Cervical curvature judgment method based on in-vitro photographing

Abstract

The invention discloses a cervical curvature judgment method based on in-vitro photographing. The method comprises the following steps: preprocessing an original neck in-vitro image of a user to obtain an equalized image; determining a single local feature of each pixel at different positions under different convolution sums; extracting a previous-layer feature map in the single local features through U-Net, and performing segmentation to obtain a current-layer feature map; performing up-sampling operation on the current-layer feature map to obtain a prediction size of the front-layer feature map, and determining cross entropy loss in combination with the prediction size; a key point set in the original neck in-vitro image is extracted through a total loss function, and the total length of a fitting curve is calculated; determining the curve of the cervical vertebra shape according to the total length of the fitting curve; determining the curvature of the position of the key point in the key point set and the correction length of the curve of the cervical vertebra shape; therefore, the degree of cervical curvature abnormity is determined. According to the invention, a deep learning technology is combined, and doctors are assisted in diagnosis in a radiation-free manner.

Description

Technical Field

[0001] This invention relates to the field of cervical curvature assessment technology, and more particularly to a method for assessing cervical curvature based on in vitro photography. Background Technology

[0002] Currently, the diagnosis of cervical curvature mainly relies on imaging technologies such as X-rays, CT, or MRI. While these technologies provide accurate images of the cervical spine, they carry certain radiation exposure risks, especially for patients requiring long-term follow-up, where cumulative radiation exposure may pose health risks. Furthermore, traditional imaging technologies are costly, require complex equipment, and are not suitable for large-scale screening and continuous monitoring, exhibiting the following drawbacks: 1. Radiation risk; 2. Strong equipment dependence; 3. High cost and complex operation. In recent years, with the development of artificial intelligence technology, image processing algorithms have been able to extract more information from ordinary photographs. However, current technologies have not fully utilized these algorithms for cervical curvature assessment, exhibiting the following drawbacks: 1. Failure to consider the characteristics of medical imaging; 2. Lack of X-ray simulation technology; 3. Simple diagnostic process; 4. Limited application scenarios. Summary of the Invention

[0003] Therefore, it is necessary to propose a method for judging cervical curvature based on external photography to address the above problems.

[0004] A method for determining cervical curvature based on external imaging, the method comprising: Acquire the user's original external neck image and preprocess the original external neck image to obtain an equalized image; By equalizing the neighboring pixel values ​​of each pixel in the image in the CNN convolution and combining the weight parameters of each pixel in the CNN convolution, we can determine the single local feature of each pixel at different positions under different convolutions. The previous layer feature map is extracted from a single local feature using U-Net; the previous layer feature map is then segmented to obtain the current layer feature map; An upsampling operation is performed on the current layer feature map to obtain the predicted size of the segmented current layer feature map; Obtain the actual label size and determine the cross-entropy loss based on the predicted size; Obtain the predicted segmentation region set and the actual segmentation region set of the original neck in vitro image, and determine the Dice loss function based on the predicted segmentation region set and the actual segmentation region set; determine the total loss function based on the cross-entropy loss and the Dice loss function; Then, the set of key points in the original neck in vitro image is extracted using the total loss function; the total length of the fitted curve is calculated for the key points in the set of key points. The curve of the cervical spine shape is determined based on the total length of the fitted curve; the curvature of the key point location in the key point set and the correction length of the curve of the cervical spine shape are determined based on the curve of the cervical spine shape. The degree of cervical curvature abnormality is determined based on the corrected length and the key points in the key point set.

[0005] In one embodiment, preprocessing the original neck in vitro image to obtain an equalized image includes: The user's original neck external image is denoised using an adaptive median filtering method to obtain the filtered pixel result corresponding to the user's original neck external image; The filtered pixel results are then subjected to grayscale conversion to obtain a grayscale image; The grayscale image is subjected to histogram equalization to obtain an equalized image.

