A neurodegenerative disease screening method based on medical image segmentation

Abstract

The application discloses a neurodegenerative disease screening method based on medical image segmentation. First, the original brain MRI image is preprocessed, unified into a standard template space, and the maximum transverse cross section is found for training the image segmentation model. Then, under the condition of a small sample case, the image segmentation model is used to realize accurate segmentation of the brain MRI image, and then the neurodegenerative disease screening of different indexes is completed.

Description

Technical Field

[0001] This invention belongs to the field of medical image processing technology, and more specifically, relates to a method for screening neurodegenerative diseases based on medical image segmentation. Background Technology

[0002] Parkinson's disease (PD) and Alzheimer's disease (AD) are two common neurodegenerative diseases. Parkinson's disease is a degenerative disease of the central nervous system, characterized by motor disorders such as tremor, rigidity, and bradykinesia. Alzheimer's disease is a neurodegenerative disease characterized by memory loss, cognitive decline, and behavioral abnormalities. Both Parkinson's and Alzheimer's diseases primarily affect the elderly. Currently, there are over 50 million people with PD and AD worldwide, and the number of new cases is increasing every year. Early diagnosis of AD and PD is of paramount importance.

[0003] Currently, clinical diagnosis primarily relies on symptom presentation, neurological function assessment, and cranial MRI (Magnetic Resonance Imaging), as well as positron emission tomography (PET) scans using tracers such as β-amyloid plaques or dopamine transporters. Several important measurement parameters exist for brain MRI images, such as the Evans index (the ratio of the maximum width of the lateral ventricle to the maximum width of the inner table of the skull on the same plane), the corpus callosum angle, and the brain-to-ventricle ratio. Currently, the measurement of these parameters remains manual, leading to cumbersome procedures and introducing subjective bias. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a screening method for neurodegenerative diseases based on medical image segmentation, which can automatically measure several important indicators related to the diagnosis of AD and PD, thereby completing the early screening of neurodegenerative diseases.

[0005] To achieve the above-mentioned objectives, the present invention provides a method for screening neurodegenerative diseases based on medical image segmentation, characterized by comprising the following steps:

[0006] (1) Download multiple MRI images of neurodegenerative diseases from the database, then register each MRI image to a unified template space to obtain standardized images, and then find the largest cross-sectional section on each standardized image.

[0007] (2) Train the image segmentation model;

[0008] (2.1) Train an image segmentation model with respect to the Evans exponent using the maximum cross-sectional area until the image segmentation model converges;

[0009] (2.2) Train an image segmentation model for the brain / ventricle ratio (BVR) index using the maximum cross-sectional area until the image segmentation model converges;

[0010] (2.3) Train an image segmentation model with respect to the CA index of the corpus callosum angle using the largest cross-sectional section until the image segmentation model converges;

[0011] (3) Register the target MRI image to a uniform template space and find the largest cross-sectional section, which is denoted as the target section;

[0012] (4) Manually retrieve the target cross section, select the corresponding image segmentation model from step (2), and then input the target cross section into the converged image segmentation model to obtain the mask image with the key regions marked;

[0013] (5) Screening for neurodegenerative diseases based on different indicators using mask images;

[0014] (5.1) Calculate the Evans index based on the mask image. If the Evans index is greater than the threshold, the patient is at higher risk of having neurodegenerative diseases.

[0015] (5.2) Calculate the brain / ventricle ratio (BVR) index based on the mask image. If the BVR index is less than the threshold, the patient is screened out as having idiopathic normal pressure hydrocephalus.

[0016] (5.3) Calculate the CA index of the corpus callosum angle based on the mask image. If the BVR index is less than the threshold, the patient is at higher risk of having neurodegenerative diseases.

[0017] The objective of this invention is achieved as follows:

[0018] This invention relates to a neurodegenerative disease screening method based on medical image segmentation. First, the original brain MRI images are preprocessed and unified into a standard template space, and the largest cross-sectional area is found for training the image segmentation model. Then, in a small sample of cases, the image segmentation model achieves accurate segmentation of brain MRI images, thereby completing the screening of neurodegenerative diseases with different indicators.

