KR20210065768A - System for Processing Medical Image and Clinical Factor for Individualized Diagnosis of Stroke - Google Patents

Abstract

The present embodiments provides a system for processing a medical image and a clinical factor for individualized diagnosis of stroke that performs image registration using a coordinate system between a CT image and an MRI image, performs registration using a coordinate system between a stroke clinical evaluation (NIHSS) and a brain image, generates a reference image for each image, and applies deep learning-based semantic segmentation, thereby minimizing the loss of image information, accurately predicting a lesion, and comprehensively evaluating individual conditions for stroke. The system includes a data acquisition part and an image processing part.

Description

{System for Processing Medical Image and Clinical Factor for Individualized Diagnosis of Stroke}

The technical field to which the present invention pertains relates to a medical image processing system for stroke diagnosis.

The content described in this section merely provides background information for the present embodiment and does not constitute the prior art.

The evaluation of the lesion status using images of stroke patients is an important factor in setting a treatment policy. Stroke disease has differences in symptoms, severity, and prognosis depending on the time of occurrence, location, and size of the lesion. Therefore, various images should be combined for judgment. The golden time is important for stroke treatment, and it is necessary to quickly and accurately identify the patient's condition. Medical images used for stroke diagnosis have different diagnostic purposes and test methods.

Imaging judgment of stroke disease is a visual analysis by a radiologist, and since the report is a descriptive, narrative, and qualitative evaluation, it is limited to analyzing quantitative information.

Existing stroke automation evaluation programs use computed tomography (CT) images to apply quantitative computed tomography scores (Alberta Stroke Program Early CT Scores, ASPECTS) to specific brain regions or use ADC MRI (Apparent Diffusion Coefficient Magnetic Resonance Imaging) images. It uses lesion recognition and extraction technology. Current technologies use a single image-based evaluation method rather than a comprehensive evaluation of multiple images, or apply a subjective evaluation method based on the experience of medical staff rather than an objective evaluation method.

Korean Patent Publication No. 10-1754291 (2017.06.29) Korean Patent Publication No. 10-2021541 (2019.09.06) Korean Patent Publication No. 10-1611489 (2016.04.05)

Embodiments of the present invention perform image registration (Coregistration) between a CT image and an MRI image using a coordinate system, and perform registration (Coregistration) using a coordinate system between a clinical stroke clinical evaluation (NIHSS) and a brain image, and a reference image for each image. The main purpose of the invention is to minimize the loss of image information, to accurately predict the lesion, and to enable a comprehensive individual status evaluation for stroke by generating and applying deep learning-based semantic segmentation.

Other objects not specified in the present invention may be additionally considered within the scope that can be easily inferred from the following detailed description and effects thereof.

According to an aspect of the present embodiment, in a medical image processing apparatus for diagnosing a stroke, a data acquisition unit for acquiring a plurality of types of medical images and a clinical evaluation of stroke, and a plurality of types of standardized medical images for the plurality of types of medical images and an image processing unit configured to extract image features of lesions for each of the standardized medical images by applying a lesion recognition model to the plurality of types of standardized medical images.

According to another aspect of the present embodiment, in a medical image processing method by a medical image processing apparatus, acquiring a plurality of types of medical images and a stroke clinical evaluation, and generating a plurality of types of standardized medical images for the plurality of types of medical images generating, applying a lesion recognition model to the plurality of types of medical standardized images to extract image features for lesions for each of the medical standardized images; Pattern of the image features according to the clinical evaluation of stroke for the medical standardized image analyzing the stroke clinical evaluation, mapping image features according to the stages of the stroke clinical evaluation, and stroke clinical factors and storing them in a medical image database, and in the stages of the stroke clinical evaluation and the stroke clinical evaluation Provided is a medical image processing method comprising outputting an individual stroke evaluation based on image characteristics according to the present invention.

As described above, according to the embodiments of the present invention, image registration using a coordinate system is performed between a CT image and an MRI image, and registration is performed between a clinical stroke clinical evaluation (NIHSS) and a brain image using a coordinate system, and each image By generating a reference image and applying deep learning-based semantic segmentation, loss of image information can be minimized, lesions can be accurately predicted, and a comprehensive individual status evaluation for stroke is effective.

Even if it is an effect not explicitly mentioned herein, the effects described in the following specification expected by the technical features of the present invention and their potential effects are treated as if they were described in the specification of the present invention.

1 is a block diagram illustrating a medical image processing system according to embodiments of the present invention.
2 is a block diagram illustrating a medical image processing apparatus according to an embodiment of the present invention.
3 is a flowchart illustrating a medical image processing method according to another embodiment of the present invention.
4 is a flowchart illustrating that a medical image processing method generates a standardized medical image according to another embodiment of the present invention.
5 is a flowchart illustrating segmentation of a medical image by a medical image processing method according to another embodiment of the present invention.
6 is a flowchart illustrating generation of a brain structure normalization region by a medical image processing method according to another embodiment of the present invention.
7 is a flowchart illustrating generation of a standardized image based on a stroke symptom by a medical image processing method according to another embodiment of the present invention.
8 and 9 are diagrams illustrating a medical image and a standardized image processed by a medical image processing method according to another embodiment of the present invention.
10 is a diagram illustrating a lesion detection model applied by a medical image processing method according to another embodiment of the present invention.
11 and 12 are diagrams illustrating analysis of patterns of image features by a medical image processing method according to another embodiment of the present invention.
13 to 15 are diagrams illustrating an example of outputting an individual stroke evaluation by a medical image processing method according to another embodiment of the present invention.

Hereinafter, in the description of the present invention, when it is determined that the subject matter of the present invention may be unnecessarily obscured as it is obvious to those skilled in the art with respect to related known functions, the detailed description thereof will be omitted, and some embodiments of the present invention will be described. It will be described in detail with reference to exemplary drawings.

1 is a block diagram illustrating a medical image processing system according to embodiments of the present invention.

