A synthetic MRI method for differentiating squamous cell carcinoma of the cervix from adenocarcinoma of the cervix

Abstract

The application relates to the field of synthetic MRI identification technology, in particular to a method for identifying cervical squamous cell carcinoma and cervical adenocarcinoma by using synthetic MRI. The method comprises the following steps: scanning a cervical lesion area, obtaining local gray scale tracks of structural weighted images, lesion signal weighted images and diffusion weighted images, fitting the structural weighted images and the lesion signal weighted images into nonlinear tracks, reconstructing virtual control MRI images, projecting the virtual images to the scanning area, identifying the main direction extension mode and the secondary direction distortion mode of the lesion, constructing diffusion restriction curves in combination with the diffusion weighted images, determining the multi-distribution evolution curves of the lesion, synchronously enhancing different weighted images according to the multi-distribution evolution curves, generating global enhancement MRI data, and identifying the cervical lesion area. The application realizes clearer lesion identification criteria, improves the contrast and resolution of the images, and enhances the sensitivity of the lesion detection.

Description

Technical Field

[0001] This invention relates to the field of synthetic MRI identification technology, and more particularly to a method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma using synthetic MRI. Background Technology

[0002] Traditional imaging methods for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma often rely on a single MRI imaging technique, lacking comprehensive analysis of multiple imaging information. This leads to limitations in lesion identification and classification, especially when differentiating between cervical squamous cell carcinoma and cervical adenocarcinoma. Existing technologies often fail to fully capture the morphology and expansion characteristics of the lesion, resulting in the risk of misdiagnosis and missed diagnosis. Insufficient image contrast and resolution make it difficult to accurately identify subtle lesions, affecting the accuracy and effectiveness of the judgment. Existing technologies also have shortcomings in the fusion analysis of multiple weighted images, lacking real-time monitoring of the dynamic evolution of lesions and failing to effectively integrate information from structurally weighted images, lesion signal-weighted images, and diffusion-weighted images. This results in a lack of comprehensiveness in the lesion identification process, especially in complex lesion cases. Traditional methods fail to reflect the main directional expansion pattern and secondary directional distortion pattern of the lesion in a timely manner, limiting the in-depth understanding and analysis of lesion characteristics. Summary of the Invention

[0003] Therefore, it is necessary to provide a method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma using synthetic MRI, in order to solve at least one of the aforementioned technical problems.

[0004] To achieve the above objective, a method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma using synthetic MRI includes the following steps:

[0005] Step S1: By scanning the cervical lesion area, the local grayscale trajectory of the structure-weighted image, the lesion signal-weighted image and the diffusion-weighted image are obtained simultaneously;

[0006] Step S2: Fit the structure-weighted image and the lesion signal-weighted image into a nonlinear trajectory, and perform morphological inversion through the nonlinear trajectory to reconstruct the virtual control MRI image;

[0007] Step S3: Project the virtual control MRI image onto the scanned cervical region to identify the main direction expansion pattern and secondary direction distortion pattern of the lesion; construct a diffusion restriction curve by combining the local gray-level trajectory of the diffusion-weighted image and the secondary direction distortion pattern, and determine the multi-distribution evolution curve of the lesion through the diffusion restriction curve and the main direction expansion pattern.

[0008] Step S4: Based on the main peak and secondary peak of the multi-distribution evolution curve, the structure-weighted image, lesion signal-weighted image, and diffusion-weighted image are simultaneously enhanced to obtain global enhanced MRI data;

[0009] Step S5: Identify cervical lesion regions based on a pre-defined dataset of known cervical squamous cell carcinoma and cervical adenocarcinoma, as well as global enhanced MRI data.

[0010] The beneficial effects of this invention are as follows: On the one hand, by fitting the nonlinear trajectory of the structure-weighted image and the lesion signal-weighted image and performing morphological inversion, the precise reconstruction of the cervical lesion area is achieved. The generated virtual control MRI image provides a clearer benchmark for lesion identification, greatly improves the contrast and resolution of the image, and effectively enhances the sensitivity of lesion detection. Combined with the local gray-scale trajectory of the diffusion-weighted image, it can comprehensively capture the subtle features of the lesion, thereby improving the accuracy of lesion identification and providing a more reliable data foundation for subsequent image analysis. It innovatively combines multiple image information to form a more comprehensive analytical perspective, improves the depth and breadth of imaging analysis, and promotes the application of MRI images in lesion identification.

[0011] On the other hand, the generation of multi-distribution evolution curves combines the simultaneous enhancement of the main peak and secondary peaks, achieving global enhancement of structure-weighted images, lesion signal-weighted images, and diffusion-weighted images. The enhanced MRI data exhibits higher detail resolution and contrast, fully showcasing the morphological characteristics of the lesion area, reducing the risk of misjudgment due to image noise, optimizing the image quality of cervical lesion areas, enhancing the ability to distinguish different types of lesions, and thus improving overall efficiency. Combined with the pre-set known cervical squamous cell carcinoma and cervical adenocarcinoma datasets, a highly efficient classification mechanism is formed, making the identification of lesion areas more accurate and reliable.

[0012] On the other hand, the identification method based on global enhanced MRI data has significantly improved in terms of processing speed and accuracy. By using advanced image processing algorithms, it can quickly analyze and classify cervical lesion types, reduce the need for manual intervention, enhance the level of automation, reduce the impact of human factors on the results, promote the deep integration of imaging and artificial intelligence technologies, and provide new ideas and methods for future imaging research and applications, thus improving the scientificity and practicality of cervical lesion identification as a whole. Attached Figure Description

[0013] Figure 1 A schematic diagram of the steps involved in a synthetic MRI method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma;

[0014] Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S2.

[0015] Figure 3 This is a schematic diagram of the multi-distribution evolution curve;

[0016] Figure 4 This is a schematic diagram of the cervical lesion area collected.

[0017] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0018] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0019] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods.

