X-ray machine intelligent inspection scheme creating method based on pet identity recognition

Abstract

The invention discloses an X-ray machine intelligent examination scheme creation method based on pet identity recognition, and relates to the technical field of veterinary diagnostic images.The method has the advantages that variety-specific artifacts are accurately inhibited, modeling is performed through breathing movement driven by skeleton points, high-frequency vibration noise is separated, dynamic blurring of a trachea area of a short-nose dog is eliminated, and the accuracy of the examination scheme is improved. The tiny lesion detection capability is improved; the limitation of a single angle threshold value is avoided by adopting a composite geometric criterion, so that pets with different body types can be clearly imaged under abnormal body positions; full-process identity data binding, biological characteristic penetration scanning protocol formulation, radiation protection and data storage links are provided, real-time association of images and individual medical records is ensured, and manual checking errors are avoided; a block chain evidence storage mechanism is introduced to meet the electronic medical record compliance requirement of a diagnosis and treatment institution, and cross-institution data security sharing is realized; historical parameters are automatically loaded, repeated debugging is reduced, new pets are rapidly initialized through a variety parameter library, and the equipment preparation time is remarkably shortened.

Description

Technical Field

[0001] This invention relates to the field of veterinary diagnostic imaging technology, and in particular to a method for creating an intelligent X-ray examination scheme based on pet identification. Background Technology

[0002] In veterinary clinical imaging examinations, the demand for respiratory diagnosis in brachynose breeds, such as French Bulldogs and Pugs, has increased significantly. These animals often require X-ray evaluation to assess lesions such as tracheal collapse and soft palate hypertrophy due to congenital anatomical abnormalities. During the imaging process, the animals cannot remain still on their own, and respiratory tremors and stress-induced body shaking cause dynamic blurring of the projected images. In particular, image distortion of fine structures such as the trachea may mask early lesion characteristics, affecting the accuracy of clinical decision-making.

[0003] Current mainstream devices employ two types of motion artifact suppression schemes: one is based on marker tracking, such as some new pet DR devices that attach lead points to the animal's body surface and compensate for motion by tracking the displacement of the markers; the other uses global motion estimation, such as video frame interpolation technology based on deep learning, to reconstruct clear images from a continuous low-dose frame sequence. The former is difficult to reliably attach markers in areas covered by pet fur and is prone to causing discomfort and increased movement in the animal; the latter is not sensitive enough to non-periodic local tremors, such as local twitching of the trachea during breathing, and is prone to misjudging high-frequency tremors as noise filtering.

[0004] Some solutions introduce breed-specific movement models, such as predicting the breathing patterns of brachynose dogs through pre-trained neural networks; however, they face two major obstacles in practical applications: first, there are significant differences in breathing among individuals of the same breed, such as the waveforms being different in excited and anesthetized states, resulting in insufficient model generalization ability; second, multiple scans are required to obtain sufficient training data, which violates the clinical low-dose principle; in order to balance clarity and radiation safety, existing technologies often have to sacrifice some dynamic structural resolution, resulting in unclear display of key features such as the tracheal wall edge. Summary of the Invention

[0005] In view of the aforementioned existing problems, the present invention is proposed.

[0006] This invention provides a method for creating an intelligent X-ray examination scheme based on pet identification to solve the problem of blurred images caused by respiratory tremors in brachynose dogs, where existing compensation techniques rely on marker points and cannot adapt to individual differences.

[0007] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0008] This invention provides a method for creating an intelligent X-ray inspection scheme based on pet identification, comprising:

[0009] Step S1: Acquire the pet's biometric data through a biometric acquisition module integrated into the rack. The biometric data includes nose prints and iris images.

[0010] Step S2: Based on the biometric data associated with the pre-stored medical record database, call the pet's historical best scanning parameters, which include tube voltage, frame rate, and region of interest;

[0011] Step S3: Perform a low-dose preview scan to generate a positioning image. Based on the positioning image, extract key skeletal points through a pose analysis network. The key skeletal points include spinal curvature and joint angles.

