Artificial intelligence-based optic nerve sheath positioning system and method

Abstract

The present invention relates to the technical field of optic nerve sheath measurement, and disclosed are an artificial intelligence-based optic nerve sheath positioning system and method. The inventiveness of the present invention lies in performing ONSD measurement work on the basis of ultrasonic video for the first time. In the past, experts typically manually selected image-level data for measurement, whereas the present invention uses an automated approach. The research method of the present invention is mainly divided into two stages: optimal frame selection and optimal frame-based ONSD measurement. In terms of optimal frame selection, according to the present invention, a ground truth value is set as a good representation of ONSD, and a superpixel-based method is used for computing overlapping between a manually set ground truth value and a segmentation result so as to infer the quality of each frame. In this way, it is possible to determine which frames are the optimal frames for further measurement. In the optimal frame-based ONSD measurement stage, the boundary of an optic nerve sheath is accurately detected by using a boundary adjustment method, thereby enabling accurate measurement.

Description

An artificial intelligence-based optic nerve sheath positioning system and method Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to an optic nerve sheath positioning system and method based on artificial intelligence. Background Art

[0002] In the medical field, intracranial hypertension is a very dangerous symptom. In the current field of medical technology, intracranial pressure measurement can be divided into invasive and non-invasive methods. The invasive method mainly includes additional ventricular drainage, which can accurately measure ICP. However, the invasive method may cause external infection, bleeding, displacement and obstruction. Therefore, non-invasive ICP measurement shows the advantages of safety and speed. The non-invasive method mainly uses ultrasound to measure the diameter of the optic nerve sheath to reflect intracranial pressure. The present invention mainly describes a system and method for automatically retrieving and locating the optic nerve sheath in ultrasound video based on artificial intelligence. It further processes the video image, tracks the best frame of the image sequence through KCF, ensures that the selected image has a clear optic nerve sheath image, and realizes the automatic positioning of the optic nerve sheath by the SLIC segmentation algorithm, so as to achieve accurate ONSD measurement.

[0003] Summary of the Invention

[0004] The purpose of the present invention is to provide an optic nerve sheath positioning system and method based on artificial intelligence to solve the problems raised in the above background technology.

[0005] The innovation of this invention lies in the first on-line scintigraphy (ONSD) measurement based on ultrasound video. While previously, experts typically manually selected image-level data for measurement, this invention utilizes an automated approach. The research method is primarily divided into two phases: optimal frame selection and ONSD measurement based on the optimal frame. For optimal frame selection, the present invention meticulously crafts a true value as a good representation of ONSD and uses a superpixel approach to calculate the overlap between the manually set true value and the segmentation result to infer frame quality. This approach allows for the determination of which frames are optimal for further measurement. During the optimal frame-based ONSD measurement phase, a boundary adjustment method is employed to precisely detect the boundaries of the optic nerve sheath, enabling accurate measurement.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] An artificial intelligence-based optic nerve sheath positioning system, the optic nerve sheath positioning system includes an optic nerve sheath ultrasound image acquisition module, an image optimal frame search module, an optic nerve sheath positioning module, and an optic nerve sheath diameter measurement module;

[0008] The optic nerve sheath ultrasound image acquisition module is used to acquire an ultrasound image of the optic nerve sheath of the eye for the system and display the acquired ultrasound image of the optic nerve sheath;

[0009] Ultrasonic waves are emitted by ultrasonic probes to detect objects. When passing through the boundaries of objects, the ultrasonic waves are reflected, refracted, and lost. The reflected ultrasonic waves are received by the piezoelectric chip and converted into electrical signals. The electrical signals are then transmitted to the display end and converted into image data for output.

[0010] The image best frame search module is used to detect each frame of the optic nerve sheath ultrasound image using the KCF tracking algorithm, and the image with the largest two fluctuations in pixel values ​​is the best frame;

[0011] The target is tracked to determine the best frame within the optic nerve sheath ultrasound image. During the tracking process, a target detector is first trained to detect the current frame, and then the target is detected again in the next frame. The target detector is updated with these new detection results, and the process is repeated until the image containing the area with the most significant pixel value changes is found.