[0006] In one embodiment, the equalized image is obtained by the following expression: in, The result is the filtered pixel count. For the front filtering window The median of the pixel values, which is the value in the middle position after sorting, is used to reduce the impact of salt and pepper noise; For the current filtering window The minimum pixel value in; For the current filtering window The maximum pixel value in; Original in vitro image of the neck; To convert the image to grayscale; For position The pixel value of the red channel; For position The pixel value of the green channel; For position The pixel value of the blue channel; To equalize the image; grayscale image In the middle, the grayscale value is The frequency of pixels; This represents the total number of pixels. The grayscale level is the number of levels.

[0007] In one embodiment, the determination of a single local feature at different positions of each pixel under different convolutions by equalizing the neighboring pixel values ​​corresponding to each pixel in the CNN convolution and combining the weight parameters of each pixel in the CNN convolution is achieved by the following expression: in, For position grayscale image at [location] The equalized image obtained after histogram equalization; It is a single local feature; For the convolution kernel at position Weight parameters; This is the bias term for the convolution kernel; This is the activation function.

[0008] In one embodiment, The current layer feature map is obtained by segmenting the previous layer feature map using the following expression: in, This is the feature map of the current layer; This is the feature map of the previous layer; For convolution kernel; For bias terms; It is a non-linear activation function; This is a convolution operation; The predicted size of the segmented current layer feature map is obtained by upsampling the current layer feature map using the following expression: in, For predicted dimensions; This is the feature map of the current layer; This is an upsampling operation; The process of obtaining the actual label size and determining the cross-entropy loss by combining it with the predicted size is achieved through the following expression: in, Cross-entropy loss; For predicted dimensions; This is the actual label size; The number of samples; The determination of the Dice loss function based on the predicted segmented region set and the actual segmented region set is achieved by the following expression: in, To predict the set of segmented regions; For a set of real partitioned regions; To predict the number of pixels in the segmented region; The number of pixels in the actual segmented region; The Dice loss function; The determination of the total loss function based on the cross-entropy loss and the Dice loss function is achieved by the following expression: in, This is the total loss function; These are weighting coefficients; Cross-entropy loss; This is the Dice loss function.

[0009] In one embodiment, the calculation of the total length of the fitted curve for the key points in the key point set is achieved by the following expression: in, This represents the total length of the fitted curve; Set of key points One of the key points; Set of key points Another key point; This represents the total number of discrete points on the curve.

[0010] In one embodiment, the curve for determining the cervical spine morphology based on the total length of the fitted curve is implemented using the following expression: in, The curve representing the shape of the cervical spine; , , and These are the coefficients for spline interpolation; This is a correction item; The independent variable in the interpolation calculation ranges from arrive The coordinates between them.

[0011] In one embodiment, the determination of the curvature of the key point locations in the key point set based on the curve of the cervical spine morphology and the correction length of the curve of the cervical spine morphology are achieved by the following expression: in, For curvature; The derivative of the curve representing the cervical spine shape; To correct the length; denoted as the slope of the curve.

[0012] In one embodiment, the determination of the degree of cervical curvature abnormality based on the corrected length and the key points in the key point set is achieved by the following expression: in, The degree of abnormal cervical curvature; The total number of discrete points; Points on the curve of the cervical spine; For the first The ordinate values ​​of each point on the standard curve; The weighting coefficients for adjustment; To correct the length.

[0013] Therefore, this invention utilizes intelligent device imaging and image analysis technology to determine the curvature of the cervical spine, overcoming the shortcomings of traditional techniques. Combined with deep learning technology and a radiation-free approach, it assists doctors in diagnosis. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] in: Figure 1 This is an application environment diagram of a cervical curvature determination method based on in vitro photography in one embodiment. Figure 2 This is a flowchart of a method for determining cervical curvature based on in vitro photography in one embodiment; Figure 3 This is a structural block diagram of a computer device in one embodiment. Detailed Implementation

[0016] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0017] Currently, the diagnosis of cervical curvature mainly relies on imaging technologies such as X-rays, CT, or MRI. While these technologies provide accurate cervical spine images, they carry certain radiation exposure risks, especially for patients requiring long-term follow-up, where cumulative radiation exposure may pose health risks. Furthermore, traditional imaging technologies are costly, require complex equipment, and are not suitable for large-scale screening and continuous monitoring, exhibiting the following drawbacks: 1. Radiation risk; 2. Strong equipment dependence; 3. High cost and complex operation. In recent years, with the development of artificial intelligence technology, image processing algorithms have been able to extract more information from ordinary photographs. However, existing technologies have not fully utilized these algorithms for cervical curvature assessment, exhibiting the following drawbacks: 1. Failure to consider the characteristics of medical imaging; 2. Lack of X-ray simulation technology; 3. Simple diagnostic process; 4. Limited application scenarios. To address the above technical problems, this application provides a method for determining cervical curvature based on in vitro imaging.