[0012] Step S4, dynamically adjust the scanning protocol according to the key skeletal points, including:

[0013] The gantry tilt angle is compensated based on the deviation of spinal curvature.

[0014] Based on joint angle adaptation X-ray energy spectrum;

[0015] Step S5: Perform the formal scan using the adjusted scanning protocol, and decompose motion artifacts in real time during the scan, including:

[0016] Separate respiratory-related motion vectors from tremor noise vectors in a continuous frame sequence.

[0017] Perform inverse compensation on the jitter noise vector to generate a corrected frame;

[0018] Step S6: Bind and store the corrected frame with the pet's biometric data.

[0019] As a preferred embodiment of the intelligent X-ray inspection scheme creation method based on pet identification described in this invention, step S2, which calls the historical optimal scanning parameters, includes:

[0020] If no historical data is found, default parameters are loaded from a preset parameter library based on the pet breed. The preset parameter library contains a mapping relationship between dog and cat breeds and scanning parameters.

[0021] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, step S4, which involves adapting the X-ray energy spectrum based on joint angles, includes:

[0022] When excessive cervical flexion is detected in brachynose dogs, the tube voltage is increased and the frame rate is reduced.

[0023] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, in step S4, the excessive buckling determination method is as follows:

[0024] The coordinates of four key skeleton points P0, P1, P2, and P3 were obtained from the pose analysis network. They are the cranial vertex, the spinous process of the cervical vertebra, the center of the shoulder joint, and the spinous process of the first thoracic vertebra, respectively. The units are all mm.

[0025] Calculate the cervical flexion angle:

[0026]

[0027] Where γ represents the cervical flexion angle in degrees, a = P0 - P1 is the cervical vertebrae vector at the cranial verge in mm, b = P2 - P1 is the cervical vertebrae vector at the shoulder joint in mm, and a·b is the dot product of the two vectors in mm. 2 ||a|| and ||b|| are the Euclidean norms of vectors a and b, respectively, in mm;

[0028] Calculate the distance between bone points:

[0029] d ij =||P i -P j ||,

[0030] Where, d ij Represents skeletal point P i With P j The Euclidean distance, in mm, P i ,P j These are the coordinates of the skeletal points;

[0031] Calculate the local radius of curvature:

[0032]

[0033] Where Δ is the area of ​​the triangle formed by P1, P2, and P3, in mm. 2 d 12 ,d 23 ,d 31 Let R be the distances from P1 to P2, P2 to P3, and P3 to P1, respectively, in mm. c The radius of curvature of the cervical segment is in mm;

[0034] Calculate adaptive weights:

[0035]

[0036] w1 = 0.5 + 0.3σ(B) c -0.4), w2 = 1 - w1,

[0037] Where σ(x) is a logic function, dimensionless, and B c, where w1 and w2 are the dimensionless morphological coefficients of the breed, pre-stored in the breed database and retrieved through the pet breed code index; w1 and w2 are the adaptive weights of the angle and curvature terms, respectively, and are dimensionless.

[0038] The composite judgment index is defined as follows:

[0039]

[0040] Among them, F flex The excessive buckling criterion index is dimensionless, γ. min R represents the lower limit of the buckling angle for this variety, in degrees. th This is the lower limit of the radius of curvature for this variety, in mm;

[0041] If F flex >1 indicates excessive cervical flexion, triggering voltage increase and frame rate reduction operations.

[0042] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, wherein: the posture analysis network output in step S3 contains a topology map with at least 15 skeletal points, the skeletal points being selected from the cranial vertex, cervical spinous processes, shoulder joints, and hip joints.

[0043] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, step S5, motion artifact decomposition, includes:

[0044] The global motion vector is calculated using an optical flow algorithm.

[0045] Respiratory-related motion vectors are extracted based on the displacement of the key skeletal points.

