image-based lesion recognition deep neural network) that performs lesion analysis on brain images, and the first artificial intelligence model includes an original By inputting a brain image, a lesion image corresponding to the original brain image can be output.

Specifically, the first analysis unit 120 converts the original brain image into a two-dimensional original brain image, and then applies the two-dimensional original brain image to the first artificial intelligence model to analyze lesions on the converted two-dimensional original brain image. Video can be input. The first analysis unit 120 may detect the lesion location and lesion shape characteristics as an output of the first artificial intelligence model, and output a 3D lesion image displaying the detected lesion location and lesion shape characteristics. The number of output 3D lesion images may correspond to the number of original brain images.

The structure of the first artificial intelligence model of the first analysis unit 120 will be described with reference to FIG. 2.

Figure 2 is a diagram schematically showing the structure of the first artificial intelligence model 121 according to an embodiment of the present invention.

As shown in FIG. 2, the structure of the first artificial intelligence model 121 uses a 3D original brain image as input, unlike the conventional UNet structure that handles 2D input images or the 3DUNet structure that handles 3D input images. There is an abstraction process that extracts the features of the lesion location and lesion shape after compressing it into a two-dimensional structure inside the model. That is, unlike the existing UNet structure, the calculation process of the first artificial intelligence model 121 may involve converting the dimension of the input image from 3D to 2D and then from 2D to 3D. The first artificial intelligence model 121 is based on the characteristics of the lesion location and lesion shape compressed into a two-dimensional structure, and can finally output a three-dimensional lesion image in which the characteristics of the lesion location and lesion shape are displayed. Deep learning analysis of the first artificial intelligence model 121 may be performed through prior learning on learning information stored in the END prediction device 100 or a related server.

In addition, in the case of the existing UNet, since the input is one 2D image, the output is also one 2D image, but in the case of the original brain image, it is a 3D image composed of multiple images, so the dimensions of the input and output images are 3D. It can be increased. The hidden layer can process 3D images without significantly increasing the number of network parameters by using the 2D structure as is and making full connections to each other. As a result, it is possible to improve both performance and functionality by enabling 3D image calculation while securing calculation speed and memory usage efficiency.

The second analysis unit 130 according to one embodiment may input the original brain image and the lesion image into the second artificial intelligence model and output a first probability value, which is the probability of occurrence of the patient's END.

Specifically, the second analysis unit 130 may generate a synthetic input image by splicing the original brain image and the corresponding lesion image to the original brain image. The second analysis unit 130 may input the obtained synthetic input image into the second artificial intelligence model and output the first probability value as the first END occurrence probability value and the first END non-occurrence probability value. The second analysis unit 130 may normalize the output probability value of 1st END occurrence and the probability value of non-occurrence of 1st END so that the sum is 1.

This will be explained with reference to FIG. 3.

Figure 3 is a diagram schematically showing the structure of the first artificial intelligence model and the structure of the second artificial intelligence model 131 according to an embodiment of the present invention.

Referring to FIG. 3, the second artificial intelligence model 131 may use as input a composite input image based on the original brain image and the lesion image that is the output of the first artificial intelligence model. The second artificial intelligence model 131 may perform deep learning analysis on the synthetic input image and output the first END occurrence probability value and the first END non-occurrence probability value. The second artificial intelligence model 131 may be prepared through prior learning using learning data stored in the END prediction device 100 or a related server.

In addition, the conventional model for video analysis uses one image as input, but the second artificial intelligence model 131 has an input dimension so that the number of input images can be used as input as much as a preset number (e.g., 60). This is an improved model. In addition, the second artificial intelligence model 131 sets the output value to determine only two cases: when END occurs in the input synthetic input image (probability value of first END occurrence) and when it does not (probability value of non-occurrence of first END). It can be reduced compared to before.

Additionally, the second analysis unit 130 may normalize the sum of the first END occurrence probability value and the first END non-occurrence probability value to be 1. Accordingly, both the first END occurrence probability value and the first END non-occurrence probability value can be determined to be values between 0 and 1. For example, the probability value of occurrence of the first END may be determined to be 0.8, and the probability value of non-occurrence of the first END may be determined to be 0.2. In this case, the sum of the two values is always 1, so the first probability value, which is the probability value for predicting the occurrence of END, can be expressed as a relative number.

The third analysis unit 140 according to one embodiment may input clinical information into a third artificial intelligence model and output a second probability value, which is the probability of occurrence of the patient's END.

