Medical image fraud identification method and device, equipment and storage medium

Abstract

The invention discloses a medical image fraud identification method and device, equipment and a storage medium. The method comprises the following steps: acquiring a to-be-analyzed medical image pair; registering the fixed image with a smaller size by using the moving image with a larger size, determining an image range registered with the fixed image from the moving image, and intercepting the moving image by using the image range to obtain a registered image corresponding to the size of the fixed image; evaluating the similarity between the registration image and the fixed image to obtain a similarity representation value; and if the similarity characterization value satisfies a similar identification condition, identifying the to-be-analyzed medical image pair as a suspicious fraud image pair. And if the two images have a multiplexing part or are multiplexed after being tampered, the similarity between the two images can be effectively identified and whether the two images are suspicious fraud image pairs or not can be distinguished through the registration operation and the similarity evaluation operation of matching the large image with the small image, so that the identification capability of medical image fraud behaviors is improved, and the probability that the medical images are illegally used is reduced.

Description

Technical Field

[0001] The present application relates to the field of medical image analysis technology, and in particular to a medical image fraud identification method, apparatus, device and storage medium. Background Art

[0002] Imaging tests (such as X-rays, CT scans, MRI scans, and ultrasound) are crucial technologies that help doctors determine the nature and extent of a disease and assist in determining treatment plans. When patients experience unexplained pain, masses, bleeding, or functional impairment, or when conventional examinations are unable to diagnose a disease, imaging tests can help locate lesions (such as tumors, fractures, and organ damage). For established diseases (such as cancer and cardiovascular disease), imaging tests can monitor lesion size, metastasis, and treatment effectiveness. Before surgery or interventional therapy, imaging tests can provide detailed images of anatomical structures, aiding in planning.

[0003] However, fraudulent activities involving the reuse or manipulation of medical images before reporting them to relevant departments are common. This issue has drawn the attention of the State Administration of Social Security, who believe that identifying fraudulent medical image applications is crucial to preventing such violations.

[0004] Currently, identifying fraudulent medical image applications is challenging. Using only a portion of an individual's image sequence as the reported image requires comparing the entire image with the partial image. This comparison may yield a low similarity, making identification difficult and hindering the accurate identification of fraudulent medical image applications. Summary of the Invention

[0005] Based on the above problems, the present application provides a medical image fraud identification method, device, equipment and storage medium, the purpose of which is to improve the ability to identify medical image fraud, more accurately identify medical images that are directly reused or reused after tampering, and reduce the chance of their illegal use.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, the present application provides a method for identifying medical image fraud, the method comprising:

[0008] Acquire a pair of medical images to be analyzed, wherein the pair of medical images to be analyzed includes two images of different sizes; wherein the image with the smaller size is used as a fixed image during registration, and the image with the larger size is used as a moving image during registration;

[0009] Registering the fixed image using the moving image, determining an image range from the moving image to be registered with the fixed image, and intercepting the moving image using the image range to obtain a registered image corresponding to the size of the fixed image;

[0010] Evaluating the similarity between the registered image and the fixed image to obtain a similarity representation value;

[0011] If the similarity representation value satisfies the similarity identification condition, the medical image pair to be analyzed is identified as a suspected fraudulent image pair.

[0012] In an optional implementation, obtaining the medical image pair to be analyzed includes:

[0013] Acquire an initial medical image pair; the initial medical image pair includes a first image of smaller size and a second image of larger size;

[0014] Sampling the first image and the second image using the same resolution to obtain a first sampled and processed image corresponding to the first image and a second sampled and processed image corresponding to the second image;

[0015] The first sampled and processed image and the second sampled and processed image are used as a medical image pair to be analyzed; wherein the first sampled and processed image is used as a fixed image during registration, and the second sampled and processed image is used as a moving image during registration.

[0016] In an optional implementation, sampling and processing the first image and the second image using the same resolution to obtain a first sampled and processed image corresponding to the first image and a second sampled and processed image corresponding to the second image includes:

[0017] acquiring a shape of the first image and a resolution of the first image;

[0018] Calculating a pseudo-sampling resolution based on the shape of the first image, the resolution of the first image, and the pseudo-sampled shape of the first image;

[0019] The first image is sampled and processed based on the pseudo-sampling resolution to obtain a first sampled and processed image; and the second image is sampled and processed based on the pseudo-sampling resolution to obtain a second sampled and processed image.