[0018] Figure 1 This is a diagram illustrating the application environment of a cervical curvature determination method based on in vitro photography in one embodiment. (Refer to...) Figure 1This method for judging cervical curvature based on external imaging is applied to a system for judging cervical curvature based on external imaging. The system includes a terminal 110 and a server 120. The terminal 110 and server 120 are connected via a network. The terminal 110 can be a desktop terminal or a mobile terminal, specifically a mobile phone, tablet, or laptop. The server 120 can be a standalone server or a server cluster consisting of multiple servers. The terminal 110 is used to acquire the user's original external cervical images and preprocess them to obtain an equalized image. The server 120 is used to determine the single local features of each pixel at different positions under different convolutions by combining the neighboring pixel values ​​of each pixel in the equalized image with the weight parameters of each pixel in the CNN convolution; extract the previous layer feature map from the single local features using U-Net; segment the previous layer feature map to obtain the current layer feature map; perform an upsampling operation on the current layer feature map to obtain the predicted size of the segmented current layer feature map; obtain the actual label size and determine the cross-entropy loss based on the predicted size; and obtain the original cervical curvature. The system generates a predicted segmentation region set and a real segmentation region set from the external image of the neck, and determines the Dice loss function based on the predicted segmentation region set and the real segmentation region set; it determines the total loss function based on the cross-entropy loss and the Dice loss function; it extracts a set of key points from the original external image of the neck using the total loss function; it calculates the total length of the fitted curve for the key points in the set of key points; it determines the curve of the cervical spine morphology based on the total length of the fitted curve; it determines the curvature of the key point locations in the set of key points and the correction length of the curve of the cervical spine morphology based on the curve of the cervical spine morphology; and it determines the degree of abnormal cervical curvature based on the correction length and the key points in the set of key points.

[0019] like Figure 2 As shown, in one embodiment, a method for determining cervical curvature based on external imaging is provided. This method can be applied to both terminals and servers; this embodiment illustrates its application to a terminal. The method for determining cervical curvature based on external imaging specifically includes the following steps: A method for determining cervical curvature based on external imaging, such as Figure 2 As shown, the method includes: S10: Acquire the user's original external neck image and preprocess the original external neck image to obtain an equalized image; S20: By equalizing the neighboring pixel values ​​of each pixel in the image in the CNN convolution and combining the weight parameters of each pixel in the CNN convolution, we determine the single local feature of each pixel at different positions under different convolutions; S30: Extract the previous layer feature map from a single local feature using U-Net; segment the previous layer feature map to obtain the current layer feature map; S40: Perform an upsampling operation on the current layer feature map to obtain the predicted size of the segmented current layer feature map; S50: Obtain the actual label size and determine the cross-entropy loss based on the predicted size; S60: Obtain the predicted segmentation region set and the actual segmentation region set of the original neck in vitro image, and determine the Dice loss function based on the predicted segmentation region set and the actual segmentation region set; determine the total loss function based on the cross-entropy loss and the Dice loss function; S70: Extract the set of key points from the original neck in vitro image using the total loss function; calculate the total length of the fitted curve for the key points in the set of key points; S80: Determine the cervical spine morphology curve based on the total length of the fitted curve; determine the curvature of the key point locations in the key point set and the correction length of the cervical spine morphology curve based on the cervical spine morphology curve. S90: Determine the degree of cervical curvature abnormality based on the corrected length and the key points in the key point set.

