AI-based large-scale medical equipment performance evaluation system and method

Abstract

The invention discloses a performance evaluation system and method for large medical equipment based on AI, and relates to the technical field of artificial intelligence, and the method comprises the steps: building a neural network model of a defect type, carrying out the training, calling a historical reshoot record of image medical equipment, and extracting a target record according to the neural network model; extracting the image shooting time of the defect image, and calling a monitoring video; the pixel gray level of the shot focus part is obtained, and the first stability degree of the examined patient is calculated; obtaining an edge contour in each frame of picture, and calculating second stability; obtaining a clustering center, judging a movement condition, and calculating third stability; and calculating the total stability degree, and intelligently studying and judging the performance evaluation index of the medical imaging equipment. According to the method, the historical reshoot records are analyzed, the fault condition of the medical imaging equipment is judged and used for performance evaluation indexes, the performance evaluation of the medical equipment is more accurate and objective, data distortion is avoided, and the accuracy of performance evaluation is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, specifically to a performance evaluation system and method for large-scale medical equipment based on AI. Background Technology

[0002] As the proportion of large medical equipment in hospital fixed assets continues to rise, quantitative evaluation of the utilization of large medical equipment and the establishment of a sound performance evaluation system have a profound impact on equipment procurement, daily use and economic benefits. They are directly related to the improvement of hospital diagnosis and treatment levels and the control of operating costs. Through scientific and efficient performance evaluation, hospital managers can be provided with accurate decision-making basis, which can help hospitals achieve high-quality development and enhance their overall strength.

[0003] For large-scale medical imaging equipment, image quality is a key factor affecting equipment performance. When the image quality is poor, re-enhancing the image is necessary, which lowers the performance of the large-scale medical imaging equipment, leading to longer waiting times for subsequent patients and even affecting the scheduling of emergency examinations for emergency patients. However, poor patient cooperation can also lead to image re-enhancing. Currently, the determination of the cause of image re-enhancing is based on human judgment and experience, which is subjective and inefficient. If the cause of image re-enhancing can be determined intelligently and quickly, the performance evaluation of medical equipment can be more accurate and objective, avoiding data distortion, improving the accuracy and authenticity of performance evaluation, and effectively enhancing the scientific nature and application value of the evaluation. Summary of the Invention

[0004] The purpose of this invention is to provide an AI-based performance evaluation system and method for large-scale medical equipment to address the problems raised in the prior art.

[0005] To solve the above-mentioned technical problems, the present invention provides the following technical solution: The performance evaluation method for large-scale medical equipment based on AI includes the following steps: Establish a neural network model corresponding to each defect type, acquire defect images of different defect levels for each defect type, and train the neural network model; retrieve historical re-enhancing records from imaging medical equipment, and extract the examinee, defect type, defect image, and qualified image from the re-enhancing records; based on the trained neural network model and the pixel saturation values ​​in the defect and qualified images, determine and extract the target record from the re-enhancing records. Extract the image capture time of the defect image corresponding to the target record, retrieve the monitoring video of the image capture time, and calculate the first stability of the examined patient based on the pixel grayscale changes of the lesion site in multiple consecutive frames. Using an edge detection algorithm, all edge contours in each frame are obtained. Based on the changes in the number of edge contours and the number of pixels within the edge contours, the second stability of the examined patient is calculated. A planar coordinate system is established, and a clustering algorithm is used to obtain several cluster centers in the defect image. Based on the movement of the cluster centers in the monitoring video, the movement vector corresponding to each cluster center is obtained, and then the third stability of the examined patient is calculated. The system presets the weight of each stability level, calculates the overall stability of the examined patient, determines the fault status of the imaging medical equipment based on the overall stability, and intelligently evaluates the performance indicators of the imaging medical equipment.