[0046] The difference between the global motion vector and the breathing-related motion vector is defined as the tremor noise vector.

[0047] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, wherein: in step S5, in consecutive frames F k ,F k+1 A three-stage decomposition is performed to obtain the jitter noise vector field V. n (x), where x∈Ω are pixel coordinates in mm, Ω is the imaging field of view coordinate domain, and the acquisition time interval δt is in seconds. The steps for defining the jitter noise vector include:

[0048] Solve for the global motion vector field:

[0049]

[0050] Among them, V g (x)=(ug ,v g ) represents the global optical flow vector, in mm, λ g These are the smoothness regularity coefficients, dimensionless. The gradient of the vector field is in mm. -1 ;

[0051] Construct the respiration-related displacement field:

[0052] a. Define the displacement of key skeleton points:

[0053]

[0054] Where, d i =(Δx) i ,Δy i ) represents the skeletal point Q. i The inter-frame displacement, in mm. and N represents the coordinates of the i-th bone point in frame (k+1) and frame (k), respectively. s =15 represents the number of skeleton points output by the pose analysis network, which is dimensionless;

[0055] b. Perform thin-plate spline interpolation:

[0056]

[0057] Among them, V r (x)=(u r ,v r α is the respiratory displacement vector, in mm. i =(α ix ,α iy () represents the radial basis function coefficients, in mm. For thin plate spline basis functions, mm 2 , A∈R 2×2 Let b be an affine matrix, dimensionless, and b∈R. 2 Let mm be the translation vector;

[0058] c. Solve for the coefficients:

[0059]

[0060] in, The unit is mm. 2 , The unit is mm, d q This is a column vector of bone displacements, in mm.

[0061] Extracting high-frequency weighting functions:

[0062]

[0063] Among them, W h (x) is a high-frequency weighting function, dimensionless. For frame difference gradient, HU / s, β h S is the frequency domain sensitization coefficient, dimensionless. max The maximum frame difference gradient in the region is HU / s;

[0064] Define the jitter noise vector:

[0065] V n (x)=W h (x)[V g (x)-V r (x)],

[0066] Among them, V n (x)=(u n ,v n ) represents the jitter noise vector, in mm.

[0067] When ||V n (x)||>ε n At that time, for V n Take the inverse compensation to generate the correction frame, ε n This represents the flutter noise amplitude threshold, in mm.

[0068] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, it further includes:

[0069] Before the formal scan, radiation shielding of the thyroid projection area is activated based on the pet's biometric data.

[0070] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, the biometric acquisition in step S1 includes:

[0071] The near-infrared light source is triggered when the rack rotates to the ready position.

[0072] Hair occlusion segmentation is performed on the original biological image.

[0073] When the proportion of effective feature regions after segmentation is lower than the threshold, multi-angle supplementary sampling is initiated.

[0074] As a preferred embodiment of the method for creating an intelligent X-ray inspection scheme based on pet identification as described in this invention, the binding storage in step S6 includes:

[0075] The corrected frame, biometric hash value, and dynamically adjusted parameters are encapsulated into a data block with a timestamped signature chain.

[0076] The beneficial effects of this invention are as follows: It accurately suppresses breed-specific artifacts, separates high-frequency vibration noise through skeletal point-driven respiratory motion modeling, eliminates dynamic blurring in the tracheal region of short-nosed dogs, and improves the detection capability of subtle lesions; it employs composite geometric criteria to avoid the limitations of single-angle thresholds, enabling clear imaging of pets of different sizes even in abnormal positions; it provides end-to-end identity data binding, with biometrics integrated into the scanning protocol formulation, radiation protection, and data storage stages, ensuring real-time association between images and individual medical records and avoiding errors from manual verification; it introduces a blockchain evidence storage mechanism to meet the compliance requirements of electronic medical records in medical institutions, achieving secure data sharing across institutions; automatic loading of historical parameters reduces redundant debugging, and new pets are quickly initialized through the breed parameter library, significantly shortening equipment preparation time; intelligent shielding for sensitive organs such as the thyroid reduces radiation risk, balancing image quality and pet safety; thin-plate spline interpolation integrates prior anatomical knowledge, overcoming the failure problem of traditional optical flow methods in areas with missing textures; hair occlusion segmentation and multi-angle supplementary sampling ensure the success rate of biometric identification, adapting to the complex clinical operating environment of pets. Attached Figure Description

[0077] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. 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.