[0020] In an optional implementation, before acquiring the initial medical image pair, the method further includes:

[0021] Read the medical images of each target object from the medical image directory and put them into the initial image list;

[0022] For each medical image in the initial image list, use a combination calculation method to combine two of them, and add the formed image combination pairs to the list of image pairs to be analyzed;

[0023] The obtaining of the initial medical image pair is specifically as follows:

[0024] An initial medical image pair is obtained from the list of image pairs to be analyzed.

[0025] In an optional implementation, obtaining an initial medical image pair from the list of image pairs to be analyzed includes:

[0026] Obtain a pair of image combination pairs from the list of image pairs to be analyzed;

[0027] If the target object information fields of the two medical images in the acquired image combination pair are the same, filtering the image combination pair;

[0028] If the target object information fields of the two medical images in the acquired image combination pair are different, the image combination pair is regarded as an initial medical image pair.

[0029] In an optional implementation, evaluating the similarity between the registered image and the fixed image to obtain a similarity representation value includes:

[0030] Obtaining a pixel value array of the registered image and a pixel value array of the fixed image;

[0031] Calculating an average pixel value of the registered image according to the pixel value array of the registered image, and calculating an average pixel value of the fixed image according to the pixel value array of the fixed image;

[0032] A similarity representation value between the registered image and the fixed image is calculated according to the pixel value array of the registered image, the average pixel value of the registered image, the pixel value array of the fixed image, and the average pixel value of the fixed image.

[0033] In an optional implementation, calculating the similarity representation value between the registered image and the fixed image based on the pixel value array of the registered image, the average pixel value of the registered image, the pixel value array of the fixed image, and the average pixel value of the fixed image includes:

[0034] Calculating, based on the pixel value array of the fixed image and the average pixel value of the fixed image, a first difference between the pixel value of each pixel point at each position of the fixed image and the average pixel value of the fixed image; and calculating, based on the pixel value array of the registered image and the average pixel value of the registered image, a second difference between the pixel value of each pixel point at each position of the registered image and the average pixel value of the registered image;

[0035] Based on the correspondence between the pixel positions of the fixed image and the registered image, obtaining a first product of the first difference and the second difference of the pixel points at each corresponding position, and summing the calculated first products to obtain a first summation result;

[0036] Calculating a first square value of each of the first differences and summing the calculated first square values ​​to obtain a second summation result; calculating a second square value of each of the second differences and summing the calculated second square values ​​to obtain a third summation result;

[0037] Calculating a second product of the second summation result and the third summation result;

[0038] The arithmetic square root of the second product is used as a denominator and the first summation result is used as a numerator to calculate a similarity representation value between the registered image and the fixed image.

[0039] A second aspect of the present application provides a medical image fraud identification device, the device comprising:

[0040] An image acquisition module is used to acquire a medical image pair to be analyzed, wherein the medical image pair to be analyzed includes two images of different sizes; wherein the smaller image is used as a fixed image during registration, and the larger image is used as a moving image during registration;

[0041] an image registration module, configured to register the fixed image using the moving image, determine an image range from the moving image to be registered with the fixed image, and intercept the moving image using the image range to obtain a registered image corresponding to the size of the fixed image;

[0042] A similarity evaluation module is used to evaluate the similarity between the registered image and the fixed image to obtain a similarity representation value;

[0043] The suspicious identification module is configured to identify the medical image pair to be analyzed as a suspicious fraud image pair if the similarity representation value satisfies a similarity identification condition.

[0044] A third aspect of the present application provides a medical image fraud identification device, the device comprising: a processor and a memory communicatively connected to each other;

[0045] The memory stores a computer program;

[0046] The processor is used to run the computer program to implement the medical image fraud identification method as described in any implementation of the first aspect.

[0047] In a fourth aspect, the present application provides a computer-readable storage medium storing a computer program, which, when processed, implements the steps of the medical image fraud identification method as described in any implementation of the first aspect.

[0048] Compared with the prior art, this application has the following beneficial effects:

[0049] In the technical solution of the present application, a pair of medical images to be analyzed is first obtained; a larger moving image is used to register a smaller fixed image, the image range to be registered with the fixed image is determined from the moving image, and the moving image is intercepted using the image range to obtain a registered image corresponding to the size of the fixed image; the similarity between the registered image and the fixed image is evaluated to obtain a similarity representation value; if the similarity representation value meets the similarity recognition condition, the medical image pair to be analyzed is identified as a suspected fraudulent image pair. If two images have reused parts or are reused after tampering, the degree of similarity between the two images can be effectively identified through the registration operation of matching the larger with the smaller and the similarity evaluation operation, and it can be determined whether it is a suspected fraudulent image pair, thereby improving the ability to identify medical image fraud and reducing the chance of medical images being used in violation of regulations. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative labor.

