Method for calculating dementia-related information using volume predicted by brain CT and analysis device thereof

Abstract

The method for calculating dementia-related information using a volume predicted by a brain computed tomography (CT) comprises the steps of: receiving, by an analysis device, a brain CT image of a subject; inputting, by the analysis device, the brain CT image to a learned segmentation model to extract regions of interest; inputting, by the analysis device, pixel information of the regions of interest to a pre-trained first learning model to predict a volume of at least one region among the regions of interest; and inputting, by the analysis device, the volume of the at least one region to a pre-trained second learning model to calculate dementia-related information of the subject.

Description

Method and analysis device for obtaining dementia-related information using volume predicted from brain CT

[0001] The technology described below is a technique for predicting dementia-related information of a subject using brain CT images.

[0002] Dementia is a syndrome that causes cognitive impairment, including memory, language, and judgment. Alzheimer's disease is the most common type of dementia. Brain imaging techniques such as Magnetic Resonance Imaging (MRI) and PET scans are used to diagnose Alzheimer's disease.

[0003] 3D (dimensional) MRI and PET-CT (positron emission tomography-computed tomography) are time-consuming and expensive to obtain. In contrast, CT has no limitations, such as the presence of metal within the body. Compared to MRI, it can be obtained much more quickly and is relatively inexpensive.

[0004] The technology described below aims to provide a method for producing dementia-related information, such as the presence of dementia, dementia risk, and brain atrophy, using only brain CT images.

[0005] A method for producing dementia-related information using a volume predicted from a brain CT comprises a step of an analysis device receiving a brain CT image of a subject, a step of the analysis device inputting the brain CT image into a learned segmentation model to extract regions of interest, a step of the analysis device inputting pixel information of the regions of interest into a first learning model learned in advance to predict the volume of at least one region among the regions of interest, and a step of the analysis device inputting the volume of the at least one region into a second learning model learned in advance to produce dementia-related information of the subject.

[0006] An analysis device for calculating dementia-related information using a volume predicted from a brain CT comprises: an interface device for receiving a brain CT image of a subject; a segmentation model for extracting a region of interest from the brain CT image; a storage device for storing a first learning model for predicting a volume by receiving information on a region of interest of the brain; and a second learning model for calculating dementia-related information; and a calculation device for calculating dementia-related information of the subject by inputting the received brain CT image into the segmentation model to extract regions of interest, inputting pixel information of the extracted regions of interest into the first learning model to predict a volume of at least one region among the regions of interest, and inputting the volume of the at least one region into the second learning model.

[0007] The technique described below extracts a region of interest from a CT image and derives dementia-related information based on its volume. Using only CT images, the technique described below can quickly and inexpensively derive meaningful dementia-related information.

[0008] Figure 1 is an example of a system that produces dementia-related information based on CT images.

[0009] Figure 2 is an example of a process for producing dementia-related information using brain CT images.

[0010] Figure 3 is an example of the process of learning a segmentation model that extracts a region of interest from a brain CT image.

[0011] Figure 4 is an example of the learning process of a learning model that predicts the volume of a region of interest based on the extracted region of interest.

[0012] Figure 5 is an example of the learning process of a learning model that produces dementia-related information based on the volume of a brain region of interest.

[0013] Figure 6 is an example of an analysis device that produces dementia-related information of a subject using a brain CT image.

[0014] The technology described below is susceptible to various modifications and embodiments. Specific embodiments are illustrated and described in detail in the drawings. However, this does not limit the technology described below to specific embodiments, and it should be understood that all modifications, equivalents, and alternatives fall within the spirit and scope of the technology described below.

[0015] Terms such as first, second, A, and B may be used to describe various components, but these components are not limited by these terms and are used solely to distinguish one component from another. For example, without departing from the scope of the technology described below, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. The term "and / or" includes a combination of a plurality of related described items or any of a plurality of related described items.

[0016] As used herein, the singular expressions should be understood to include the plural expressions unless the context clearly dictates otherwise, and the term "comprises" and the like should be understood to mean the presence of a described feature, number, step, operation, component, part, or combination thereof, but not to exclude the presence or addition of one or more other features, numbers, steps, operations, components, parts, or combinations thereof.

[0017] Before going into a detailed description of the drawings, it should be made clear that the division of components in this specification is merely a division based on the main function of each component. In other words, two or more components described below may be combined into a single component, or a single component may be further subdivided into two or more components with more detailed functions. In addition to its own main function, each component described below may additionally perform some or all of the functions of other components, and of course, some of the main functions of each component may be exclusively performed by other components.

