Method for assisting doctor to quickly search for anomaly in fMRI image

Abstract

The invention discloses a method for assisting a doctor in quickly searching for anomalies in an fMRI image, and belongs to the crossing field of medicine and computers. Firstly, feature points are detected in two images by using an SURF algorithm; then, randomly selecting a point from a neighborhood around each feature point, comparing the gray values of the points, and coding a comparison result as 0 or 1; in this way, each feature point can generate a binary code. And the similarity between the two images is measured by calculating the Hamming distance of the codes. And finally, abnormal characteristic values in the image are screened and marked in the pathological image to assist a doctor to quickly find a lesion area, so that the diagnosis time is shortened and the diagnosis accuracy is improved. According to the method, feature point matching can be carried out under different scales, rotation and illumination conditions, and certain robustness is achieved.

Description

Technical Field

[0001] The invention relates to a method for assisting doctors to quickly find abnormalities in fMRI images, and belongs to the technical field of medicine and computer image processing. Background Art

[0002] Functional magnetic resonance imaging (fMRI) is a new non-invasive technology that studies brain function with high resolution and real-time tracking of signal changes in recent years. It is currently recognized as one of the most effective ways to diagnose cancer and complex diseases. Functional magnetic resonance imaging diagnosis has low radiation and can achieve good diagnosis of the patient's delicate structure, providing an important basis for clinical diagnosis and treatment, and has been widely used in clinical practice. Moreover, with the update of process level and computer technology, fMRI has developed rapidly.

[0003] However, fMRI images also have some defects. The contrast of fMRI images is relatively low, and the differences between different brain regions are not obvious. This may make it difficult for doctors to accurately identify and locate specific brain regions or abnormalities. In addition, the amount of data in fMRI images is huge, containing a large amount of spatial and temporal information. Doctors need to use complex data analysis methods and models to extract meaningful information. Therefore, it is of great practical value and significance to develop a method that can quickly assist doctors in determining whether fMRI images have abnormalities and quickly locate abnormal areas. Summary of the invention

[0004] In view of the problems arising from the above background, the object of the present invention is to provide a method for assisting doctors in quickly finding abnormalities in fMRI images.

[0005] In order to achieve the above-mentioned technical effects, the present invention provides an image processing method to assist doctors in rapid judgment and positioning, wherein the process mainly includes: obtaining feature points in different scale spaces in fMRI images: comparing the feature values ​​of two images at the same spatial scale and encoding and storing the feature values: marking points with abnormal feature values ​​detected in the images: counting the points with abnormal feature values ​​and outputting the Hamming distance between images: outputting an image with marked abnormal feature points.

[0006] The specific implementation steps are as follows: S01: Obtain the patient's fMRI image and the normal fMRI image, and use the Gaussian pyramid to construct different scale spaces of the two images; First, a Gaussian filter is applied to the original image to smooth the image and reduce noise. Second, the smoothed image is downsampled to reduce the resolution of the image. This can be achieved by sampling or interpolating each pixel in the image with the surrounding pixels. Finally, the above two steps are repeated to build a multi-layer Gaussian pyramid, where each layer corresponds to a different scale of the original image. Each layer is generated by applying a Gaussian filter and downsampling on the previous layer.

[0007] S02 uses the SURF (Speeded-Up Robust Features) algorithm to extract feature points from images at different scales and encode them into description operator vectors; The Hessian matrix is ​​obtained by calculating the second-order derivative of the image matrix on images of each scale, and the determinant of the Hessian matrix is ​​used to obtain the feature points on images of different scales. The features including grayscale value, gradient direction and gradient amplitude are stored in the computer unit, and then the gradient direction histogram of the area around the feature point is calculated to capture the texture and shape information of the area around the feature point.

[0008] S03 compares the similarity of the feature point description operators between the two images and encodes the comparison result; uses the Hamming distance to measure the difference between the two feature point description operators; S04 calculates the Hamming distance between the codes to obtain the similarity between the images, and screens out abnormal feature values ​​in the detected image compared with the normal image; The threshold for judging whether the disease is serious is obtained through the ROC curve and the case fMRI images and normal fMRI images stored in the database. The threshold is used to divide the detection image into normal images and abnormal images, and the threshold is used to judge the abnormal feature points of the detection image.

[0009] S05 marks the different feature points in the two compared images and outputs them together with the evaluation results. Beneficial Effects

[0010] The present invention is a method for assisting doctors to quickly find abnormalities in fMRI images. Different scale spaces of images are constructed by a Gaussian pyramid method. SURF algorithm is used to obtain feature points and obtain texture information and shape information of the image in different scale spaces. Binary encoding method is used to store information and Hamming is used to judge the similarity of images. Its superiority lies in that the algorithm can match feature points under different scales, rotations and illumination conditions in images and has certain robustness. In addition, the operation speed is fast, which can assist doctors to quickly determine whether lesions occur in images, and quickly mark areas suspected of lesions, thereby improving diagnostic efficiency. The method is of great significance for improving the diagnostic efficiency of doctors, saving diagnostic time and improving the accuracy of diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] Figure 1 A flowchart of a method for assisting doctors in quickly finding abnormalities in fMRI images according to the present invention; Figure 2 To divide the image into 4 images of different scales; Figure 3 The process of obtaining image feature points for the SURF algorithm; Figure 4 Label the fMRI horizontal plane images and abnormal point images; Figure 5 Annotate images of fMRI sagittal images and abnormal points; Figure 6 fMRI coronal images and abnormal point annotation images. DETAILED DESCRIPTION

[0012] The following is a further detailed description of the specific implementation of the present invention in conjunction with the accompanying drawings and specific implementations. The following examples or drawings are used to illustrate the present invention, but the protection scope of the present invention is not limited to the contents described.