The medical image processing system includes a medical image diagnosis apparatus 10 , a medical image processing apparatus 20 , and a medical image database 30 .

The medical image processing system is a personalized automatic stroke diagnosis assistance program. For an objective and precise diagnosis of stroke disease, the medical image processing system uses artificial intelligence (AI) to comprehensively analyze the patterns of multiple images and clinical factors for each risk stage of the disease. , apply a quantitative scoring method using brain region recognition technology and lesion detection technology. An AI-CT image emphasizing stroke lesion characteristics is generated using artificial intelligence pattern recognition technology from a stroke disease database.

The medical imaging apparatus 10 refers to CT, X-ray, MRI equipment, and the like, and may acquire medical images such as a CT image and an MRI image by photographing a brain region of a subject.

CT images are used to detect cerebral infarction and hemorrhage, and have high contrast of cerebral hemorrhagic lesions. However, CT images have low contrast of cerebral infarct lesions.

MRI images are used to detect cerebral infarction and hemorrhage, and have a high contrast of cerebral infarction and hemorrhagic lesions. However, the speed and accessibility of image acquisition are limited in MRI images, and cerebral hemorrhagic lesions are overestimated.

Brain images are generated using a combined pulse sequence according to the needs of each image. Accordingly, the brain images may have different modalities. ADC is an image generated by calculation based on DWI.

In order to obtain an image with MRI, a large number of RF pulses must be applied. TR (repetition time) is a time interval for applying RF pulses of the same size, and is a repetition time. TE (time to echo) is delay time or echo time. Since the signal cannot be measured immediately after the RF pulse, a short time is waited, and the signal is measured after a certain period of time has elapsed. This short time is called TE. TR and TE can be controlled by the examiner. By appropriately adjusting TR and TE, it is possible to obtain a T1-weighted image or a T2-weighted image that can be applied to clinical practice.

Diffusion Weighted Imaging (DWI) is an image mapping the diffusion motion of molecules, particularly water molecules, in living tissue. The diffusion of water molecules within the tissue is not free. DWI reflects the impact of water molecules on fibrous tissue or membranes. The diffusion pattern of water molecules indicates the normal or abnormal state of the tissue. DWI can well represent the normal and abnormal state of the fibrous structure of the brain's white matter or gray matter.

The principle of diffusion-weighted imaging (DWI) is to sequentially give two motion probe gradient pulses with the same intensity but opposite directions before and after the 180-degree pulse to detect the movement of water. , the second applies a re-phase pulse in the opposite direction according to the polarity changed by 180 degrees. When a signal is received to obtain an image, the signal is lost as much as the phase recovery is less due to the random movement of water molecules, thereby obtaining contrast for each tissue. In this case, the intensity of the gradient pulse is referred to as the b value, which is a major variable affecting the signal intensity of the diffusion-weighted image. Since the signal intensity of the diffusion-weighted image can be affected by the signal intensity of the T2-weighted image, tissues showing high signal intensity in the T2-weighted image may appear as high signal intensity in the diffusion-weighted image as well.

The Apparent Diffusion Coefficient (ADC) is a function of temperature as the diffusion coefficient. Because the cell wall exists in the body and the temperature is non-uniform, ADC can be calculated using DWI. DWI and ADC are out of phase. In the infarct area, the diffusion of water outside the cell is reduced due to the expansion of the cell. The area with reduced diffusion becomes an area with small signal degradation when DWI is taken on the B1000, and it appears brightly in the DWI image. On the other hand, the region with reduced diffusion appears darker than normal in ADC. Water, such as Cerebrospinal Fluid (CSF), is a free diffusion region where ADC is bright and DWI is dark.

Perfusion Weighted Imaging (PWI) is a perfusion image showing blood flow. For example, when gadolinium is injected as a contrast agent and an MRI image is taken, parameters such as blood flow volume, blood flow velocity, mean transit time (MTT), and time to peak (TTP) may be acquired.

A FLAIR image is an image in which the signal from the fluid is nullified. FLAIR imaging is used to suppress the effects of cerebrospinal fluid in images when acquiring brain images with MRI. FLAIR images show the anatomy of the brain well. Signals from a specific tissue can be suppressed by well-selecting the inversion time according to the tissue.

The T1- and T2-weighted image is an image in which the T1 or T2 effect of a specific tissue is emphasized by controlling TR and TE. In the course of MRI, when the applied RF (radio frequency) pulse is blocked, the protons of the tissue are rearranged in the external magnetic field (B0) direction (Z-axis direction) while releasing the absorbed energy to the surrounding tissue. T1 is the time constant for the realignment of the proton's spindle along the Z-axis, that is, the time constant of the curve where the Z-direction magnetization is recovered. T1 is a time constant of magnetization recovery and is called a longitudinal relaxation time or a spin-lattice relaxation time. On the other hand, when the RF pulse is blocked, the XY component of magnetization decays. T2 is a time constant of the XY component decay curve of magnetization and is called a lateral relaxation time or a spin-spin relaxation time. T1 and T2 are intrinsic values of tissues, and have different values for each water, solid, fat, and protein. Here, when TR is lengthened, the T1 effect is reduced. Conversely, when TR is shortened, the T1 effect (contrast) is increased, that is, a T1-weighted image is obtained. When TE is shortened, the T2 effect is reduced, and when TE is lengthened, the T2 effect is increased, that is, a T2-weighted image is obtained.

An MRA image is an image of blood flow by MRI using a contrast agent.

The medical image processing apparatus 20 includes a data acquisition unit 21 and an image processing unit 22 .

The data acquisition unit 21 acquires a plurality of types of medical images and stroke clinical evaluation. The image processing unit 22 generates multiple types of standardized medical images for multiple types of medical images, and extracts image features of lesions for each standardized medical image by applying a lesion recognition model to the multiple types of standardized medical images.

The medical image processing apparatus 20 acquires a plurality of types of medical images from the medical image diagnosis apparatus 10 . The medical image processing apparatus 20 acquires clinical evaluation factor data from the database.