[0018] Additionally, in performing a method or method of operation, each process constituting the method may occur in a different order than the stated order, unless the context clearly indicates a specific order. That is, each process may occur in the same order as the stated order, may be performed substantially simultaneously, or may be performed in the opposite order.

[0019] The technology described below is a technology that produces dementia-related information from brain CT.

[0020] Dementia below includes Alzheimer's dementia.

[0021] Dementia-related information refers to information related to dementia diagnosis or prediction. For example, dementia-related information may include at least one of the following: dementia onset, dementia risk, dementia likelihood, dementia-related scores, dementia prognosis prediction, degree of brain atrophy, beta-amyloid (Aß) positivity, tau protein positivity, and brain age.

[0022] Hereinafter, a device that analyzes brain CT scans to derive dementia-related information about a subject is called an "analysis device." The analysis device can take the form of a computer device such as a PC, a smart device, a network server, or a dedicated data processing chipset.

[0023] The analysis device can use multiple learning models to derive dementia-related information from brain images. The analysis device can use a segmentation model to extract regions of interest (ROIs) from CT scans. Furthermore, the analysis device can use a classification model to derive dementia-related information from CT scans or ROIs.

[0024] Figure 1 is an example of a system (100) that produces dementia-related information based on CT images. In Figure 1, the analysis device is illustrated as an example of a computer terminal (130) and a server (140).

[0025] A CT imaging device (110) captures a CT image of a subject. The CT image is an image of a brain region (brain CT image). The brain CT image generated by the CT imaging device (110) can be stored in a separate database, such as an Electronic Medical Record (EMR, 120).

[0026] A computer terminal (130) can receive a brain CT image of a subject from a CT imaging device (110) or an EMR (120) via a wired or wireless network. In some cases, the computer terminal (130) may be a device physically connected to the CT imaging device (110). The computer terminal (130) can preprocess the brain CT image in a predetermined manner.

[0027] A computer terminal (130) extracts regions of interest (ROIs) from a brain CT image. The computer terminal (130) can extract regions of interest (ROIs) from the brain CT image using a learned segmentation model. The computer terminal (130) can estimate the volume of the regions of interest (ROIs) using the learned model. The computer terminal (130) can use the learned model to derive dementia-related information about the subject based on the volume of the regions of interest (ROIs). User A can check dementia-related information through the computer terminal (130).

[0028] The server (140) can receive a brain CT image of a subject from a CT imaging device (110) or EMR (120) via a wireless network. The server (140) can preprocess the brain CT image on a regular basis.

[0029] The server (140) extracts regions of interest (ROIs) from a brain CT image. The server (140) can extract the ROIs from the brain CT image using a learned segmentation model. The server (140) can estimate the volume of the ROIs using the learned model. The server (140) can use the learned model to derive dementia-related information about the subject based on the volume of the ROIs using the learned model. The server (140) can transmit the dementia-related information to a user terminal. User A can check the dementia-related information through the user terminal.

[0030] Figure 2 is an example of a process (200) for producing dementia-related information using brain CT images. The analysis device can produce dementia-related information about a subject based on brain CT images.

[0031] The analysis device receives a brain CT image of a subject (210). The brain CT may be composed of multiple slices. The object analyzed by the analysis device may be all of the multiple slices or at least one of the multiple slices.

[0032] The analysis device can uniformly preprocess brain CT images.

[0033] For example, an analysis device can extract entire brain regions from brain CT images. In this process, the analysis device can utilize a pre-trained segmentation model. Using the CIVET pipeline, the analysis device extracts entire brain regions from brain MRI images and generates an answer. The segmentation model can then be trained to extract entire brain regions from brain CT images based on the answer (the entire brain region in the brain MRI image). The analysis device can then use the trained segmentation model to extract entire brain regions from brain CT images.

[0034] Additionally, the analysis device can normalize the size or resolution of the subject's brain CT image to a certain size.

[0035] The analysis device extracts regions of interest (ROI) from the input brain CT image (220). The analysis device can extract the ROI using a pre-trained segmentation model. The analysis device can extract multiple ROIs using the segmentation model. These ROIs correspond to the areas used to calculate the volume of the corresponding regions.

[0036] The region of interest may include at least one of a cerebrospinal fluid (CSF) region and a ventricle region. The region of interest may include at least one of a frontal CSF region, a temporal CSF region, a parietal CSF region, an occipital CSF region, an anterior lateral ventricle region, a posterior lateral ventricle region, and an around hippocampal CSF region.