[0020] It should be understood that although the terms "first," "second," etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are used merely to distinguish one unit from another. For example, without departing from the scope of the exemplary embodiments, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

[0021] To achieve the above objectives, please refer to Figures 1 to 4 A method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma using synthetic MRI includes the following steps:

[0022] Step S1: By scanning the cervical lesion area, the local grayscale trajectory of the structure-weighted image, the lesion signal-weighted image and the diffusion-weighted image are obtained simultaneously;

[0023] Please see Figure 4 In one embodiment of the present invention, the cervical lesion area is acquired in its entirety using an MRI scanning device. Structural weighted images, lesion signal weighted images, and diffusion weighted images are acquired simultaneously in the same coordinate system. The images are then sampled pixel by pixel to record local grayscale trajectory data.

[0024] In another embodiment of the present invention, a block window method is used when sampling grayscale pixel by pixel, that is, the cervical region is divided into multiple regular grid blocks, and the average grayscale trajectory is collected for each grid block to reduce random noise and improve the stability of the grayscale trajectory.

[0025] Step S2: Fit the structure-weighted image and the lesion signal-weighted image into a nonlinear trajectory, and perform morphological inversion through the nonlinear trajectory to reconstruct the virtual control MRI image;

[0026] In one embodiment of the present invention, nonlinear fitting is performed on the grayscale data of corresponding pixels in the structure-weighted image and the lesion signal-weighted image to obtain a nonlinear trajectory curve, and morphological inversion is performed based on the nonlinear trajectory to reconstruct a virtual control MRI image on the basis of the original data.

[0027] It should be noted that when performing nonlinear fitting, a polynomial fitting algorithm is used in conjunction with a residual threshold to screen the fitting accuracy, and points that deviate too much are removed to ensure that the fitted nonlinear trajectory is smoother and better reflects the morphological change law of the real tissue.

[0028] Step S3: Project the virtual control MRI image onto the scanned cervical region to identify the main direction expansion pattern and secondary direction distortion pattern of the lesion; construct a diffusion restriction curve by combining the local gray-level trajectory of the diffusion-weighted image and the secondary direction distortion pattern, and determine the multi-distribution evolution curve of the lesion through the diffusion restriction curve and the main direction expansion pattern.

[0029] Please see Figure 3 In this embodiment of the invention, the reconstructed virtual control MRI image is projected onto the cervical scanning area through synchronous coordinate indexing, and the main direction expansion mode and secondary direction distortion mode of the lesion are identified according to the gradient field of the projected image. The diffusion restriction curve is constructed by combining the local gray-level trajectory of the diffusion-weighted image and the secondary direction distortion mode, and the multi-distribution evolution curve of the lesion is obtained by the joint analysis of this curve and the main direction expansion mode.

[0030] It should be noted that when identifying the main direction expansion pattern of lesions, the sliding window method is used to extract the local main direction vector, and then the global main direction is synthesized by weighted averaging to avoid the deviation of a single local abnormality from the overall direction identification.

[0031] Step S4: Based on the main peak and secondary peak of the multi-distribution evolution curve, the structure-weighted image, lesion signal-weighted image, and diffusion-weighted image are simultaneously enhanced to obtain global enhanced MRI data;

[0032] In this embodiment, based on the main peak and secondary peak of the multi-distribution evolution curve obtained in the preceding step, the corresponding grayscale intensities of the structure-weighted image, lesion signal-weighted image and diffusion-weighted image are synchronously enhanced to generate global enhanced MRI data containing features of the main direction and secondary direction.

[0033] It should be noted that during the enhancement process, the intensity of the main peak of the curve is increased by linear gain, while the local response of the secondary peak is enhanced by contrast stretching, so as to ensure that the main direction expansion mode is prominent and the secondary direction distortion mode is clear.

[0034] Step S5: Identify cervical lesion regions based on a pre-defined dataset of known cervical squamous cell carcinoma and cervical adenocarcinoma, as well as global enhanced MRI data.

[0035] In this embodiment, a pre-set dataset of known cervical squamous cell carcinoma and cervical adenocarcinoma images is used as a control sample library. Globally enhanced MRI data is input into the classification and analysis module, and the identification of cervical lesion areas is completed through feature comparison and pattern matching.

[0036] It should be noted that, during feature comparison, the global enhanced MRI data is first standardized to ensure consistency with the feature range of the dataset. Then, a similarity-based method is used to compare the differences between the main direction expansion pattern and the secondary direction distortion pattern of the enhanced image and the known samples.

[0037] Preferably, step S2 includes the following steps:

[0038] Step S21: Project the structure-weighted image and the lesion signal-weighted image onto the same voxel coordinate system and record the synchronous coordinate index;

[0039] Step S22: Identify the nonlinear pattern of signal variation with space based on synchronous coordinate index, and identify multiple nonlinear variation points;

[0040] Step S23: Connect all nonlinear change points to fit a nonlinear trajectory; derive the spatial deformation parameters of the cervical lesion region through the nonlinear trajectory;

[0041] Step S24: Based on the spatial deformation parameters, the structural weighted image and the lesion signal weighted image are reverse-restored to their pre-deformation state to reconstruct the virtual control MRI image.

[0042] In one implementation of this invention, the structure-weighted image and the lesion signal-weighted image are projected onto the same voxel coordinate system, and a synchronous coordinate index is recorded at each voxel position. During the projection process, a three-dimensional interpolation method is used to fill in the pixel values ​​of images with different resolutions. Before projection, the images are rigidly registered, that is, the key anatomical structures of the two images are made to overlap through rotation and translation operations, and then voxel projection is performed.

[0043] In this embodiment, the signal value after projection is analyzed based on the synchronous coordinate index to determine the pattern of change with spatial position, detect nonlinear change feature points, and mark these points as nonlinear change points. When detecting nonlinear change points, a sliding window scanning method is used, that is, the voxel data is moved block by block according to a fixed-size cube window, and the first and second derivatives of the signal curve in each window are calculated. When the derivative suddenly exceeds the threshold, the point is marked as a nonlinear change point. Before identifying nonlinear change points, the signal curve is smoothed by using mean filtering or Gaussian filtering methods to make the curve more stable before the change point detection is performed.