Multiple types of medical imaging include: Computed Tomography (CT) image, CT Perfusion image, Multimodal CT image, Perfusion Weighted Imaging (PWI) image, and liquid attenuation. Fluid Attenuated Inversion Recovery (FLAIR) image, T1-weighted image, T2-weighted image, Low b-Value Diffusion Weighted Image, High b-Value Diffusion Weighted Imaging, A medical image selected from among a DWI) image and an apparent diffusion coefficient (ADC) image may be included.

Different types of medical imaging have different characteristics and advantages and disadvantages.

CT images are capable of fast imaging time (Widespread Access, Speed Of Acquisition), ease of access, primary differentiation between ischemic stroke and hemorrhagic stroke, and primary differentiation from diseases other than stroke (cerebral hemorrhage, brain tumor, etc.).

CT perfusion image provides CBF (Cerebral Blood Flow), CBV (Cerebral Blood Volume), and MTT (Mean Transit Time) information, and it is possible to distinguish between the penumbra and the infarcted core region.

Multimodal CT imaging includes pre-enhanced CT (non-contrast CT), post-contrast CT (post-contrast CT), CT perfusion, and CT angiography, etc. It is possible to check the presence or absence of circulation (Collateral Vessel).

FLAIR imaging is easy to distinguish the extent of stroke, and it is easy to see brain lesions other than stroke.

The DWIb0 (low b-value) image is easy to see the T2 effect.

The DWIb1000 (high b-value) image has the highest sensitivity for stroke diagnosis. Stroke can be diagnosed within 30 minutes of symptom onset.

The ADC map image distinguishes between Cytotoxic Edema and Vasogenic Edema, both of which have the same high signal intensity in DWI, and provides quantitative information on the degree of ischemia. A distinction can be made between acute and subacute stroke.

PWI image provides information similar to CTP in case of DSC-PWI. In the case of CASL (Continuous Arterial Spin Labeling), CBF information is provided without using a contrast agent, and the PWI-DWI mismatch area is viewed as a penumbra, which helps to determine the treatment direction as a perfusion treatment area possible.

The clinical evaluation of stroke is the National Institutes of Health Stroke Scale (NIHSS) evaluation, and the level of consciousness, gaze, visual field, facial paralysis, upper extremity motor ability, lower extremity motor ability, limb ataxia ( Ataxia), sensory, speech, dysarthria (Dysarthria), motor extinction (extinCtion) and attention loss (Inattention) consists of 11 items. For each item, a normal score is 0, and a score from 1 to 4 is given depending on the severity and quantified. The more severe the clinical condition, the higher the score, with a maximum score of 42 points.

The image processing unit performs a precise diagnosis of stroke by inputting multiple images. Stroke diagnosis is limited to an accurate diagnosis with only a single image, so a precise diagnosis is performed by combining the advantages of each image. The input image may be CT + Multi-Phase CT + (CT-Perfusion) + DWIb0 (low b-value) + DWIb1000 (high b-value) + ADC map + FLAIR + MR-PWI + NIHSS score. The analysis is performed by combining CT and multiple MR images. Also, CT-Perfusion and MR-Perfusion imaging characteristics are reflected.

The image processing unit is capable of quantitative measurement of lesions based on artificial intelligence. When the volume measurement method through CT or ADC value post-processing technology is applied, it is very difficult to establish an absolute standard for each protocol and subject, so there is a problem that there is a difference between medical institutions and medical equipment. The medical image processing apparatus according to the present embodiment uses an artificial intelligence technique showing high accuracy in image lesion extraction to learn the characteristics of various brain image lesions, and then extracts lesions in units of pixels.

The image processing unit uses a pre-made brain function region standardized image (Template) to recognize location and extract a specific brain region from multiple images. The existing stroke analysis technique was a volume-based evaluation of the ischemic center and ischemic pannumbra, but the cellular composition of each brain region is different, the selective vulnerability in the ischemic or hemorrhagic pathology environment is different, and the functional significance is different. The medical image processing apparatus according to the present embodiment enables precise diagnosis and treatment by providing additional information to the existing volume-based evaluation as a structure-based brain classification method.

The image processing unit divides a plurality of types of medical images into gray matter, white matter, and cerebrospinal fluid images, generates a plurality of types of brain structure normalization regions based on correlation factors for the gray matter, white matter, and cerebrospinal fluid images, and generates a plurality of types of brain structures The plurality of types of standardized medical images are generated by matching the standardized area with the stroke symptom-based functional area.

The image processing unit corrects motion and removes noise for a plurality of types of medical images, divides brain tissue regions according to the distribution of signal strength based on different voxel sizes for each type, and performs normalization on signal strength within voxels, , distort the signal intensity of the segmented brain tissue region, perform smoothing on the signal intensity of the segmented brain tissue region, perform normalization on the average image of the signal intensity of the segmented brain tissue region, and change the segmentation of the medical image. Repeat the splitting process until there are none.

Input images include CT, Multi-Phase CT, CT-Perfusion, low b-value DWI, high b-value DWI, ADC map, FLAIR, MR-PWI, etc., and performs motion compensation and noise removal. The three brain tissue segmentation is performed based on the Gaussian distribution of signal intensity based on the voxel size for each input image. For example, DWI / ADC is 1.8x1.8x5 mm ^3, T2WI is 0.56x0.75x5.0 mm ^3, FLAIR is 0.65x0.9x5.0 mm ^3, PWI will be 1.72x1.75x4.5mm ³ and , the input voxel size may be changed according to required design matters.

Affine regularization is performed to optimize the signal intensity in the corresponding voxel, and the signal intensity is distorted and registered in the voxel for registration of the divided tissue. FWHM (Full Width Half Maximum) can be smoothed by overlapping integration with 70mm. Normalization of the segmented gray matter, white matter, and cerebrospinal fluid signals is performed on the average image in a non-linear manner, and the process is repeated until there is no change in image segmentation. If there is a change, the process from segmentation to normalization is repeatedly performed. As the output image, the gray matter, white matter, and cerebrospinal fluid images divided into the same voxel size as the input image are output.

