Continuous brain magnetic resonance image analysis system and method

Abstract

The present invention relates to a continuous brain magnetic resonance image (MRI) analysis system and method capable of detecting and visualizing a brain lesion and providing results thereof by comparing and analyzing continuous brain MRI images of the same patient. An electronic device comprising at least one processor for continuous brain MRI analysis comprises: an image input unit for inputting a reference FLAIR MRI image and a follow-up FLAIR MRI image in order to automatically detect a new brain lesion from continuous FLAIR MRI images of the same patient; a preprocessing unit that performs image normalization and image registration on the reference FLAIR MRI image and the follow-up FLAIR MRI image and removes noise; a deep learning-based analysis unit that detects a new lesion by analyzing the difference between the reference image and the follow-up image on the basis of deep learning and measures the location and size of the lesion; and a visualization output unit that highlights the detected lesion, quantifies the amount of change, maps the lesion by location, and outputs a visualization thereof.

Description

Continuous Brain Magnetic Resonance Imaging Analysis System and Method

[0001] The present invention relates to medical image analysis, and specifically to a continuous brain magnetic resonance imaging analysis system and method capable of detecting and visualizing brain lesions by comparing and analyzing continuous brain magnetic resonance imaging (MRI) of the same patient.

[0002] Cerebral infarction is a disease in which a portion of the blood vessels distributed in the brain becomes blocked, causing impairment of brain function. Most patients exhibit symptoms such as physical paralysis and speech disorders.

[0003] However, there are cases where lesions occur in brain regions that do not cause symptoms, resulting in no outward symptoms; this is called 'asymptomatic cerebral infarction'.

[0004] Asymptomatic cerebral infarction appears healthy because there are no immediate symptoms, but there is a high likelihood that cognitive decline and dementia will suddenly appear in the future.

[0005] As such, asymptomatic cerebral infarction is a cerebral infarction that occurs without clinical symptoms and is found in about 20% of healthy older adults, and up to 50% in certain study groups. It is known to be a significant risk factor that increases the risk of future stroke.

[0006] Currently, the diagnosis of asymptomatic cerebral infarction is primarily made using FLAIR MRI imaging, but the following limitations exist.

[0007] Radiological judgment is a visual analysis by radiologists, and since the report is a descriptive, narrative, and qualitative evaluation, there are limitations in analyzing quantitative information, making it difficult to detect subtle changes in lesions and requiring significant time and effort for interpretation.

[0008] In particular, inter-examiner diagnostic agreement is low, and it is difficult to perform precise analysis within limited consultation time.

[0009] Therefore, there is a need for the development of new technology that can automatically detect new cerebral infarction lesions from acquired medical imaging information to support accurate and consistent diagnosis of asymptomatic cerebral infarction.

[0010] The present invention aims to solve the problems of conventional medical image analysis technology by providing a continuous brain magnetic resonance imaging analysis system and method capable of detecting and visualizing brain lesions by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) of the same patient.

[0011] The purpose of the present invention is to provide a continuous brain magnetic resonance imaging analysis system and method capable of detecting and visualizing Silent Brain Infarction (SBI) by comparing and analyzing continuous brain magnetic resonance imaging (MRI) of the same patient.

[0012] The present invention aims to provide a continuous brain magnetic resonance imaging analysis system and method that automatically detects new brain lesions in continuous FLAIR MRI images by inputting a reference FLAIR MRI image and a tracking FLAIR MRI image, and by analyzing the difference between the reference and tracking images based on deep learning to measure the location and size of the detected lesion.

[0013] The purpose of the present invention is to provide a continuous brain magnetic resonance imaging analysis system and method that enables accurate and consistent diagnostic support for asymptomatic cerebral infarction and efficient decision-making support for medical staff through visualization output that highlights detected lesions, quantifies changes, and maps lesions by location.

[0014] The present invention aims to provide a continuous brain magnetic resonance imaging analysis system and method that can improve the accuracy of difference analysis between reference and tracking images based on deep learning by performing image normalization and image registration and noise removal when reference FLAIR MRI image and tracking FLAIR MRI image data are input.

[0015] Other objectives of the present invention are not limited to those mentioned above, and other unmentioned objectives will be clearly understood by those skilled in the art from the description below.