[0037] The analysis device can estimate the volume of the extracted region(s) of interest (230). The analysis device can estimate the volume of the specific region of interest using a learning model (e.g., a regression analysis model) that has learned in advance the relationship between the number of pixels for the specific region of interest and the volume of the region. To this end, the analysis device needs to uniformly normalize the size of the brain CT image or the region of interest. The analysis device can estimate the volume of the specific region of interest based on the uniformly normalized brain CT image. At this time, the learning model can be prepared in advance for each of multiple regions of interest. For example, the analysis device can individually estimate the volume of the corresponding region using a learning model for each of the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the cerebrospinal fluid region around the hippocampus.

[0038] Meanwhile, the analysis device can also estimate the volume of the region of interest by using additional clinical information (gender, age, etc.) of the subject in addition to the brain CT image.

[0039] The analysis device can derive dementia-related information about the subject based on the volume of the region(s) of interest (240). The analysis device can derive dementia-related information based on the volume of the region(s) of interest using a learning model. Furthermore, the analysis device can derive dementia-related information about the subject using additional clinical information (age, gender, APOE4 genotype, etc.) in addition to the volume of the region(s) of interest.

[0040] The researcher constructed the aforementioned learning models using data from a population of patients visiting the affiliated medical institution (Samsung Seoul Hospital). The population consisted of dementia patients (Alzheimer's disease dementia (ADD)) and normal controls (NC). The population was selected from subjects who underwent both MRI and CT scans within a certain period (one year). The population excluded young healthy individuals and patients with dementia and cardiovascular disease. The data included MRI, CT, clinical information (gender, age, APOE4 phenotype), and dementia status (label value). The total number of subjects in the population was 895. Table 1 below presents the population information. Of the 895 subjects, 716 (80%) were used as training data, and 179 (20%) were used as test data. Validation was performed using 5-fold cross-validation.

[0041] TotalNCADDSubjects, N (%)895 (100.0)457 (51.1)438 (48.9)Age, mean years (SD)67.4 (11.4)66.8 (12.5)68.0 (10.2)Female, N (%)541 (60.4)279 (60.9)262 (59.8)APOEε4 carriers, N (%)322 (36.0)86 (18.8)236 (51.5)Frontal cerebrospinal fluid region, mean volume (mm) 3 )61052.550380.972187.0Temporal cerebrospinal fluid area, mean volume (mm) 3 )36534.931169.542133.0Cerebral spinal fluid area, mean volume (mm) 3 )36045.529592.542778.3Occipital cerebrospinal fluid area, mean volume (mm) 3 )14828.413150.016579.6Anterior lateral ventricle area, mean volume (mm) 3)17687.313955.521581.1 Posterior lateral ventricle area, mean volume (mm) 3 )20476.314904.426289.8Cebrospinal fluid area around the hippocampus, mean volume (mm) 3 )7577.05562.89678.6

[0042] Figure 3 illustrates an example of the process (300) for learning a segmentation model that extracts regions of interest from brain CT images. The researcher normalized the brain CT images used as training data to a certain size. Furthermore, the researcher augmented the training data by adjusting brightness and performing horizontal flips on the images.

[0043] The learning process of the model below is described as being performed by a learning device. The researcher built a segmentation model using the learning device. A learning device is a computer capable of processing image data and performing the training process for a machine learning model.

[0044] The learning device learns a segmentation model using medical images of multiple subjects. However, Figure 3 illustrates the process of learning a segmentation model using data from subject i as an example.

[0045] The learning device constructs learning data for learning the learning model (310). The learning data can be constructed from brain images of a population (including dementia patients and normal individuals). The CT image database (database) stores brain CT images of the population. The MRI image database stores brain MRI images of the population. The MRI image database contains information on regions of interest (ROIs) in brain images. The ROIs can be ROIs labeled automatically or manually by an expert. The learning device constructs learning data by extracting CT images and MRI images (including ROI information) of the same subject.

[0046] The learning device performs a process of learning a segmentation model that extracts regions of interest using learning data (320). The learning device performs learning based on data about a specific patient in the learning data. The learning device can repeatedly perform the learning process using data about multiple patients in the learning data.

[0047] The learning device receives a 3D MRI image (including region of interest information) of subject i (311). Furthermore, the learning device receives a 2D or 3D CT image of subject i (312). The region of interest may be a region of interest (ROI) in a 2D image or a volume of interest (VOI) in a 3D image. The learning device aligns the brain MRI image and the brain CT image (320). The learning device aligns images of the same brain region of the same subject.