The image processing unit evaluates the correlation factor in consideration of the location difference between images divided based on age, gender, and disease group, and generates the brain structure standardization region through regression analysis and pairwise approach using the evaluated correlation factor, Combine labeled images with brain structure normalization regions.

The image processing unit performs image evaluation in consideration of correlation factors. For the divided multi-brain region images, evaluation is performed in consideration of the location difference between the images separated based on age, gender, and disease group. Analysis between influence factors may use a general linear model equation.

In the disease group, the size and pattern of stroke vary according to toast classfication. (i) In the case of large vessel atherosclerosis, a relatively large-sized stroke occurs along the area of the blood vessel because the embolism travels from the large artery to the artery and invades a large blood vessel such as the middle cerebral artery within the skull. (ii) In the case of small vessel occlusion, a stroke smaller than 5 mm is characterized by being limited to a specific location area (basal ganglia, thalamus, medulla oblongata, cerebellum). In the case of Cardioembolic Stroke, since the embolus of the heart flies to the head, it appears in the form of a small-sized stroke along the end artery.Also, in the case of moyamoya disease or vasculitis, strokes with a specific area and size appear in the form

The image processing unit matches the brain structure normalization region with the functional region based on the stroke symptom, and uses a distribution histogram to correct the interference error between adjacent tissues for each region, using a portion of the signal strength of each region to correct the image. Only 95% of the signal strength of each area can be used by using a Gaussian distribution histogram to correct errors caused by interference between adjacent tissues in each area.

Multiple types of medical standardized images are: Computed Tomography (CT) standardized image, CT Perfusion standardized image, Multimodal CT (Multimodal CT) standardized image, and Perfusion Weighted Imaging (PWI) standardized image. A standardized image selected from a plurality of images, a Fluid Attenuated Inversion Recovery (FLAIR) normalized image, a T1-weighted normalized image, a T2-weighted normalized image, and a Diffusion Weighted Image normalized image may be included.

The medical image processing device creates a new brain symptom-function matching map tailored to cerebral infarction by matching the functions in charge according to the location of the infarction onset and the symptoms of a possible patient. Visually, the location of the lesion - the function of the brain - the patient's symptoms can be quickly and accurately determined, and the severity of the pathological environment can be quantitatively evaluated. Conventional methods have matched the T1 MNI reference image, but since CT and MR images have different dimensions, information loss in each image occurs when the reference image is matched. All medical images have been optimized to more precisely evaluate the target disease, so it is necessary to minimize information loss. Since the medical image processing apparatus according to the present embodiment creates each reference image, information loss can be minimized.

The image processing unit analyzes a pattern of image features according to a stroke clinical evaluation with respect to a medical standardized image. The lesion is extracted using the lesion detection model in which the disease lesion characteristics are learned. The lesion detection model receives all matched medical images (CT, Multi-Phase CT, CT-Perfusion, low b-value DWI, high b-value DWI, ADC map, FLAIR, MR-PWI, etc.) Output the size, location, and signal intensity of the lesion extracted from the fields. Due to each reference image, it is possible to maintain the dimension of each image as it is, and to minimize loss of information about size, position, and signal strength.

The image processing unit extracts the size, position, and signal intensity of the lesion for each medical standardized image, and generates a stroke feature combination set by combining the size, location, and signal intensity of the lesion according to stroke clinical evaluation, and stroke feature combination An individual pattern is analyzed among the sets and a matching stroke feature combination set is output.

The image processing unit extracts a lesion volume and an average signal intensity for each image. Optimized Feature Combination-Set according to clinical evaluation factors (NHISS) is performed. For example, the total number of feature factors may be set to 624 (functional area 39 * number of images (8) * number of image features (2)). The feature factor is optimized using the Feature Dimension Reduction Algorithm. The feature combination can be optimized by creating a new set of feature combinations within the range of 2 to 10.

The image processing unit performs AI pattern analysis of NHISS-image feature factors. Individual pattern analysis is performed with the learned ensemble combination model based on the Multiple-CNN model. The ensemble combination model receives the size, location, and signal intensity of the lesion extracted from each image, and outputs a high-accuracy feature combination set (generated as 2 to 10 sets).

Symptoms and prognosis vary depending on the size of the lesion, as well as the difference between the location of the lesion and the signal size (ex. ADC value). Even if a cerebral infarction with the same size of 1 cm occurs, it does not leave significant sequelae when it occurs in the right cerebral hemisphere, but when it occurs in the left cerebral hemisphere, it is highly likely to cause sequelae of motor dysfunction depending on the location. NHISS-image feature factors can be used to generate a set of feature combinations according to individual pattern analysis.

The image processing unit maps the stroke clinical evaluation, image features according to stages of stroke clinical evaluation, and stroke clinical factors and stores the mapping in the medical image database. Establish a database of NHISS clinical evaluation factors and stroke clinical characteristic factors. The database stores NHISS step-by-step image volumes and signal values.

The database includes clinical factors (patient demographics (age, gender, etc.), stroke risk factors (hypertension, diabetes, history of cerebral infarction, cardiovascular disease, smoking, drinking, hyperlipidemia, atrial fibrillation, etc.), NIHSS at the time of admission, time after symptom onset, Treatment start time, imaging time, image reading findings (CT, Multi-Phase CT, CT-Perfusion, low b-value DWI, high b-value DWI, ADC map, FLAIR, MR-PWI), treatment (IV-tPA) (Intravenous tissue Plasminogen Activator), IAT (Intra-Arterial chemical Thrombolysis)), prognostic factors (3-month mRS (modified Rankin Scale)), etc. are input, and the input data is managed as a database.