[0006] Preferably, the target record in the re-enactment record is identified and extracted, including: The defect type refers to the specific reason for re-taking the defective image selected by relevant personnel. However, the selected defect type may be incorrect, leading to misjudgment. The purpose of this step is to screen these misjudged records so that the defective images in the target records actually have defects. The advantage of doing this is that it makes the performance evaluation of medical equipment more accurate and objective, and the subsequent calculation results more reliable. The specific screening process is as follows: Establish a neural network model corresponding to each defect type. There are several defect types. Extract multiple defect images belonging to a certain defect type with different defect levels. The defect level value ranges from 0 to 1. Substitute the defect images into the neural network model of a certain defect type for training. The trained neural network model is used to determine the defect level of the input image in a certain defect type. Extract the defect type Q, defect image Img1, and qualified image Img2 from a certain re-enactment record. Input the defect region C1 in the defect image Img1 into the neural network model corresponding to each defect type to obtain the defect level corresponding to each defect type. Obtain the qualified region C2 corresponding to the defect region C1 in the qualified image Img2. Input the qualified region C2 into the neural network model corresponding to each defect type to obtain the defect level corresponding to each defect type. Obtain the saturation value of each pixel in the defective region C1 and the qualified region C2, obtain several saturation intervals, classify the saturation values ​​into the corresponding saturation intervals, and establish saturation histograms corresponding to the defective region C1 and the qualified region C2 respectively. Calculate the histogram similarity between the two saturation histograms. If the defective region C1 has the highest defect level in defect type Q, and the qualified region C2 has a defect level less than a preset level threshold in every defect type, and the histogram similarity is less than a preset similarity threshold, then a retake record is taken as the target record.

[0007] Preferably, the calculation of the patient's first degree of stability includes: Extract the examination patient Per and defective image Img corresponding to a certain target record, as well as the image capture time T1 of the defective image Img. Retrieve the monitoring segments with a duration of D before and after time T1. The monitoring segments are videos recorded on the lesion site of the examination patient Per. Extract any two consecutive frames of monitoring screens F1 and F2 from the monitoring segments, obtain the defective region in the defective image Img, and mark the corresponding defective regions R1 and R2 in monitoring screens F1 and F2 respectively. Obtain the grayscale value of each pixel in the defect regions R1 and R2, obtain several grayscale intervals, classify the grayscale values ​​into the corresponding grayscale intervals, establish grayscale histograms H1 and H2 corresponding to the defect regions R1 and R2 respectively, and calculate the histogram similarity between grayscale histograms H1 and H2; then, based on multiple consecutive two-frame monitoring images, obtain multiple histogram similarities, calculate the average value to obtain the first stability degree of the examined patient Per.

[0008] Preferably, the calculation of the patient's second degree of stability includes: Using an edge detection algorithm, all edge contours in monitoring images F1 and F2 are obtained. Binary segmentation is then used to divide the edge contours. The number of common edge contours in monitoring images F1 and F2 is designated as M, the number of unique edge contours in monitoring image F1 is designated as A, and the number of unique edge contours in monitoring image F2 is designated as B. Based on the number of edge contours in monitoring images F1 and F2, and the number of pixels contained within each edge contour, the similarity between monitoring images F1 and F2 is calculated as follows: Where S is the similarity between monitoring images F1 and F2, α and β are the first and second weights respectively, min() calculates the minimum value, max() calculates the maximum value, and X m Y represents the number of pixels contained in the m-th common edge contour in monitoring screen F1. m Let Z be the number of pixels contained in the m-th common edge contour in the monitoring screen F2, e be the natural constant, and Z be the number of pixels contained in the edge contour. n Let α be the number of pixels contained in the nth unique edge contour, 0 < α < β < 1, α + β = 1; The second stability of the examined patient Per is obtained by averaging the similarity of multiple consecutive two-frame monitoring images.

[0009] Preferably, the calculation of the third degree of stability of the examined patient includes: All edge contour pixels in the defect image Img are obtained, a plane coordinate system is established, and the edge contour pixels are labeled in the plane coordinate system. The number of cluster centers is set to k, and a clustering algorithm is used to obtain k cluster centers. In the planar coordinate system, obtain the coordinates P1 and P2 of a certain cluster center k1 in the first two frames of the monitoring segment, Pic1 and Pic2. Obtain the vector pointing from coordinate P1 to coordinate P2, and use this vector as the movement vector of cluster center k1. Then, obtain the movement vector of each cluster center, sum the movement vectors, and calculate the average to obtain the target vector V between monitoring frames Pic1 and Pic2. 12 Set the initial value of the target distance L to |V 12 |; Similarly, based on the coordinates of the cluster centers in monitoring images Pic2 and Pic3, the target vector V between monitoring images Pic2 and Pic3 is obtained. 23 , target vector V 12 and target vector V 23 The angle between them is taken as θ, and the value of the target distance L is added to |V. 23 |(cosθ+3) / 2, and so on, to obtain the final target distance, and thus the third stability degree e of examining patient Per. -L .