Specifically, the third analysis unit 140 may input the patient's clinical information into the third artificial intelligence model and output the second probability value as the second END occurrence probability value and the second END non-occurrence probability value. The third analysis unit 140 may normalize the output second END occurrence probability value and the second non-occurrence probability value so that the sum is 1.

This will be explained with reference to FIG. 4.

Figure 4 is a diagram schematically showing the structure of the third artificial intelligence model 141 according to an embodiment of the present invention.

Referring to FIG. 4, the third analysis unit 140 inputs clinical information including items such as age, BMI, and fasting blood sugar into the third artificial intelligence model 141 to obtain a second probability value as an output. . Here, clinical information may be composed of clinical information items corresponding to each patient. For example, if a patient satisfies a specific condition, clinical information including clinical information items corresponding thereto may be obtained. Specific conditions may include, for example, whether a patient hospitalized with a brain disease arrived within 7 days or later of the onset of disease-related symptoms, whether he or she received IV thrombolytics or intravascular treatment, etc., and may include specific conditions. Items included in clinical information may be determined depending on whether the conditions are satisfied. Deep learning analysis of the third artificial intelligence model 141 may be performed through prior learning on learning information stored in the END prediction device 100 or a related server.

In addition, the third artificial intelligence model 141 may be composed of a total of 10 hidden layers, an input layer, and an output layer, and may be a model that uses techniques such as batchnormalization, dropout (50% ratio), Xavier initialization, and Adam optimization. . Since clinical information can all consist of numbers, the accuracy of output values can be improved when using the above techniques.

The third artificial intelligence model 141 may output the probability of occurrence of the 2nd END and the probability of non-occurrence of the 2nd END as a result of analysis of clinical information. The third artificial intelligence model 141 can reduce the output value compared to the past to determine only two cases, such as when END occurs (probability value of second END occurrence) and when not (probability value of non-occurrence of second END).

Additionally, the third analysis unit 140 may normalize the sum of the second END occurrence probability value and the second END non-occurrence probability value to be 1. Accordingly, both the 2nd END occurrence probability value and the 2nd END non-occurrence probability value can be determined to be values between 0 and 1. For example, the probability value of occurrence of the 2nd END may be determined to be 0.4, and the probability value of non-occurrence of the 2nd END may be determined to be 0.6. In this case, the sum of the two values is always 1, so the second probability value, which is the probability value for predicting the occurrence of END, can be expressed as a relative number.

The final probability value analysis unit 150 according to one embodiment may calculate a final probability value, which is the probability of occurrence of END of the patient, based on the first probability value and the second probability value.

The process by which the final probability value analysis unit 150 calculates the final probability value will be described with reference to FIG. 5.

Figure 5 is a diagram schematically showing the entire process of deriving the final probability value using the first artificial intelligence model 121 to the third artificial intelligence model 141 according to an embodiment of the present invention.

Referring to FIG. 5, the final probability value analysis unit 150 calculates the final probability value using the first probability value obtained in the second analysis unit 130 and the second probability value obtained in the third analysis unit 140. It can operate as .

Specifically, the final probability value analysis unit 150 calculates the final END occurrence probability value and the final END occurrence probability value using a weighted average value of the 1st END occurrence probability value, the 2nd END occurrence probability value, the 1st END non-occurrence probability value, and the 2nd END non-occurrence probability value. The probability value of END non-occurrence can be calculated. Generally, the ratio of applied weights may be a 1:1 ratio, but in some cases, the ratio of applied weights may be determined differently.

For example, the final probability value analysis unit 150 may determine the risk of the lesion according to the location and shape of the lesion displayed in the lesion image, and may variably determine the weight applied to the first END occurrence probability value based on the determined risk. there is.

Additionally, for example, higher importance may be assigned to items of clinical information to correspond to patient characteristics (e.g., age, related treatment history, etc.), and weights may be variably determined based on the assigned importance. If a high level of importance is given to age and fasting blood sugar according to the patient's characteristics, the weight to be applied to the second END occurrence probability value determined based on the level of importance can be variably determined.

The weights applied to the first END non-occurrence probability value and the second END non-occurrence probability value may be determined by applying a method opposite to the above method.

The final probability value analysis unit 150 calculates the final END occurrence probability value as the average value of the weighted first END occurrence probability value and the second END occurrence probability value, and calculates the final END occurrence probability value as the average value of the first END non-occurrence probability value and the second END non-occurrence probability value. The probability value of END non-occurrence can be calculated.