[0048] The learning device builds a segmentation model using aligned brain CT and MRI images. The segmentation model can be a semantic segmentation model. For example, the segmentation model can be a fully convolutional network (FCN)-based model such as U-net.

[0049] The segmentation model predicts regions of interest (ROIs) based on brain CT images. During the training process, the segmentation model learns to extract the same ROIs from brain CT images by referencing the correct ROIs from brain MRI images.

[0050] As described above, regions of interest can be multiple regions. Regions of interest can include frontal cerebrospinal fluid regions, temporal cerebrospinal fluid regions, parietal cerebrospinal fluid regions, occipital cerebrospinal fluid regions, anterior lateral ventricle regions, posterior lateral ventricle regions, and perihippocampal cerebrospinal fluid regions. The learning device can build a segmentation model for each region of interest.

[0051] The researcher also built multiple segmentation models for each region of interest. The researcher built a U-net-based segmentation model. The researcher built a segmentation model (a 2D image-based segmentation model) using each axial slice in the training data. In addition, the researcher built a segmentation model (a 3D image-based segmentation model) using 3D CT of the training data. Table 2 below shows the performance evaluation results of the built segmentation models. The researcher performed 5-fold cross-validation on the validation data to evaluate the performance of the 2D image-based segmentation model and the 3D image-based segmentation model.

[0052] 2D image-based segmentation model 2D (n=179) Frontal CSF region Temporal CSF region Parietal CSF region Occipital CSF region Anterior lateral ventricle region Posterior lateral ventricle region Perihippocampal CSF region 1 Fold 0.5686 0.483 20.571 10.429 40.884 80.857 30.651 12 Fold 0.5682 0.478 60.568 60.424 60.883 30.856 30.647 63 Fold 0.581 50.494 90.581 30.444 30.881 60.855 70.644 24 Fold0.58360.49380.58350.44340.88220.85480.64355 Fold0.57110.48720.57140.43110.88490.85840.66133D image-based segmentation model (n=179) Frontal CSF regionTemporal CSF regionParietal CSF regionOccipital CSF regionAnterior lateral ventricle regionPosterior lateral ventricle regionPerihippocampal CSF region1 Fold0.58270.48850.59030.44490.88510.86030.62782 Fold0.58650.49070.59220.4480.88370.8590.62733 Fold0.5940.49670.60370.46080.88320.85930.62014 Fold0.60260.50230.60630.46130.88230.85630.61815 Fold0.58640.49070.5940.44940.88620.86110.6357

[0053] Figure 4 illustrates an example of a learning process (400) of a learning model that predicts the volume of a region of interest based on an extracted region of interest. In Figure 3, the learning model predicts the volume of a region of interest based on the aforementioned region of interest.

[0054] The learning in Figure 4 assumes that the aforementioned region of interest has been extracted from a brain CT image of a subject belonging to the population. In other words, the learning device can utilize the region of interest derived using the segmentation model in Figure 3.

[0055] The learning device extracts input data based on regions of interest (ROI) extracted from brain CT images (410). The input data includes the ROI(s) extracted by the segmentation model. The input data includes the actual volume (correct value) of a specific ROI. The actual volume can be derived from an MRI. Furthermore, the input data may also include clinical information about the subject (gender, age).

[0056] The learning device performs a process of learning a learning model using the extracted input data (420). The learning device repeats the learning model training process using input data extracted from brain CT images of multiple subjects. The learning device can build a learning model that predicts volume for each region of interest.

[0057] The learning device calculates the total number of pixels for regions of interest. As described above, the regions of interest may be the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the cerebrospinal fluid region around the hippocampus.

[0058] The learning device builds a learning model by comparing the number of pixels in a region of interest with the volume (correct answer) of that region of interest. Through this process, the learning model learns to predict the volume of the region of interest. The correct answer is the volume of a specific region of interest derived from brain MRIs of subjects in the population.

[0059] Learning models can be built as individual models depending on the information to be predicted. Learning models are machine learning models and can be implemented as one of various types of models. The researchers used a regression model. The researchers individually built a model to predict the volume of the anterior CSF (anterior CSF model), a model to predict the volume of the temporal CSF (temporal CSF model), a model to predict the volume of the parietal CSF (parietal CSF model), a model to predict the volume of the occipital CSF (occipital CSF model), a model to predict the volume of the anterior lateral ventricle (lateral ventricle model), a model to predict the volume of the posterior lateral ventricle (posterior lateral ventricle model), and a model to predict the volume of the cerebrospinal fluid region around the hippocampus (hippocampal CSF model).