[0010] The value of (cosθ+3) / 2 ranges from 1 to 2, and the value of the included angle θ ranges from 0 to 180 degrees. When θ is closer to 0, the value of (cosθ+3) / 2 is closer to 2. This is because if the movement vector of the cluster center k1 in two consecutive frames is in the same direction, it means that the patient is moving significantly. Therefore, to deal with this situation, the value of the target distance L needs to be larger. When θ is closer to 180 degrees, the value of (cosθ+3) / 2 is closer to 1, which is the opposite of the above situation. Therefore, this setting is reasonable and reliable in this scheme.

[0011] Preferably, the performance evaluation of intelligent medical imaging equipment includes: The overall stability of the examined patient is obtained by combining the first, second, and third stability levels. If the overall stability level is greater than a preset threshold, it is determined that the imaging medical equipment has a malfunction when capturing images. The failure rate of the imaging medical equipment is calculated by combining the judgment results of several target records and used as a performance evaluation indicator for the imaging medical equipment.

[0012] In this scheme, a higher first level of stability indicates greater patient stability during the imaging process, primarily based on grayscale value changes; a higher second level of stability indicates greater patient stability during the imaging process, primarily based on the number of edge contours and the number of pixels within them; a higher third level of stability indicates greater patient stability during the imaging process, primarily based on the movement of cluster centers; and a higher overall level of stability indicates greater patient stability during the imaging process, meaning a higher probability of equipment malfunction. Therefore, if multiple retakes are all due to equipment malfunction, it indicates a relatively high failure rate as a performance evaluation indicator. This invention enables more accurate and objective evaluation of medical equipment performance, avoids data distortion, improves the accuracy and authenticity of performance evaluation, and effectively enhances the scientific nature and application value of the evaluation.

[0013] The AI-based performance evaluation system for large-scale medical equipment includes a target record extraction module, a stability calculation module, and a performance evaluation indicator analysis module. Target record extraction module: used to establish a neural network model corresponding to each defect type, acquire defect images of different defect levels for various defect types, and train the neural network model; retrieve historical re-enhancing records from imaging medical equipment, and extract the examined patient, defect type, defect image, and qualified image from the re-enhancing records; based on the trained neural network model and the saturation values ​​of pixels in the defect images and qualified images, judge and extract the target record from the re-enhancing records. Stability calculation module: used to extract the image capture time of the defect image corresponding to the target record, retrieve the monitoring video of the image capture time, and calculate the first stability of the examined patient based on the pixel grayscale changes of the lesion site in multiple consecutive frames. Using an edge detection algorithm, all edge contours in each frame are obtained. Based on the changes in the number of edge contours and the number of pixels within the edge contours, the second stability of the examined patient is calculated. A planar coordinate system is established, and a clustering algorithm is used to obtain several cluster centers in the defect image. Based on the movement of the cluster centers in the monitoring video, the movement vector corresponding to each cluster center is obtained, and then the third stability of the examined patient is calculated. Performance evaluation indicator analysis module: used to preset the weight of each stability level, calculate the overall stability of the examined patient, determine the fault status of the imaging medical equipment based on the overall stability level, and intelligently analyze the performance evaluation indicators of the imaging medical equipment.

[0014] Preferably, the target record extraction module includes a neural network model training unit and a target record extraction unit; Neural network model training unit: used to build a neural network model corresponding to each defect type, extract multiple defect images of different defect levels belonging to a certain defect type, and input the defect images into the neural network model of a certain defect type for training. Target record extraction unit: used to extract the defect type, defect image and qualified image of a certain re-shot record. The defect image and qualified image are input into the neural network model corresponding to each defect type. Based on the output results and the saturation value of the pixels in the defect image and qualified image, the target record in the re-shot record is judged and extracted.