[0020] This invention utilizes intelligent device imaging and image analysis technology to determine the curvature of the cervical spine, overcoming the shortcomings of traditional techniques. Combined with deep learning technology and a radiation-free approach, it assists doctors in diagnosis.

[0021] In one embodiment, the preprocessing of the original neck in vitro image to obtain an equalized image in step S10 includes: S101: Denoise the user's original neck external image by using an adaptive median filtering method to obtain the filtered pixel result corresponding to the user's original neck external image; S102: Perform grayscale conversion processing on the filtered pixel results to obtain a grayscale image; S103: Perform histogram equalization on the grayscale image to obtain an equalized image.

[0022] In one embodiment, raw images of the user's neck outside the body are acquired via a smart device (such as a smartphone or digital camera). The captured color photographs were then denoised and enhanced using an improved adaptive median filtering algorithm. Next, image grayscale conversion, brightness enhancement, and contrast adjustment were applied to simulate the visual effect of cervical spine X-ray images. The improved adaptive median filtering algorithm was used for image denoising, primarily targeting salt-and-pepper noise and Gaussian noise.

[0023] The equalized image obtained in step S10 is obtained by the following expression: (1) (2) (3) in, The filtered pixel results are filtered based on the median of the pixel neighborhood, effectively removing salt-and-pepper noise from the image while enhancing image smoothness and detail preservation. (The coefficients are...) These coefficients reflect the human eye's sensitivity to different colors. The green channel contributes the most to brightness, followed by red, while blue contributes the least. For the front filtering window The median of the pixel values, which is the value in the middle position after sorting, is used to reduce the impact of salt and pepper noise; For the current filtering window The minimum pixel value in the filter window is used to determine whether the range of pixel values ​​within the filter window is reasonable. For the current filtering window The maximum pixel value in the filter window is used to determine whether there are outliers within the filter window. When the median condition of the original neck in vitro image is not met, the original pixel values ​​are used to preserve the image structure. A grayscale image is a single-channel brightness information in a neck image, typically ranging from grayscale to grayscale. ; For position The pixel values ​​of the red channel on the screen, typically ranging from 100 to 1000 pixels. ; For position The pixel values ​​of the green channel on the screen, typically ranging from 100 to 1000 pixels. ; For position The pixel values ​​of the blue channel on the screen, typically ranging from 100 to 1000 pixels. ; To equalize the image; grayscale image In the middle, the grayscale value is The frequency of pixels; This represents the total number of pixels. The gray level is the number of gray levels (usually the gray range of the image, typically 256). : Represents the filtering window region (usually a two-dimensional matrix, such as ), containing pixels and the surrounding pixel values; conditions : Used to determine whether the median in the current filtering window is within a reasonable range. If the condition is met, it indicates that the median is reliable, and the median is output as the filtering result. When the conditions are not met, the original neck image is used directly at the pixel location. By adjusting the size of the filtering window, the filter can adapt to noise of different densities and preserve the edge information of the neck image; The calculation of the cumulative distribution function (CDF) represents the gray levels in the image. The frequency of pixels. By accumulating all pixels less than or equal to the gray level. It is obtained from the pixel frequency; This is a normalization constant used to adjust the gray levels of the image so that the final gray values ​​fall within a certain range. Within the range.

[0024] In one embodiment, after the grayscale image is equalized, key points (such as edges, joints, etc.) in the cervical spine image are identified through manual annotation. This annotation data is used to train a CNN model. Subsequently, the CNN automatically extracts key features (such as edges, textures, shapes, etc.) of the image through convolution operations, improving the model's accuracy and robustness, thereby accurately annotating key points in new images.

[0025] The step S20, which involves equalizing the neighboring pixel values ​​of each pixel in the CNN convolution and combining them with the weight parameters of each pixel in the CNN convolution to determine the single local feature of each pixel at different positions under different convolutions, is achieved through the following expression: (4) (5) in, For position grayscale image at [location] The equalized image obtained through histogram equalization is the neighborhood pixel value of the current pixel (the pixel covered by the receptive field of the convolutional kernel). All pixel values ​​input into the CNN are grayscale values ​​after histogram equalization, i.e., each... ; It is a single local feature; For the convolution kernel at position The weight parameters are used to perform multiplication with the input values ​​of the image to extract... Image features after equalization processing; The bias term for the convolution kernel is used to adjust the output value; For activation functions; For all positions within the convolution kernel window Summation of .