[0060] Furthermore, the learning model can be trained using clinical information (age and gender) in addition to the size of the region of interest.

[0061] The researchers evaluated the performance of the constructed volume prediction model.

[0062] The researchers compared the predicted volume of the region of interest extracted from brain CT images using the aforementioned segmentation model with the correct values ​​measured from 3D MRI. The researchers segmented the region of interest using each of the two segmentation models they constructed, and compared the predicted volume based on these results with the correct values. The performance evaluation results of the volume prediction models are shown in Table 3 below. The performance indicator is the Pearson Correlation Coefficient. The researchers repeated 10 rounds of cross-validation (permutation). The volume of the region of interest predicted by the learning model showed a significant correlation with the actual correct values.

[0063] Region of Interest Correlation r (Volume prediction of region of interest segmented based on 2D images) Correlation r (Volume prediction of region of interest segmented based on 3D images) Frontal CSF area 0.838 0.868 Temporal CSF area 0.827 0.878 Parietal CSF area 0.868 0.905 Occipital CSF area 0.76 10.824 Anterior lateral ventricle area 0.98 4 0.984 Posterior lateral ventricle area 0.98 1 0.981 Perihippocampal CSF area 0.94 5 0.944

[0064] Figure 5 illustrates an example of the learning process (500) of a learning model that generates dementia-related information based on the volume of a brain region of interest. The learning model that generates dementia-related information can be implemented using any of a variety of machine learning model types. The researcher implemented a model that generates dementia-related information using a deep learning model.

[0065] The learning in Figure 5 assumes that the volume of each region of interest and clinical information for each subject within the population have been acquired. For example, the learning device can utilize the volume of each region of interest for each subject calculated using the learning model in Figure 4.

[0066] The learning device acquires input data, which is learning data (510).

[0067] The learning device extracts the volume of specific regions of interest (ROIs) from brain CT images of subjects in the population. These brain region volumes may include the volumes of the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the perihippocampal cerebrospinal fluid region. Furthermore, the learning device can build a learning model to generate dementia-related information based on a subset of the ROIs among multiple ROIs.

[0068] Additionally, the learning device acquires clinical information about the subjects. This clinical information may include age, gender, and APOE4 phenotype. The learning device can build a learning model that utilizes some of the clinical information, in addition to the volume of the region(s) of interest, to derive dementia-related information.

[0069] The training data may include the volume of a specific brain region calculated using the learning model of Figure 3. Alternatively, the training data may include cortical thickness and the volume of a specific brain region calculated from brain images of actual subjects. The researcher used the volume of a region of interest calculated from 3D MRI analysis of an actual subject as the training data.

[0070] The learning device performs a process of learning a learning model using the extracted input data (520). The learning device repeats the learning process of the learning model using input data extracted from multiple subjects.

[0071] The learning device performs learning by inputting extracted input data into a learning model and comparing the output (predicted value) of the learning model with the correct answer. The learning model is trained to predict dementia-related information for the subject. The correct answer is the subject's dementia-related information.

[0072] The researcher developed a learning model to classify dementia patients or healthy individuals. The researcher used 90% of the population data as training data and 10% as validation data. The multiple learning models developed by the researcher are shown in Table 3 below. The learning models were broadly divided into a model that uses only the volume of regions of interest (Model A), a model that uses the volume of regions of interest and clinical information (age and gender) (Model B), and a model that uses the volume of regions of interest and clinical information (age, gender, and APOE4 phenotype) (Model C). Each model group was developed by dividing it into models that use 1, 2, 3, 4, 6, and 7 (total) regions of interest.

[0073] Table 4 shows the model groups using four different ROIs and the training data used to build the different models in each group. The model in Table 4 below predicts whether a subject has dementia (dementia or normal). In other words, the training model in Figure 4 corresponds to a binary classification model for whether the subject has dementia or normal.

