Stroke assessment apparatus based on fractal dimensions

Abstract

A stroke assessment apparatus, comprising: an image acquisition unit (102), which is configured to acquire target images, wherein the target images are a brain structure image and a brain blood vessel image; a feature extraction unit (104), which is configured to acquire spectral shape parameters and fractal dimensions of the target images; and a stroke prediction unit (106), which is a machine learning model and is configured to determine a stroke assessment result on the basis of the spectral shape parameters and the fractal dimensions. As the fractal characteristics of medical images reflect the degree of heterogeneity inside an object to be analyzed, an important reference is provided for disease diagnosis, and an assessment result obtained in this way is more accurate.

Description

Stroke assessment device based on fractal dimension Technical Field

[0001] The present application relates to the field of medical devices, and in particular to a stroke assessment device. Background Art

[0002] Stroke is a common neurological disease, usually caused by rupture or blockage of blood vessels in the brain, resulting in ischemia or oxygen deprivation of brain tissue. Stroke is one of the leading causes of death and disability worldwide, making early diagnosis and prevention crucial.

[0003] Traditional stroke diagnosis methods typically use medical imaging techniques (such as magnetic resonance imaging and computed tomography) to observe changes in brain structure and blood vessels to diagnose stroke. However, these methods often require interpretation by professional doctors and are expensive, limiting their application in large-scale screening and early diagnosis.

[0004] In recent years, with the development of machine learning and computer vision technology, more and more studies have begun to explore the use of computer-aided diagnosis technology to achieve automated and intelligent diagnosis of stroke. This invention captures brain structure and morphological characteristics by extracting spectral parameters and fractal dimensions of images, providing a new perspective for stroke prediction.

[0005] Spectral shape parameters and fractal dimensions are concepts based on fractal geometry and are used to describe the fractal characteristics of complex systems. In brain image analysis, spectral shape parameters and fractal dimensions can be used to describe the structure and morphology of brain tissue, thereby providing informative features for the early diagnosis of stroke.

[0006] Summary of the Invention

[0007] Therefore, it is necessary to address the above-mentioned technical issues and provide a stroke assessment device. This device, based on spectral shape parameters and fractal dimensions, aims to provide an automated, highly interpretable, and efficient stroke diagnosis solution. This device combines image processing, feature extraction, and machine learning techniques to extract spectral shape parameters and fractal dimensions from brain images and utilize these features to assess stroke.

[0008] In a first aspect, an embodiment of the present application provides a stroke assessment device, characterized by comprising:

[0009] an image acquisition unit configured to acquire a target image, wherein the target image is a brain structure image and a brain blood vessel image;

[0010] a feature extraction unit configured to obtain spectral shape parameters and fractal dimensions of a target image;

[0011] The stroke prediction unit is a machine learning model configured to determine a stroke assessment result based on the spectral parameters and the fractal dimension.

[0012] In one embodiment, the feature extraction unit includes:

[0013] The fractal dimension calculation module is configured to calculate the fractal dimension of the target image. The calculation formula is:

[0014] Among them D Q (r) is the fractal dimension, r is the splitting ratio, r represents the ratio of the small blood flow to the large blood flow in the two sub-branches of the blood vessel, and r is a value between 0 and 1.

[0015] In another embodiment, the feature extraction unit includes:

[0016] The fractal dimension calculation module is configured to calculate the fractal dimension of the target image. The calculation formula is:

[0017] N(∈) is the number of boxes covering the image with scale ε, p i is the number of pixels in the i-th box, q is the weight factor, and D(q) is the fractal dimension.

[0018] In one embodiment, the feature calculation unit includes a spectral shape parameter calculation module, which is configured to calculate the multifractal spectrum of the target image and calculate the spectral shape parameters of the target image based on the multifractal spectrum. The spectral shape parameters include at least one of the width of the multifractal spectrum, the height of the multifractal spectrum, and the slope of the multifractal spectrum.

[0019] In one embodiment, the feature extraction unit further includes

[0020] The feature dimension reduction unit is configured to perform dimension reduction processing on the fractal dimension and spectral shape parameters of the target image and output the spectral shape parameters and fractal dimension after the dimension reduction processing.

[0021] In one embodiment, the stroke prediction unit is any one of the machine learning models including random forest, support vector machine, and gradient boosting tree.

[0022] In one embodiment, the stroke assessment result is the severity of the stroke.

[0023] In one embodiment, a brain structure image is input into a blood vessel extraction model to obtain a brain blood vessel image, and the blood vessel extraction model is a neural network model.