[0012] The optic nerve sheath positioning module is used to segment the image using the SLIC algorithm after obtaining the best frame of the optic nerve sheath ultrasound image, calculate the sum of the pixel values ​​of each row and column to obtain a one-dimensional signal based on the pixel values, and then use the Gaussian mixture model to calculate the initial search point to locate the position of the optic nerve sheath;

[0013] The optic nerve sheath diameter measurement module is used to calculate the diameter of the optic nerve sheath after locating the position of the optic nerve sheath;

[0014] The diameter of the optic nerve sheath is measured as the distance between the two pixel peaks of the optic nerve sheath. When the SLIC algorithm locates the position of the optic nerve sheath, the two maximum values ​​on both sides of i are measured, the points of the two maximum values ​​are determined, and the distance formula between the two points is used for calculation.

[0015] The optic nerve sheath ultrasound image acquisition module includes an ultrasound transmitting unit, a transmission unit and a display unit;

[0016] The ultrasonic wave transmitting unit is used to transmit ultrasonic waves to detect objects. When the ultrasonic waves encounter the interface of the object, they are reflected and refracted. After receiving the reflected ultrasonic waves, they are converted into electrical signals and sent to the transmission unit.

[0017] The transmission unit is used to transmit the electrical signal generated by the conversion of the received reflected wave to the display port;

[0018] The display unit is used to receive the electrical signal transmitted by the transmission unit, and convert the received electrical signal into image data for output and display.

[0019] The image key frame search module uses the KCF tracking algorithm to compare each frame of the optic nerve sheath ultrasound image and detects the image with the two largest fluctuations in pixel values ​​as the best frame. The KCF tracking algorithm uses the filtering algorithm to construct a circulant matrix to achieve target tracking.

[0020] The optic nerve sheath localization module uses the SLIC algorithm to segment the image into z superpixel regions where the pixel values ​​of adjacent regions vary by no more than 10, maps the pixels into the superpixel grid, and then uses a Gaussian mixture model to binarize the image, calculate the initial search point, and determine the location of the optic nerve sheath.

[0021] After the SLIC algorithm completes the segmentation of the optic nerve sheath ultrasound image, the optic nerve sheath diameter measurement module calculates the diameter distance of the optic nerve sheath based on the position of the optic nerve sheath.

[0022] In order to solve the above technical problems, the present invention provides an artificial intelligence-based optic nerve sheath positioning method, the specific technical method steps are as follows:

[0023] S1. Acquiring an ultrasound image of the optic nerve sheath of the eye. Specifically, an ultrasound probe transmits ultrasound waves to detect an object. When passing through the boundary of the object, the ultrasound waves are reflected, refracted, and lost. The reflected ultrasound waves are received by a piezoelectric chip and converted into electrical signals. The electrical signals are then transmitted to a display terminal and converted into image data for output.

[0024] S2. Use the KCF tracking algorithm to track the target after comparing the images of each frame to determine the best frame in the optic nerve sheath ultrasound image; during the tracking process, first train a target detector to detect the current frame, perform target detection again in the next frame, use these new detection results to update the target detector, and execute the loop until the image containing the area with the most significant pixel value changes is found. Through continuous iteration, detect the area with the most significant pixel value changes, and compare the area with the most significant pixel value changes with the previous frame, gradually determine the frame with the largest pixel value change, and thus find the best frame containing the optic nerve sheath image.

[0025] S3. Using the SLIC algorithm, the best frame image is segmented into z super-pixel images. By calculating the sum of the pixel values ​​of each row and each column respectively, a one-dimensional pixel function based on the pixel value is obtained; the sum of the pixels of each row and each column of the image is calculated respectively to detect the position of the optic nerve sheath;

[0026] Assume that the original image without any processing is an N×M two-dimensional array, and the pixel at (n, m) is represented as P(n, m). The one-dimensional signal can be calculated using the following formula:

[0027] The calculation method of v(n) above, where n ranges from 1 to N and m ranges from 1 to M, means that for each fixed value of n, P(n,m) is added from m=1 to m=M to obtain the value of v(n). The result of the calculation is a one-dimensional signal formed by the sum of the pixel values ​​of each column in the image.