[0044] In this embodiment, all nonlinear change points are connected in the order of synchronous coordinate index to fit a complete nonlinear trajectory. Based on this trajectory, spatial deformation parameters of the cervical lesion area are extracted. When connecting the change points, a polynomial fitting method is used, and the fitting curve is calculated by the least squares method. The offset and curvature of the nonlinear trajectory in each segment are calculated. These indicators are used as quantitative parameters of deformation to obtain a set of spatial deformation parameters. The parameter results of different segments are weighted and averaged. The weighting coefficient is determined according to the density of nonlinear change points in the segment. The denser the points, the higher the weight.

[0045] In this embodiment, spatial deformation parameters are used to perform inverse recovery operations on the structure-weighted image and the lesion signal-weighted image to eliminate the offset caused by tissue morphology changes, obtain the image before deformation, and reconstruct the virtual control MRI image. A voxel-by-voxel coordinate backtracking method is adopted, that is, the original position of each voxel is calculated according to the spatial deformation parameters of each voxel, and the gray value is mapped back to the original position. For the uncovered voxels, the gray value is filled in by the neighborhood interpolation method.

[0046] Preferably, step S23 includes:

[0047] Read the nonlinear change points, sort all the nonlinear change points according to the synchronous coordinate index, and connect the nonlinear change points in sequence according to the sorting result to establish an initial trajectory segment set;

[0048] Correcting the continuity and smoothness of the initial trajectory segment set yields a complete nonlinear trajectory;

[0049] Calculate the local curvature of the trajectory based on the complete nonlinear trajectory; identify the trajectory direction vector in the complete nonlinear trajectory;

[0050] The spatial offset characteristics of the trajectory are determined by the local curvature of the trajectory and the trajectory direction vector.

[0051] The spatial offset features of the trajectory are transformed into spatial deformation parameters of the cervical lesion region.

[0052] In one implementation of this invention, nonlinear change points are obtained by calculating the locations of signal intensity abrupt changes. After sorting these points in numerical order using synchronous coordinate indices, several initial trajectory segments are obtained by connecting them sequentially. Then, the breaks in the trajectory segments are corrected by interpolation methods to ensure that the curve is continuous and smooth, thereby obtaining a complete nonlinear trajectory. Finally, the offset features of the trajectory space are extracted by the local curvature and direction changes of the curve, and these offset features are converted into spatial deformation parameters.

[0053] For example, suppose there are 10 nonlinear change points scanned in the cervical lesion area. After being sorted by synchronous coordinate index, they are connected in sequence to form a set of trajectory segments. The initial trajectory breaks between the 3rd and 4th points. The gap is filled by cubic spline interpolation to finally obtain a complete trajectory curve. The curve has the greatest local curvature at the 6th point. The corresponding direction vector offset is X-axis +2, Y-axis -1, and Z-axis +3. After converting the offset into a spatial deformation parameter matrix, it can be used for subsequent virtual control MRI image restoration.

[0054] It should be noted that the reading process can identify jump points that exceed a preset threshold by scanning the grayscale changes row by row in the image matrix. These points are non-linear change points, and their positions are recorded along with their indices.

[0055] In this embodiment, cubic spline interpolation is used to adjust the connection part of the trajectory segment to avoid abrupt angles and smooth the overall trajectory, making its spatial distribution more in line with the actual deformation trend.

[0056] It should be noted that the local curvature is estimated by calculating the difference in tangent slope between adjacent points at each trajectory point, and then the curvature of the region is obtained by combining the average results of multiple points. The greater the curvature, the more significant the deformation.

[0057] In this embodiment, by selecting two adjacent points on the trajectory curve and using the vector between the two points as the local direction, the set of direction vectors for the entire curve is continuously calculated.

[0058] In this embodiment, by setting the components of the offset features, such as the offset in the X-axis direction, the offset in the Y-axis direction, and the offset in the Z-axis direction, these components are combined into a three-dimensional vector, which is used as a spatial deformation parameter for inverse image recovery.

[0059] Of particular importance is the conversion of trajectory spatial offset features into spatial deformation parameters of the cervical lesion region, including:

[0060] Calculate local deformation descriptors based on spatial offset features of the trajectory;

[0061] Based on the spatial offset characteristics of the nonlinear trajectory, the local displacement vector, local bending intensity and direction variation distribution are calculated along the sampling points of each trajectory segment;

[0062] The local displacement vector, local bending strength and direction change are distributed at each corresponding synchronous coordinate index and summarized into a local deformation descriptor dataset;

[0063] Based on spatial topological relationships, local descriptors are fused into neighborhoods and aggregated at scale to extract local strain trends and the ratio of primary to secondary deformation amplitudes.

[0064] The spatial deformation parameters of the cervical lesion area are determined by the ratio of the regional displacement field pattern to the primary and secondary deformation amplitudes.

[0065] In one implementation of this invention, each point in the nonlinear trajectory is extracted sequentially, and the displacement of these points in the three-dimensional coordinate system is calculated. The bending strength is obtained by the angle between the three points, and the directional change angle of adjacent segments is calculated. All these data are summarized into a table to form a local deformation descriptor dataset. Subsequently, these descriptors are merged according to the proximity relationship between trajectory points, and a multi-scale method is used for aggregation to obtain the overall strain trend. The spatial deformation parameters of the cervical lesion are derived based on the statistically obtained ratio of the displacement in the main direction and the secondary direction.

[0066] It should be noted that by importing the three-dimensional coordinate data of the trajectory points into tools such as Excel or Matlab, calculating the coordinate difference between adjacent points, saving the result as a local displacement vector, and repeating this operation for all sampling points.

[0067] In this embodiment, three consecutive trajectory points are selected, the angle between the two preceding and following vector segments is calculated, and the angle is used as the bending strength value of that point. This operation is repeated point by point along the entire trajectory to obtain a complete bending strength sequence.