Existing previous studies have weak versatility for commonly suggested standards. That is, although prognostic prediction in a single institution may be effective, there is a limit to applying the same standard to other institutions. Since most studies were conducted by extracting only a few factors rather than a comprehensive approach, the problem of bias is always presented. This patent is meaningful in describing and organizing comprehensive information on diagnostic methods, treatment guidelines, and prognosis prediction.

The image processing unit outputs a stroke clinical evaluation and an individual stroke evaluation based on image characteristics according to the stages of the stroke clinical evaluation.

The image processing unit outputs a correlation between the stroke clinical evaluation and image features according to the stages of the stroke clinical evaluation. Through an AI model in which a pattern is learned from the stroke characteristic database, the result is expressed as a percentage of the correlation of each individual characteristic factor. NIHSS items (1a. Level of consciousness, 1b. Questions, 1c. Commands, 2. Best gaze, 3. Visual fields, 4. Facial palsy5a. Left motor arm, 5b. Right motor arm, 6a. Left motor leg, 6b. Right motor leg, 7. Limb ataxia, 8. Sensory, 9. Best language, 10. Dysarthria, 11. Extinction and inattention) were input, and lesion size, signal intensity and input clinical factors were extracted from multiple images by location. Output the highest correlation with Pearson correlation coefficient or Pearson's r value.

The clinical/imaging core factors and NIHSS clinical evaluation factors are built together as a DB, and correlation analysis through AI models collects big data to extract various features, minimizes the influence of unnecessary variables, and calculates weights for important factors. can give If existing studies set the reference point (approximately 70-74 ml) of the volume of cerebral infarction in DWI and provided cut-off values of good and bad prognosis, this patent is a more continuous and long-term data can be provided.

Looking at the stroke severity classification criteria using the NIHSS evaluation method, group 1 represents no stroke symptoms with a score of 0. Group 2 represents a minor stroke with a score of 1-4. Group 3 represents a moderate stroke with 5-15 points. Group 4 represents a Moderate to severe stroke with a score of 16-20. Group 5 represents a severe stroke with a score of 21-42.

The image processing unit outputs scores obtained by calculating the volume of the stroke lesion relative to the brain region for the plurality of types of medical images. The original image and the extracted lesion image are visualized in 3D (axi, sag, cor) form to show the result. The relative volume of the stroke lesion compared to the lesion detection brain region is expressed as a relative percentage, and additional verification can be performed for lesions with a relative numerical value of less than 1%. The total number of brain regions based on stroke function (78 - left and right separately evaluated) is given a total score of 78 points, and the area in which the lesion is detected is deducted from the total score for scoring. Receives pattern analysis results and signals processed from the database and generates 3D lesion images (CT, Multi-Phase CT, CT-Perfusion, CT angiography, low b-value DWI, high b-value DWI, ADC map, FLAIR, MR-PWI). Information expressed as a continuous variable can be fine-tuned through artificial intelligence learning to suit the user's environment (characteristics of the patient group, characteristics of imaging equipment, characteristics of imaging protocol, and characteristics of treatment protocol), thereby improving the accuracy of the result output unit in the future. can

The medical image processing apparatus processes continuous variables of standardized signals as well as volume information and outputs them as relative numerical values. That is, the pixel information is expressed as a continuous variable without binarizing it as {1,0}, and the relationship with the surrounding pixels is statistically summarized and expressed. By measuring the weights of various 3D lesion images, it is possible to generate a 3D map that can be used even in a limited environment (eg, small and medium hospitals) where various imaging techniques do not exist.

The image processing unit captures multiple types of medical images at multiple viewpoints, and time-series changes in size, location, and signal strength of the lesion are tracked.

The image processing unit measures the longitudinal micro change of the stroke image. For continuous imaging tests of stroke patients, the location, size, and average signal value of the lesion are quantitatively indicated. Based on the results of the first examination, the characteristics of the lesion for the follow-up examination are expressed as a relative percentage. Visually indicate the follow-up test results as a graph. The signals generated from the patient's past images, initial hospitalization images, images immediately after hospitalization, images just before discharge, and images for rehabilitation after discharge are expressed in a time series technique.

This patent visually displays time series changes because the continuous variable information is output as a continuous variable without bisecting or truncating it. In the method of classifying the diagnosis, the uncertainty of the TOAST classification of stroke causes (Trial of Org 10172 in Acute Stroke Treatment), rather than the existing definitive classification, is expressed as a probability, helping stroke specialists to make decisions. The severity prediction method is expressed with a different probability from the existing ordinal classification method (NIHSS 1-4, moderate low; NIHSS 5-15, moderate moderate; NIHSS 16-20, moderate risk; NIHSS 21-42, moderate severe). It helps stroke specialists make decisions. Through the graph of time series information expressed as probability, non-specialists such as the patient himself/herself, guardians, and caregivers can understand the trend of test results.

The image processing unit outputs an expected state in text form in consideration of changes in size, location, and signal strength of the tracked lesion.

Since the existing reports are descriptive, narrative, and qualitative evaluations, there are limitations to the analysis of the following quantitative information. The image processing unit automatically describes the location of the stroke lesion for each image, the relative volume ratio of the lesion, the signal intensity, and the lesion evaluation score in text form. In addition to the current patient status, the degree of change of the lesion compared with the previous examination values is described together with quantitative values. Outputs the presence or absence of acute cerebral infarction, the location and size of cerebral infarction. Automated readings perform CT scoring and MRI scoring, indicate NIHSS-IMAGING ROIs correlation quantitative %, indicate follow-up management values, and output expected patient status and prognosis.

The image processing unit generates an AI-CT image emphasizing stroke MR features. The image processing unit extracts a suspected lesion region from the tomography image based on the medical standardized image according to the clinical evaluation of stroke, and the size and signal intensity of the lesion in the tomography image through a feature generation model between the tomography image and the magnetic resonance image Outputs an enhanced tomography image that emphasizes

The image processing unit analyzes the pattern by analyzing the correlation characteristics between the NIHSS score + the CT image and the MR image. Using the information of the NIHSS score, a region suspected of a lesion is detected and extracted. Using pattern recognition AI, we learn the difference in stroke lesion characteristics between CT and MRI. The trained AI model extracts the suspected lesion area by disease severity from the NIHSS score. After extracting the brain region containing the lesion in the CT using a standardized image, the lesion region expansion and signal emphasis are applied through the learned feature difference generation algorithm. The feature difference generation algorithm may use a modeling algorithm as a nonlinear equation, or a generative adversarial network (GAN) algorithm may be used.