[0028] Calculate v(n) using ultrasound images of the optic nerve sheath that have not been processed by the KCF algorithm. Since the v(n) signal is located at the valley between two peaks in the center of the optic nerve sheath, the specific location of the optic nerve sheath is to find a signal v(n) with n = i:

[0029] In the above formula, x is the location of the local maximum, i is the location of the valley between the two peaks, and v(n) is the minimum value in the valley. Local minimun is the local minimum, and exist a local maximun means there is a local minimum.

[0030] For the optic nerve sheath ultrasound image that has not been processed by the KCF algorithm, there is a local maximum in the image. When the position x of the maximum value is to the right of position i, the value of the variable v(n) is set to the pixel value of the position i of the local maximum value, where position i represents the specific position of the pixel; if the position x of the maximum value is to the left of position i, the variable v(n) still takes the pixel value of the position i of the local maximum value, where i represents the specific position of the pixel. When the position x of the local maximum value is to the right and left of position i, the value of the variable v(n) remains consistent and is equal to the pixel value of the position i of the local maximum value.

[0031] S4. After the outline of the optic nerve sheath is detected, the image is processed using a Gaussian mixture model to segment the image into a background and foreground with only 0 and 255 values. An initial search point can be obtained through calculation to detect the position of the searched optic nerve sheath.

[0032] The Gaussian mixture model is based on the Gaussian density function. The binary segmentation mask of the Gaussian mixture model is set to M(n,m), which contains the values ​​of (n,m) as 255 and 0. The signal k(n) is then obtained based on the segmentation mask.

[0033] The above calculation formula calculates the sum of M(n,m) from m=1 to m=M divided by the product of the sum of M(n,m) from m=1 to m=M and the sum of M(n,m) from n=1 to n=N, and finally obtains the value of k(n). k(n) calculates the ratio of the number of pixels with a pixel value of 255 in each row of the binary segmentation mask to the total number of pixels in the row. The position of the k(n) value is used as the initial search point to confirm the location of the optic nerve sheath;

[0034] The optimized initial search point d is the center of the signal k(n). Efficient search is achieved by solving the equation k(n) = 0.5. The center of the optic nerve sheath is located in the depression of the signal v(n). There are two peaks on the left and right sides. The position of the optic nerve sheath can be determined by finding the only depression. The initial search point n = d, the center of the optic nerve sheath being searched is v, and the signal d is calculated. center The current gradient, the specific gradient calculation formula is v (n＝d) =v d -v d-1 , v d is the value of the next gradient during gradient calculation, v d-1 is the value of the previous gradient; if v (n＝d) If negative, search to the left. (n＝d) Is positive, search to the right; according to the obtained center point d center , determine d by observing the first peak on the left and right left and d right , d center is the center point, d left is the peak on the left, d right It is the peak on the right, and two pixel peaks of the optic nerve sheath image are obtained in sequence.

[0035] S5. After detecting the best frame of the optic nerve sheath ultrasound image and locating the image of the optic nerve sheath, the diameter of the optic nerve sheath is measured. The diameter of the optic nerve sheath is measured as the distance between the two pixel peaks of the optic nerve sheath. When the SLIC algorithm locates the position of the optic nerve sheath, the two maximum values ​​on both sides of i are measured, and the diameter of the optic nerve sheath is calculated. The specific calculation formula is:

[0036] d n is the diameter of the optic nerve sheath, (e1, f1) and (e2, f2) are the two maximum points on both sides of i, and the distance formula between the two points is used to calculate the diameter of the optic nerve sheath.

[0037] Compared with the prior art, the present invention has the following beneficial effects:

[0038] 1. The present invention uses the KCF algorithm to track video images and intelligently selects the best frame in the video to ensure that the selected image has a complete and clear image of the optic nerve sheath, making the subsequent positioning of the optic nerve sheath more accurate and stable.