[0016] A continuous brain magnetic resonance imaging analysis system according to the present invention for achieving the above-mentioned purpose comprises an electronic device including at least one processor for continuous brain magnetic resonance imaging analysis, wherein the electronic device comprises: an image input unit that inputs a reference FLAIR MRI image and a tracking FLAIR MRI image to automatically detect a new brain lesion in a continuous FLAIR MRI image of the same patient; a preprocessing unit that performs normalization and image registration of the reference FLAIR MRI image and the tracking FLAIR MRI image and performs noise removal; a deep learning-based analysis unit that detects a new lesion by analyzing the difference between the reference and tracking images based on deep learning and measures the location and size of the lesion; and a visualization output unit that highlights the detected lesion, quantifies the amount of change, maps the lesion by location, and outputs a visualization.

[0017] Here, the deep learning-based analysis unit is characterized by detecting Silent Brain Infarction (SBI) by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) of the same patient.

[0018] It is further characterized by including a deep learning model construction and verification unit that constructs and verifies a deep learning model for analyzing the difference between reference and tracking images based on deep learning.

[0019] And the deep learning model construction and verification unit is characterized by selecting a patient who has undergone at least two consecutive brain magnetic resonance imaging scans to create training data for building a deep learning model, wherein the first MRI scan image becomes the reference FLAIR MRI image (Base FLAIR) and the subsequent MRI scan image is used as the follow-up FLAIR MRI image (Follow-up FLAIR).

[0020] It is also characterized by creating a single paired slice by matching a base slice and a follow-up slice at the same anatomical level, and ensuring that each patient has multiple such paired slices.

[0021] It is also characterized by creating high-accuracy training data by labeling all generated paired slices as 'Changed' if a new brain lesion is present and 'No change' if no new lesion is present.

[0022] In addition, to input into a deep learning model, it is characterized by resampling to 256 x 256 pixels, performing skull stripping, and histogram matching between time points to merge a base slice and a follow-up slice to create a 2-channel image.

[0023] In addition, the image input unit is characterized by including a reference image data input unit that inputs a reference FLAIR MRI image to automatically detect a new brain lesion in a continuous FLAIR MRI image, and a tracking image data input unit that inputs a tracking FLAIR MRI image to automatically detect a new brain lesion in a continuous FLAIR MRI image.

[0024] The preprocessing unit is characterized by including an image normalization unit that normalizes a reference FLAIR MRI image and a tracking FLAIR MRI image input through an image input unit to detect new lesions by analyzing the difference between reference and tracking images based on deep learning, an image matching unit that matches the images normalized by the image normalization unit, and a noise removal unit that removes noise from the images matched by the image matching unit so that the input to the deep learning model can be used.

[0025] And the deep learning-based analysis unit is characterized by including an image difference analysis unit that analyzes the difference between a reference FLAIR MRI image and a tracking FLAIR MRI image using a deep learning model, a lesion detection unit that detects new lesions in the tracking FLAIR MRI image using the analysis results from the image difference analysis unit, and a lesion location and size measurement unit that measures the location and size of the lesions detected by the lesion detection unit.

[0026] And the visualization output unit is characterized by including a lesion highlighting unit that highlights new lesions detected and measured by the deep learning-based analysis unit, a change amount quantification unit that quantifies the change amount of new lesions detected and measured by the deep learning-based analysis unit, and a lesion mapping unit that provides lesion mapping by location.

[0027] A method for analyzing continuous brain magnetic resonance imaging according to the present invention for achieving other purposes is characterized by performing an operation for analyzing continuous brain magnetic resonance imaging in an electronic device comprising at least one processor, and comprising the steps of: inputting a reference FLAIR MRI image and inputting a tracking FLAIR MRI image to automatically detect a new brain lesion in a continuous FLAIR MRI image; a data preprocessing step of performing image normalization and image registration and noise removal; a step of detecting a new lesion by analyzing the difference between the reference and tracking images using a deep learning model and measuring the location and size of the lesion; and a step of highlighting the detected lesion, quantifying the amount of change, mapping the lesion by location, and producing a visualization output.

[0028] In addition, during the step of measuring the location and size of the lesion, it is characterized by detecting Silent Brain Infarction (SBI) by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) of the same patient.