[0068] In this embodiment, the descriptor of each point is weighted and averaged with the descriptors of adjacent points to eliminate noise at individual points. By repeating the same operation at different scales, such as averaging once at a range of 3 points and once at a range of 5 points, an aggregated result that has both local details and overall trends is obtained.

[0069] In this embodiment, the components of all local displacement vectors in the principal and secondary directions are statistically analyzed, and their average values ​​are taken respectively. Finally, the average value in the principal direction is divided by the average value in the secondary direction to obtain a ratio. This ratio is combined with the overall displacement field pattern of the region to finally determine the spatial deformation parameters of the cervical lesion region.

[0070] For example, assuming the trajectory has 15 sampling points, the three-dimensional coordinates of each point are obtained through MRI data. After calculating the displacement difference between adjacent points, a series of local displacement vectors are obtained. Then, the bending intensity is calculated by the angle between the three points, and the directional change angle of adjacent vectors is recorded. Next, these data are organized into a table in Excel by point number to form a local deformation descriptor dataset. Subsequently, neighborhood weighted averages are performed on these data in the range of 3 points and 5 points respectively to obtain a smoothed strain trend curve. Statistical analysis shows that the average displacement in the main direction is 2 mm, the average displacement in the secondary direction is 0.8 mm, and the ratio of main to secondary is 2.5 to 1. Combined with the displacement field pattern, the spatial deformation parameters of the cervical lesion area are finally determined to be a displacement of 2 mm in the main direction and a displacement of 0.8 mm in the secondary direction.

[0071] Preferably, in step S3, projecting the virtual control MRI image onto the scanned cervical region and identifying the main directional expansion pattern and secondary directional distortion pattern of the lesion includes:

[0072] Based on synchronous coordinate indexing, virtual control MRI images are projected back to the scanned cervical region, and the local structural gradient of the lesion after projection is calculated.

[0073] The principal direction vector of each local window is estimated by using the local structural gradient of the lesion.

[0074] Determine the main direction expansion mode based on the main direction vector;

[0075] Deformation differences are analyzed based on the principal direction vector and nonlinear deformation in the nonlinear trajectory.

[0076] Determine the deformation in the secondary direction by observing the deformation differences;

[0077] Identify secondary distortion clusters based on secondary deformation characteristics;

[0078] Determine the sub-direction distortion mode based on sub-direction distortion clusters.

[0079] In one implementation of this invention, a virtual control MRI image and an actual scan image are loaded into computer software and aligned using a synchronous coordinate index. The lesion region is extracted from the aligned image, the local pixel grayscale difference is calculated to obtain the local structural gradient, the gradient direction distribution is statistically analyzed within each small window region, the principal direction vector is estimated, and the overall principal direction expansion mode is determined accordingly. A nonlinear trajectory is read in the same image, the difference between the principal direction and the trajectory deformation is compared to obtain the deformation difference, the displacement change in the secondary direction is derived from the deformation difference result to form the secondary direction deformation, and similar distortion distribution regions are clustered in the secondary direction results to obtain secondary direction distortion clusters. The secondary direction distortion mode is determined by analyzing the directional consistency of these clusters.

[0080] In this embodiment, the gray-level gradient of the lesion region in the projected image is calculated pixel by pixel using the existing Sobel or Scharr operators in Matlab or Python to obtain the horizontal and vertical gradient components, and then they are synthesized into a local structure gradient map.

[0081] In this embodiment, the gradient directions of all pixels within each local window are averaged to obtain the main direction vector of that window. The results of all windows are then aggregated to form the overall main direction distribution of the lesion.

[0082] It should be noted that by standardizing the displacement magnitude in the secondary deformation case, and then using the K-means clustering method to divide similar distortion points into several classes, each class is a secondary distortion cluster.

[0083] In this embodiment, the directional consistency index of each distortion cluster is calculated, for example, by calculating the average value of the directional vectors within each cluster, and the final sub-directional distortion mode is determined based on the distribution pattern between clusters.

[0084] For example, assuming the cervical region has a 200×200 pixel matrix, the virtual control MRI image and the actual scan image are first aligned in the software. Then, the lesion region is extracted and the local structural gradient is calculated using the Sobel operator. Each 16×16 window region yields an average direction vector, which is then aggregated to obtain the main direction expansion pattern. Subsequently, an average difference of 0.5 mm between the main direction and the actual deformation is detected in the nonlinear trajectory. Based on this, the displacement of the secondary direction is obtained. These displacement points are divided into three clusters, and finally, the overall pattern is determined to be a "multi-cluster secondary direction distortion pattern".

[0085] Of particular importance is the analysis of deformation differences based on the principal direction vector and nonlinear deformation in the nonlinear trajectory, including:

[0086] The direction change rate of the principal direction vector and the nonlinear deformation in the nonlinear trajectory are matched point by point, and the difference between the local curvature of the trajectory in the principal direction and the gray intensity change rate is calculated.

[0087] By combining the differences into tensors, a deformation tensor containing curvature, elongation, and local offset is obtained.

[0088] Decompose the eigenvalues ​​of the deformation tensor and combine them with the nonlinear trajectory to identify the first deformation feature;

[0089] Deformation is simulated using the principal direction vector, and secondary deformation features are identified.

[0090] By comparing the first deformation feature and the second deformation feature, the deformation difference can be obtained.

[0091] In one implementation of this invention, an equidistant sampling point method is used to compare the direction angle of the main direction vector at each point with the tangent direction angle of the nonlinear trajectory, calculate the direction difference as the direction change rate, and combine it with the gray intensity change rate at that point to ensure that the quantifiable curvature and gray intensity change rate difference can be obtained.

[0092] In this embodiment, a three-dimensional tensor matrix is ​​established, with local curvature as the first dimension, elongation as the second dimension, and local offset as the third dimension. The matrix superposition method is used to store the three-dimensional information of each point in a unified manner.

[0093] In this embodiment, the Jacobi iterative method is used to decompose the deformation tensor to obtain principal eigenvalues ​​and secondary eigenvalues. The principal eigenvalues ​​represent the main deformation trend, and the secondary eigenvalues ​​represent the local perturbation trend. Then, the nonlinear offset of the overall trajectory is combined for cross-validation to determine the first deformation feature.