[0039] 2. The present invention uses the SLIC algorithm to cut the image of the optic nerve sheath to form a super-pixel image. By calculating the sum of the pixels in each row and column, the boundary of the optic nerve sheath is located. The diameter of the optic nerve sheath can be calculated in a non-invasive and non-invasive manner, avoiding the problems of infection, bleeding, displacement and blockage that may be caused by invasive measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0041] FIG1 is a module structure diagram of an artificial intelligence-based optic nerve sheath positioning system of the present invention;

[0042] FIG2 is a positioning model diagram of an artificial intelligence-based optic nerve sheath positioning method of the present invention;

[0043] FIG3 is a table showing average error comparison results of an artificial intelligence-based optic nerve sheath positioning method of the present invention;

[0044] FIG4 is a Bland-Altman diagram of an artificial intelligence-based optic nerve sheath positioning method of the present invention;

[0045] FIG5 is a boxplot of an artificial intelligence-based optic nerve sheath positioning method of the present invention. DETAILED DESCRIPTION

[0046] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0047] In order to solve the above problems, the present invention provides a new system:

[0048] An artificial intelligence-based optic nerve sheath positioning system includes an optic nerve sheath ultrasound image acquisition module, an image best frame search module, an optic nerve sheath positioning module, and an optic nerve sheath diameter measurement module;

[0049] The optic nerve sheath ultrasound image acquisition module is used to acquire an ultrasound image of the optic nerve sheath of the eye for the system and display the acquired ultrasound image of the optic nerve sheath;

[0050] The image best frame search module is used to detect each frame of the optic nerve sheath ultrasound image using the KCF tracking algorithm, and the image with the largest two fluctuations in pixel values ​​is the best frame;

[0051] The optic nerve sheath positioning module is used to segment the image using the SLIC algorithm after obtaining the best frame of the optic nerve sheath ultrasound image, calculate the sum of the pixel values ​​of each row and column to obtain a one-dimensional signal based on the pixel values, and then use the Gaussian mixture model to calculate the initial search point to locate the position of the optic nerve sheath;

[0052] The optic nerve sheath diameter measurement module is used to calculate the diameter of the optic nerve sheath after locating the position of the optic nerve sheath.

[0053] The optic nerve sheath ultrasound image acquisition module includes an ultrasound transmitting unit, a transmission unit and a display unit;

[0054] The ultrasonic wave transmitting unit is used to transmit ultrasonic waves to detect objects. When the ultrasonic waves encounter the interface of the object, they are reflected and refracted. After receiving the reflected ultrasonic waves, they are converted into electrical signals and sent to the transmission unit.

[0055] The transmission unit is used to transmit the electrical signal generated by the conversion of the received reflected wave to the display port;

[0056] The display unit is used to receive the electrical signal transmitted by the transmission unit, and convert the received electrical signal into image data for output and display.

[0057] The image key frame search module uses the KCF tracking algorithm to compare each frame of the optic nerve sheath ultrasound image and detects the image with the two largest fluctuations in pixel value as the best frame. The KCF tracking algorithm uses the filtering algorithm to construct a circulant matrix to achieve target tracking.

[0058] The optic nerve sheath localization module uses the SLIC algorithm to segment the image into z superpixel regions where the pixel values ​​of adjacent regions vary by no more than 10, maps the pixels into the superpixel grid, and then uses a Gaussian mixture model to binarize the image, calculate the initial search point, and determine the location of the optic nerve sheath.

[0059] After the SLIC algorithm completes the segmentation of the optic nerve sheath ultrasound image, the optic nerve sheath diameter measurement module calculates the diameter distance of the optic nerve sheath based on the position of the optic nerve sheath.

[0060] Through the embodiments and drawings, the present invention provides the following technical solutions:

[0061] An artificial intelligence-based optic nerve sheath positioning method, the specific steps of the method are:

[0062] S1. Acquiring an ultrasound image of the optic nerve sheath of the eye. Specifically, an ultrasound probe transmits ultrasound waves to detect an object. When passing through the boundary of the object, the ultrasound waves are reflected, refracted, and lost. The reflected ultrasound waves are received by a piezoelectric chip and converted into electrical signals. The electrical signals are then transmitted to a display terminal and converted into image data for output.