[0051] Figure 1 A flowchart of a medical image fraud identification method provided in an embodiment of the present application;

[0052] Figure 2 This is a diagram illustrating an implementation architecture of a medical image fraud identification method provided in an embodiment of the present application;

[0053] Figure 3 This is a diagram illustrating an implementation architecture for initial medical image sampling processing provided in an embodiment of the present application;

[0054] Figure 4 A schematic diagram of the structure of a medical image fraud identification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0055] As previously mentioned, insurance fraud through the reuse of medical images is a common occurrence. This type of fraudulent use of medical images requires precise analysis and identification. One difficulty in assessing image similarity is that if a partial image, rather than the entire image, is used, the similarity assessment may be low. This makes it difficult to accurately identify illegal image reuse, leading to the continued uncertainty surrounding medical image fraud.

[0056] To address the above issues, the inventors have developed a method, device, equipment, and storage medium for identifying medical image fraud. By registering a larger image with a smaller one, capturing the registered image, and then evaluating the similarity between the registered image and the smaller image, the inventors can pinpoint areas in the larger image that are associated with the smaller image and accurately assess their similarity with the smaller image. This avoids identification loopholes caused by comparing areas of the larger image with a lower correlation with the smaller image. This improves the ability to identify medical image fraud and reduces the risk of medical image misuse.

[0057] In order to help those skilled in the art better understand the present invention, the following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of this application.

[0058] See also Figure 1 , which is a flow chart of a medical image fraud identification method provided by an embodiment of the present application. Figure 1 As shown, the medical image fraud identification method includes:

[0059] S101. Obtain a pair of medical images to be analyzed.

[0060] In this application, before formally carrying out the medical image fraud identification link, it is first necessary to obtain a pair of image materials to be analyzed, that is, the medical image pair to be analyzed mentioned in this step. The medical image pair to be analyzed includes two images of different sizes. Among them, the image with a smaller size is used as a fixed image during registration, and the image with a larger size is used as a moving image during registration. For example, the medical image pair to be analyzed includes image A and image B. Among them, the sizes of image A and image B can be read or calculated. On the premise that the sizes of the two images are known, it is natural to determine which image is larger and which image is smaller, and then determine the image with a larger size as the moving image during registration, and the image with a smaller size as the fixed image.

[0061] S102 , registering the fixed image using the moving image, determining an image range to be registered with the fixed image from the moving image, and intercepting the moving image using the image range to obtain a registered image corresponding to the size of the fixed image.

[0062] In the embodiments of the present application, it is particularly pointed out that a large-size image is registered with a small-size image in a registration method: the smaller-size image is kept in a fixed position as a fixed image, and the larger-size image (that is, the moving image referred to above) is moved, and the moving image and the fixed image are registered with each other by moving.

[0063] As an optional implementation, you can use a registration tool, such as ants.registration(), and set the image transformation type to 'Rigid'. After performing the registration operation, you can determine the image range from the moving image to be registered with the fixed image. This image range indicates the area of ​​the moving image that should be cut out as the registration image.

[0064] ANTs (Advanced Normalization Tools) is a C++ library designed for medical imaging, providing tools for image registration and segmentation. ants.registration() is a function in the ANTs library used for image registration. The main parameters of the ants.registration() function include fixed (fixed image, i.e., the target image for registration), moving (moving image, i.e., the source image to be registered), and type_of_transform (transform type). The ants.registration() function supports various transformation types, such as translation (translation), rigid (rigid transformation, including rotation and translation), and similarity (similarity transformation, including rotation, translation, and scaling).

[0065] Because rigid registration involves only two basic transformations, translation and rotation, the transformation model is simpler than that of non-rigid registration. This means that during the registration process, rigid registration requires fewer parameters and lowers computational complexity, enabling faster completion of the registration task. Furthermore, due to the simplicity of the transformation model, rigid registration converges more quickly to the optimal solution, reducing the number of iterations and computational time.

[0066] Rigid registration does not involve complex deformations, so it has good robustness to noise and local deformations in the image. Even if there are some slight noise or deformations in the medical image, rigid registration can still maintain a good registration effect. Non-rigid registration criteria may overfit the local deformations in the image, resulting in inaccurate registration results. Rigid registration criteria can avoid this problem because they limit the degrees of freedom of transformation. In the embodiment of the present application, rigid registration is used to limit image deformation, which can align the moving image and the fixed image well, which is very important for subsequent similarity evaluation. Figure 2 This is a diagram of the implementation architecture of a medical image fraud identification method provided in an embodiment of the present application. Figure 2 The process of registering a moving image with a fixed image and obtaining a registered image from the moving image is shown in FIG.