[0094] In this embodiment, an affine transformation matrix is ​​established, and the principal direction vector is introduced into the principal axis direction of the transformation matrix to simulate the local stretching and compression effects of the region under different loads. Then, the deformation features caused by the affine transformation are extracted as the source of the second deformation features.

[0095] For example, when the directional angle of a local trajectory in the scanned cervical region undergoes a 15-degree bending change between consecutive sampling points, and the local grayscale intensity change rate is 0.3, the deformation tensor at this point is decomposed into a principal eigenvalue of 2.5 and a secondary eigenvalue of 0.8 after tensor combination. By comparing it with the second deformation feature obtained by affine transformation simulation, the deformation difference value at this point is found to be 1.7, and it is finally classified as a deformation mode of significant principal direction stretching accompanied by slight secondary direction offset.

[0096] Preferably, in step S3, the diffusion-limited curve is constructed by combining the local gray-level trajectory of the diffusion-weighted image with the secondary distortion pattern, and the multi-distribution evolution curve of the lesion is determined by the diffusion-limited curve and the primary direction expansion pattern, including:

[0097] Extract texture feature vectors from the main direction expansion pattern;

[0098] The main diffusion direction of the lesion is determined based on the texture feature vector and the local gray-level trajectory of the diffusion-weighted image;

[0099] Identify the secondary diffusion direction of lesions in secondary distortion patterns;

[0100] Project the main and secondary diffusion directions of the lesion onto the same coordinate system to determine the adversarial data of the main and secondary diffusion directions of the lesion;

[0101] Construct diffusion-limiting curves using data on the primary and secondary directions of lesion spread;

[0102] Based on the diffusion-limited curve, the evolution pattern of each edge point in the lesion is predicted by the main direction expansion pattern of the lesion;

[0103] Multi-distribution evolution curves are constructed by analyzing the evolution patterns of each edge point in the lesion.

[0104] In one implementation of this invention, a gray-level co-occurrence matrix method is used to calculate parameters such as energy, contrast, and entropy within each directional window, and these parameters are combined into a vector to ensure that the texture features can reflect the directionality and local differences of the lesion tissue.

[0105] In this embodiment, the primary diffusion direction and the secondary diffusion direction are unified into a synchronous coordinate indexing system based on cervical region scanning. Then, the angle relationship and vector difference between the primary diffusion direction and the secondary diffusion direction are calculated to obtain the primary and secondary diffusion direction adversarial data.

[0106] It should be noted that by utilizing the angle and vector differences in the primary and secondary directional adversarial data, the spread restriction of lesions in different directions can be represented as a non-linear curve. The trough of the curve represents the direction with the strongest restriction, and the peak represents the direction with relatively free spread.

[0107] It should be noted that the edge of the lesion is divided into equidistant sampling points. Under the guidance of the diffusion restriction curve, the direction of each sampling point is fitted and the intensity is corrected to obtain the local evolution trend of that point. Then, the trends of all sampling points are merged into a multi-distribution evolution curve of the lesion as a whole.

[0108] For example, suppose the texture vector extracted from the main direction expansion pattern of a lesion indicates that its main direction is horizontal, and the calculation result of the local gray-scale trajectory of the diffusion-weighted image matches it, identifying the main diffusion direction as horizontal extension to the right. At the same time, the secondary direction distortion pattern indicates that there is slight diffusion in the vertical direction. Projecting the two into a unified coordinate system, we obtain adversarial data with an angle close to 90 degrees. The constructed diffusion-limited curve has an "L" shaped trend. The curve shows that horizontal diffusion is free while vertical diffusion is limited. The final predicted edge point evolution pattern shows that the lesion expands rapidly outward at the horizontal boundary while remaining stable at the vertical boundary. The resulting multi-distribution evolution curve accurately describes the diffusion characteristics of the lesion.

[0109] Preferably, the primary and secondary diffusion directions of the lesion are projected onto the same coordinate system, and the data for determining the adversarial relationship between the primary and secondary diffusion directions of the lesion includes:

[0110] By projecting the main diffusion direction and the secondary diffusion direction of the lesion onto the same coordinate system, we can obtain the main diffusion direction and multiple secondary diffusion directions of the lesion in the same plane.

[0111] Calculate the angle between the main diffusion direction of the lesion and the secondary diffusion directions of all lesions in the same plane;

[0112] The interaction parameters of the main diffusion direction of the lesion and the secondary diffusion direction of the lesion corresponding to the included angle are calculated by the angle relationship.

[0113] The interaction parameters are used to determine the primary and secondary spread directions of lesions and the counteracting data.

[0114] In one implementation of this invention, the main diffusion direction data and multiple secondary diffusion direction data of the cervical lesion region are read, and all directional data are uniformly mapped to the same spatial coordinate system so that they are in the same plane for easy subsequent calculation. The angle between the main diffusion direction and each secondary diffusion direction is calculated one by one, and the corresponding interaction parameters are calculated based on the size of the angle and the directional intensity. The strength of the interaction is judged by setting a threshold, thereby obtaining the adversarial data of the main and secondary diffusion directions of the lesion. A unified synchronous coordinate index based on cervical region scanning is established, and all directional data are standardized to this index system to ensure that the main direction and secondary directions are vectorized and compared in the same reference plane.

[0115] In this embodiment, the cosine of the included angle is obtained by the ratio of the vector inner product to the modulus, and then the included angle is obtained by the inverse cosine function, thereby accurately obtaining the included angle value between the main diffusion direction and each secondary diffusion direction.

[0116] In this embodiment, the angle size is combined with the directional strength. For example, when the angle is close to 90 degrees, the interaction is given a higher weight, and when the angle is close to 0 degrees or 180 degrees, a lower weight is given, thereby forming a set of parameter values ​​that can characterize the strength of the primary and secondary diffusion directions.