The image processing unit receives the NIHSS score, CT image, and ADC map image, and outputs a CT image with an intensified stroke lesion feature in which the size and signal intensity of the target lesion in the CT image are emphasized. By expressing the lesion features similar to those of the MR image only with the CT image, it enables precise diagnosis compared to the short image examination time.

The medical image database 30 refers to a data storage type in which data can be freely searched, extracted, deleted, edited, added, and the like. Databases are Oracle, Infomix, Sybase, Relational Data Base Management System (RDBMS), Gemston, Orion, object-oriented database management system ( An Object Oriented Database Management System (OODBMS) or a distributed cloud may be used to suit the purpose of the present embodiment.

2 is a block diagram illustrating a medical image processing apparatus according to an embodiment of the present invention.

The medical image processing apparatus 20 includes at least one processor 23 , a computer-readable storage medium 24 , and a communication bus 28 .

The data acquisition unit 21 of the medical image processing apparatus 20 may correspond to the input/output interface 26 or the communication interface 27 , and the image processing unit 22 may correspond to the processor 23 .

The processor 23 may control to operate as the medical image processing apparatus 20 . For example, the processor 23 may execute one or more programs stored in the computer-readable storage medium 24 . The one or more programs may include one or more computer-executable instructions, which, when executed by the processor 23 , configure the medical image processing apparatus 20 to perform operations according to the exemplary embodiment. can be

Computer-readable storage medium 24 is configured to store computer-executable instructions or program code, program data, and/or other suitable form of information. The program 25 stored in the computer-readable storage medium 24 includes a set of instructions executable by the processor 23 . In one embodiment, computer-readable storage medium 24 includes memory (volatile memory, such as random access memory, non-volatile memory, or a suitable combination thereof), one or more magnetic disk storage devices, optical disk storage devices, It may be flash memory devices, other types of storage media accessed by the medical image processing apparatus 20 and capable of storing desired information, or a suitable combination thereof.

The communication bus 28 interconnects various other components of the medical image processing apparatus 20 including the processor 23 and the computer-readable storage medium 24 .

The medical image processing apparatus 20 may also include one or more input/output interfaces 26 and one or more communication interfaces 27 that provide interfaces for one or more input/output devices. The input/output interface 26 and the communication interface 27 are connected to the communication bus 28 . The input/output device may be connected to other components of the medical image processing apparatus 20 through the input/output interface 26 .

3 is a flowchart illustrating a medical image processing method according to another embodiment of the present invention. The medical image processing method may be performed by a medical image processing apparatus.

A method of processing a medical image by a medical image processing apparatus includes obtaining a plurality of types of medical images and a clinical stroke clinical evaluation (S310), generating a plurality of types of standardized medical images for the plurality of types of medical images (S320), a plurality of Step of applying the lesion recognition model to the type of medical standardized image to extract image features of the lesion for each of the medical standardized images (S330), analyzing the pattern of the image features according to the stroke clinical evaluation for the medical standardized image (S330) S340), stroke clinical evaluation, image characteristics according to the stages of stroke clinical evaluation, and stroke clinical factors are mapped and stored in a medical image database (S350), and image characteristics according to the stages of stroke clinical evaluation and stroke clinical evaluation and outputting an individual stroke evaluation based on ( S360 ).

4 is a flowchart illustrating that a medical image processing method generates a standardized medical image according to another embodiment of the present invention.

The step of generating a plurality of types of standardized medical images (S320) is a step of dividing the plurality of types of medical images into gray matter, white matter, and cerebrospinal fluid images (S410), gray matter, white matter, and cerebrospinal fluid images based on correlation factors. Generating the plurality of types of brain structure standardization regions (S420), and generating the plurality of types of standardized medical images by matching the functional regions based on stroke symptoms to the plurality of types of brain structure standardization regions (S430) do.

5 is a flowchart illustrating segmentation of a medical image by a medical image processing method according to another embodiment of the present invention.

The step of segmenting the medical image ( S410 ) includes correcting motion and removing noise for a plurality of types of medical images ( S510 ), and segmenting the brain tissue region according to the distribution of signal strength based on different voxel sizes for each type. (S520), performing normalization on the signal strength within the voxel (S530), distorting the signal strength of the divided brain tissue region (S540), and performing smoothing on the signal strength of the divided brain tissue region (S550), performing normalization on the average image of the signal intensity of the divided brain tissue region (S560), and determining the segmentation process until there is no change in segmentation of the medical image (S570). If there is a division change, steps S520 to S560 are repeated.

6 is a flowchart illustrating generation of a brain structure normalization region by a medical image processing method according to another embodiment of the present invention.

The step of generating the brain structure standardization region (S420) is the step of evaluating the correlation factor in consideration of the position difference between the images divided based on age, gender, and disease group (S610), regression analysis using the evaluated correlation factor, and pair Generating the brain structure normalization region through a matching approach (S620), and combining the labeled image with the brain structure normalization region (S630).

7 is a flowchart illustrating generation of a standardized image based on a stroke symptom by a medical image processing method according to another embodiment of the present invention.

The step of generating the standardized image based on stroke symptoms (S430) is the step of matching the brain structure standardized region with the functional region based on the stroke symptoms (S710), using a distribution histogram to correct the interference error between adjacent tissues for each region. and correcting the image using a portion of the signal strength of each region ( S720 ).

Stroke brain functional regions are represented in Table 1.

Stroke symptom-based function - Medical images and standardized images required to match brain regions structural regions are shown in FIGS. 8 and 9 .