[0026] In one embodiment, after image processing by CNN, the U-Net deep learning model is used for image segmentation to handle complex backgrounds and irregular boundaries, ensuring the accuracy of subsequent fitting and analysis. Specifically, single local features extracted by CNN are used as input to U-Net and passed to the encoder part of U-Net for further processing. U-Net extracts multi-scale features of the image through multiple convolutional units, ultimately completing fine segmentation of the image.

[0027] The step S30, which involves segmenting the previous layer feature map to obtain the current layer feature map, is achieved through the following expression: (6) in, The current layer feature map is a collection of multiple single-channel feature maps. The result obtained later; This is the previous layer feature map, derived from the image after CNN processing and feature extraction. This represents the first layer of feature maps. Subsequent layers extract deeper image features by continuously stacking convolutional and activation units. The convolution kernel is used to extract specific features; This is a bias term used to adjust the result after convolution, providing additional flexibility. For non-linear activation functions, it is defined as: ; Convolution is a convolutional operation. In convolutional neural networks, convolutional operations extract specific features by sliding convolutional kernels (or filters).

[0028] During the decoding phase of U-Net, the current layer feature map The process involves upsampling to restore the size of the segmentation result. Specifically, the upsampling operation performed on the current layer feature map in step S40, to obtain the predicted size of the segmented current layer feature map, is achieved through the following expression: (7) in, To predict the size, the spatial size is greater than the input feature map. Larger, typically close to the size of the original input image; This is the feature map of the current layer; This is an upsampling operation used to restore a low-resolution feature map to the same resolution as the input image; The step S40, which involves obtaining the actual label size and determining the cross-entropy loss based on the predicted size, is achieved through the following expression: (8) in, Cross-entropy loss represents the error between the model's prediction and the actual labels. Cross-entropy is a metric that measures the difference between the true label distribution and the predicted label distribution. The smaller the loss value, the closer the model's prediction is to the true value. For predicted dimensions; This is the actual label size. It is the label in a binary classification task, and its value is 0 or 1; The number of samples is used to standardize the loss so that the loss value is not affected by the number of samples. It is a logarithmic function used to measure the difference between probabilities.

[0029] The determination of the Dice loss function based on the predicted segmented region set and the actual segmented region set in step S60 is achieved by the following expression: (9) in, To predict the set of segmented regions; For a set of real partitioned regions; To predict the number of pixels in the segmented region; The number of pixels in the actual segmented region; The Dice loss function; : The intersection region between the predicted and ground truth segments, representing the number of pixels that overlap between the predicted and ground truth segments.

[0030] The determination of the total loss function based on the cross-entropy loss and the Dice loss function in step S60 is achieved by the following expression: (10) in, This is the total loss function; These are weighting coefficients; Cross-entropy loss; This is the Dice loss function.

[0031] In one embodiment, the calculation of the total length of the fitted curve for the key points in the key point set in step S70 is achieved by the following expression: (11) in, The total length of the fitted curve is calculated by summing the distances between all discrete points; Set of key points One of the key points; Set of key points Another key point; This represents the total number of discrete points on the curve.

[0032] In one embodiment, the curve for determining the cervical spine morphology based on the total length of the fitted curve in step S80 is achieved by the following expression: (12) in, The curve representing the shape of the cervical spine; , , and These are the coefficients for spline interpolation; This is a correction term, associated with the total length of the curve, which modifies the length of the curve. Adding it to the cubic spline interpolation formula ensures that the fitted curve is not only smooth in the local area, but also matches the length of the actual curve as a whole. The independent variable in the interpolation calculation ranges from arrive The coordinates between them.