[0117] It should be noted that when judging adversarial data based on interaction parameters, an adversarial threshold can be set. When the interaction parameters are greater than the threshold, it is judged as a strong adversarial zone, and when the parameters are less than the threshold, it is judged as a weak adversarial zone, thus forming the distribution result of adversarial data in the primary and secondary diffusion directions.

[0118] For example, suppose the primary diffusion direction of a lesion is horizontal, while there are two secondary diffusion directions: one at 45 degrees and the other vertical. After unifying them to the same coordinate system, the angle between the primary direction and the 45-degree secondary direction is calculated to be 45 degrees, and the angle between the primary direction and the vertical direction is 90 degrees. Through interaction parameter calculation, the antagonistic strength of the vertical direction is significantly higher than that of the 45-degree direction. Therefore, the final antagonistic data results show that there is strong antagonism between the horizontal and vertical directions, while there is weak antagonism between the horizontal and the 45-degree direction. Such results can intuitively reflect the primary and secondary directional relationship of the lesion during diffusion.

[0119] Preferably, constructing a diffusion-limiting curve using data on the primary and secondary directions of lesion spread includes:

[0120] The resistance parameters of each secondary diffusion direction to the primary diffusion direction of the lesion are calculated based on the data on the resistance between the primary and secondary diffusion directions of the lesion.

[0121] Identify situations where diffusion is restricted in the main direction by using adversarial parameters;

[0122] Integrate all diffusion-limited cases in the main directions and construct diffusion-limited curves.

[0123] In one implementation of this invention, the data on the primary and secondary diffusion directions of the lesion are read. Each secondary diffusion direction of the lesion is compared with the primary diffusion direction of the lesion in the same coordinate system. The resistance parameter of each secondary diffusion direction to the primary diffusion direction is calculated in turn. The degree of diffusion restriction of the primary direction under the action of the secondary diffusion direction is determined based on the calculated resistance parameter. The restriction of all secondary diffusion directions is summarized and integrated into a continuous diffusion restriction curve by smooth interpolation or curve fitting to represent the restriction distribution of the primary direction of the lesion under the constraints of each direction.

[0124] It should be noted that in the calculation of the adversarial parameters, the magnitude of the included angle and the directional intensity in the adversarial data are combined and weighted to obtain the constraint coefficient of each secondary diffusion direction, thereby quantifying the constraint effect of each direction on the main diffusion direction.

[0125] It should be noted that when judging the degree of diffusion restriction in the main direction, a threshold rule is set. For example, if the adversarial parameter is ≥0.7, the diffusion in this direction is considered severely restricted, indicating that the secondary direction has a strong constraint on the main diffusion direction; if the adversarial parameter is 0.4-0.7, the diffusion in this direction is considered moderately restricted, indicating that the secondary direction has a certain constraint on the main diffusion direction; if the adversarial parameter is <0.4, the diffusion in this direction is considered weakly restricted, indicating that the secondary direction has a very small constraint on the main diffusion direction.

[0126] In this embodiment, the degree of constraint in each direction is mapped to the amplitude value on the curve, and then interpolation is used to smooth and form a continuous curve to more intuitively represent the constraint distribution.

[0127] For example, if the main diffusion direction of the lesion is horizontal to the right, and the secondary diffusion directions include vertical upward, 45-degree upward to the right, and horizontal to the left, the corresponding calculated restriction coefficients are 0.9, 0.6, and 0.8, respectively. Then, a curve is plotted based on each restriction coefficient. The curve has the largest amplitude in the vertical direction, indicating that the main diffusion direction is most severely restricted in the vertical direction, while it is slightly lower in the 45-degree direction, and then in the horizontal to the left direction. This curve can intuitively reflect the distribution of diffusion constraints of the lesion in each direction.

[0128] Preferably, based on the diffusion-limited curve, predicting the evolution pattern of each edge point in the lesion through the main directional expansion pattern of the lesion includes:

[0129] Mark each edge point in the lesion;

[0130] The global extension vector field of the main direction extension mode is projected onto each edge point, and a local coordinate system for the edge point is constructed.

[0131] The principal direction expansion intensity and expansion direction angle of each edge point are calculated based on the local coordinate system of the edge point and the global expansion vector field.

[0132] The diffusion constraint conditions at each edge point are calculated using the diffusion-limited curve.

[0133] Reduce the corresponding main direction expansion intensity according to the diffusion constraints and expansion direction angle of each edge point, and predict the evolution mode of each edge point in the lesion.

[0134] In one implementation of this invention, edge detection is performed on the lesion region, each edge point in the lesion is marked, the global expansion vector field generated by the main direction expansion pattern of the lesion is mapped to the position of each edge point, and a local coordinate system is established around each edge point. In the local coordinate system of the edge point, the main direction expansion intensity and the corresponding expansion direction angle of each edge point are calculated in combination with the global expansion vector field. The diffusion constraint condition of each edge point is calculated according to the constructed diffusion constraint curve, and the main direction expansion intensity of each edge point is reduced in combination with the diffusion constraint condition. The evolution mode of each edge point in the lesion is predicted based on the reduced main direction expansion intensity. For example, if a lesion has an edge point with the main direction vector pointing outward, the calculated expansion intensity is 0.8 and the diffusion constraint is 0.6. Then the reduced expansion intensity is 0.32, and the predicted evolution trend of this edge point is local extension.

[0135] It should be noted that local gray-level gradients can be extracted around edge points using high-resolution MRI images to correct local biases in the global extended vector field, thereby enhancing the accuracy of edge point evolution prediction.

[0136] In this embodiment, the calculation of diffusion constraints can take into account the interaction between edge points and secondary direction distortion modes, and the reduction quantization of edge points can be adjusted by the influence coefficient of secondary direction on the expansion intensity of primary direction.

[0137] It should be noted that by combining MRI scan data from multiple time points, time-series predictions can be made on the evolution patterns of edge points. For example, if three consecutive scans show that the main direction expansion intensity of the same edge point is 0.32, 0.35, and 0.37 respectively, then its future evolution trend is predicted to be continuous outward expansion.