[0078] Figure 1 This is a flowchart illustrating the method for creating an intelligent X-ray inspection scheme based on pet identification in Example 1. Detailed Implementation

[0079] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0080] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0081] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0082] Example 1, referring to Figure 1 This embodiment provides a method for creating an intelligent X-ray inspection scheme based on pet identification, including the following steps:

[0083] Step S1: Acquire the pet's biometric data through a biometric acquisition module integrated into the rack. The biometric data includes nose print and iris images.

[0084] Step S1, biometric data collection, includes:

[0085] The near-infrared light source is triggered when the rack rotates to the ready position.

[0086] Hair occlusion segmentation is performed on the original biological image.

[0087] When the proportion of effective feature regions after segmentation is lower than the threshold, multi-angle supplementary sampling is initiated.

[0088] Step S2: Based on the biometric data and the pre-stored medical record database, call the pet's historical best scanning parameters, which include tube voltage, frame rate and region of interest.

[0089] Step S2, which calls the historical best scan parameters, includes:

[0090] If no historical data is found, default parameters are loaded from the preset parameter library based on the pet breed. The preset parameter library contains the mapping relationship between dog and cat breeds and scanning parameters.

[0091] Step S3: Perform a low-dose preview scan to generate a positioning image. Based on the positioning image, extract key skeletal points through a pose analysis network. Key skeletal points include spinal curvature and joint angles.

[0092] In step S3, the pose analysis network outputs a topology map containing at least 15 skeletal points, which are selected from the cranial vertex, cervical spinous processes, shoulder joints, and hip joints.

[0093] Step S4, dynamically adjust the scanning protocol based on key skeletal points, including:

[0094] The gantry tilt angle is compensated based on the deviation of spinal curvature.

[0095] Based on joint angle adaptation X-ray energy spectrum;

[0096] Step S4, based on the joint angle adaptation X-ray energy spectrum, includes:

[0097] When excessive cervical flexion is detected in brachynose dogs, the tube voltage is increased and the frame rate is decreased;

[0098] In step S4, the method for determining excessive buckling is as follows:

[0099] The coordinates of four key skeleton points P0, P1, P2, and P3 were obtained from the pose analysis network. They are the cranial vertex, the spinous process of the cervical vertebra, the center of the shoulder joint, and the spinous process of the first thoracic vertebra, respectively. The units are all mm.

[0100] Calculate the cervical flexion angle:

[0101]

[0102] Where γ represents the cervical flexion angle in degrees, a = P0 - P1 is the cervical vertebrae vector at the cranial verge in mm, b = P2 - P1 is the cervical vertebrae vector at the shoulder joint in mm, and a·b is the dot product of the two vectors in mm. 2 ||a|| and ||b|| are the Euclidean norms of vectors a and b, respectively, in mm;

[0103] Calculate the distance between bone points:

[0104] d ij =||P i -P j ||,

[0105] Where, d ij Represents skeletal point P i With P j The Euclidean distance, in mm, P i ,P j These are the coordinates of the skeletal points;

[0106] Calculate the local radius of curvature:

[0107]

[0108] Where Δ is the area of ​​the triangle formed by P1, P2, and P3, in mm. 2 d 12 ,d 23 ,d 31 Let R be the distances from P1 to P2, P2 to P3, and P3 to P1, respectively, in mm. c The radius of curvature of the cervical segment is in mm;

[0109] Calculate adaptive weights:

[0110]

[0111] w1 = 0.5 + 0.3σ(B) c -0.4), w2 = 1 - w1,

[0112] Where σ(x) is a logic function, dimensionless, and R c , where w1 and w2 are the morphological coefficients of the variety, dimensionless, and the adaptive weights of the angle and curvature terms, dimensionless.