[0063] S2. Use the KCF tracking algorithm to track the target after comparing the images of each frame to determine the best frame in the optic nerve sheath ultrasound image; during the tracking process, first train a target detector to detect the current frame, perform target detection again in the next frame, use these new detection results to update the target detector, and execute the loop until the image containing the area with the most significant pixel value changes is found. Through continuous iteration, detect the area with the most significant pixel value changes, and compare the area with the most significant pixel value changes with the previous frame, gradually determine the frame with the largest pixel value change, and thus find the best frame containing the optic nerve sheath image.

[0064] S3. Using the SLIC algorithm, the best frame image is segmented into z super-pixel images. By calculating the sum of the pixel values ​​of each row and each column respectively, a one-dimensional pixel function based on the pixel value is obtained; the sum of the pixels of each row and each column of the image is calculated respectively to detect the position of the optic nerve sheath;

[0065] Assume that the original image without any processing is an N×M two-dimensional array, and the pixel at (n, m) is represented as P(n, m). The one-dimensional signal can be calculated using the following formula:

[0066] The calculation method of v(n) above, where n ranges from 1 to N and m ranges from 1 to M, means that for each fixed value of n, P(n,m) is added from m=1 to m=M to obtain the value of v(n). The result of the calculation is a one-dimensional signal formed by the sum of the pixel values ​​of each column in the image.

[0067] Calculate v(n) using ultrasound images of the optic nerve sheath that have not been processed by the KCF algorithm. Since the v(n) signal is located at the valley between two peaks in the center of the optic nerve sheath, the specific location of the optic nerve sheath is to find a signal v(n) with n = i:

[0068] In the above formula, x is the location of the local maximum, i is the location of the valley between the two peaks, and v(n) is the minimum value at the valley. Local minimun is the local minimum, and exista local maximun means there is a local minimum.

[0069] For the optic nerve sheath ultrasound image that has not been processed by the KCF algorithm, there is a local maximum in the image. When the position x of the maximum value is to the right of position i, the value of the variable v(n) is set to the pixel value of the position i of the local maximum value, where position i represents the specific position of the pixel; if the position x of the maximum value is to the left of position i, the variable v(n) still takes the pixel value of the position i of the local maximum value, where i represents the specific position of the pixel. When the position x of the local maximum value is to the right and left of position i, the value of the variable v(n) remains consistent and is equal to the pixel value of the position i of the local maximum value.

[0070] S4. After the outline of the optic nerve sheath is detected, the image is processed using a Gaussian mixture model to segment the image into a background and foreground with only 0 and 255 values. An initial search point can be obtained through calculation to detect the position of the searched optic nerve sheath.

[0071] The Gaussian mixture model is based on the Gaussian density function. The binary segmentation mask of the Gaussian mixture model is set to M(n,m), which contains the values ​​of (n,m) as 255 and 0. The signal k(n) is then obtained based on the segmentation mask.

[0072] The above calculation formula calculates the sum of M(n,m) from m=1 to m=M divided by the product of the sum of M(n,m) from m=1 to m=M and the sum of M(n,m) from n=1 to n=N, and finally obtains the value of k(n). k(n) calculates the ratio of the number of pixels with a pixel value of 255 in each row of the binary segmentation mask to the total number of pixels in the row. The position of the k(n) value is used as the initial search point to confirm the position of the optic nerve sheath;

[0073] The optimized initial search point d is the center of the signal k(n). Efficient search is achieved by solving the equation k(n) = 0.5. The center of the optic nerve sheath is located in the depression of the signal v(n). There are two peaks on the left and right sides. The position of the optic nerve sheath can be determined by finding the only depression. The initial search point n = d, the center of the optic nerve sheath being searched is v, and the signal d is calculated. center The current gradient, the specific gradient calculation formula is v (n＝d) =v d -v d-1 , v d is the value of the next gradient during gradient calculation, v d-1 is the value of the previous gradient; if v (n＝d) If negative, search to the left. (n＝d) Is positive, search to the right; according to the obtained center point d center , determine d by observing the first peak on the left and right left and d right , d center is the center point, d left is the peak on the left, dright It is the peak on the right, and two pixel peaks of the optic nerve sheath image are obtained in sequence.