[0029] And in the step of detecting new lesions and measuring the location and size of the lesions, in the first stage, global average pooling is performed in the encoder of a 3-layer CNN having 128 filters each, and supervised contrastive learning is performed to pull identical labels closer and different labels are pushed away in the embedding space, and in the second stage, binary classification of (Change / No change) is performed through a classifier with a Frozen encoder and a single output node structure, and training and evaluation are performed.

[0030] In addition, to verify the ability of the deep learning model to detect asymptomatic cerebral infarction and to evaluate whether the detection of asymptomatic cerebral infarction effectively predicts the clinical outcome of stroke patients, the diagnostic yield is evaluated by analyzing the time of event occurrence of selected patients, excluding patients with symptomatic cerebral infarction on the index date, and the evaluation indicators include AUC (Area Under the Curve), Sensitivity, Specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value).

[0031] The continuous brain magnetic resonance imaging analysis system and method according to the present invention, as described above, has the following effects.

[0032] First, it enables the detection and visualization of brain lesions by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) scans of the same patient.

[0033] Second, it enables the detection and visualization of Silent Brain Infarction (SBI) by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) scans of the same patient.

[0034] Third, to automatically detect new brain lesions in continuous FLAIR MRI images, reference FLAIR MRI images and tracking FLAIR MRI images are input, and the difference between the reference and tracking images is analyzed based on deep learning to measure the location and size of the detected lesions, thereby enabling the automatic detection of new brain lesions.

[0035] Fourth, by providing a visualization output that highlights detected lesions, quantifies changes, and maps lesions by location, it enables accurate and consistent diagnostic support for asymptomatic cerebral infarction and efficient decision-making support for medical staff.

[0036] Fifth, when reference FLAIR MRI image and tracking FLAIR MRI image data are input, image normalization and image registration are performed, and noise is removed to preprocess the data, thereby improving the accuracy of the difference analysis between the reference and tracking images based on deep learning.

[0037] FIG. 1 is a configuration diagram of a continuous brain magnetic resonance imaging analysis system according to the present invention.

[0038] Figure 2 is a detailed configuration diagram of the image input unit.

[0039] Figure 3 is a detailed configuration diagram of the preprocessing unit.

[0040] Figure 4 is a detailed configuration diagram of the deep learning-based analysis unit.

[0041] Figure 5 is a detailed configuration diagram of the visualization output section.

[0042] FIG. 6 is a flowchart illustrating a continuous brain magnetic resonance imaging analysis method according to the present invention.

[0043] FIG. 7 is a flowchart illustrating an example of a deep learning model construction and verification process according to the present invention.

[0044] FIGS. 8A and 8B are image configuration diagrams illustrating an example of lesion change detection during the deep learning model construction and verification process according to the present invention.

[0045] Figures 9a and 9b are performance comparison graphs between internal and external datasets.

[0046] FIGS. 10a and 10b are image configuration diagrams illustrating an example of a mapping result according to continuous brain magnetic resonance imaging analysis according to the present invention.

[0047] Hereinafter, preferred embodiments of the continuous brain magnetic resonance imaging analysis system and method according to the present invention will be described in detail as follows.

[0048] The features and advantages of the continuous brain magnetic resonance imaging analysis system and method according to the present invention will become apparent from the detailed description of each embodiment below.

[0049] Figure 1 is a configuration diagram of a continuous brain magnetic resonance imaging analysis system according to the present invention.

[0050] The terms used in this disclosure have been selected to be as widely used and general as possible, taking into account their functions within this disclosure; however, these terms may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been selected at the applicant's discretion, and in such cases, their meanings will be described in detail in the relevant description of the invention. Therefore, terms used in this disclosure should be defined not merely by their names, but based on their meanings and the overall content of this disclosure.

[0051] When a part of a specification is described as "including" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part" or "module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software.

[0052] In particular, units that process at least one function or operation may be implemented as an electronic device including at least one processor, and at least one peripheral device may be connected to the electronic device depending on the method of processing the function or operation.

[0053] In the following description, "brain lesion" may include, but is not limited to, Silent Brain Infarction (SBI) and new brain infarctions passing through the subacute phase.

[0054] In addition, while the use of FLAIR MRI images for brain magnetic resonance imaging was described as an example, it is not limited to this, and it is obvious that other images, including T1-weighted images, can be used.