[0019] Meanwhile, the neurodegenerative disease screening method based on medical image segmentation of this invention also has the following beneficial effects:

[0020] (1) MRI images used clinically for screening neurodegenerative diseases have significant individual differences, which increases the difficulty of manual examination. This invention registers the original brain MRI images to a unified template space, efficiently completes data preprocessing, and facilitates subsequent screening.

[0021] (2) This invention uses an effective segmentation model structure and loss function to train an image segmentation network that is adapted to three measurement indicators under small sample cases, and intelligently locks the key areas of brain MRI images used for screening.

[0022] (3) The present invention removes sensitive noise that may affect subsequent measurement results, such as the edge of a single key area and the boundary line between different key areas, through multiple targeted smoothing processes, thereby further ensuring the screening results.

[0023] (4) For each indicator, a corresponding automated measurement algorithm is developed, and the average error between the measurement results and the manually labeled results is less than 0.05. Attached Figure Description

[0024] Figure 1 This is a flowchart of the neurodegenerative disease screening method based on medical image segmentation of the present invention;

[0025] Figure 2 A diagram illustrating the Evans index;

[0026] Figure 3 A schematic diagram of the brain / ventricle ratio (BVR) index;

[0027] Figure 4 A schematic diagram of the CA index of the corpus callosum angle;

[0028] Figure 5 This is a flowchart illustrating the measurement of the Evans index according to the present invention;

[0029] Figure 6 This is a flowchart illustrating the brain-to-ventricle ratio (BVR) index of this invention.

[0030] Figure 7 This is a flowchart illustrating the CA index for measuring the corpus callosum angle according to the present invention. Detailed Implementation

[0031] The specific embodiments of the present invention will now be described with reference to the accompanying drawings to enable those skilled in the art to better understand the invention. It should be particularly noted that in the following description, detailed descriptions of known functions and designs that might obscure the main content of the invention will be omitted here.

[0032] Example

[0033] In this embodiment, as Figure 1As shown, the present invention provides a method for screening neurodegenerative diseases based on medical image segmentation, comprising the following steps:

[0034] S1, MRI image download and preprocessing;

[0035] Multiple MRI images of neurodegenerative diseases were downloaded from the database. Each MRI image was then registered to a uniform template space to obtain a standardized image. The largest cross-sectional area was then found on each standardized image.

[0036] In this embodiment, the original MRI image is a T1 structural image, and the template space we selected is a T1-weighted image based on the MNI (Montreal Neurological Institute) space provided in the spm12 software;

[0037] In this embodiment, the main indicators for neurodegenerative diseases studied are the Evans index, the brain / ventricle ratio (BVR) index, and the corpus callosum angle (CA) index.

[0038] The Evans index represents the ratio of the maximum width of the anterior horn of the lateral ventricle to the maximum width of the inner table of the skull at the same level; for example... Figure 2 As shown.

[0039] The brain-to-ventricle ratio (BVR) is expressed as the reciprocal of the ratio of the maximum height of the anterior horn of the lateral ventricle to the distance from the highest point of the ipsilateral lateral ventricle to the intracranial table. Figure 3 As shown.

[0040] The corpus callosum angle (CA index) is represented by the angle between the roof walls of the lateral ventricles on coronal T1WI or CT-MPR; for example... Figure 4 As shown.

[0041] We can download relevant MRI images of the three indicators from medical databases to facilitate subsequent model training;

[0042] S2. Train the image segmentation model;

[0043] S2.1. Train an image segmentation model with respect to the Evans exponent using the maximum cross-sectional area until the image segmentation model converges.

[0044] S2.2. Train an image segmentation model for the brain / ventricle ratio (BVR) index using the maximum cross-sectional area until the image segmentation model converges.