[0015] Preferably, the performance evaluation indicator analysis module includes a performance evaluation indicator analysis unit; Performance evaluation indicator judgment unit: It is used to combine the first stability level, the second stability level and the third stability level to obtain the overall stability level of the examined patient. If the overall stability level is greater than the preset stability threshold, it is judged that the imaging medical equipment has a fault when taking pictures. By combining the judgment results of several target records, the failure rate of the imaging medical equipment is calculated and used as the performance evaluation indicator of the imaging medical equipment.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: This invention provides a performance evaluation system and method for large-scale medical equipment based on AI, including: establishing and training a neural network model of defect types; retrieving historical re-enhancing records of the imaging medical equipment; extracting target records based on the neural network model; extracting the image capture time of defective images; retrieving monitoring videos; obtaining the pixel grayscale of the captured lesion area; calculating the first stability degree of the examined patient; obtaining the edge contour in each frame; calculating the second stability degree; obtaining cluster centers and judging movement; calculating the third stability degree; calculating the overall stability degree; and intelligently judging the performance evaluation indicators of the imaging medical equipment. This invention, through analysis of historical re-enhancing records, determines the fault status of the imaging medical equipment for performance evaluation indicators, which helps to make the performance evaluation of medical equipment more accurate and objective, avoids data distortion, and improves the accuracy of performance evaluation. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart illustrating the performance evaluation method for large-scale medical equipment based on AI according to the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example: Figure 1 As shown, this invention provides a technical solution for performance evaluation of large-scale medical equipment based on AI, including the following steps: (1) Establish a neural network model corresponding to each defect type, obtain defect images of different defect levels for various defect types, and train the neural network model; retrieve the historical re-capture records of the imaging medical equipment, and extract the examination patient, defect type, defect image and qualified image of the re-capture record; based on the trained neural network model and the saturation value of pixels in the defect image and qualified image, judge and extract the target record in the re-capture record.

[0021] In this embodiment, the defect type refers to the specific reason for re-enhancing the defective image selected by the relevant personnel, such as artifacts, streaks, blurring, and misalignment. However, the selected defect type may be incorrect, leading to misjudgment. The purpose of this step is to screen these misjudged records so that the defective images in the target records actually have defects. The advantage of doing so is that it makes the performance evaluation of medical equipment more accurate and objective, and the subsequent calculation results more reliable. The specific screening process is as follows: Establish a neural network model corresponding to each defect type. There are several defect types. Extract multiple defect images belonging to a certain defect type with different defect levels. The defect level value ranges from 0 to 1. Substitute the defect images into the neural network model of a certain defect type for training. The trained neural network model is used to determine the defect level of the input image in a certain defect type. Extract the defect type Q, defect image Img1, and qualified image Img2 from a certain re-enactment record. Input the defect region C1 in the defect image Img1 into the neural network model corresponding to each defect type to obtain the defect level corresponding to each defect type. Obtain the qualified region C2 corresponding to the defect region C1 in the qualified image Img2. Input the qualified region C2 into the neural network model corresponding to each defect type to obtain the defect level corresponding to each defect type. Obtain the saturation value of each pixel in the defective region C1 and the qualified region C2, obtain several saturation intervals, classify the saturation values ​​into the corresponding saturation intervals, and establish saturation histograms corresponding to the defective region C1 and the qualified region C2 respectively. Calculate the histogram similarity between the two saturation histograms. If the defective region C1 has the highest defect level in defect type Q, and the qualified region C2 has a defect level less than a preset level threshold in every defect type, and the histogram similarity is less than a preset similarity threshold, then a retake record is taken as the target record.

[0022] Saturation determination is an existing technique and will not be elaborated here. In this embodiment, the saturation is normalized, with a value ranging from 0 to 1. Five saturation intervals are set: 0 to 0.2, 0.2 to 0.4, 0.4 to 0.6, 0.6 to 0.8, and 0.8 to 1. Then, the saturation of each pixel in the defective region C1 and the qualified region C2 is assigned to the corresponding saturation interval. Saturation histograms are then constructed for the defective region C1 and the qualified region C2, respectively, and the results are calculated based on cosine similarity. The algorithm calculates the histogram similarity. A higher histogram similarity indicates a closer proximity between the defective region C1 and the acceptable region C2, while a lower histogram similarity indicates a greater deviation between them. In this scheme, a greater deviation between the defective region C1 and the acceptable region C2 suggests that the defective image Img1 is more likely to actually contain a defect. The steps described below for establishing a grayscale histogram based on grayscale values ​​and grayscale ranges, and then calculating the histogram similarity, are similar to this step and will not be elaborated further.