For example, if the 1st END occurrence probability value is 0.8, the 1st END non-occurrence probability value is 0.2, the 2nd END occurrence probability value is 0.4, and the 2nd END non-occurrence probability value is 0.6, (0.8+0.4) Thus, the final END occurrence probability value can be determined to be 60%, and (0.2+0.6)×2=0.4, so the final END non-occurrence probability value can be determined to be 40%.

According to the above-described embodiment, the END prediction device 100 can be viewed as a configuration that obtains the final probability value using the ensemble technique of the output values of the first to third artificial intelligence models 141. In this case, the accuracy of the final probability value calculated through the combination of multiple artificial intelligence models can be significantly improved compared to the prior art.

Figure 6 is a flowchart showing each step in which the END prediction device operates according to an embodiment of the present invention.

Since the END prediction method according to an embodiment of the present invention is carried out on substantially the same configuration as the END prediction device 100 shown in FIG. 1, the same reference numerals refer to the same components as the END prediction device 100 shown in FIG. will be given, and repeated explanations will be omitted.

The END prediction method of the present invention includes the steps of acquiring the original brain image and clinical information (S210), outputting the lesion image of the original brain image (S220), outputting the first probability value (S230), and calculating the second probability value. It may include an output step (S240) and a step of calculating the final probability value (S250).

The step of acquiring the original brain image and clinical information (S210) may be a step of acquiring the patient's original brain image and clinical information from the server through the communication unit 110.

In the step of outputting the lesion image of the original brain image (S220), the original brain image is input to the first artificial intelligence model that performs lesion analysis on the brain image through the first analysis unit 120 to output the lesion image corresponding to the original brain image. This may be the step of outputting a lesion image.

In addition, in the step of outputting the lesion image of the original brain image (S220), the first analysis unit 120 converts the original brain image into a two-dimensional original brain image and converts the original brain image into a two-dimensional original brain image through the first artificial intelligence model. By performing lesion analysis, the lesion location and lesion shape characteristics can be detected, and a 3D lesion image displaying the detected lesion location and lesion shape characteristics can be output.

Meanwhile, the step of outputting the first probability value (S230) inputs the original brain image and lesion image into the second artificial intelligence model through the second analysis unit 130 and outputs the first probability value, which is the probability of occurrence of the patient's END. It may be a step.

In addition, in the step of outputting the first probability value (S230), the second analysis unit 130 generates a synthetic input image by splicing the lesion image corresponding to the original brain image to the original brain image, and converts the synthetic input image into a second artificial image. By inputting the first probability value into the intelligence model, the first probability value can be output as the probability value of the occurrence of the 1st END and the probability value of the non-occurrence of the 1st END, and the sum of the probability value of the occurrence of the 1st END and the probability value of the non-occurrence of the 1st END can be normalized to be 1.

Meanwhile, the step of outputting the second probability value (S240) may be a step of inputting clinical information into a third artificial intelligence model through the third analysis unit 140 and outputting a second probability value, which is the probability of occurrence of the patient's END. .

In addition, in the step of outputting the second probability value (S240), the third analysis unit 140 inputs clinical information into the third artificial intelligence model and outputs the second probability value as the probability value of the occurrence of the 2nd END and the probability value of non-occurrence of the 2nd END. And, the sum of the probability value of occurrence of the 2nd END and the probability value of non-occurrence of the 2nd END can be normalized to be 1.

The step of calculating the final probability value (S250) may be a step of calculating the final probability value, which is the probability of occurrence of the patient's END, based on the first probability value and the second probability value through the final probability value analysis unit 150.

In addition, in the step of calculating the final probability value (S250), the final probability value analysis unit 150 calculates the final END occurrence probability value as a weighted average value of the first END occurrence probability value and the second END occurrence probability value, and calculates the final END occurrence probability value as the weighted average value of the first END occurrence probability value and the second END occurrence probability value. The final END non-occurrence probability value can be calculated as the average value of the probability value and the weighted average value of the second END non-occurrence probability value.

The END prediction method of the present invention can be implemented in the form of program instructions that can be executed through various computer components and recorded on a computer-readable recording medium. The computer-readable recording medium may include program instructions, data files, data structures, etc., singly or in combination.

Program instructions recorded on the computer-readable recording medium may be specially designed and configured for the present invention, or may be known and usable by those skilled in the computer software field.

Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specifically configured to store and perform program instructions, such as ROM, RAM, flash memory, etc.

Examples of program instructions include not only machine language code such as that created by a compiler, but also high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules to perform processing according to the invention and vice versa.