[0074] S5. After detecting the best frame of the optic nerve sheath ultrasound image and locating the image of the optic nerve sheath, the diameter of the optic nerve sheath is measured. The diameter of the optic nerve sheath is measured as the distance between the two pixel peaks of the optic nerve sheath. When the SLIC algorithm locates the position of the optic nerve sheath, the two maximum values ​​on both sides of i are measured, and the diameter of the optic nerve sheath is calculated. The specific calculation formula is:

[0075] d n is the diameter of the optic nerve sheath, (e1, f1) and (e2, f2) are the two maximum points on both sides of i, and the distance formula between the two points is used to calculate the diameter of the optic nerve sheath.

[0076] There is a binary segmentation mask that represents the areas that are not easily penetrated by ultrasound and other areas. This segmentation mask only contains the pixels of the foreground and background. When using the Gaussian mixture model to segment the image into foreground and background, the data is clustered into five components. The specific steps are:

[0077] D1. Collect sample data and prepare a set of sample data representing different areas in the image. These data can be pixel values, color values, and other vector values.

[0078] D2. Initialize the Gaussian mixture model and determine the initial parameters of the Gaussian mixture model. The specific parameters include the number, mean, covariance and mixing weight of the Gaussian distribution;

[0079] D3, iterative training process, using the expectation maximization algorithm to iteratively update the model parameters until the convergence condition is reached;

[0080] a. Calculate the posterior probability of each sample belonging to each Gaussian parameter based on the current model parameters;

[0081] b. Update the mean, covariance, and mixing weight of the Gaussian distribution based on the posterior probability;

[0082] c. Repeat steps a and b above until the model converges;

[0083] D4. Thresholding: After training, select a suitable threshold to separate the image into foreground and background. The most commonly used method is to select the peak value of the probability density function in the Gaussian distribution as the segmentation threshold.

[0084] D5, mark pixels greater than the threshold as 255, and pixels less than the threshold as 0;

[0085] Based on the segmentation results, a binary segmentation mask can be obtained. The area that is not easily penetrated by ultrasound is marked as 255 foreground, and other areas are marked as 0 background. This binary segmentation mask has the same size as the image.

[0086] Example 1

[0087] To demonstrate the effectiveness of the proposed method, we calculated the average error between the proposed method and the manual measurements of two experts using the following formula:

[0088] Where n represents the number of ultrasound images, ONSD1 is the measurement result calculated by our method, and ONSD2 is the average manual measurement result of two experts. The mean square error is used to calculate the difference between the two experts' measurement results.

[0089] Furthermore, we confirmed the correlation coefficient and 95% confidence interval between the automatically measured diameters and the true diameters.

[0090] The results of the average error are shown in Figure 3. Comparing ONSD1 and ONSD2, the average error of the proposed method is 0.044 and 0.043 respectively. We also calculated the mean squared error to evaluate the difference between the two experts' measurements. The mean squared error of the two experts' measurements is 0.027, the intraclass correlation coefficient of the two experts' measurements is 0.952, and the ICC between the algorithm's measurement results and the average measurement of the two experts is 0.782. The Bland-Altman plots of ONSD1, ONSD2, the average, and the proposed method are shown in Figure 4. The boxplots of ONSD1, ONSD2, the average, and the proposed method are shown in Figure 5.

[0091] The results of the examples show that there is no significant difference or variability between the measurement method of the present invention and manual measurement. Our algorithm demonstrates comparable reliability and consistency to experts in comparison. However, our algorithm has certain requirements for ultrasound image quality, and data deviations may occur in the presence of noise and image blur. Our KCF tracking algorithm for selecting the best image frame can intuitively display the differences between different frames, providing a better basis for selection.