[0023] (2) Extract the image capture time of the defect image corresponding to the target record, retrieve the monitoring video of the image capture time, and calculate the first stability of the examined patient based on the pixel grayscale changes of the lesion site captured in multiple consecutive frames.

[0024] Extract the examination patient Per and defective image Img corresponding to a certain target record, as well as the image capture time T1 of the defective image Img. Retrieve the monitoring segments with a duration of D before and after time T1. The monitoring segments are videos recorded on the lesion site of the examination patient Per. Extract any two consecutive frames of monitoring screens F1 and F2 from the monitoring segments, obtain the defective region in the defective image Img, and mark the corresponding defective regions R1 and R2 in monitoring screens F1 and F2 respectively. Obtain the grayscale value of each pixel in the defect regions R1 and R2, obtain several grayscale intervals, classify the grayscale values ​​into the corresponding grayscale intervals, establish grayscale histograms H1 and H2 corresponding to the defect regions R1 and R2 respectively, and calculate the histogram similarity between grayscale histograms H1 and H2; then, based on multiple consecutive two-frame monitoring images, obtain multiple histogram similarities, calculate the average value to obtain the first stability degree of the examined patient Per.

[0025] (3) Using an edge detection algorithm, all edge contours in each frame are obtained. Based on the changes in the number of edge contours and the number of pixels within the edge contours, the second stability of the patient is calculated.

[0026] Using an edge detection algorithm, all edge contours in monitoring images F1 and F2 are obtained. Binary segmentation is then used to divide the edge contours. The number of common edge contours in monitoring images F1 and F2 is designated as M, the number of unique edge contours in monitoring image F1 is designated as A, and the number of unique edge contours in monitoring image F2 is designated as B. Based on the number of edge contours in monitoring images F1 and F2, and the number of pixels contained within each edge contour, the similarity between monitoring images F1 and F2 is calculated as follows: Where S is the similarity between monitoring images F1 and F2, α and β are the first and second weights respectively, min() calculates the minimum value, max() calculates the maximum value, and X m Y represents the number of pixels contained in the m-th common edge contour in monitoring screen F1. m Let Z be the number of pixels contained in the m-th common edge contour in the monitoring screen F2, e be the natural constant, and Z be the number of pixels contained in the edge contour. n Let α be the number of pixels contained in the nth unique edge contour, 0 < α < β < 1, α + β = 1; The second stability of the examined patient Per is obtained by averaging the similarity of multiple consecutive two-frame monitoring images.

[0027] (4) Establish a planar coordinate system, use a clustering algorithm to obtain several cluster centers in the defect image, and obtain the movement vector corresponding to each cluster center based on the movement of the cluster center in the monitoring video, and then calculate the third stability of the examined patient.

[0028] All edge contour pixels in the defect image Img are obtained, a plane coordinate system is established, and the edge contour pixels are labeled in the plane coordinate system. The number of cluster centers is set to k, and a clustering algorithm is used to obtain k cluster centers. In the planar coordinate system, obtain the coordinates P1 and P2 of a certain cluster center k1 in the first two frames of the monitoring segment, Pic1 and Pic2. Obtain the vector pointing from coordinate P1 to coordinate P2, and use this vector as the movement vector of cluster center k1. Then, obtain the movement vector of each cluster center, sum the movement vectors, and calculate the average to obtain the target vector V between monitoring frames Pic1 and Pic2. 12 Set the initial value of the target distance L to |V 12 |; Similarly, based on the coordinates of the cluster centers in monitoring images Pic2 and Pic3, the target vector V between monitoring images Pic2 and Pic3 is obtained. 23 , target vector V 12 and target vector V 23 The angle between them is taken as θ, and the value of the target distance L is added to |V. 23 |(cosθ+3) / 2, and so on, to obtain the final target distance, and thus the third stability degree e of examining patient Per. -L .