[0055] The continuous brain magnetic resonance imaging analysis system and method according to the present invention is capable of detecting and visualizing Silent Brain Infarction (SBI) by comparing and analyzing continuous brain magnetic resonance imaging (MRI) of the same patient.

[0056] To this end, the present invention may include a configuration that automatically detects new brain lesions in continuous FLAIR MRI images by inputting a reference FLAIR MRI image and a tracking FLAIR MRI image, and by analyzing the difference between the reference and tracking images based on deep learning to measure the location and size of the detected lesion.

[0057] The present invention may include a configuration that enables accurate and consistent diagnostic support for asymptomatic cerebral infarction and efficient decision-making support for medical staff through a visualization output that highlights detected lesions, quantifies the amount of change, and maps lesions by location.

[0058] The present invention may include a configuration that, when reference FLAIR MRI image and tracking FLAIR MRI image data are input, performs image normalization and image registration and noise removal to preprocess the data, thereby increasing the accuracy of the difference analysis between reference and tracking images based on deep learning.

[0059] As shown in FIG. 1, the continuous brain magnetic resonance imaging analysis system according to the present invention includes an image input unit (10) that inputs a reference FLAIR MRI image and a tracking FLAIR MRI image to automatically detect a new brain lesion in a continuous FLAIR MRI image of the same patient, a preprocessing unit (20) that performs data preprocessing by normalizing and registering the reference FLAIR MRI image and the tracking FLAIR MRI image and removing noise, a deep learning-based analysis unit (30) that detects a new lesion by analyzing the difference between the reference and tracking images based on deep learning and measures the location and size of the lesion, and a visualization output unit (40) that highlights the detected lesion, quantifies the amount of change, maps the lesion by location, and outputs a visualization.

[0060] Here, the continuous brain magnetic resonance imaging analysis system according to the present invention further includes a deep learning model construction and verification unit (50) that performs deep learning model construction and verification for analysis of differences between reference and tracking images based on deep learning.

[0061] And the deep learning-based analysis unit (30) detects silent brain infarction (SBI) by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) of the same patient.

[0062] In the present invention, a deep learning model mounted on a deep learning-based analysis unit (30) performs the function of detecting a change (FLAIR change) by comparing a baseline and a follow-up in a FLAIR image, and is configured to detect Silent Brain Infarction (SBI).

[0063] The continuous brain magnetic resonance imaging analysis system according to the present invention, having such a configuration, achieves the effects of improving the early detection and management of asymptomatic cerebral infarction, enhancing diagnostic accuracy and consistency, and increasing the work efficiency of medical staff from a medical perspective.

[0064] In terms of technical aspects, it offers automated lesion detection and analysis, the ability to measure quantitative changes, and the effect of ensuring high reproducibility.

[0065] From an industrial perspective, it has the effect of reducing medical costs, decreasing social costs through enhanced preventive treatment, and increasing the potential for entering global markets.

[0066] The detailed configuration of the video input unit (10) is as follows.

[0067] Figure 2 is a detailed configuration diagram of the video input section.

[0068] As shown in FIG. 2, the image input unit (10) includes a reference image data input unit (11) that inputs a reference FLAIR MRI image to automatically detect a new cerebral infarction lesion in a continuous FLAIR MRI image, and a tracking image data input unit (12) that inputs a tracking FLAIR MRI image to automatically detect a new cerebral infarction lesion in a continuous FLAIR MRI image.

[0069] The detailed configuration of the preprocessing unit (20) is as follows.

[0070] Figure 3 is a detailed configuration diagram of the preprocessing unit.

[0071] As shown in FIG. 3, the preprocessing unit (20) includes an image normalization unit (21) that normalizes a reference FLAIR MRI image and a tracking FLAIR MRI image input through an image input unit (10) to detect new lesions by analyzing the difference between the reference and tracking images based on deep learning, an image matching unit (22) that matches the images normalized by the image normalization unit (21), and a noise removal unit (23) that removes noise from the images matched by the image matching unit (22) so that the input to the deep learning model can be used.

[0072] The detailed configuration of the deep learning-based analysis unit (30) is as follows.

[0073] Figure 4 is a detailed configuration diagram of the deep learning-based analysis unit.