[0113] The composite judgment index is defined as follows:

[0114]

[0115] Among them, F flex The excessive buckling criterion index is dimensionless, γ. min R represents the lower limit of the buckling angle for this variety, in degrees. th This is the lower limit of the radius of curvature for this variety, in mm;

[0116] If F flex If the value is >1, it is determined to be excessive cervical flexion. In subsequent scanning protocols, the tube voltage is increased and the frame rate is decreased.

[0117] Specifically, this step utilizes both buckling angle and radius of curvature in a single step to avoid the incompatibility of a single threshold with pets of different body sizes; the weights are self-adjusted by the breed morphology through a logical function, making special body types such as brachynose dogs more sensitive to abnormal neck postures; the circumcircle of a triangle method can calculate the radius of curvature within a single frame, reducing the dependence on inter-frame smoothing; the combined index uses normalized difference accumulation, and the threshold of 1 has an intuitive interpretation, consistent with the subsequent protocol call interface;

[0118] Step S5: Perform the formal scan using the adjusted scanning protocol, and decompose motion artifacts in real time during the scan, including:

[0119] Separate respiratory-related motion vectors from tremor noise vectors in a continuous frame sequence.

[0120] Perform inverse compensation on the jitter noise vector to generate a corrected frame;

[0121] Step S5, motion artifact decomposition, includes:

[0122] The global motion vector is calculated using an optical flow algorithm.

[0123] Respiratory-related motion vectors are extracted based on the displacement of key skeletal points.

[0124] The difference between the global motion vector and the breathing-related motion vector is defined as the tremor noise vector;

[0125] In step S5, in consecutive frames F k ,F k+1 A three-stage decomposition is performed to obtain the jitter noise vector field V. n (x), where x∈Ω are pixel coordinates in mm, Ω is the imaging field of view coordinate domain, and the acquisition time interval δt is in seconds. The steps for defining the jitter noise vector include:

[0126] Solve for the global motion vector field:

[0127]

[0128] Among them, V g (x)=(u g ,v g ) represents the global optical flow vector, in mm, λ g These are the smoothness regularity coefficients, dimensionless. The gradient of the vector field is in mm. -1 ;

[0129] Construct the respiration-related displacement field:

[0130] a. Define the displacement of key skeleton points:

[0131]

[0132] Where, d i =(Δx) i ,Δy i ) represents the skeletal point Q. i The inter-frame displacement, in mm. and N represents the coordinates of the i-th bone point in frame (k+1) and frame (k), respectively. s =15 represents the number of skeleton points output by the pose analysis network, which is dimensionless;

[0133] b. Perform thin-plate spline interpolation:

[0134]

[0135] Among them, V r (x)=(u r ,v r α is the respiratory displacement vector, in mm. i =(α ix ,α iy () represents the radial basis function coefficients, in mm. For thin plate spline basis functions, mm 2 , A∈R 2×2 Let b be an affine matrix, dimensionless, and b∈R. 2 Let mm be the translation vector;

[0136] c. Solve for the coefficients:

[0137]

[0138] in, The unit is mm. 2 , The unit is mm, d q This is a column vector of bone displacements, in mm.

[0139] Extracting high-frequency weighting functions:

[0140]

[0141] Among them, W h (x) is a high-frequency weighting function, dimensionless. For frame difference gradient, HU / s, β h S is the frequency domain sensitization coefficient, dimensionless. max The maximum frame difference gradient in the region is HU / s;

[0142] Define the jitter noise vector:

[0143] V n (x)=W h (x)[V g (x)-V r (x)],

[0144] Among them, V n (x)=(u n ,v n ) represents the jitter noise vector, in mm.