[0024] The training steps of the stroke prediction unit in one embodiment include:

[0025] Acquire a training sample, where the training sample includes a first number of test images and a first stroke assessment result corresponding to each test image in the first number of test images;

[0026] Inputting each test image of the first number of test images into a feature extraction unit to obtain a corresponding first spectral shape parameter and a first fractal dimension;

[0027] inputting the first spectral shape parameter, the first fractal dimension and the first stroke assessment result corresponding to each test image into a stroke prediction unit;

[0028] When the stroke prediction unit meets the preset conditions, the trained stroke prediction unit is obtained.

[0029] In one embodiment, when the stroke prediction unit meets a preset condition, obtaining a trained stroke prediction unit includes:

[0030] When the loss function value of the stroke prediction unit is less than or equal to a preset threshold, or the iteration rounds of the stroke prediction unit reach the preset iteration rounds, a trained stroke prediction unit is obtained.

[0031] The above-mentioned stroke assessment device further mines the potential information in medical images and provides an important reference for the diagnosis of stroke. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] FIG1 is a diagram of a stroke assessment device according to an embodiment;

[0033] FIG2 shows the spectral shape parameters and fractal dimension of a target image obtained by a feature extraction unit in one embodiment;

[0034] FIG3 shows the multifractal spectra calculated by the feature extraction unit for different patients in one embodiment;

[0035] FIG4 is a diagram showing the verification results of the stroke assessment device. DETAILED DESCRIPTION

[0036] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0037] In one embodiment, as shown in FIG1 , a stroke assessment device is provided, comprising:

[0038] The image acquisition module 102 is configured to acquire target images, which are brain structural images and brain vascular images. The method for acquiring the brain structural image is to image the patient's brain using an image acquisition device to acquire a structural image. The structural image reflects the structural characteristics of the brain. Specifically, the image acquisition device can be one or more imaging devices such as CT (computed tomography), MRI (magnetic resonance imaging), ultrasound imaging, and PET (positron emission tomography). The brain vascular image corresponding to the brain structural image can be acquired. Specifically, the brain vascular image can be acquired by extracting blood vessels from the brain structural image using a blood vessel extraction model obtained through training of a neural network model. The brain vascular image can be a blood vessel mask. Alternatively, direct brain vascular imaging can be performed, such as by using DSA (digital subtraction angiography) or CT perfusion / enhanced imaging.

[0039] The feature extraction unit 104 is configured to obtain the spectral shape parameters and fractal dimension of the target image. In one embodiment, the calculation formula for calculating the fractal dimension of the target image is:

[0040] Among them D Q (r) is the fractal dimension, r is the splitting ratio, r represents the ratio of the small blood flow to the large blood flow in the two sub-branches of the blood vessel, and r is a value between 0 and 1.

[0041] The smaller the split ratio, the greater the heterogeneity of blood flow distribution. This is because ischemic images often exhibit greater heterogeneity compared to normal images. Normal blood flow is typically between 0.5 and 1.5, while ischemic images can range from 0 to 2. This is because ischemia can cause some vessels to have no blood flow, while others experience abnormally high blood flow. Therefore, normal individuals have a higher split ratio r than ischemic images.

[0042] In this embodiment, the fractal dimension is calculated based on a brain structural image and a brain vascular image. The brain structural image provides comprehensive brain structural information, while the brain vascular image focuses on vascular structure and blood flow. Combining these two images allows for more accurate extraction of blood flow at specific vascular locations, increasing the accuracy of the assessment results.

[0043] In another embodiment, the calculation formula for calculating the fractal dimension of the target image is:

[0044] N(ε) is the number of boxes covering the image with scale ε, p i is the number of pixels in the i-th box, q is the weight factor, and D(q) is the fractal dimension.

[0045] Different values ​​of q can be used to characterize fractal features at different scales, thus providing different descriptions of the image's multi-scale structure. When q is 0, D(0) is often called the capacity dimension; when q is 1, D(1) is often called the information dimension; and when q is 2, D(2) is often called the correlation dimension.

[0046] This embodiment provides a method for calculating fractal dimensions based on brain structural images and brain vascular images. The brain structural image provides comprehensive brain structural information, while the brain vascular image focuses on vascular structures. Combining these two images allows for more accurate vascular location extraction. The calculated fractal dimension provides additional vascular fractal features, further interpreting the brain vascular image and increasing the accuracy of the assessment results.

[0047] In another embodiment, the feature calculation unit includes a spectral shape parameter calculation module configured to calculate the multifractal spectrum of the target image, and calculate the spectral shape parameters of the target image based on the multifractal spectrum, the spectral shape parameters including at least one of the width of the multifractal spectrum, the height of the multifractal spectrum, and the slope of the multifractal spectrum.