[0045] S2.3. Train the image segmentation model about the CA index of the corpus callosum angle using the maximum cross-sectional area until the image segmentation model converges. In this embodiment, the image segmentation model selected is the Transformer-based image segmentation model SwinUnet, which is trained using the cross loss function until the image segmentation model converges. The specific training process will not be described in detail here.

[0046] The cross-entropy loss function used during training is:

[0047]

[0048] Where i represents the i-th sample, c represents the c-th class, x and y represent the horizontal and vertical coordinates on the image, N represents the total number of samples, k represents the k-th class, C represents the total number of classes, and S represents the entire image region. Let represent the probability that the point at position (x, y) on the prediction result map of the i-th sample is predicted to be of class c. Let represent the probability that the point at position (x, y) on the prediction result map of the i-th sample is predicted to be of class k, and This represents the value of the i-th sample image at coordinates (x, y) with respect to class c. The value is 1 when the sample belongs to class c, and 0 otherwise.

[0049] S3, Preprocessing of target MRI images;

[0050] Register the target MRI image to a uniform template space and find the largest cross-sectional section, which is denoted as the target section;

[0051] In this embodiment, the corresponding target MRI image is also registered to a T1-weighted image based on the MNI (Montreal Neurological Institute) space.

[0052] S4. Obtain the mask image;

[0053] The target cross section is manually retrieved, and the corresponding image segmentation model is selected from step S2. Then, the target cross section is input into the converged image segmentation model to obtain a mask image with the key regions annotated.

[0054] In this embodiment, the purpose of manually retrieving the target cross-section is to identify the image category, thereby enabling the selection of a suitable image segmentation model.

[0055] S5. Screen for neurodegenerative diseases based on different indicators using the mask image;

[0056] S5.1 Calculate the Evans index based on the mask image. If the Evans index is greater than 0.3, the patient is at higher risk of having a neurodegenerative disease.

[0057] In this embodiment, as Figure 5 As shown, the Evans index is calculated as follows:

[0058] S5.1.1 The mask image contains three regions: background, lateral ventricle, and the edge of the inner table of the skull;

[0059] S5.1.2. Following a fixed rotation step α = 1°, starting from -5°, rotate the mask image sequentially within the range of -5° to 5° to eliminate the influence of head movement.

[0060] S5.1.3 For each image after a rotation step α, scan from top to bottom using horizontal straight lines, recording two values ​​for each row during the scan. The first value is the maximum difference in pixel coordinates of the lateral ventricle in the horizontal direction, recorded as the maximum width of the lateral ventricle; the second value is the maximum difference in pixel coordinates of the inner table of the skull in the horizontal direction, recorded as the maximum width of the inner table of the skull. After the scan is completed, take the maximum width of the lateral ventricle and the maximum width of the inner table of the skull respectively. Then divide the maximum width of the lateral ventricle by the maximum width of the inner table of the skull to obtain an Evans index.

[0061] S5.1.4 After calculating the Evans exponent of the image after each rotation step α according to step S5.1.3, the average value is calculated to obtain the final Evans exponent.

[0062] S5.2 Calculate the brain / ventricle ratio (BVR) index based on the mask image. If the BVR index is less than 1.0, the patient is screened to have idiopathic normal pressure hydrocephalus.

[0063] In this embodiment, as Figure 6 As shown, the brain / ventricle ratio (BVR) index is calculated as follows:

[0064] S5.2.1 The mask image contains three regions: the lateral ventricle, the brain parenchyma, and the background;

[0065] S5.2.2 Perform smoothing and denoising processing on the mask image three times in sequence;

[0066] For the first time, the mean filtering function in the Python library was used to perform mean filtering on the mask image to alleviate the blurring of the lateral ventricle edges. The mean filtering kernel size was fixed at 2×2.

[0067] The second time, after the mean filtering, the median filtering function of the Python library is used to perform median filtering to remove noise in the brain parenchyma region. The median filtering kernel size is fixed at 3×3.

[0068] The third step involves using an anomaly detection and processing method based on Gaussian distribution to remove noise points in the lateral ventricle region. The specific process is as follows: calculate the average value μ and standard deviation σ of the coordinate points in the lateral ventricle region along the vertical axis, and then delete the anomalies located outside the regions of ||μ-3×σ, μ+3×σ||.