The lesion detection model is shown in FIG. 10 . The lesion detection model receives all matched medical images (CT, Multi-Phase CT, CT-Perfusion, low b-value DWI, high b-value DWI, ADC map, FLAIR, MR-PWI, etc.) Output the size, location, and signal intensity of the lesion extracted from the fields. Due to each reference image, it is possible to maintain the dimension of each image as it is, and it is possible to minimize loss of information about size, position, and signal strength.

11 is a diagram illustrating analysis of a pattern of an image feature by a medical image processing method according to another embodiment of the present invention.

The step of analyzing the pattern of image features according to the clinical stroke clinical evaluation with respect to the medical standardized image (S340), the step of extracting the size, location, and signal strength of the lesion for each medical standardized image (S1110), according to the clinical stroke clinical evaluation Generating a stroke feature combination set by combining the size, location, and signal strength of the lesion (S1120), outputting a matching stroke feature combination set by analyzing individual patterns among the stroke feature combination set (S1130) include Fig. 12 shows the extracted feature optimization trial and pattern analysis.

13 to 15 are diagrams illustrating an example of outputting an individual stroke evaluation by a medical image processing method according to another embodiment of the present invention.

The step of outputting the individual stroke evaluation (S360) is a step of outputting a correlation between the stroke clinical evaluation and the image features according to the stages of the stroke clinical evaluation (S1310), the stroke lesion compared to the brain region for a plurality of types of medical images Outputting the volume calculated score (S1320), taking multiple types of medical images at multiple viewpoints, and time-series tracking changes in size, location, and signal strength of the lesion (S1330), the tracked and outputting an expected state in text form in consideration of changes in size, location, and signal strength of the lesion (S1340).

The step of outputting the individual stroke evaluation (S360) is the step of extracting the suspected lesion area from the tomography image based on the medical standardized image according to the stroke clinical evaluation (S1410), the feature generation model between the tomography image and the magnetic resonance image. and outputting an emphasis tomography image emphasizing the size and signal strength of the lesion in the tomography image through the tomography image (S1420).

Referring to FIG. 15 , the medical image processing apparatus according to the present embodiment compares and analyzes a CT signal (HV) and an MRI (ADC value) by voxel-by-voxel within the extracted region. When both CT and MRI are not lesions, A=A', in CT and MRI, when non-lesions are lesions, B=α*B'+K expands the area (range), and when both CT and MRI are lesions, C= Correct (emphasize) the α*C'+K signal strength. By using linear/nonlinear data fitting, an image in which the MRI signal effect is emphasized is generated on a CT image through the completed model.

Although components included in the medical image processing system are illustrated separately in FIG. 1 , a plurality of components may be coupled to each other and implemented as at least one module. The components are connected to a communication path connecting a software module or a hardware module inside the device to operate organically with each other. These components communicate using one or more communication buses or signal lines.

The medical image processing system may be implemented in a logic circuit by hardware, firmware, software, or a combination thereof, or may be implemented using a general-purpose or special-purpose computer. The device may be implemented using a hardwired device, a field programmable gate array (FPGA), an application specific integrated circuit (ASIC), or the like. In addition, the device may be implemented as a system on chip (SoC) including one or more processors and controllers.

The medical image processing system may be implemented in a form of software, hardware, or a combination thereof in a computing device provided with hardware elements. A computing device includes all or part of a communication device such as a communication modem for performing communication with various devices or a wired/wireless communication network, a memory for storing data for executing a program, and a microprocessor for executing an operation and command by executing the program. It can mean a device.

2 to 7, 11, 13, and 14, each process is described as sequentially executed, but this is merely an exemplary description, and those skilled in the art will not deviate from the essential characteristics of the embodiment of the present invention. 2 to 7, 11, 13, 14 may be applied by changing the order described in FIGS. 2 to 7, 11, 13, 14, or by variously modifying and transforming one or more processes in parallel or adding other processes .

The operations according to the present embodiments may be implemented in the form of program instructions that can be performed through various computer means and recorded in a computer-readable medium. Computer-readable media refers to any medium that participates in providing instructions to a processor for execution. Computer-readable media may include program instructions, data files, data structures, or a combination thereof. For example, there may be a magnetic medium, an optical recording medium, a memory, and the like. A computer program may be distributed over a networked computer system so that computer readable code is stored and executed in a distributed manner. Functional programs, codes, and code segments for implementing the present embodiment may be easily inferred by programmers in the technical field to which the present embodiment pertains.

The present embodiments are for explaining the technical idea of the present embodiment, and the scope of the technical idea of the present embodiment is not limited by these embodiments. The protection scope of this embodiment should be interpreted by the following claims, and all technical ideas within the equivalent range should be interpreted as being included in the scope of the present embodiment.

10: medical imaging device
20: medical image processing device
30: medical image database

Claims (18)