[0092] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0093] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art will be able to modify the technical solutions described in the aforementioned embodiments or substitute equivalents for some of the technical features. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based optic nerve sheath positioning system, characterized in that: The optic nerve sheath positioning system comprises an optic nerve sheath ultrasound image acquisition module, an image best frame search module, an optic nerve sheath positioning module and an optic nerve sheath diameter measurement module; The optic nerve sheath ultrasonic image acquisition module is used to acquire an ultrasonic image of the optic nerve sheath of the eye for the system, and to display the acquired ultrasonic image of the optic nerve sheath; The image best frame search module is used to detect each frame of the optic nerve sheath ultrasound image using the KCF tracking algorithm, and the image with the largest two fluctuations in pixel value is the best frame; The optic nerve sheath positioning module is used to segment the image using the SLIC algorithm after obtaining the best frame of the optic nerve sheath ultrasound image, calculate the sum of the pixel values ​​of each row and column, obtain a one-dimensional signal based on the pixel value, and then use the Gaussian mixture model to calculate the initial search point to locate the position of the optic nerve sheath; The optic nerve sheath diameter measurement module is used to calculate the diameter of the optic nerve sheath after locating the position of the optic nerve sheath.

2. The optic nerve sheath positioning system based on artificial intelligence according to claim 1, characterized in that: The optic nerve sheath ultrasound image acquisition module includes an ultrasound transmitting unit, a transmission unit and a display unit; The ultrasonic wave transmitting unit is used to transmit ultrasonic waves to detect objects. When the ultrasonic waves encounter the interface of the object, they are reflected and refracted. After receiving the reflected ultrasonic waves, they are formed into electrical signals and reach the transmission unit. The transmission unit is used to transmit the electrical signal generated by the conversion of the received reflected wave to the display port; The display unit is used to receive the electrical signal transmitted by the transmission unit, and convert the received electrical signal into image data for output and display.

3. The optic nerve sheath positioning system based on artificial intelligence according to claim 1, characterized in that: The image key frame search module detects the image with the largest two fluctuations in pixel value as the best frame after comparing each frame of the optic nerve sheath ultrasound image through the KCF tracking algorithm. The KCF tracking algorithm uses a filtering algorithm to construct a circulant matrix to achieve target tracking.

4. The optic nerve sheath positioning system based on artificial intelligence according to claim 1, characterized in that The optic nerve sheath positioning module uses the SLIC algorithm to segment the image into z super-pixel areas where the pixel value variation of adjacent areas does not exceed 10, maps the pixels to the super-pixel grid, and then uses the Gaussian mixture model to binarize the image, calculate the initial search point, and determine the location of the optic nerve sheath.

5. The artificial intelligence-based optic nerve sheath positioning system according to claim 1, characterized in that: The optic nerve sheath diameter measurement module calculates the diameter distance of the optic nerve sheath according to the position of the optic nerve sheath after the SLIC algorithm completes the segmentation of the optic nerve sheath ultrasound image.

6. An artificial intelligence-based optic nerve sheath positioning method, characterized in that: The specific steps of this method are: S1, obtaining an ultrasonic image of the optic nerve sheath of the eye, specifically, the ultrasonic probe emits ultrasonic waves to detect objects, and the ultrasonic waves are reflected, refracted and lost when passing through the boundary of the object. The reflected ultrasonic waves are received by the piezoelectric chip and converted into electrical signals, and then the electrical signals are transmitted to the display end and converted into image data for output; S2, using the KCF tracking algorithm to track the target and determine the best frame in the optic nerve sheath ultrasound image after comparing each frame of the image; S3, using the SLIC algorithm, segmenting the best frame image into z super-pixel images, and obtaining a one-dimensional pixel function based on the pixel value by calculating the sum of the pixel values ​​of each row and each column respectively; S4. After the outline of the optic nerve sheath is detected, the image is processed using a Gaussian mixture model to segment the image into a background and foreground with only 0 and 255. An initial search point can be obtained by calculation to detect the position of the searched optic nerve sheath. S5. After detecting the best frame of the optic nerve sheath ultrasound image and locating the image of the optic nerve sheath, Complete the measurement of the optic nerve sheath diameter.

7. The method for locating the optic nerve sheath based on artificial intelligence according to claim 6, characterized in that: In step S2, during the tracking process, firstly, a target detector is trained to detect the current frame, and then target detection is performed again in the next frame. The target detector is updated with the new detection result, and the cycle is performed until the image containing the area with the most significant pixel value change is detected. The area with the most significant pixel value change is detected by continuous iteration, and the area with the most significant pixel value change is compared with the previous frame, and the frame with the largest pixel value change is gradually determined, thereby detecting the best frame containing the optic nerve sheath image.