[0069] S5.2.3. Based on step S5.2.2, find the highest and lowest points on the same side in the lateral ventricle region along the vertical direction, and then calculate the difference to obtain the ventricle height;

[0070] S5.2.4. At the highest point found above, extend upwards to the highest point of the brain parenchyma, and then subtract the two points to obtain the brain height;

[0071] S5.2.5. Divide the brain height by the ventricular height to obtain the brain / ventricular ratio (BVR) index.

[0072] S5.3 Calculate the corpus callosum angle CA index based on the mask image. If the BVR index is less than the threshold, the patient is at higher risk of having neurodegenerative diseases.

[0073] In this embodiment, as Figure 7 As shown, the calculation method for the CA index of the corpus callosum angle is as follows:

[0074] S5.3.1, The mask image contains two regions: the lateral ventricle and the background;

[0075] S5.3.2. Fit the coordinate points of the lateral ventricle region using a higher-order polynomial function to obtain the lateral ventricle wall curve.

[0076] In this embodiment, for the segmented lateral ventricle region, since the lateral ventricle wall has an irregular curvature, the polyfit function in the Python library is used to accurately fit it with a high-order polynomial curve, with the order fixed at 5, to obtain the lateral ventricle wall curve.

[0077] S5.3.3 For the lateral ventricle wall curve, take the lowest point of the curve as the midline and divide it into left and right parts. Find the midpoints of the left and right parts according to the x-coordinates of the pixels on the inner wall. Draw tangents along the inner wall curve at the two midpoints. The angle between the tangents is the CA index of the corpus callosum angle.

[0078] Experimental verification

[0079] The experimental conditions were set as follows: System: Ubuntu 18.04, Programming language: Python 3.8, Deep learning framework: PyTorch 11.6, GPU: NVIDIA GeForce RTX 3090.

[0080] Experiment 1: Regarding the Evans index measurement, there were 24 positive samples and 134 negative samples out of 158 data samples. A positive sample was defined as an Evans index greater than or equal to 0.3. As can be seen from Table 1, the currently measured recall rate can reach 75%, while the precision rate can reach 64%, and the average error between the 158 measurement results and the actual results is 0.023.

[0081] Experiment 2: Regarding the brain / ventricle ratio (BVR) measurement, out of 157 data samples, there were 15 positive samples and 142 negative samples. A positive sample was defined as a BVR ratio less than 1.0. As shown in Table 2, the current recall rate is 0.73, and the precision rate is 0.55. The average error between the 157 measurement results and the actual results is 0.022.

[0082] Experiment 3: Regarding the measurement of the CA index of the corpus callosum angle, there were 6 positive samples and 151 negative samples out of 157 data samples. A positive sample was defined as a CA index of less than 90°. As shown in Table 3, the current recall rate is 0.5, and the precision rate is 0.375. The average error between the 157 measurement results and the actual results is 8°.

[0083] Table 1. Evans Index Measurement Results

[0084]

[0085] Table 2. Results of Brain / Ventricular Ratio (BVR) Index Measurement

[0086]

[0087] Table 3. Measurement results of the CA index of the corpus callosum angle.

[0088]

[0089] Although the illustrative specific embodiments of the present invention have been described above to enable those skilled in the art to understand the invention, it should be understood that the invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of the present invention are protected.