[0138] Preferably, the process of reducing the corresponding main direction expansion intensity according to the diffusion constraint conditions and expansion direction angle of each edge point, and predicting the evolution mode of each edge point in the lesion includes:

[0139] Calculate the angular deviation between the expansion direction angle of each edge point and the main diffusion direction of the lesion;

[0140] The reduction diffusion intensity is calculated based on the angle deviation, where the reduction intensity = original intensity × (1 - constraint coefficient) × (1 - deviation coefficient);

[0141] Based on the diffusion constraint conditions of each edge point and the reduction of the diffusion intensity, the corresponding main direction expansion intensity is reduced to obtain the reduced main direction expansion intensity.

[0142] The position coordinates and expansion state of each edge point at future time points are calculated based on the reduction of the main direction of expansion intensity, thus obtaining the single-point evolution trajectory;

[0143] Predict the evolution pattern of each edge point in the lesion by using a single-point evolution trajectory.

[0144] In one implementation of this invention, a directional statistical matrix is ​​established for each edge point in its local neighborhood to capture the spatial distribution characteristics of the expansion of the main and secondary directions. A local constraint index is generated based on the matrix, and the original expansion intensity of the edge point is adjusted using the constraint index to form a modified expansion intensity field. The modified intensity field is mapped to the future spatial location through time step iteration, thereby obtaining the local evolution trajectory of each edge point. For example, in a certain local edge region, the expansion of points with higher constraint indices weakens, while the expansion of points with lower constraint indices is larger, thus forming an observable evolution pattern.

[0145] In this embodiment, a neighborhood weighted average method is used to smooth the local constraint index, so that the evolution intensity changes between edge points are continuous.

[0146] It should be noted that a dynamic constraint coefficient is introduced, and the correction intensity is adjusted according to the relative position of the edge point and the neighboring distortion mode to reflect the impact of spatial heterogeneity on evolution.

[0147] In this embodiment, the evolution trajectories of each edge point are superimposed to form a local evolution curve, which makes it easier to observe the evolution trend of the edge point group.

[0148] It should be noted that the local evolution trajectory is verified by combining actual MRI scan sequences. For example, by comparing and analyzing the results of continuous scans of the same region, the consistency between the predicted trajectory and the actual expansion can be verified, thereby further adjusting the constraint indicators and correcting the intensity parameters.

[0149] Preferably, the position coordinates and expansion state of each edge point at future time points are calculated based on the reduction of the main direction of expansion intensity, resulting in a single-point evolution trajectory including:

[0150] Based on the reduction of the main direction of expansion intensity, establish the initial position coordinates and the current expansion state for each edge point;

[0151] The future displacement increment of each edge point is calculated based on the reduction of the main direction of expansion intensity and the main diffusion direction vector, and the predicted position coordinates of the edge point at the next time point are updated.

[0152] The updated edge point location coordinates are combined with the corresponding shaving principal direction expansion intensity to determine the expansion state of each edge point;

[0153] The evolution trajectory of a single point is determined by the extended state of each edge point.

[0154] In one implementation of this invention, a record table containing the initial coordinates and current expansion state is generated for each edge point. The displacement increment is calculated using the reduced main direction expansion intensity and the local direction vector where the edge point is located. The displacement increment is superimposed on the initial coordinates to generate the predicted position. At the same time, the expansion state of each point is determined according to the updated coordinates and the corresponding expansion intensity, thereby forming a single-point evolution trajectory. For example, in a dense area of ​​edge points, points with higher intensity will move further along the main diffusion direction, while points with lower intensity will move less, forming an observable local evolution pattern.

[0155] In this embodiment, a neighborhood smoothing operation is added to the predicted position of each edge point, and the movement trends of the surrounding neighboring edge points are weighted and averaged.

[0156] It should be noted that by introducing a dynamic time step, the time step speed of the edge points is adjusted according to the magnitude of the reduction and expansion intensity. This allows for smaller time steps for points that expand faster to improve accuracy, and slightly larger time steps for points that expand slower to improve efficiency.

[0157] It should be noted that a local evolution vector field can be generated at each time step by combining the edge point expansion state, which can be used to visualize the expansion direction and magnitude of the edge point population.

[0158] For example, suppose 100 edge points are selected in a local area of ​​a cervical lesion, with initial coordinates recorded in millimeters. The reduction direction expansion intensity of each point ranges from 0.2 to 0.8. The points are moved along the main diffusion direction, and the predicted positions of each point in the next three frames are calculated. After applying neighborhood smoothing, a continuous and smooth single-point evolution trajectory is obtained.

[0159] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0160] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma using synthetic MRI, characterized in that, Includes the following steps: Step S1: By scanning the cervical lesion area, the local grayscale trajectory of the structure-weighted image, the lesion signal-weighted image and the diffusion-weighted image are obtained simultaneously; Step S2: Fit the structure-weighted image and the lesion signal-weighted image into a nonlinear trajectory, and perform morphological inversion through the nonlinear trajectory to reconstruct the virtual control MRI image; Step S3: Project the virtual control MRI image onto the scanned cervical region to identify the main directional expansion pattern and secondary directional distortion pattern of the lesion; construct a diffusion-limiting curve by combining the local gray-level trajectory of the diffusion-weighted image with the secondary directional distortion pattern, and determine the multi-distribution evolution curve of the lesion through the diffusion-limiting curve and the main directional expansion pattern; wherein step S3, constructing the diffusion-limiting curve by combining the local gray-level trajectory of the diffusion-weighted image with the secondary directional distortion pattern, and determining the multi-distribution evolution curve of the lesion through the diffusion-limiting curve and the main directional expansion pattern, includes: Extract texture feature vectors from the main direction expansion pattern; The main diffusion direction of the lesion is determined based on the texture feature vector and the local gray-level trajectory of the diffusion-weighted image; Identify the secondary diffusion direction of lesions in secondary distortion patterns; Projecting the primary and secondary directions of lesion spread onto the same coordinate system, the opposing data of the primary and secondary lesion spread directions are determined. This determination includes: By projecting the main diffusion direction and the secondary diffusion direction of the lesion onto the same coordinate system, we can obtain the main diffusion direction and multiple secondary diffusion directions of the lesion in the same plane. Calculate the angle between the main diffusion direction of the lesion and the secondary diffusion directions of all lesions in the same plane; The interaction parameters of the main diffusion direction of the lesion and the secondary diffusion direction of the lesion corresponding to the included angle are calculated by the angle relationship. Based on the interaction parameters, determine the primary and secondary directions of lesion spread and the data on the resistance to such spread; Construct diffusion-limiting curves using data on the primary and secondary directions of lesion spread; Based on the diffusion-limited curve, the evolution pattern of each edge point in the lesion is predicted by the main direction expansion pattern of the lesion; Multi-distribution evolution curves were constructed by analyzing the evolution patterns of each edge point in the lesion; Step S4: Based on the main peak and secondary peak of the multi-distribution evolution curve, the structure-weighted image, lesion signal-weighted image, and diffusion-weighted image are simultaneously enhanced to obtain global enhanced MRI data; Step S5: Identify cervical lesion regions based on a pre-defined dataset of known cervical squamous cell carcinoma and cervical adenocarcinoma, as well as global enhanced MRI data.

2. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Project the structure-weighted image and the lesion signal-weighted image onto the same voxel coordinate system and record the synchronous coordinate index; Step S22: Identify the nonlinear pattern of signal variation with space based on synchronous coordinate index, and identify multiple nonlinear variation points; Step S23: Connect all nonlinear change points to fit a nonlinear trajectory; derive the spatial deformation parameters of the cervical lesion region through the nonlinear trajectory; Step S24: Based on the spatial deformation parameters, the structural weighted image and the lesion signal weighted image are reverse-restored to their pre-deformation state to reconstruct the virtual control MRI image.

3. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 2, characterized in that, Step S23 includes: Read the nonlinear change points, sort all the nonlinear change points according to the synchronous coordinate index, and connect the nonlinear change points in sequence according to the sorting result to establish an initial trajectory segment set; Correcting the continuity and smoothness of the initial trajectory segment set yields a complete nonlinear trajectory; Calculate the local curvature of the trajectory based on the complete nonlinear trajectory; identify the trajectory direction vector in the complete nonlinear trajectory; The spatial offset characteristics of the trajectory are determined by the local curvature of the trajectory and the trajectory direction vector. The spatial offset features of the trajectory are transformed into spatial deformation parameters of the cervical lesion region.

4. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 1, characterized in that, In step S3, the virtual control MRI image is projected onto the scanned cervical region to identify the main directional expansion pattern and secondary directional distortion pattern of the lesion, including: Based on synchronous coordinate indexing, virtual control MRI images are projected back to the scanned cervical region, and the local structural gradient of the lesion after projection is calculated. The principal direction vector of each local window is estimated by using the local structural gradient of the lesion. Determine the main direction expansion mode based on the main direction vector; Deformation differences are analyzed based on the principal direction vector and nonlinear deformation in the nonlinear trajectory. Determine the deformation in the secondary direction by observing the deformation differences; Identify secondary distortion clusters based on secondary deformation characteristics; Determine the sub-direction distortion mode based on sub-direction distortion clusters.

5. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 1, characterized in that, Constructing diffusion-limiting curves based on data from the primary and secondary directions of lesion spread includes: The resistance parameters of each secondary diffusion direction to the primary diffusion direction of the lesion are calculated based on the data on the resistance between the primary and secondary diffusion directions of the lesion. Identify situations where diffusion is restricted in the main direction by using adversarial parameters; Integrate all diffusion-limited cases in the main directions and construct diffusion-limited curves.

6. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 1, characterized in that, Based on the diffusion-limited curve, the evolution pattern of each edge point in the lesion is predicted by the main directional expansion pattern of the lesion, including: Mark each edge point in the lesion; The global extension vector field of the main direction extension mode is projected onto each edge point, and a local coordinate system for the edge point is constructed. The principal direction expansion intensity and expansion direction angle of each edge point are calculated based on the local coordinate system of the edge point and the global expansion vector field. The diffusion constraint conditions at each edge point are calculated using the diffusion-limited curve. Reduce the corresponding main direction expansion intensity according to the diffusion constraints and expansion direction angle of each edge point, and predict the evolution mode of each edge point in the lesion.

7. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 6, characterized in that, Based on the diffusion constraints and expansion direction angles of each edge point, the corresponding main direction expansion intensity is reduced, and the evolution pattern of each edge point in the lesion is predicted, including: Calculate the angular deviation between the expansion direction angle of each edge point and the main diffusion direction of the lesion; The reduction diffusion intensity is calculated based on the angle deviation, where the reduction intensity = original intensity × (1 - constraint coefficient) × (1 - deviation coefficient); Based on the diffusion constraint conditions of each edge point and the reduction of the diffusion intensity, the corresponding main direction expansion intensity is reduced to obtain the reduced main direction expansion intensity. The position coordinates and expansion state of each edge point at future time points are calculated based on the reduction of the main direction of expansion intensity, thus obtaining the single-point evolution trajectory; Predict the evolution pattern of each edge point in the lesion by using a single-point evolution trajectory.

8. The method for differentiating cervical squamous cell carcinoma from cervical adenocarcinoma according to claim 7, characterized in that, Based on the calculation of the expansion intensity along the main direction of reduction, the position coordinates and expansion state of each edge point at future time points are obtained, resulting in the single-point evolution trajectory, including: Based on the reduction of the main direction of expansion intensity, establish the initial position coordinates and the current expansion state for each edge point; The future displacement increment of each edge point is calculated based on the reduction of the main direction of expansion intensity and the main diffusion direction vector, and the predicted position coordinates of the edge point at the next time point are updated. The updated edge point location coordinates are combined with the corresponding shaving principal direction expansion intensity to determine the expansion state of each edge point; The evolution trajectory of a single point is determined by the extended state of each edge point.

Images