[0029] The value of (cosθ+3) / 2 ranges from 1 to 2, and the value of the included angle θ ranges from 0 to 180 degrees. When θ is closer to 0, the value of (cosθ+3) / 2 is closer to 2. This is because if the movement vector of the cluster center k1 in two consecutive frames is in the same direction, it means that the patient is moving significantly. Therefore, to deal with this situation, the value of the target distance L needs to be larger. When θ is closer to 180 degrees, the value of (cosθ+3) / 2 is closer to 1, which is the opposite of the above situation. Therefore, this setting is reasonable and reliable in this scheme.

[0030] (5) Preset the weight of each stability level, calculate the total stability of the examined patient, determine the fault status of the imaging medical equipment based on the total stability level, and intelligently evaluate the performance evaluation indicators of the imaging medical equipment.

[0031] The overall stability of the examined patient is obtained by combining the first, second, and third stability levels. If the overall stability level is greater than a preset threshold, it is determined that the imaging medical equipment has a malfunction when capturing images. The failure rate of the imaging medical equipment is calculated by combining the judgment results of several target records and used as a performance evaluation indicator for the imaging medical equipment.

[0032] In this embodiment, the higher the values ​​of the first stability level (ranging from -1 to 1), the second stability level (ranging from 0 to 1), and the third stability level (ranging from 0 to 1), the more stable the patient is during the imaging process. The sum of the weights of the first, second, and third stability levels is set to 1. Therefore, in this scheme, the closer the total stability level is to 1, the more stable the patient is during the imaging process. In this embodiment, the preset threshold is set to 0.8. That is, when the total stability level is greater than 0.8, it indicates that the re-shooting was caused by a equipment malfunction. By integrating multiple target records, the performance evaluation of medical equipment can be more accurate and objective, avoiding data distortion, improving the accuracy and authenticity of performance evaluation, and effectively enhancing the scientific nature and application value of the evaluation.

[0033] This embodiment also provides an AI-based performance evaluation system for large-scale medical equipment, including a target record extraction module, a stability calculation module, and a performance evaluation index judgment module. The target record extraction module includes a neural network model training unit and a target record extraction unit, and the performance evaluation index judgment module includes a performance evaluation index judgment unit. When the system executes the computer program, it implements the above-mentioned AI-based performance evaluation method for large-scale medical equipment. Since the AI-based performance evaluation method for large-scale medical equipment has been described in detail above, it will not be repeated here.