Claims

1. A method for screening neurodegenerative diseases based on medical image segmentation, characterized in that, Includes the following steps: (1) Download multiple MRI images of neurodegenerative diseases from the database, then register each MRI image to a unified template space to obtain standardized images, and then find the largest cross-sectional section on each standardized image. (2) Train the image segmentation model; (2.1) Train an image segmentation model with respect to the Evans exponent using the maximum cross section until the image segmentation model converges. The image segmentation model selected is the Transformer-based image segmentation model SwinUnet. (2.2) Train an image segmentation model for the brain / ventricle ratio (BVR) index using the largest cross-sectional area until the image segmentation model converges; (2.3) Train an image segmentation model for the CA index of the corpus callosum angle using the largest cross-sectional section until the image segmentation model converges; (3) Register the target MRI image to a uniform template space and find the largest cross-sectional section, which is recorded as the target section; (4) Manually retrieve the target cross section, select the corresponding image segmentation model from step (2), and then input the target cross section into the converged image segmentation model to obtain the mask image with the key regions marked; (5) Screening for neurodegenerative diseases based on different indicators using mask images; (5.1) Calculate the Evans index based on the mask image. If the Evans index is greater than the threshold, the patient is at higher risk of having neurodegenerative diseases. (5.2) Calculate the brain / ventricle ratio (BVR) index based on the mask image. If the BVR index is less than the threshold, the patient is screened out as having idiopathic normal pressure hydrocephalus. (5.3) Calculate the CA index of the corpus callosum angle based on the mask image. If the BVR index is less than the threshold, the patient is at higher risk of having neurodegenerative diseases.

2. The method for screening neurodegenerative diseases based on medical image segmentation according to claim 1, characterized in that, The Evans index is calculated as follows: (5.1.1) The mask image contains three regions: the background, the lateral ventricle, and the edge of the inner table of the skull; (5.1.2) Follow the fixed rotation step Starting from -5°, the mask image is rotated sequentially within the range of -5° to 5°; (5.1.3) For each rotational step The image was then scanned from top to bottom using horizontal lines. Two values ​​were recorded for each row during the scan. The first value was the maximum difference in pixel coordinates of the lateral ventricle in the horizontal direction, which was recorded as the maximum width of the lateral ventricle. The second value is the maximum difference in pixel coordinates of the inner table of the skull in the horizontal direction, which is recorded as the maximum width of the inner table of the skull. After the scan is completed, the maximum width of the lateral ventricle and the maximum width of the inner table of the skull are taken respectively. Then, the maximum width of the lateral ventricle is divided by the maximum width of the inner table of the skull to obtain an Evans index. (5.1.4) Calculate each rotation step according to step (5.1.3). After calculating the Evans index of the image, the average of these values ​​is taken to obtain the final Evans index.

3. The method for screening neurodegenerative diseases based on medical image segmentation according to claim 1, characterized in that, The brain / ventricular ratio (BVR) index is calculated as follows: (5.2.1) The mask image contains three regions: the lateral ventricle, the brain parenchyma, and the background; (5.2.2) Perform smoothing and denoising processing on the mask image three times in sequence; The first step involves applying a mean filter to the mask image to alleviate the blurring of the lateral ventricle edges. The second step involves mean filtering followed by median filtering to remove noise from the brain parenchyma. Third, the average value of the coordinates of the contralateral ventricle region along the vertical axis is calculated. and standard deviation Then delete the location. Anomalies outside the designated area; (5.2.3) Based on step (5.2.2), find the highest and lowest points on the same side in the lateral ventricle region along the vertical direction, and then calculate the difference to obtain the ventricle height; (5.2.4) At the highest point found above, extend upwards to the highest point of the brain parenchyma, and then subtract the two points to obtain the brain height; (5.2.5) Divide the brain height by the ventricular height to obtain the brain / ventricular ratio (BVR) index.

4. The method for screening neurodegenerative diseases based on medical image segmentation according to claim 1, characterized in that, The method for calculating the CA index of the corpus callosum angle is as follows: (5.3.1) The mask image contains two regions: the lateral ventricle and the background; (5.3.2) The coordinate points of the lateral ventricle region are fitted using a higher-order polynomial function to obtain the lateral ventricle wall curve; (5.3.3) For the lateral ventricle wall curve, take the lowest point of the curve as the midline and divide it into left and right parts. Find the midpoints of the left and right parts according to the x-coordinates of the inner wall pixels. Draw tangents along the inner wall curve at the two midpoints. The angle between the tangents is the CA index of the corpus callosum angle.

Images