[0145] When ||V n (x)||>ε n At that time, for V n Take the inverse compensation to generate the correction frame, ε n This is the threshold value for jitter noise amplitude, in mm;

[0146] Specifically, this decomposition process first obtains the global optical flow using an energy model, then expands the skeletal point displacements into a continuous breathing field using thin-plate splines. This breathing field aggregates in the low-frequency domain but is insensitive to local high frequencies, making it easily distinguishable from random jitter. The high-frequency weighting function is adaptively normalized according to the frame difference gradient, avoiding threshold adjustments under different exposure conditions. The residual is then processed by W... h After modulation, it is defined as a jitter noise vector, which can locate high-amplitude jitter regions at the pixel level;

[0147] Step S6: Bind and store the corrected frame with the pet's biometric data;

[0148] The binding storage in step S6 includes:

[0149] The correction frame, biometric hash value, and dynamic adjustment parameters are encapsulated into a data block with a timestamped signature chain;

[0150] The method in this embodiment further includes:

[0151] Before the formal scan, radiation shielding of the thyroid projection area was activated based on the pet's biometric data.

[0152] The aforementioned fixed coefficients (such as 0.5, 0.3, etc.) are empirically optimized values ​​obtained after parameter sensitivity optimization of large sample pet image data. They can be adjusted linearly or nonlinearly as needed under the technical concept of this invention, and are not limited to the specific values ​​listed.

[0153] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for creating an intelligent inspection scheme of an X-ray machine based on pet identification, characterized in that, The method comprises the following steps: Step S1, acquiring biological feature data of the pet through a biological feature acquisition module integrated in the gantry, wherein the biological feature data comprises nose print and iris image; Step S2, associating a pre-stored medical record database based on the biological feature data, and calling historical optimal scanning parameters of the pet, wherein the historical optimal scanning parameters comprise tube voltage, frame rate and region of interest; Step S3, performing a low-dose preview scan to generate a positioning image, and extracting key skeletal points including spinal curvature and joint angle through a posture analysis network based on the positioning image; Step S4, dynamically adjusting a scanning protocol according to the key skeletal points, comprising: compensating the gantry tilt angle based on the spinal curvature deviation amount, adapting the ray spectrum based on the joint angle; Step S5, performing a formal scan by using the adjusted scanning protocol, and decomposing motion artifacts in real time during the scanning process, comprising: separating a breathing-related motion vector and a tremor noise vector from a continuous frame sequence, performing reverse compensation on the tremor noise vector to generate a corrected frame; Step S6, binding and storing the corrected frame and the biological feature data of the pet.

2. The pet identification-based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, The calling of the historical optimal scanning parameters in step S2 comprises: if no historical data is matched, loading default parameters from a pre-set parameter library according to the pet breed, wherein the pre-set parameter library contains a mapping relationship between dog and cat breeds and scanning parameters.

3. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, The adapting of the ray spectrum based on the joint angle in step S4 comprises: when the cervical vertebra of a brachycephalic dog is identified to be excessively flexed, the tube voltage is increased and the frame rate is decreased.

4. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 3, wherein, In step S4, the excessive flexion determination method is: four key skeletal point coordinates P0, P1, P2, P3 are obtained from the posture analysis network, which are in turn the top of the skull, the cervical spine process, the shoulder joint center and the first thoracic spine process, and the unit is mm; the cervical flexion angle is calculated: wherein γ represents the flexion angle of the cervical segment, in °, a = P0-P1 is the cranial top cervical vector, in mm, b = P2-P1 is the shoulder joint cervical vector, in mm, a.b is the dot product of the two vectors, in mm 2 ||a||, ||b|| are the Euclidean norms of the vectors a, b, respectively, in mm; the distance between skeletal points is calculated: d ij =||P i -P j ||, where d ij represents the Euclidean distance between the skeletal point P i and P j , in mm, P i , P j are skeletal point coordinates; the local curvature radius is calculated: Wherein, Δ is the area of the triangle formed by P1, P2, P3, unit: mm 2 , d 12 , d 23 , d 31 are the distances of P1-P2, P2-P3, P3-P1 respectively, mm, R c is the local radius of curvature of the neck, mm; the adaptive weight is calculated: w1 = 0.5 + 0.3σ(B c -0.4), w2 = 1 - w1, wherein σ(x) is a logistic function, dimensionless, B c is a breed shape coefficient, dimensionless, pre-stored in a breed database and obtained through a pet breed code index, and w1 and w2 are adaptive weights of the angle term and the curvature term, dimensionless. the composite determination index is defined as: wherein F flex is the hyperflexion index, dimensionless, γ min is the lower limit of the flexion angle for the breed, in °, R th is the lower limit of the radius of curvature for the breed, in mm; If F flex >1, it is determined that the cervical vertebrae are excessively flexed, and a frame rate reduction operation is triggered.

5. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, In step S3, the posture analysis network outputs a topology graph containing at least 15 skeletal points selected from the top of the skull, the cervical spine process, the shoulder joint and the hip joint.

6. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, In step S5, the motion artifact decomposition comprises: calculating a global motion vector by an optical flow algorithm, extracting a breathing-related motion vector based on the displacement of the key skeletal points, defining the difference between the global motion vector and the breathing-related motion vector as a tremor noise vector.

7. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 6, wherein, In step S5, three-stage decomposition is performed between consecutive frames F k ,F k+1 to obtain a tremor noise vector field V n (x), where x ∈ Ω is the pixel coordinate with unit mm, Ω is the coordinate domain of the imaging field of view, the time interval δt is the unit s, and the definition step of the tremor noise vector includes: solving the global motion vector field: where V g (x) = (u g , v g ) is the global optical flow vector in mm, λ g is the smoothing regularization coefficient, dimensionless, is the gradient of the vector field in mm -1 ; constructing a breathing-related displacement field: a. defining the displacement of key skeletal points: where d i = (Δx i , Δy i ) is the inter-frame displacement of the skeletal point Q i , in mm, and are the coordinates of the i-th skeletal point in the k+1-th frame and the k-th frame, respectively, N s = 15 is the number of skeletal points output by the pose analysis network, dimensionless. b. performing thin-plate spline interpolation: where V r (x) = (u r ,v r ) is the respiratory displacement vector in mm, a i = (a ix ,a iy ) are the radial basis function coefficients in mm, is the thin plate spline basis function in mm 2 , A e R 2×2 is the affine matrix, dimensionless, b e R 2 is the translation vector in mm; c. solving the coefficients: wherein, in mm 2 , in mm, d q is the bone displacement column vector in mm, d a = [b T A 11 A 12 A 21 A 22 ] T ; extracting a high-frequency weight function: where W h (x) is a high frequency weight function, dimensionless, is the frame difference gradient, HU / s, β h is a frequency domain sensitivity coefficient, dimensionless, S max is the regional maximum frame difference gradient, HU / s; defining the tremor noise vector: V n (x) = W h (x) [V g (x) - V r (x)], where V n (x) = (u n ,v n ) is the jitter noise vector in mm, When ||V n (x)||>ε n At that time, for V n Take the inverse compensation to generate the correction frame, ε n This represents the flutter noise amplitude threshold, in mm.

8. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, Further comprising: before the formal scan, activating the radiation shielding of the thyroid projection region based on the biological feature data of the pet.

9. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, In step S1, the biological feature acquisition comprises: triggering a near-infrared light source when the gantry rotates to a standby position, performing hair occluder segmentation on the original biological image, starting multi-angle supplementary acquisition when the proportion of the segmented effective feature region is lower than a threshold.

10. The pet identification based X-ray machine intelligent inspection scheme creation method of claim 1, wherein, The binding storage in step S6 comprises: encapsulating the corrected frame, the biological feature hash value and the dynamic adjustment parameter into a data block of a time-stamped signature chain.

Images