[0067] because Figure 2 The image range shown by the dotted line is obtained by registering the two images, and the fixed image is used as the target image for registration. Therefore, the image range and the fixed image have the same size. Figure 2 The size of the registered image obtained by intercepting the image range shown by the dotted line is also the same as the size of the fixed image. For images of the same size, similarity evaluation can be performed in the following step S103.

[0068] S103: Evaluate the similarity between the registered image and the fixed image to obtain a similarity representation value.

[0069] In the embodiment of the present application, the similarity between the two images is evaluated mainly based on the pixel values ​​in the images. In an optional implementation, the similarity between the registered image and the fixed image is evaluated to obtain a similarity representation value, including:

[0070] First, the pixel value array of the registration image and the pixel value array of the fixed image are obtained; wherein the pixel value array of the registration image includes the pixel values ​​of the pixel points at each position of the registration image, and similarly, the pixel value array of the fixed image includes the pixel values ​​of the pixel points at each position of the fixed image.

[0071] Next, the average pixel value of the registered image is calculated based on the pixel value array of the registered image, and the average pixel value of the fixed image is calculated based on the pixel value array of the fixed image. The average pixel value of the registered image is calculated by averaging the pixel values ​​of each pixel point in the pixel value array of the registered image. The average pixel value of the registered image represents the overall pixel value level of the registered image. Similarly, the average pixel value of the fixed image is calculated by averaging the pixel values ​​of each pixel point in the pixel value array of the fixed image. The average pixel value of the fixed image represents the overall pixel value level of the fixed image.

[0072] In the embodiment of the present application, the similarity between the registered image and the fixed image is evaluated, not only the pixel values ​​​​specific to each pixel point in the two images are taken into account, but also the overall pixel value level of each of the two images is taken into account. Therefore, based on the pixel value array of the registered image, the average pixel value of the registered image, the pixel value array of the fixed image, and the average pixel value of the fixed image, the similarity representation value between the registered image and the fixed image is calculated. The following is a specific calculation method of the similarity representation value using the expression:

[0073] Formula (1)

[0074] If I f Represents an array of pixel values ​​of a fixed image, I m Represents the pixel value array of the registered image. In formula (1), I f_k Represents the pixel value of the k-th pixel in the pixel value array of the fixed image, I m_k Represents the pixel value of the k-th pixel in the pixel value array of the moving image. represents the average pixel value of a fixed image, Represents the average pixel value of the registered images.

[0075] In an optional implementation, based on the pixel value array of the fixed image and the average pixel value of the fixed image, the first difference between the pixel value of each pixel point at each position of the fixed image and the average pixel value of the fixed image is calculated. Referring to formula (1), the pixel value I of the pixel point at the kth position in the pixel value array of the fixed image is f_k , calculate its The first difference (I f_k - ). In addition, according to the pixel value array of the registered image and the average pixel value of the registered image, the second difference between the pixel value of each position of the registered image and the average pixel value of the registered image is calculated. Referring to formula (1), the pixel value I of the pixel at the kth position in the pixel value array of the registered image is m_k , calculate its The second difference (Im_k - ).

[0076] Based on the correspondence between the pixel positions of the fixed image and the registered image, the first product of the first difference and the second difference of the pixel at each corresponding position is obtained, and the calculated first products are summed to obtain a first summation result. See the numerator of formula (1), which represents the process of summing the first products of the first difference and the second difference.

[0077] Calculate each first difference (I f_k - ), and sum the calculated first square values ​​to obtain the second summation result ; Calculate each second difference (I m_k - ), and sum the calculated second square values ​​to obtain the third summation result. .

[0078] Calculate the second product of the second summation result and the third summation result. See the content within the square root symbol in the denominator of formula (1) . Finally, the arithmetic square root of the second product is used as the denominator and the first summation result is used as the numerator to calculate the similarity representation value between the registered image and the fixed image. The ratio of the numerator to the denominator is expressed as shown in formula (1). The similarity representation value finally calculated can be expressed as a percentage, which is more intuitive to understand. In the example calculation method of formula (1), the larger the similarity representation value, the more similar the registered image and the fixed image are; the smaller the similarity representation value, the greater the difference between the registered image and the fixed image.