[0048] The multifractal spectrum f(α) can be obtained by the Legendre transformation of the fractal dimension D(q), specifically: f(α) = qα - D(q)

[0049] f(α) is the multifractal spectrum, and α is the Holder index. max -α min The width of the multifractal spectrum can also be obtained, as can the height of f(α) and the slope of f(α) at ​​a specific location. Furthermore, the spectral symmetry and shape of f(α) can also be determined. The above analysis yields the multifractal spectrum and corresponding spectral shape parameters, providing additional cerebrovascular parameters and greater interpretability for stroke diagnosis.

[0050] FIG2 shows a list of characteristic information calculated by the applicant for a patient based on the patient's brain vascular mask and brain structure image, including spectral parameters such as the total width, which is the width of the multifractal spectrum, f(q=+10), etc.; the fractal dimension is also provided.

[0051] Figure 3 shows the multifractal spectra of different patients. Different patients have different r and multifractal spectra. As shown in Figure 4, patients with and without stroke have different spectral shape parameters, such as different multifractal spectrum widths. Therefore, from Figures 3 and 4, it can be concluded that spectral shape parameters and fractal dimension can be used to assess patients for stroke.

[0052] The present invention provides an embodiment for assessing whether a patient has suffered a stroke. A stroke, also known as a "stroke," is a serious medical emergency in which a part of the brain is damaged due to a disruption in blood supply. Therefore, stroke examination and assessment should be performed as quickly as possible. In stroke treatment, the principle "time is brain" is generally followed. This means that from the moment stroke symptoms begin to appear, every minute is crucial. The optimal treatment window for a stroke is often referred to as the "golden hour." During this time, doctors typically immediately perform imaging studies (such as CT scans or MRIs) to determine the type and extent of the stroke. For ischemic stroke, this window is typically between 3 and 4.5 hours after onset. Within this time, thrombolytic drugs (typically tissue plasminogen activator (tPA)) can effectively dissolve blood clots blocking cerebral blood vessels, thereby restoring blood flow. The earlier treatment is initiated, the greater the likelihood of recovery and the fewer the sequelae. Therefore, early prediction of stroke and its severity are crucial. A CT scan or MRI of the patient's brain is performed to obtain images of brain structure. Blood vessels are extracted from the brain structure image to obtain a brain structure blood vessel image. The brain structure image and the brain blood vessel image are input into a stroke assessment device, which outputs an assessment result indicating whether the patient has suffered a stroke.

[0053] In an embodiment of the present application, preferably, the structural image and the vascular image are simultaneously input into a preset evaluation model to calculate the fractal dimension and spectral shape parameters of the brain structural image and the brain vascular image. The spectral shape parameters refer to specific numerical features extracted from the multifractal spectrum. The present invention also provides an embodiment for performing dimensionality reduction processing on the fractal dimension and spectral shape parameters. Dimensionality reduction processing is intended to reduce the dimensionality of the feature data set while preserving as much information as possible from the original data. When processing high-dimensional data, feature dimensionality reduction is particularly important for improving computational efficiency and classifier performance.

[0054] The fractal dimension characterizes the combination of vascular structural characteristics and blood flow distribution characteristics. Specifically, the fractal dimension reflects the degree of blood flow heterogeneity within a specific vascular tree, with lower fractal dimensions resulting from a more heterogeneous blood flow distribution. The spectral shape parameter essentially describes the degree of heterogeneity within the fractal object and increases with increasing blood flow heterogeneity. In other words, the height of the multifractal spectrum reflects the heterogeneity or asymmetry of the vascular structure. The higher the multifractal spectrum, the closer the vascular tree approaches a perfect binary tree. Using the fractal dimension, variations in blood flow heterogeneity within a specific vascular tree can be observed. The multifractal spectrum can be used to assess the blood flow distribution and structural heterogeneity of the corresponding vascular tree by considering both its width and height, ultimately achieving the goal of assessing blood flow heterogeneity within different vascular trees. The method proposed in this invention provides an effective tool for describing the multiscale characteristics of blood flow distribution. Simultaneously inputting two images (structural and vascular images) is intended to extract density features of vascular locations, thereby increasing diagnostic accuracy. The structural image provides comprehensive brain structural information, while the addition of the vascular image allows for a focus on vascular structure and density features. The combination of the fractal features corresponding to the two images helps to more accurately assess the risk of stroke and the extent of damage. The invention has practical value for clinical applications.