[0074] Model ClassificationModel NameInput FeaturesModel AModel A13 ROIs (aLV / pLV / HippCSF)Model A24 ROIs (F / T / P / O)Model A36 ROIs (F / T / P / O / aLV / pLV)Model A47 ROIs (F / T / P / O / aLV / pLV / HippCSF)Model A52 ROIs (aLV / pLV)Model A61 ROIs (HippCSF)Model BModel B13 ROIs (aLV / pLV / HippCSF)+(Age, Gender)Model B24 ROIs (F / T / P / O)+(Age, Gender)Model B36 ROIs (F / T / P / O / aLV / pLV)+(Age, Gender)Model B47 ROIs (F / T / P / O / aLV / pLV / HippCSF)+(Age, Gender)Model B52 ROIs (aLV / pLV)+(age, sex, APOE4) Model B61 ROIs (HippCSF)+(age, sex, APOE4) Model C13 ROIs (aLV / pLV / HippCSF)+(age, sex, APOE4) Model C24 ROIs (F / T / P / O)+(age, sex, APOE4) Model C36 ROIs (F / T / P / O / aLV / pLV)+(age, sex, APOE4) Model C47 ROIs (F / T / P / O / aLV / pLV / HippCSF)+(age, sex, APOE4) Model C52 ROIs (aLV / pLV)+(age, sex, APOE4) Model C61 ROIs (HippCSF)+(age, sex, APOE4)

[0075] In Table 4, regions of interest (ROIs) are abbreviated. F stands for frontal CSF, T for temporal CSF, P for parietal CSF, O for occipital CSF, aLV for anterior lateral ventricle, pLV for posterior lateral ventricle, and HippCSF for perihippocampal CSF.

[0076] The researcher verified the constructed learning model. Meanwhile, during the verification process, the researcher used the cerebral cortex thickness and brain region volume predicted by the learning model in Fig. 4 as input data for the learning model in Fig. 5. The researcher performed 10-fold cross-validation. The verification results are shown in Table 5 below. In Table 5, 3D CIVET is the classification result using the volume values ​​obtained from CIVET. Table 5 shows the classification results using the region of interest segmented by the 2D image-based segmentation model and the classification results using the region of interest segmented by the 3D image-based segmentation model. The verification compared the results predicted by the learning model with the correct answer. The performance indicators used were the area under the receiver operating characteristic curve (AUC) and the area under the precision-recall curve (AUPRC).

[0077] Model Type 3D CIVET2D Classification Performance Using Image-Based Segmentation Model Classification Performance Using Image-Based Segmentation Model AUCAUPRCAUCAUPRCAUCAUPRCModel A10.7910.7720.7830.7670.7840.769Model A20.7670.7370.7380.7300.7720.735Model A30.7790.7440.7460.7370.7790.738Model A40.7800.7460.7460.7380.7730.734Model A50.8050.7720.8010.7760.8000.776Model A60.8000.7840.7660.7660.7770.763Model Model B10.9240.9030.8940.8930.9000.892 Model B20.8820.8700.8380.8370.8840.881 Model B30.9100.8970.8720.8660.9020.899 Model B40.9300.9130.9080.8990.9160.907 Model B50.8630.8700.8540.8650.8620.874 Model B60.9120.8820.8850.8790.8850.872 Model C10.9450.9030.9200.8970.9340.899 Model C20.9260.8990.8770.8690.9370.927 Model C30.9350.9150.8930.8860.9370.929 Model C40.9440.9130.9190.9050.9420.917 Model C50.9030.9100.9050.9140.9040.915 Model C60.9350.8930.9070.8890.9210.887

[0078] Looking at Table 5, overall, there was no significant difference in performance between the results using 3D CIVET and the results using the segmentation model. In addition, the results of classification using the volume of the region of interest extracted using the 2D image-based segmentation model were also sufficiently significant in obtaining dementia-related information. Looking at Table 5, Model A, Model B, and Model C all showed significant performance, and overall, the model using additional clinical information showed slightly higher performance. Furthermore, looking at Table 5, the model using the anterior lateral ventricle region (aLV), posterior lateral ventricle region (pLV), and hippocampal cerebrospinal fluid region (hipp CSF) showed relatively high performance.

[0079] Figure 6 illustrates an example of an analysis device (600) that uses brain CT images to derive dementia-related information for a subject. The analysis device (600) can be physically implemented in various forms. For example, the analysis device (600) may take the form of a computer device such as a PC, a smart device, a network server, a data processing chipset, or the like. Alternatively, the analysis device (600) may be connected to brain imaging equipment or may be an integrated device.

[0080] The analysis device (600) may include a storage device (610), a memory (620), a computing device (630), an interface device (640), a communication device (650), and an output device (660).

[0081] The storage device (610) can store a brain CT image of a subject generated by a CT imaging device.