[0033] In one embodiment, the determination of the curvature of the key point location in the key point set and the correction length of the cervical spine curve based on the curve of the cervical spine morphology in step S80 is achieved by the following expression: (13) (14) in, The curvature is used to adjust the effect of the curvature at each point on length calculation; The derivative of the curve representing the cervical spine shape represents the instantaneous slope or velocity of the curve at a certain point. To correct the length; The slope of the curve represents the minute horizontal increment at each point on the curve. The denominator in curvature calculation contains the square of the curve's slope and is used to standardize the curvature calculation. For local length elements of the curve, based on the curve slope This expression is used to calculate the local length of each point.

[0034] In one embodiment, the overall shape of the cervical spine is quantified by calculating the cervical curvature length (CCL), providing a quantitative basis for assessing curvature abnormalities. The deviation from the standard curve is used to assess the deviation between the individual curve and a standard curve, further determining the degree of cervical curvature abnormality. Deviation from the standard curve: The degree of abnormality is assessed by calculating the deviation between the individual curve and the standard physiological curve. The determination of the degree of cervical curvature abnormality based on the corrected length and the key points in the key point set in step S90 is achieved by the following expression: (15) in, The degree of cervical curvature abnormality is indicated by a combination of deviations from the standard curve. and curve length The total metric. It represents the degree of deviation between the cervical curvature and the standard curve, used to assess the extent of cervical curvature abnormality. (Based on a large sample size collected previously, statistical indicators...) Based on the distribution, establish reference intervals or thresholds; set definition criteria: < Considered within the normal range; ≤ < Mild abnormality; ≥ Obvious abnormalities; these can be used for automated assessments and also to assist doctors in determining the severity of cervical curvature abnormalities. For example... <1%, 1%≤ ≤5%, >5%) Used for standard curve deviation That is, to calculate each point on the curve. Compared with the standard curve The absolute difference between them; Indicates the first A practical point The standard curve on the corresponding x-axis The vertical coordinate value on The absolute difference between them; This represents the total number of discrete points, i.e., the number of points obtained after the curve is sampled or discretized. Points on the curve of the cervical spine; For the first The ordinate value of each point on the standard curve represents the position of the standard curve on the corresponding curve. Theoretical value at the location; The weighting coefficient is used to adjust the total length of the curve. Impact on the final result; To correct the length, it is used to quantify the overall shape of the cervical spine curve.

[0035] Users can take photos of their necks using smart devices and upload them to a cloud or local processing system. The system preprocesses the photos to generate simulated X-ray images; a deep learning model segments the cervical spine region and extracts curvature information; the system generates a report on cervical curvature, health status classification, and risk of curvature abnormalities. This invention uses smart device photography and image analysis technology to determine cervical curvature, overcoming the shortcomings of traditional techniques. Combined with deep learning technology and a radiation-free approach, it assists doctors in diagnosis.

[0036] Figure 3 An internal structural diagram of a computer device in one embodiment is shown. This computer device can specifically be a terminal or a server. Figure 3 As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and may also store a computer program. When executed by the processor, this computer program enables the processor to implement a method for determining cervical curvature based on external imaging. The internal memory may also store a computer program, which, when executed by the processor, enables the processor to implement the method for determining cervical curvature based on external imaging. Those skilled in the art will understand that... Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0037] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments described above. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and RAMbus dynamic RAM (RDRAM), etc.

[0038] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0039] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for determining cervical curvature based on external imaging, characterized in that, The method includes: Acquire the user's original external neck image and preprocess the original external neck image to obtain an equalized image; By equalizing the neighboring pixel values ​​of each pixel in the image in the CNN convolution and combining the weight parameters of each pixel in the CNN convolution, we can determine the single local feature of each pixel at different positions under different convolutions. The previous layer feature map is extracted from a single local feature using U-Net; the previous layer feature map is then segmented to obtain the current layer feature map; An upsampling operation is performed on the current layer feature map to obtain the predicted size of the segmented current layer feature map; Obtain the actual label size and determine the cross-entropy loss based on the predicted size; Obtain the predicted segmentation region set and the actual segmentation region set of the original neck in vitro image, and determine the Dice loss function based on the predicted segmentation region set and the actual segmentation region set; determine the total loss function based on the cross-entropy loss and the Dice loss function; Then, the set of key points in the original neck in vitro image is extracted using the total loss function; the total length of the fitted curve is calculated for the key points in the set of key points. The cervical spine morphology curve is determined based on the total length of the fitted curve; the curvature of the key points in the key point set and the correction length of the cervical spine morphology curve are determined based on the cervical spine morphology curve. The degree of cervical curvature abnormality is determined based on the corrected length and the key points in the key point set.

2. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The preprocessing of the original neck in vitro image to obtain the equalized image includes: The user's original neck external image is denoised using an adaptive median filtering method to obtain the filtered pixel result corresponding to the user's original neck external image; The filtered pixel results are then subjected to grayscale conversion to obtain a grayscale image; The grayscale image is subjected to histogram equalization to obtain an equalized image.

3. The method for determining cervical curvature based on external imaging according to claim 2, characterized in that, The equalized image is obtained through the following expression: in, The result is the filtered pixel count. For the front filtering window The median of the pixel values, which is the value in the middle position after sorting, is used to reduce the impact of salt and pepper noise; For the current filtering window The minimum pixel value in; For the current filtering window The maximum pixel value in; Original in vitro image of the neck; To convert the image to grayscale; For position The pixel value of the red channel; For position The pixel value of the green channel; For position The pixel value of the blue channel; To equalize the image; grayscale image In the middle, the grayscale value is The frequency of pixels; This represents the total number of pixels. The grayscale level is the number of levels.

4. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The method of determining the single local feature of each pixel at different positions under different convolutions by equalizing the neighboring pixel values ​​of each pixel in the CNN convolution and combining the weight parameters of each pixel in the CNN convolution is achieved by the following expression: in, For position grayscale image at [location] The equalized image obtained after histogram equalization; It is a single local feature; For the convolution kernel at position Weight parameters; This is the bias term for the convolution kernel; This is the activation function.

5. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The current layer feature map is obtained by segmenting the previous layer feature map using the following expression: in, This is the feature map of the current layer; This is the feature map of the previous layer; For convolution kernel; For bias terms; It is a non-linear activation function; This is a convolution operation; The predicted size of the segmented current layer feature map is obtained by upsampling the current layer feature map using the following expression: in, For predicted dimensions; This is the feature map of the current layer; This is an upsampling operation; The process of obtaining the actual label size and determining the cross-entropy loss by combining it with the predicted size is achieved through the following expression: in, Cross-entropy loss; For predicted dimensions; This is the actual label size; The number of samples; The determination of the Dice loss function based on the predicted segmented region set and the actual segmented region set is achieved by the following expression: in, To predict the set of segmented regions; For a set of real partitioned regions; To predict the number of pixels in the segmented region; The number of pixels in the actual segmented region; The Dice loss function; The determination of the total loss function based on the cross-entropy loss and the Dice loss function is achieved by the following expression: in, This is the total loss function; These are weighting coefficients; Cross-entropy loss; This is the Dice loss function.

6. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The calculation of the total length of the fitted curve for the key points in the key point set is achieved by the following expression: in, This represents the total length of the fitted curve; Set of key points One of the key points; Set of key points Another key point; This represents the total number of discrete points on the curve.

7. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The curve used to determine the cervical spine morphology based on the total length of the fitted curve is achieved through the following expression: in, The curve representing the shape of the cervical spine; , , and These are the coefficients for spline interpolation; This is a correction item; The independent variable in the interpolation calculation ranges from arrive The coordinates between them.

8. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The curvature of the key points in the key point set and the correction length of the cervical spine curve are determined based on the curve of the cervical spine morphology. This can be achieved using the following expression: in, For curvature; The derivative of the curve representing the cervical spine shape; To correct the length; denoted as the slope of the curve.

9. The method for determining cervical curvature based on external imaging according to claim 1, characterized in that, The determination of the degree of cervical curvature abnormality based on the corrected length and the key points in the key point set is achieved by the following expression: in, The degree of abnormal cervical curvature; The total number of discrete points; Points on the curve of the cervical spine; For the first The ordinate values ​​of each point on the standard curve; The weighting coefficients for adjustment; To correct the length.

Images