[0079] The above describes only one example implementation method for calculating the similarity representation value between the registered image and the fixed image in this step through equation (1). In practical applications, the implementation method of step S103 is not limited to this, and the above example is not intended to limit the implementation method of this step. Other pixel-based calculation methods can also be used, such as mean square error, peak signal-to-noise ratio, structural similarity index, etc.

[0080] S104: Determine whether the similarity representation value meets the similarity recognition condition. If so, proceed to S105.

[0081] In the present application, the similarity recognition condition is a condition associated with the similarity characterization value, which is used to determine whether two images with specific similarity characterization values ​​are similar enough to be identified as a pair of suspicious fraudulent images. The similarity recognition condition is: the similarity characterization value between the registered image and the fixed image is greater than or equal to the preset threshold. As an example, the preset threshold is 80%. It should be noted that in actual applications, the preset threshold in the similarity recognition condition can be set based on actual needs. For example, if the method is required to have a more sensitive recognition ability, a smaller preset threshold is set to identify as many pairs of suspicious fraudulent images as possible; if the method is required to have more accurate and reliable recognition results, a larger preset threshold is set to reduce the chance of misjudgment.

[0082] S105: Identify the medical image pair to be analyzed as a suspected fraudulent image pair.

[0083] In the technical solution of the present application, a pair of medical images to be analyzed is first obtained; a larger moving image is used to register a smaller fixed image, the image range to be registered with the fixed image is determined from the moving image, and the moving image is intercepted using the image range to obtain a registered image corresponding to the size of the fixed image; the similarity between the registered image and the fixed image is evaluated to obtain a similarity representation value; if the similarity representation value meets the similarity recognition condition, the medical image pair to be analyzed is identified as a suspected fraudulent image pair. If two images have reused parts or are reused after tampering, the degree of similarity between the two images can be effectively identified through the registration operation of matching the larger with the smaller and the similarity evaluation operation, and it can be determined whether it is a suspected fraudulent image pair, thereby improving the ability to identify medical image fraud and reducing the chance of medical images being used in violation of regulations.

[0084] Image pairs identified as suspected fraudulent can be marked to distinguish them from other image pairs. Furthermore, the image range within the mobile medical image in the suspected fraudulent image pair that is registered with the fixed image can be marked to facilitate visual identification by relevant personnel. Suspicious situations regarding the medical image pair to be analyzed can be output externally via a message, for example, generating a suspicious analysis report for the medical image pair, noting the similarity representation value in the report.

[0085] In the above embodiments, the implementation method of identifying whether the medical image pair to be analyzed is a suspected fraudulent image pair is introduced. Next, the process of obtaining the medical image pair to be analyzed in the early stage is explained. In the present application, the initial medical image pair is first obtained. Here, the initial medical image pair can be understood as the original medical image directly obtained by reading the medical imaging data of the target object (as an example, the target object can be understood as a patient who receives medical treatment and takes medical images in a medical institution). In actual applications, considering that in medical image fraud, it is also possible that the resolution of the image is modified, making it difficult to identify the fraudulent image and the reused image as highly similar, the inventor proposes that it is necessary to re-acquire the original medical image at a uniform resolution and use the acquired image pair as the medical image pair to be analyzed.

[0086] Figure 3 The embodiment of the present application provides an implementation architecture diagram of an initial medical image pair sampling process. An initial medical image pair is obtained, which includes a first image of smaller size and a second image of larger size. Using the same resolution (i.e. Figure 3 The first image and the second image are each sampled and processed at the simulated sampling resolution shown in , obtaining a first sampled and processed image corresponding to the first image and a second sampled and processed image corresponding to the second image. Since the first image is originally smaller in size, its smaller features are retained after sampling at the same resolution. Therefore, after the first and second sampled and processed images are considered as a medical image pair to be analyzed, the first sampled and processed image serves as the fixed image during registration, and the second sampled and processed image serves as the moving image during registration.

[0087] The following describes a method for calculating the pseudo-sampling resolution used in sampling processing. In an optional implementation, sampling the first image and the second image using the same resolution to obtain a first sampled image corresponding to the first image and a second sampled image corresponding to the second image includes:

[0088] Obtain the shape and resolution of the first image; and calculate the pseudo-sampling resolution based on the shape, resolution, and pseudo-sampling shape of the first image. The pseudo-sampling shape of the first image can be preset. For example, set the pseudo-sampling shape of the first image to new_img_shape = [25, 25, 25] as required.