8. The method for locating the optic nerve sheath based on artificial intelligence according to claim 6, characterized in that: In step S3, the sum of pixels in each row and column of the image is calculated to detect the position of the optic nerve sheath; Assume that the original image without any processing is an N×M two-dimensional array, and the pixel at (n, m) is represented as P(n, m). The one-dimensional signal can be calculated using the following formula: The above calculation method of v(n), where n ranges from 1 to N and m ranges from 1 to M, means that for each fixed value of n, P(n, m) is added from m=1 to m=M to obtain the value of v(n). The calculation result is a one-dimensional signal formed by the sum of the pixel values ​​of each column in the image. The optic nerve sheath ultrasound image that has not been processed by the KCF algorithm is used to calculate v(n). Since the v(n) signal is located at the trough between the two peaks in the center of the optic nerve sheath, the specific location of the optic nerve sheath is to find a signal v(n) with n=i: In the above formula, x is the location of the local maximum value, i is the location of the valley between the two peaks, and v(n) is the minimum value at the valley; local minimun is the local minimum value, and there is a local maximun There is a local minimum; For the optic nerve sheath ultrasound image that has not been processed by the KCF algorithm, there is a local maximum in the image. When the position x of the maximum value is to the right of position i, the value of the variable v(n) is set to the pixel value of the position i of the local maximum value, where position i represents the specific position of the pixel; if the position x of the maximum value is to the left of position i, the variable v(n) still takes the pixel value of the position i of the local maximum value, where i represents the specific position of the pixel. When the position x of the local maximum value is to the right and left of position i, the value of the variable v(n) remains consistent, which is equal to the pixel value of the position i of the local maximum value.

9. The method for locating the optic nerve sheath based on artificial intelligence according to claim 6, characterized in that: In step S4, the Gaussian mixture model is based on a Gaussian density function, and the binary segmentation mask of the Gaussian mixture model is set to M(n,m), which contains the values ​​of (n,m) as 255 and 0, and then the signal k(n) is obtained according to the segmentation mask; The above calculation formula calculates the sum of M(n,m) from m=1 to m=M divided by the product of the sum of M(n,m) from m=1 to m=M and the sum of M(n,m) from n=1 to n=N, and finally obtains the value of k(n). k(n) calculates the ratio of the number of pixels with a pixel value of 255 in each row of the binary segmentation mask to the total number of pixels in the row. The position of the k(n) value is used as the initial search point to confirm the position of the optic nerve sheath; The optimized initial search point d is the center of the signal k(n). Efficient search is achieved by solving the equation k(n) = 0.

5. The center of the optic nerve sheath is located at the depression of the signal v(n). There are two peaks on the left and right sides. The location of the optic nerve sheath can be determined by finding the only depression. The initial search point n = d, the center of the optic nerve sheath being searched is v, and the signal d is calculated. center The current gradient of (n＝d) =v d -v d-1 , v d is the value of the next gradient when calculating the gradient, v d-1 is the value of the previous gradient; if v (n＝d) If negative, search to the left. (n＝d) is positive, search to the right; according to the obtained center point d center , determine d by observing the first peak on the left and right left and d right , d center is the center point, d left is the peak on the left, d right It is the peak on the right, and two pixel peaks of the optic nerve sheath image are obtained in turn.

10. The method for locating the optic nerve sheath based on artificial intelligence according to claim 6, characterized in that: In step 5, the measurement of the diameter of the optic nerve sheath is the distance between the two pixel peaks of the optic nerve sheath. When the SLIC algorithm locates the position of the optic nerve sheath, the two maximum values ​​on both sides of i are measured to calculate the diameter of the optic nerve sheath. The specific calculation formula is: d n is the diameter of the optic nerve sheath, (e1, f1) and (e2, f2) are the two maximum points on both sides of i, and the distance formula between the two points is used to calculate the diameter of the optic nerve sheath.

Images