[0074] As shown in FIG. 4, the deep learning-based analysis unit (30) includes an image difference analysis unit (31) that analyzes the difference between a reference FLAIR MRI image and a tracking FLAIR MRI image using a deep learning model, a lesion detection unit (32) that detects a new lesion in the tracking FLAIR MRI image using the analysis results from the image difference analysis unit (31), and a lesion location and size measurement unit (33) that measures the location and size of the lesion detected by the lesion detection unit (32).

[0075] The detailed configuration of the visualization output unit (40) is as follows.

[0076] Figure 5 is a detailed configuration diagram of the visualization output section.

[0077] As shown in FIG. 5, the visualization output unit (40) includes a lesion highlighting unit (41) that highlights new lesions detected and measured by the deep learning-based analysis unit (30), a change amount quantification unit (42) that quantifies the change amount of new lesions detected and measured by the deep learning-based analysis unit (30), and a lesion mapping unit (43) that provides lesion mapping by location.

[0078] The continuous brain magnetic resonance imaging analysis method according to the present invention is described in detail as follows.

[0079] FIG. 6 is a flowchart illustrating a continuous brain magnetic resonance imaging analysis method according to the present invention.

[0080] As shown in FIG. 6, the continuous brain magnetic resonance imaging analysis method according to the present invention includes a step (S601) of inputting a reference FLAIR MRI image and a tracking FLAIR MRI image to automatically detect a new cerebral infarction lesion in a continuous FLAIR MRI image, a data preprocessing step (S602) of performing image normalization and image registration and noise removal, a step (S603) of detecting a new lesion and measuring the location and size of the lesion by analyzing the difference between the reference and tracking images based on deep learning, and a step (S604) of highlighting the detected lesion, quantifying the amount of change, and mapping the lesion by location to produce a visualization output.

[0081] The method according to such an embodiment of the present invention can be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium.

[0082] In this case, the medium may continuously store a program executable by a computer, or temporarily store it for execution or download. Additionally, the medium may be various recording or storage means in the form of a single or multiple combined hardware components; it is not limited to a medium directly connected to a computer system, but may also exist distributed across a network.

[0083] The process of constructing and verifying a deep learning model according to the present invention is specifically described as follows.

[0084] FIG. 7 is a flowchart illustrating an example of a deep learning model construction and verification process according to the present invention.

[0085] In the present invention, training data for building a deep learning model is selected based on the following criteria, and a process to increase accuracy is carried out.

[0086] First, patients are selected who have undergone at least two consecutive brain magnetic resonance imaging scans.

[0087] The first MRI scan image serves as the base FLAIR MRI image, and subsequent MRI scan images are used as follow-up FLAIR MRI images.

[0088] Then, a one-paired slice is created by matching a base slice and a follow-up slice at the same anatomical level, and multiple such paired slices can be created per patient.

[0089] The reason for creating paired slices in this way is to enable the direct comparison of the same brain regions at different time points.

[0090] In the process of generating training data for building such a deep learning model, images that are not axial plane FLAIR images or do not cover the entire brain are excluded, and images of major new lesions not attributable to cerebral infarction (including white degeneration, tumors, or intracranial hemorrhage) are excluded.

[0091] FIGS. 8A and FIGS. 8B are image configuration diagrams illustrating an example of lesion change detection in the process of building and verifying a deep learning model according to the present invention.

[0092] All generated paired slices are labeled 'Changed' if a new stroke lesion is present and 'No change' if no new lesion is present to create high-accuracy training data.

[0093] Then, to input into a deep learning model, the image is resampled to 256 x 256 pixels, skull stripping is performed, and histogram matching between time points is conducted to merge the base slice and follow-up slice to create a 2-channel image.

[0094] The structure of a deep learning model that performs learning using data created in this way is explained as follows.

[0095] In the first stage, global average pooling is performed on the encoders of a 3-layer CNN each having 128 filters, and supervised contrastive learning is performed to pull identical labels closer together in the embedding space and push different labels away.

[0096] In the second stage, binary classification of (Change / No change) is performed through a classifier with a frozen encoder and a single output node structure.

[0097] Training and evaluation are conducted through this process.

[0098] Then, the ability of the deep learning model constructed by the present invention to detect asymptomatic cerebral infarction is verified, and whether the detection of asymptomatic cerebral infarction effectively predicts the clinical outcome of stroke patients is evaluated.