[0082] The storage device (610) can store clinical information of the subject. The clinical information can include at least one of age, gender, and APOE4 genotype.

[0083] The storage device (610) can store a segmentation model for extracting a region of interest from a brain CT image (slice).

[0084] The storage device (610) can store a learning model that predicts the volume of a region of interest based on information about the region of interest. The learning model that predicts the volume of a specific brain region of interest is called a first learning model. The region of interest includes at least one of a frontal cerebrospinal fluid region, a temporal cerebrospinal fluid region, a parietal cerebrospinal fluid region, an occipital cerebrospinal fluid region, an anterior lateral ventricle region, a posterior lateral ventricle region, and a cerebrospinal fluid region around the hippocampus. As described above, the first learning model can be prepared for each region of interest.

[0085] In addition, the storage device (610) can store a learning model that produces dementia-related information using the volume of the region of interest. The learning model that produces dementia-related information is called a second learning model. As described above, the second learning model can be implemented as various models depending on the type of input data. The second learning model can be at least one of various types, such as (i) a model that uses only the volume of the region of interest as input data (Model A), (ii) a model that uses the volume of the region of interest and clinical information (age, gender) as input data (Model B), or (iii) a model that uses the volume of the region of interest and clinical information (age, gender, and APEO4) as input data (Model C). In addition, the region of interest used by the learning model may be all or a part of the seven regions of interest as described in Table 3.

[0086] The dementia-related information inferred by the second learning model may include at least one of the following: dementia onset, dementia risk, dementia likelihood, dementia-related score, dementia prognosis prediction, degree of brain atrophy, beta-amyloid (Aß) positivity, tau protein positivity, brain age, etc. The dementia-related information may vary depending on the type of training data (correct answer value) used by the second learning model.

[0087] The memory (620) can store data and information generated during the process of the analysis device (600) producing dementia-related information from a brain CT image.

[0088] The interface device (640) is a device that receives certain commands and data from an external source. The interface device (640) can receive a subject's brain CT image from a physically connected input device or an external storage device. The interface device (640) can also transmit dementia-related information predicted based on the brain CT image to an external entity.

[0089] The communication device (650) refers to a configuration that receives and transmits certain information via a wired or wireless network. The communication device (650) can receive a brain CT image of a subject from an external entity. The communication device (650) can also transmit dementia-related information predicted based on the brain CT image to an external entity, such as a user terminal.

[0090] The interface device (640) may be a device that internally transmits data received from the communication device (650).

[0091] The output device (660) is a device that outputs certain information. The output device (660) can output interfaces required for data processing, brain images, regions of interest identified from brain images, dementia-related information calculated based on regions of interest, etc.

[0092] The computational device (630) can preprocess the brain CT image of the subject in a predetermined manner. For example, the computational device (630) can normalize the brain CT image to a predetermined size or resolution.

[0093] The computational device (630) can input a brain CT image of a subject into a learned segmentation model to extract a region of interest. The computational device (630) can input a brain CT image of a subject into a learned segmentation model slice by slice to extract a region of interest.

[0094] The region of interest may be at least one of the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the parahippocampal cerebrospinal fluid region.

[0095] The computing device (630) can extract size information for the region(s) of interest. The computing device (630) can calculate the sum of the number of pixels for each region of interest. The computing device (630) can estimate the volume of each region of interest based on the number of pixels in the region of interest. The computing device (630) can calculate the volume of the region of interest based on the number of pixels in the region of interest using the first learning model described above. The computing device (630) can estimate the volume of each of a plurality of regions of interest using the first learning model constructed for each region of interest.

[0096] The computing device (630) can input the volume of the region(s) of interest predicted by the first learning model into the second learning model to produce dementia-related information of the subject. For example, the computing device (630) can input the volume of at least one region among the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the cerebrospinal fluid region around the hippocampus into the second learning model to produce dementia-related information of the subject. In addition, the computing device (630) can input clinical information (at least one of age, gender, and APOE4 genotype) in addition to the volume of the region of interest into the second learning model to produce dementia-related information of the subject. This process is as described in FIG. 5.

[0097] The computing device (630) may be a device such as a processor, AP, or chip embedded with a program that processes data and performs certain operations.

[0098] Additionally, the brain CT image analysis method or dementia-related information generation method described above may be implemented as a program (or application) including an executable algorithm that can be executed on a computer. The program may be stored and provided on a temporary or non-transitory computer-readable medium.