[0055] The training steps of the stroke prediction unit include: obtaining training samples, the training samples including a first number of test images and a first stroke assessment result corresponding to each test image in the first number of test images; inputting each test image in the first number of test images into a feature extraction unit to obtain a corresponding first spectral parameter and a first fractal dimension; inputting the first spectral parameter, first fractal dimension, and first stroke assessment result corresponding to each test image into the stroke prediction unit; when the loss function value of the stroke prediction unit is less than or equal to a preset threshold, or the number of iterations of the stroke prediction unit reaches a preset number of iterations, a trained stroke prediction unit is obtained. The stroke prediction unit is a machine learning model, such as a random forest, XGBoost (eXtreme Gradient Boosting), SVM (Support Vector Machine), etc.

[0056] The above training process for the stroke prediction unit enables it to learn the characteristics of spectral parameters and fractal dimensions of stroke patients and, based on these characteristics, predict stroke. This evaluation result can indicate lesion type, disease severity, and other factors. Because medical image characteristics reflect the internal heterogeneity of the object being analyzed, they provide an important reference for disease diagnosis. This method provides more accurate evaluation results.

[0057] The present invention also provides a computer program product. The principles and processes of implementing the various embodiments can be found in the description of the medical image processing method in the aforementioned embodiments, and will not be repeated here.

[0058] It should be noted that the data involved in this application (including but not limited to data used for analysis, stored data, displayed data, etc.) are all information and data that are authorized or fully authorized by all parties.

[0059] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, database or other media used in the embodiments provided in this application may include at least one of non-volatile and volatile memory. Non-volatile memory may include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory may include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processor involved in the various embodiments provided herein may be, but are not limited to, a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic unit, a data processing logic unit based on quantum computing, and the like.

[0060] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0061] The above embodiments merely illustrate several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the spirit of the present invention, all of which fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A stroke assessment device, characterized in that, Comprising: An image acquisition unit configured to acquire a target image, where the target image is a brain structure image and a brain blood vessel image; A feature extraction unit configured to acquire the spectral shape parameters and fractal dimensions of the target image; A stroke prediction unit is a machine learning model configured to determine a stroke evaluation result according to the spectral shape parameters and the fractal dimensions.

2. The device according to claim 1, characterized in that, The feature extraction unit includes a fractal dimension calculation module configured to calculate the fractal dimension of the target image, and the calculation formula is: Among which D Q (r) is the fractal dimension, r is the splitting ratio, r represents the ratio of the smaller blood flow rate to the larger blood flow rate in two sub-branches of a blood vessel, and r is a value between 0 and 1.

3. The device according to claim 1, characterized in that The feature extraction unit includes a fractal dimension calculation module configured to calculate the fractal dimension of the target image, and the calculation formula is: N(ε) is the number of boxes covering the target image at scale ε, p i is the number of pixels in the i-th box, q is the weight factor, and D(q) is the fractal dimension.

4. The device according to claim 1, characterized in that, The feature calculation unit includes a spectral shape parameter calculation module configured to calculate the multifractal spectrum of the target image and calculate the spectral shape parameters of the target image according to the multifractal spectrum. The spectral shape parameters include at least one of the width of the multifractal spectrum, the height of the multifractal spectrum, and the slope of the multifractal spectrum.

5. The device according to claim 1, characterized in that The feature extraction unit further includes A feature dimensionality reduction unit configured to perform dimensionality reduction processing on the fractal dimensions and spectral shape parameters of the target image and output the spectral shape parameters and fractal dimensions after dimensionality reduction processing.

6. The device according to claim 1, characterized in that, The stroke prediction unit is any one of a random forest, a support vector machine, and a gradient boosting tree machine learning model.

7. The device according to claim 1, characterized in that, The stroke evaluation result is the severity of stroke.

8. The device according to claim 1, characterized in that, Comprising: Inputting the brain structure image into a blood vessel extraction model to obtain the brain blood vessel image, where the blood vessel extraction model is a neural network model.

9. The device according to claim 1, characterized in that The training steps of the stroke prediction unit include: Acquiring training samples, where the training samples include a first number of test images and a first stroke evaluation result corresponding to each of the first number of test images; Inputting each of the first number of test images into the feature extraction unit respectively to obtain corresponding first spectral shape parameters and first fractal dimensions; Inputting the first spectral shape parameters, the first fractal dimensions, and the first stroke evaluation result corresponding to each test image into the stroke prediction unit; When the stroke prediction unit meets a preset condition, obtaining a trained stroke prediction unit.

10. The device according to claim 9, characterized in that, The step of when the stroke prediction unit meets a preset condition and obtaining a trained stroke prediction unit includes: When the loss function value of the stroke prediction unit is less than or equal to a preset threshold, or The iteration round of the stroke prediction unit reaches a preset iteration round, obtaining the trained stroke prediction unit.

Images