A medical image processing apparatus for stroke diagnosis, comprising:
a data acquisition unit that acquires multiple types of medical images and stroke clinical evaluation; and
An image processing unit that generates multiple types of standardized medical images for the multiple types of medical images, and extracts image features of lesions for each of the medical standardized images by applying a lesion recognition model to the multiple types of standardized medical images
A medical image processing device comprising a. According to claim 1,
The plurality of types of medical images,
Computed Tomography (CT) Image, CT Perfusion Image, Multimodal CT Image, Perfusion Weighted Imaging (PWI) Image, Fluid Attenuated Inversion Recovery , FLAIR image, T2-weighted image, Low b-Value Diffusion Weighted Image, High b-Value Diffusion Weighted Imaging (DWI) image, and Apparent Diffusion A medical image processing apparatus comprising a plurality of medical images selected from coefficient (ADC) images. According to claim 1,
The plurality of types of standardized medical images are
Computed Tomography (CT) standardized image, CT Perfusion standardized image, Multimodal CT (Multimodal CT) standardized image, Perfusion Weighted Imaging (PWI) standardized image, Fluid A medical image processing apparatus, comprising: a standardized image selected from a plurality of Attenuated Inversion Recovery, FLAIR standardized images, T2-weighted standardized images, and Diffusion Weighted Image standardized images. According to claim 1,
The image processing unit,
The plurality of types of medical images are divided into gray matter, white matter, and cerebrospinal fluid images, and the plurality of types of brain structure normalization regions are generated based on correlation factors for the gray matter, white matter, and cerebrospinal fluid images, and the plurality of types of brain The medical image processing apparatus according to claim 1, wherein the plurality of types of standardized medical images are generated by matching the structural standardized area with the stroke symptom-based functional area. 5. The method of claim 4,
The image processing unit,
Correcting motion and removing noise for the plurality of types of medical images,
Segmenting the brain tissue region according to the distribution of signal intensity based on different voxel sizes for each type,
Normalize the signal strength within the voxel,
Distorting the signal strength of the divided brain tissue region,
Smoothing is performed on the signal strength of the divided brain tissue region,
Normalization is performed on the average image of the signal intensity of the divided brain tissue region,
and repeating the segmentation process until there is no change in segmentation of the medical image. 5. The method of claim 4,
The image processing unit,
Evaluate the correlation factor by considering the location difference between images divided based on age, sex, and disease group,
Generating the brain structure normalization region through regression analysis and pairwise approach using the evaluated correlation factor,
A medical image processing apparatus characterized in that the labeled image is combined with the brain structure normalization region. 5. The method of claim 4,
The image processing unit,
matching the brain structure standardization region with the stroke symptom-based functional region,
A medical image processing apparatus, characterized in that the image is corrected using a portion of the signal strength of each region using a distribution histogram in order to correct an interference error between adjacent tissues for each region. According to claim 1,
The image processing unit,
A medical image processing apparatus, characterized in that the pattern of the image feature is analyzed with respect to the standardized medical image according to the stroke clinical evaluation. 9. The method of claim 8,
The image processing unit,
Extracting the size, position, and signal intensity of the lesion for each medical standardized image,
generating a stroke feature combination set by combining size, location, and signal intensity for the lesion according to the stroke clinical evaluation;
A medical image processing apparatus, characterized in that it outputs a matching stroke feature combination set by analyzing an individual pattern from among the stroke feature combination set. According to claim 1,
The image processing unit,
The medical image processing apparatus according to claim 1, wherein the stroke clinical evaluation, image features according to the stages of the stroke clinical evaluation, and stroke clinical factors are mapped and stored in a medical image database. According to claim 1,
The image processing unit,
The medical image processing apparatus of claim 1, wherein the individual stroke evaluation is output based on the image characteristics according to the stroke clinical evaluation and the stages of the stroke clinical evaluation. 12. The method of claim 11,
The image processing unit,
output the correlation between the image features according to the stage of the stroke clinical evaluation and the stroke clinical evaluation,
Outputs a score calculated by the volume of the stroke lesion compared to the brain region for the plurality of types of medical images,
Taking the plurality of types of medical images at multiple time points and tracking changes in size, location, and signal strength of the lesion in time series;
The medical image processing apparatus of claim 1, wherein an expected state is outputted in text form in consideration of changes in size, location, and signal strength of the tracked lesion. 12. The method of claim 11,
The image processing unit,
According to the stroke clinical evaluation, a suspected lesion region is extracted from the tomography image based on the medical standardized image, and the size and signal intensity of the lesion in the tomography image are determined through a feature generation model between the tomography image and the magnetic resonance image. A medical image processing apparatus for outputting a highlighted tomography image. A medical image processing method by a medical image processing apparatus, comprising:
acquiring multiple types of medical images and stroke clinical assessments;
generating a plurality of types of standardized medical images for the plurality of types of medical images;
extracting image features of a lesion for each standardized medical image by applying a lesion recognition model to the plurality of types of standardized medical images;
analyzing the pattern of the image feature according to the stroke clinical evaluation with respect to the medical standardized image;
mapping image features according to the stroke clinical evaluation, the stages of the stroke clinical evaluation, and stroke clinical factors and storing the mapping in a medical image database; and
Outputting an individual stroke evaluation based on the image characteristics according to the stroke clinical evaluation and the stage of the stroke clinical evaluation
A medical image processing method comprising a. 15. The method of claim 14,
The generating of the plurality of types of standardized medical images includes:
The plurality of types of medical images are divided into gray matter, white matter, and cerebrospinal fluid images, and the plurality of types of brain structure normalization regions are generated based on correlation factors for the gray matter, white matter, and cerebrospinal fluid images, and the plurality of types of brain A method for processing a medical image, characterized in that the plurality of types of standardized medical images are generated by matching the structural standardized area with the stroke symptom-based functional area. 15. The method of claim 14,
The step of analyzing the pattern of the image feature comprises:
Extracting the size, position, and signal intensity of the lesion for each medical standardized image,
generating a stroke feature combination set by combining size, location, and signal intensity for the lesion according to the stroke clinical evaluation;
A medical image processing method, characterized in that outputting a matching stroke feature combination set by analyzing an individual pattern from among the stroke feature combination set. 15. The method of claim 14,
The step of outputting the individual stroke evaluation comprises:
output the correlation between the image features according to the stage of the stroke clinical evaluation and the stroke clinical evaluation,
Outputs a score calculated by the volume of the stroke lesion compared to the brain region for the plurality of types of medical images,
Taking the plurality of types of medical images at multiple time points and tracking changes in size, location, and signal strength of the lesion in time series;
A medical image processing method, characterized in that outputting an expected state in a text form in consideration of changes in size, position, and signal strength of the tracked lesion. 15. The method of claim 14,
The step of outputting the individual stroke evaluation comprises:
According to the stroke clinical evaluation, a suspected lesion region is extracted from the tomography image based on the medical standardized image, and the size and signal intensity of the lesion in the tomography image are determined through a feature generation model between the tomography image and the magnetic resonance image. A medical image processing method comprising outputting a highlighted tomography image.

Images