[0099] This evaluates the diagnostic yield by analyzing the time of event occurrence of selected patients, excluding patients with cerebral infarction who have symptoms on the index date, and the evaluation indicators may include AUC (Area Under the Curve), Sensitivity, Specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value).

[0100] Specifically, the clinical outcome prediction is explained as follows.

[0101] To evaluate the predictive clinical outcomes of selected patients, excluding patients with symptomatic cerebral infarction on the index date, the occurrence of a new clinically evident stroke after the index date is assessed.

[0102] For AI-based risk stratification, risk groups are defined through patient-level classification based on slice analysis.

[0103] Patients with at least one slice exceeding the stochastic cutoff point (indicating the presence of asymptomatic cerebral infarction) are classified as a high-risk group, and patients without a slice exceeding the stochastic cutoff point are classified as a low-risk group.

[0104] Then, analyze the time of event occurrence when the primary result is stroke recurrence.

[0105] The analysis method includes Kaplan-Meier estimation for cumulative incidence rates, Log-rank verification for group comparison, and Cox proportional hazards regression with reference to the AI ​​low-risk group and the evaluation of proportional hazard assumptions using Schoenfeld residuals.

[0106] In addition, the diagnostic yield evaluation is conducted through the calculation of diagnostic yield over time and the comparison of cumulative risk between groups (risk ratio estimation).

[0107] Table 1 shows the baseline characteristics (statistics) of the training and validation data.

[0108]

[0109] Figures 9a and 9b are graphs comparing performance between internal and external data sets.

[0110] Table 2 shows the results of the diagnostic performance evaluation.

[0111]

[0112] FIGS. 10a and FIGS. 10b are image configuration diagrams showing an example of a mapping result according to continuous brain magnetic resonance imaging analysis according to the present invention.

[0113] The continuous brain magnetic resonance imaging analysis system and method according to the present invention described above is capable of detecting and visualizing Silent Brain Infarction (SBI) by comparing and analyzing continuous brain magnetic resonance imaging (MRI) of the same patient. It is capable of automatically detecting new brain infarction lesions by inputting a reference FLAIR MRI image and a tracking FLAIR MRI image, and analyzing the difference between the reference and tracking images based on deep learning to measure the location and size of the detected lesion.

[0114] As explained above, it will be understood that the present invention is implemented in a modified form without departing from the essential characteristics of the invention.

[0115] Therefore, the described embodiments should be considered in an illustrative rather than a limiting sense, and the scope of the invention is defined by the claims rather than the foregoing description, and all variations within the equivalent scope should be interpreted as being included in the invention.

[0116] The present invention relates to medical image analysis, and specifically to a continuous brain magnetic resonance imaging analysis system and method capable of detecting and visualizing brain lesions by comparing and analyzing continuous brain magnetic resonance imaging (MRI) of the same patient.

Claims

1. An electronic device comprising at least one processor for continuous brain magnetic resonance imaging analysis, An image input unit that inputs a reference FLAIR MRI image and a tracking FLAIR MRI image to automatically detect new brain lesions in consecutive FLAIR MRI images of the same patient; A preprocessing unit that performs normalization and image registration of reference FLAIR MRI images and tracking FLAIR MRI images, and removes noise; A deep learning-based analysis unit that detects new lesions and measures the location and size of lesions by analyzing the difference between reference and tracking images based on deep learning; A continuous brain magnetic resonance imaging analysis system characterized by including a visualization output unit that highlights detected lesions, quantifies changes, and maps lesions by location to produce a visual output.

2. In claim 1, the deep learning-based analysis unit, A continuous brain magnetic resonance imaging analysis system characterized by detecting silent brain infarction (SBI) by comparing and analyzing consecutive brain magnetic resonance imaging (MRI) of the same patient.

3. A continuous brain magnetic resonance imaging analysis system according to claim 1, further comprising a deep learning model construction and verification unit that performs deep learning model construction and verification for analyzing differences between reference and tracking images based on deep learning.

4. In Paragraph 3, the deep learning model construction and verification unit, in order to create training data for constructing a deep learning model, A continuous brain magnetic resonance imaging analysis system characterized by selecting a patient who has undergone at least two consecutive brain magnetic resonance imaging scans, wherein the first MRI scan image becomes the base FLAIR MRI image and the subsequent MRI scan image is used as the follow-up FLAIR MRI image.