[0034] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments have been described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired result. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0035] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0036] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A performance evaluation method for large medical devices based on AI, characterized by, The method comprises the following steps: A neural network model corresponding to each defect type is established, defect images of different defect levels of various defect types are obtained, and the neural network model is trained; the historical re-shooting records of the image medical equipment are called, the examination patients, defect types, defect images and qualified images of the re-shooting records are extracted; the target record in the re-shooting records is judged and extracted according to the trained neural network model and the saturation values of the pixels in the defect images and the qualified images; The image shooting time of the defect image corresponding to the target record is extracted, the monitoring video of the image shooting time is called, and the first stability degree of the examination patient is calculated according to the pixel gray scale change of the lesion part in the continuous multiple frames of pictures; An edge detection algorithm is used to obtain all edge contours in each frame of picture, and the second stability degree of the examination patient is calculated according to the number of edge contours and the change of the number of pixels in the edge contours; A plane coordinate system is established, a clustering algorithm is used to obtain a plurality of clustering centers in the defect image, a moving vector corresponding to each clustering center is obtained according to the movement of the clustering center in the monitoring video, and the third stability degree of the examination patient is calculated; The weight of each stability degree is preset, the total stability degree of the examination patient is calculated, the fault condition of the image medical equipment is determined based on the total stability degree, and the performance evaluation index of the image medical equipment is intelligently judged. 2.The AI-based performance evaluation method of a large medical device according to claim 1, characterized in that, The target record in the re-shooting records is judged and extracted, including: A neural network model corresponding to each defect type is established, the defect types have a plurality of types, a plurality of defect images belonging to a certain defect type and having different defect levels are extracted, the defect level has a value of 0 to 1, the defect image is substituted into the neural network model of the certain defect type for training, and the trained neural network model is used to judge the defect level of the input image in the certain defect type; The defect type Q and the defect image Img1 and the qualified image Img2 of a certain re-shooting record are extracted, the defect area C1 in the defect image Img1 is input into the neural network model corresponding to each defect type, the defect level corresponding to each defect type is obtained, the qualified area C2 corresponding to the defect area C1 in the qualified image Img2 is also input into the neural network model corresponding to each defect type, and the defect level corresponding to each defect type is obtained; The saturation values of each pixel in the defect area C1 and the qualified area C2 are obtained, a plurality of saturation intervals are obtained, the saturation values are classified into the corresponding saturation intervals, the saturation histograms corresponding to the defect area C1 and the qualified area C2 are respectively established, the histogram similarity between the two saturation histograms is calculated, if the defect level of the defect type Q of the defect area C1 is the maximum, the defect level of the qualified area C2 in each defect type is less than a preset level threshold, and the histogram similarity is less than a preset similarity threshold, the certain re-shooting record is taken as the target record. 3.The AI-based performance evaluation method of a large medical device according to claim 1, characterized in that, The first stability degree of the examination patient is calculated, including: Extract the examination patient Per and the defect image Img corresponding to a target record, and the image shooting time T1 of the defect image Img, call the monitoring segment with a length of D before and after T1, and the monitoring segment is a video recording the lesion site of the examination patient Per; extract any two consecutive frames of monitoring pictures F1 and F2 in the monitoring segment, obtain the defect area in the defect image Img, and mark the corresponding defect areas R1 and R2 in the monitoring pictures F1 and F2 respectively; Obtain the gray value of each pixel in the defect areas R1 and R2, obtain a plurality of gray interval, classify the gray value into the corresponding gray interval, establish the gray histogram H1 and H2 corresponding to the defect areas R1 and R2 respectively, and calculate the histogram similarity between the gray histograms H1 and H2; then, according to a plurality of consecutive two frames of monitoring pictures, a plurality of histogram similarities are obtained, and the average value is obtained as the first stability degree of the examination patient Per. 4.The AI-based performance evaluation method of a large medical device according to claim 3, characterized in that, Calculate the second stability degree of the examination patient, including: Using an edge detection algorithm, obtain all edge contours in the monitoring pictures F1 and F2, using a binary segmentation algorithm, the number of edge contours common to the monitoring pictures F1 and F2 is M, the number of edge contours unique to the monitoring picture F1 is A, and the number of edge contours unique to the monitoring picture F2 is B; according to the number of edge contours in the monitoring pictures F1 and F2 and the number of pixel points contained in each edge contour, the similarity of the monitoring pictures F1 and F2 is: Wherein, S is the similarity of the monitoring pictures F1 and F2, α and β are respectively the first weight and the second weight, min() is the minimum value, max() is the maximum value, X m is the number of pixel points contained in the mth common edge contour in the monitoring picture F1, Y m is the number of pixel points contained in the mth common edge contour in the monitoring picture F2, e is a natural constant, Z n is the number of pixel points contained in the nth individual edge contour, 0<α<β<1, α+β=1; According to the similarity of a plurality of consecutive two frames of monitoring pictures, the average value is obtained as the second stability degree of the examination patient Per. 5.The AI-based performance evaluation method of a large medical device according to claim 4, characterized in that, Calculate the third stability degree of the examination patient, including: Obtain all edge contour pixel points in the defect image Img, establish a plane coordinate system, and mark the edge contour pixel points in the plane coordinate system, set the number of cluster centers to k, and use a clustering algorithm to obtain k cluster centers; In the acquisition plane coordinate system, coordinates P1 and P2 of a cluster center k1 in two frames of monitoring pictures Pic1 and Pic2 at the beginning of the monitoring segment are obtained, a vector from coordinate P1 to coordinate P2 is obtained as a moving vector of the cluster center k1, then a moving vector of each cluster center is obtained, the moving vectors are added and averaged to obtain a target vector V between the monitoring pictures Pic1 and Pic2 12 , and an initial value of a target distance L is set as |V 12 | Similarly, according to the coordinates of the cluster center in the monitoring pictures Pic2 and Pic3, the target vector V between the monitoring pictures Pic2 and Pic3 is obtained 23 , the target vector V 12 , and the target vector V 23 between the target distance L is taken as θ, the value of the target distance L is added with |V 23 |(cosθ+3) / 2, and so on, the final target distance is obtained, and then the third stability degree e of the patient Per is obtained -L . 6.The AI-based performance evaluation method of a large medical device according to claim 1, characterized in that, Intelligently research and judge the performance evaluation of the image medical equipment, including: Comprehensively, the first stability degree, the second stability degree and the third stability degree are obtained as the total stability degree of the examination patient, if the total stability degree is greater than a preset degree threshold, it is judged that the image medical equipment has a fault when shooting the image, a plurality of target record judgment results are comprehensively judged, and the fault rate of the image medical equipment is calculated as a performance evaluation index of the image medical equipment.