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

100: END prediction device
110: Department of Communications
120: first analysis unit
121: First artificial intelligence model
130: second analysis unit
131: Second artificial intelligence model
140: Third analysis unit
141: Third artificial intelligence model
150: Final probability value analysis unit

Claims (10)

In the method of predicting a patient's END using a plurality of artificial intelligence models performed in an END (Early Neurological Deterioration) prediction device,
In a communication unit, obtaining original brain images and clinical information of the patient from a server;
In a first analysis unit, inputting the original brain image to a first artificial intelligence model that performs lesion analysis on the brain image and outputting a lesion image of the original brain image;
In a second analysis unit, inputting the original brain image and the lesion image into a second artificial intelligence model to output a first probability value, which is the probability of occurrence of the END of the patient;
In a third analysis unit, inputting the clinical information into a third artificial intelligence model to output a second probability value, which is the probability of occurrence of the END of the patient; and
END prediction method comprising; calculating, in a final probability value analysis unit, a final probability value that is the probability of occurrence of the END of the patient based on the first probability value and the second probability value.
According to claim 1,
The step of outputting the lesion image is
Converting the original brain image into a two-dimensional original brain image;
performing lesion analysis on the two-dimensional original brain image using the first artificial intelligence model to detect lesion location and lesion shape characteristics; and
END prediction method comprising; outputting a 3D lesion image displaying characteristics of the detected lesion location and lesion shape.
According to claim 1,
The step of outputting the first probability value is,
generating a synthetic input image by splicing the lesion image corresponding to the original brain image to the original brain image;
Inputting the synthesized input image into the second artificial intelligence model and outputting the first probability value as a first END occurrence probability value and a first END non-occurrence probability value; and
Normalizing the sum of the probability of occurrence of the first END and the probability of non-occurrence of the first END to be 1.
According to claim 3,
The step of calculating the second probability value is
Inputting the clinical information into the third artificial intelligence model and outputting the second probability value as a second END occurrence probability value and a second END non-occurrence probability value; and
Normalizing the sum of the second END occurrence probability value and the second END non-occurrence probability value to 1. END prediction method including.
According to claim 4,
The step of calculating the final probability value is
calculating a final END occurrence probability value as a weighted average value of the first END occurrence probability value and the second END occurrence probability value; and
END prediction method comprising; calculating a final END non-occurrence probability value as a weighted average value of the first END non-occurrence probability value and the second END non-occurrence probability value.
In a device that predicts a patient's END (Early Neurological Deterioration) using multiple artificial intelligence models,
a communication unit that obtains the original brain image and clinical information of the patient from a server;
a first analysis unit that inputs the original brain image into a first artificial intelligence model that performs lesion analysis on the brain image and outputs a lesion image of the original brain image;
a second analysis unit that inputs the original brain image and the lesion image into a second artificial intelligence model and outputs a first probability value, which is the probability of occurrence of the END of the patient;
a third analysis unit that inputs the clinical information into a third artificial intelligence model and outputs a second probability value, which is the probability of occurrence of the END of the patient; and
A final probability value analysis unit that calculates a final probability value that is the probability of occurrence of the END of the patient based on the first probability value and the second probability value. END prediction device comprising a.
According to claim 6,
The first analysis unit is
Converting the original brain image into a two-dimensional original brain image,
Perform lesion analysis on the two-dimensional original brain image through the first artificial intelligence model to detect lesion location and lesion shape characteristics,
An END prediction device that outputs a 3D lesion image displaying characteristics of the detected lesion location and lesion shape.
According to claim 6,
The second analysis unit is
Creating a synthetic input image by splicing the lesion image corresponding to the original brain image to the original brain image,
Input the synthesized input image into the second artificial intelligence model and output the first probability value as a first END occurrence probability value and a first END non-occurrence probability value,
An END prediction device that normalizes the sum of the first END occurrence probability value and the first END non-occurrence probability value to 1.
According to claim 8,
The third analysis unit is
Input the clinical information into the third artificial intelligence model and output the second probability value as a second END occurrence probability value and a second END non-occurrence probability value,
An END prediction device that normalizes the sum of the second END occurrence probability value and the second END non-occurrence probability value to be 1.
According to clause 9,
The final probability value analysis unit
Calculate the final END occurrence probability value as a weighted average value of the first END occurrence probability value and the second END occurrence probability value,
An END prediction device that calculates a final END non-occurrence probability value as a weighted average value of the first END non-occurrence probability value and the second END non-occurrence probability value.

Images