5. In Clause 4, a one paired slice is created by matching a base slice and a follow-up slice that correspond at the same anatomical level, and A continuous brain magnetic resonance imaging analysis system characterized by having multiple pairs of such paired slices per patient.

6. In Clause 5, all of the paired slices made, A continuous brain magnetic resonance imaging analysis system characterized by creating high-accuracy training data by labeling the presence of a new brain lesion as 'Changed' and the absence of a new lesion as 'No change'.

7. In accordance with Clause 6, in order to input into a deep learning model, A continuous brain magnetic resonance imaging analysis system characterized by resampling to 256 x 256 pixels, performing skull stripping, and histogram matching between time points to merge a base slice and a follow-up slice to create a 2-channel image.

8. In claim 1, the image input unit, A reference image data input unit for inputting reference FLAIR MRI images to automatically detect new brain lesions in continuous FLAIR MRI images, and A continuous brain magnetic resonance imaging analysis system characterized by including a tracking image data input unit that inputs tracking FLAIR MRI images to automatically detect new brain lesions in continuous FLAIR MRI images.

9. In claim 1, the pretreatment unit, An image normalization unit that normalizes the reference FLAIR MRI image and the tracking FLAIR MRI image input through the image input unit to detect new lesions by analyzing the difference between the reference and tracking images based on deep learning, and An image matching unit that matches images normalized in an image normalization unit, and A continuous brain magnetic resonance imaging analysis system characterized by including a noise removal unit that removes noise from an image registered in an image registration unit to enable the use of input for a deep learning model.

10. In claim 1, the deep learning-based analysis unit, An image difference analysis unit that analyzes the difference between a reference FLAIR MRI image and a tracking FLAIR MRI image using a deep learning model, and A lesion detection unit that detects new lesions in tracking FLAIR MRI images using the analysis results from the image difference analysis unit, and A continuous brain magnetic resonance imaging analysis system characterized by including a lesion location and size measuring unit that measures the location and size of a lesion detected by a lesion detection unit.

11. In claim 1, the visualization output unit, A lesion highlighting unit that highlights new lesions detected and measured by a deep learning-based analysis unit, and A change quantification unit that quantifies the amount of change of new lesions detected and measured by a deep learning-based analysis unit, and A continuous brain magnetic resonance imaging analysis system characterized by including a lesion mapping unit that provides location-specific lesion mapping.

12. An operation for continuous brain magnetic resonance imaging analysis is performed in an electronic device comprising at least one processor, and A step of inputting a reference FLAIR MRI image and a tracking FLAIR MRI image to automatically detect new brain lesions in continuous FLAIR MRI images; Data preprocessing step involving image normalization, image registration, and noise removal; A step of detecting new lesions and measuring the location and size of lesions by analyzing the difference between reference and tracking images using a deep learning model; A continuous brain magnetic resonance imaging analysis method characterized by including the step of highlighting detected lesions, quantifying the amount of change, mapping lesions by location, and producing a visualized output.

13. In Clause 12, at the step of measuring the location and size of the lesion, A method for analyzing consecutive brain magnetic resonance imaging (MRI) images characterized by detecting silent brain infarction (SBI) by comparing and analyzing consecutive brain MRI images of the same patient.

14. In the step of detecting a new lesion and measuring the location and size of the lesion in accordance with Clause 12, In the first stage, global average pooling is performed on the encoders of a 3-layer CNN, each having 128 filters, and supervised contrastive learning is performed to pull identical labels closer together in the embedding space and push different labels away, and In the second stage, binary classification of (Change / No change) is performed through a Frozen encoder and a classifier with a single output node structure, and A continuous brain magnetic resonance imaging analysis method characterized by training and evaluation.

15. In accordance with Paragraph 14, in order to verify the ability of a deep learning model to detect asymptomatic cerebral infarction and to evaluate whether asymptomatic cerebral infarction detection effectively predicts clinical outcomes in stroke patients, Evaluate the diagnostic yield by analyzing the event occurrence times of selected patients, excluding symptomatic cerebral infarction patients on the index date, and A continuous brain magnetic resonance imaging analysis method characterized by including evaluation indicators such as AUC (Area Under the Curve), Sensitivity, Specificity, PPV (Positive Predictive Value), and NPV (Negative Predictive Value).

Images