[0013] like Figure 1 As shown, in this embodiment, a method for assisting doctors to quickly find abnormalities in fMRI images is provided. The main process is to obtain the patient case image and then divide it into four images of different scales using the Gaussian pyramid algorithm. The Hessian matrix is ​​obtained by calculating the second-order derivative of the image matrix on each scale of the image. The position and scale of the feature points on the images of different scales are obtained with the help of the Hessian matrix. The feature description operator that describes the image shape and texture information, that is, the feature vector, is obtained based on the obtained feature points. The feature vector is converted into a binary code and stored in a binary format, and then the Hamming distance between the feature vectors is calculated to evaluate whether the image taken by the patient is a diseased image. The specific process is as follows: S01: Obtain the patient's fMRI image and the normal fMRI image, and use the Gaussian pyramid to construct different scale spaces of the two images, such as Figure 2 shown.

[0014] First, a Gaussian filter is applied to the original image to smooth the image and reduce noise. Second, the smoothed image is downsampled to reduce the resolution of the image. This can be achieved by sampling or interpolating each pixel in the image with the surrounding pixels. Finally, the above two steps are repeated to construct a 4-layer Gaussian pyramid, where each layer corresponds to a different scale of the original image. Each layer is generated by applying a Gaussian filter and downsampling on the previous layer.

[0015] S02 uses the SURF (Speeded-Up Robust Features) algorithm to extract feature points from images at different scales and encode them into description operator vectors, such as Figure 3 As shown; The Hessian matrix is ​​obtained by calculating the second-order derivative of the image matrix on images of each scale, and the determinant of the Hessian matrix is ​​used to obtain the feature points on images of different scales. The features including grayscale value, gradient direction and gradient amplitude are stored in the computer unit, and then the gradient direction histogram of the area around the feature point is calculated to capture the texture and shape information of the area around the feature point.

[0016] S03 compares the similarity of the feature point description operators between the two images and encodes the comparison result; uses the Hamming distance to measure the difference between the two feature point description operators; S04 calculates the Hamming distance between the codes to obtain the similarity between the images, and screens out abnormal feature values ​​in the detected image compared with the normal image; The threshold for judging whether the disease is serious is obtained through the ROC curve and the case fMRI images and normal fMRI images stored in the database. The threshold is used to divide the detection image into normal images and abnormal images, and the threshold is used to judge the abnormal feature points of the detection image.

[0017] S05 marks the different feature points in the two compared images and outputs them together with the evaluation results.

[0018] The abnormal feature values ​​in the two images are marked and displayed in the fMRI images taken by the patient, such as Figure 4-6 Doctors can use the abnormal feature value areas marked in the image to conduct further analysis and diagnosis to determine whether the patient is sick.

[0019] The advantages of this method are: First, the image processing process is not easily affected by noise and blur in the image and has good robustness.

[0020] Second, the Hamming distance is used to count abnormal feature values ​​and the similarity of the two images (the image extracted from the normal fMRI image library and the fMRI image of the patient) is evaluated and the ROC curve is used to determine the threshold for judging whether the disease is present.

[0021] Third, the abnormal feature point areas will be marked in the figure, which can help doctors quickly locate suspected complications and make judgments.

[0022] Fourth, this method reduces the workload of doctors to a certain extent and improves the efficiency of diagnosis.

[0023] In summary: the present invention can effectively provide doctors with a method for quickly discovering and marking abnormal areas in images. The fMRI image of the detected area is obtained and divided into images of four different scales using a Gaussian pyramid. The SURF algorithm is used to extract feature points for each scale of the image. The image texture of the neighborhood is encoded by the extracted feature points, and the position and scale of the feature points are calculated using the Hessian matrix. The eigenvalues ​​of the two images are compared and the comparison results are encoded into 0 and 1 for storage. Finally, the Hamming distance between the stored encoding images is calculated, and a threshold is used to determine whether the image is a pathological image. The output image will be marked with abnormal eigenvalues.

Claims

1. A method for assisting doctors to quickly find abnormalities in fMRI images, characterized in that: include: S01: Obtain the patient's fMRI image and the normal fMRI image, and use the Gaussian pyramid to construct different scale spaces of the two images; S02 uses the SURF algorithm to extract feature points from images at different scales and encodes them into description operator vectors; S03 compares the similarity of the feature point description operators between the two images and encodes the comparison result; uses the Hamming distance to measure the difference between the two feature point description operators; S04 calculates the Hamming distance between the codes to obtain the similarity between the images, and screens out abnormal feature values ​​in the detected image compared with the normal image; S05 marks the different feature points in the two compared images and outputs them together with the evaluation results.

2. A method for assisting doctors to quickly find abnormalities in fMRI images according to claim 1, characterized in that: In S01, a Gaussian filter is first applied to the original image to smooth the image, and then the smoothed image is downsampled to reduce the resolution of the image. Finally, the above two steps are repeated to construct a multi-layer Gaussian pyramid, each layer corresponding to a different scale of the original image.

3. A method for assisting doctors to quickly find abnormalities in fMRI images according to claim 1, characterized in that: In step S02, the Hessian matrix is ​​obtained by calculating the second-order derivative of the image matrix on each scale of the image, and the determinant of the Hessian matrix is ​​used to obtain the feature points on the images of different scales. Then, the gradient direction histogram of the area around the feature points is calculated to capture the texture and shape information of the area around the feature points.

4. A method for assisting doctors to quickly find abnormalities in fMRI images according to claim 1, characterized in that: In step S04, a threshold for judging whether a patient is ill is obtained through the ROC curve and the case fMRI images and normal fMRI images stored in the database, and the threshold is used to judge abnormal feature points of the detection image.

Images