[0099] A non-transitory readable medium refers to a medium that stores data semi-permanently and can be read by a device, rather than a medium that stores data for a short period of time, such as a register, cache, or memory. Specifically, the various applications or programs described above may be stored and provided on a non-transitory readable medium, such as a CD, DVD, hard disk, Blu-ray disk, USB, memory card, ROM (read-only memory), PROM (programmable read only memory), EPROM (Erasable PROM, EPROM), EEPROM (Electrically EPROM), or flash memory.

[0100] Temporarily readable media refers to various types of RAM, such as Static RAM (SRAM), Dynamic RAM (DRAM), Synchronous DRAM (SDRAM), Double Data Rate SDRAM (DDR SDRAM), Enhanced SDRAM (ESDRAM), Synchronous DRAM (Synclink DRAM, SLDRAM), and Direct Rambus RAM (DRRAM).

[0101] The present embodiment and the drawings attached to the present specification only clearly illustrate a part of the technical idea included in the above-described technology, and it is obvious that all modified examples and specific embodiments that can be easily inferred by a person skilled in the art within the scope of the technical idea included in the specification and drawings of the above-described technology are included in the scope of the rights of the above-described technology.

Claims

1. A step in which the analysis device receives a subject's brain CT (Computed Tomography) image; A step in which the above analysis device inputs the brain CT image into a learned segmentation model to extract regions of interest; A step in which the analysis device inputs pixel information of the regions of interest into a first learning model that has been learned in advance to predict the volume of at least one region among the regions of interest; and The above analysis device includes a step of inputting the volume of at least one area into a second learning model that has been learned in advance to produce dementia-related information of the subject, A method for producing dementia-related information using volumes predicted from brain CT, wherein the above regions of interest include multiple regions among the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the cerebrospinal fluid region around the hippocampus.

2. In paragraph 1, The above first learning model is a method for producing dementia-related information using volume predicted from brain CT, which includes multiple learning models prepared in advance for each of the above areas of interest.

3. In paragraph 1, The second learning model is a method for producing dementia-related information using a volume predicted from a brain CT, which further receives at least one piece of information from the subject's age, the subject's sex, and the subject's APOE4 genotype, and produces dementia-related information of the subject.

4. In paragraph 1, The above dementia-related information includes the occurrence of dementia, the risk of dementia, the possibility of dementia, a dementia-related score, a prediction of dementia prognosis, the degree of brain atrophy, beta amyloid (amyloid-ß, Aß) positivity, tau protein positivity, and a method for producing dementia-related information using the volume predicted from a brain CT, which is one of the brain age.

5. In paragraph 1, The above segmentation model is a method for producing dementia-related information using volume predicted from a brain CT including multiple models prepared in advance for each region of interest.

6. Interface device for receiving brain CT (Computed Tomography) images of the subject; A storage device storing a segmentation model that extracts a region of interest from a brain CT image, a first learning model that receives information about a region of interest in the brain and predicts its volume, and a second learning model that produces dementia-related information; and Including a computing device that inputs the brain CT image received above into the segmentation model to extract regions of interest, inputs pixel information of the extracted regions of interest into the first learning model to predict the volume of at least one region among the regions of interest, and inputs the volume of the at least one region into the second learning model to produce dementia-related information of the subject, An analysis device for producing dementia-related information using volumes predicted from a brain CT, wherein the above regions of interest include multiple regions among the frontal cerebrospinal fluid region, the temporal cerebrospinal fluid region, the parietal cerebrospinal fluid region, the occipital cerebrospinal fluid region, the anterior lateral ventricle region, the posterior lateral ventricle region, and the cerebrospinal fluid region around the hippocampus.

7. In paragraph 6, The above segmentation model is an analysis device that produces dementia-related information using a volume predicted from a brain CT that includes multiple models prepared in advance for each region of interest.

8. In paragraph 6, The above first learning model is an analysis device that produces dementia-related information using a volume predicted from a brain CT that includes multiple learning models prepared in advance for each of the above areas of interest.

9. In paragraph 6, The second learning model is an analysis device that calculates dementia-related information by using a volume predicted from a brain CT that receives at least one piece of information from the subject's age, the subject's gender, and the subject's APOE4 genotype as input and calculates dementia-related information for the subject.

10. In paragraph 6, The above dementia-related information is an analysis device that uses the volume predicted from a brain CT, which is one of the following: presence of dementia, risk of dementia, possibility of dementia, dementia-related score, dementia prognosis prediction, degree of brain atrophy, beta amyloid (amyloid-ß, Aß) positivity, tau protein positivity, and brain age.

Images