7. The performance evaluation system for large medical devices based on AI for performing the performance evaluation method for large medical devices based on AI according to any one of claims 1 to 6, characterized by, The system includes a target record extraction module, a stability degree calculation module and a performance evaluation index research and judgment module; The target record extraction module is used for establishing a neural network model corresponding to each defect type, obtaining defect images of different defect levels of various defect types, training the neural network model, calling the historical re-shooting records of the image medical equipment, extracting the examination patients, defect types, defect images and qualified images of the re-shooting records, and judging and extracting the target records in the re-shooting records according to the trained neural network model and the saturation values of the pixels in the defect images and the qualified images. The stability degree calculation module is configured to extract an image shooting time of the target record corresponding to the defect image, call a monitoring video of the image shooting time, calculate a first stability degree of the patient under examination according to a pixel gray scale change of a lesion site in a plurality of continuous frames, and calculate a second stability degree of the patient under examination using an edge detection algorithm to obtain all edge contours in each frame of picture and according to a number of edge contours and a pixel number change in the edge contours. The stability degree calculation module is configured to extract an image shooting time of the target record corresponding to the defect image, call a monitoring video of the image shooting time, calculate a first stability degree of the patient under examination according to a pixel gray scale change of a lesion site in a plurality of continuous frames, and calculate a second stability degree of the patient under examination using an edge detection algorithm to obtain all edge contours in each frame of picture and according to a number of edge contours and a pixel number change in the edge contours. The stability degree calculation module is configured to extract an image shooting time of the target record corresponding to the defect image, call a monitoring video of the image shooting time, calculate a first stability degree of the patient under examination according to a pixel gray scale change of a lesion site in a plurality of continuous frames, and calculate a second stability degree of the patient under examination using an edge detection algorithm to obtain all edge contours in each frame of picture and according to a number of edge contours and a pixel number change in the edge contours. The performance evaluation index research and judgment module is configured to preset a weight value of each stability degree, calculate a total stability degree of the patient under examination, determine a fault condition of the image medical equipment based on the total stability degree, and intelligently research and judge a performance evaluation index of the image medical equipment. 8.The AI-based performance evaluation system of a large medical device according to claim 7, characterized in that, The target record extraction module includes a neural network model training unit and a target record extraction unit. The neural network model training unit is configured to establish a neural network model corresponding to each defect type, extract a plurality of defect images with different defect levels belonging to a certain defect type, and train the defect images in the neural network model of the certain defect type. The target record extraction unit is configured to extract a defect type, a defect image and a qualified image of a certain re-shooting record, input the defect image and the qualified image into the neural network model corresponding to each defect type, and judge and extract the target record in the re-shooting record according to an output result and a saturation value of pixels in the defect image and the qualified image. 9.The AI-based performance evaluation system of a large medical device according to claim 7, characterized in that, The performance evaluation index research and judgment module includes a performance evaluation index research and judgment unit. The performance evaluation index research and judgment unit is configured to comprehensively obtain a total stability degree of the patient under examination by the first stability degree, the second stability degree and the third stability degree, judge that the image medical equipment has a fault when shooting an image if the total stability degree is greater than a preset degree threshold, and calculate a fault rate of the image medical equipment as a performance evaluation index of the image medical equipment by comprehensively judging a plurality of target record judgment results.

Images