[0089] The shape and resolution of the first image can be extracted from the first image using image processing tools. For example, the SimpleITK functions image.GetSize() and image.GetSpacing() can be used. The formulas are as follows:

[0090] image_shape = image.GetSize();

[0091] image_spacing = image.GetSpacing();

[0092] Here, image_shape represents the shape of the first image; image_spacing represents the resolution of the first image.

[0093] The size of an image can be calculated by multiplying the image's shape by its resolution. The following formula shows how to calculate the actual size of the first image.

[0094] image_size = image_shape * image_spacing;

[0095] In the embodiment of the present application, the first image and the second image are resampled at a low resolution using the same resolution. In the specific implementation, the smaller image of the first image and the second image is used as the basis for calculating the proposed sampling resolution new_spacing, and the formula is as follows:

[0096] new_spacing = image_size / new_img_shape;

[0097] Then, the first image is sampled and processed based on the quasi-sampling resolution new_spacing to obtain a first sampled and processed image; and the second image is sampled and processed based on the quasi-sampling resolution new_spacing to obtain a second sampled and processed image, such as Figure 3 shown.

[0098] In an embodiment of the present application, the quasi-sampling resolution is less than the resolution of the first image and also less than the resolution of the second image. The use of a low-resolution resampling process for the first and second images can significantly improve the calculation speed while ensuring calculation accuracy. Furthermore, it can avoid calculation errors caused by directly calculating the similarity between the first and second images when the resolution of the first image is different from that of the second image. Specifically, since the fixed image (the first sampled image) and the moving image (the second sampled image) are resampled using the same resolution, i.e., the quasi-sampling resolution mentioned above, the resolutions of the fixed image and the moving image are consistent. If there is indeed reuse or tampering between the first and second images, the technical solution of the present application can also effectively identify the high similarity between the original two images. In this way, the technical solution of the present application further enhances the ability to identify medical image fraud.

[0099] Based on the introduction above, before the image similarity evaluation in this application, the first image and the second image were downsampled by downsampling. Without affecting the quality of the similarity evaluation, the amount of calculation is greatly reduced. For example, the shape of the first image is 512*512*600, and the image shape becomes 25*25*25 after sampling. When performing subsequent registration calculations of fixed and moving images, the calculation time is greatly shortened. The 5 minutes that usually takes is shortened to within 10 seconds. Due to the use of registration and recognition calculations, the amount of calculation is huge. Here, the efficient calculation promoted by downsampling is of great significance.

[0100] In an optional implementation, before obtaining the initial medical image pair, the medical image fraud identification method may further include: reading the medical image of each target object from the medical image directory and placing it in the initial image list; for each medical image in the initial image list, using a combination calculation method to combine them two by two, and adding the formed image combination pairs to the list of image pairs to be analyzed.

[0101] In this embodiment, medical images of a group of patients in a hospital are used as identification targets. Medical images are read from a medical image directory and paired up, and then the resulting image pairs are added to a list of image pairs to be analyzed. The initial medical image pair is obtained, specifically by obtaining it from the list of image pairs to be analyzed.

[0102] For example, a medical image directory might look like this: a hospital group contains a patient directory, a study directory, and an image sequence directory. All image sequence directories are retrieved and placed into the initial image list. Image pairs are then grouped together and placed into the list of image pairs to be analyzed.

[0103] In an optional implementation, obtaining an initial medical image pair from the list of image pairs to be analyzed includes: obtaining a pair of image combination pairs from the list of image pairs to be analyzed; if the target object information fields of the two medical images in the obtained image combination pair are the same, filtering the image combination pair; if the target object information fields of the two medical images in the obtained image combination pair are different, treating the image combination pair as an initial medical image pair. This process is a process of cyclically reading the list of image pairs to be analyzed. The target object information field is analyzed by string analysis to determine whether the two medical images in the image combination pair belong to the same patient. Since medical image fraud generally involves two different patients, image combination pairs of the same patient can be effectively filtered based on the target object information field, thereby improving the overall analysis and recognition efficiency.

[0104] It should be noted that in the embodiments of the present application, the integrity of the image can be further identified. For example, in the list of image pairs to be analyzed, if a pair of image combination pairs is taken out and it is found that the image sequence length (number of slices) of at least one image is less than the preset length threshold, the image is deemed to be abnormal, and an abnormal result is returned. For image combination pairs with abnormalities, since a complete image cannot be formed, they are not used as initial medical image pairs. There is no need to perform resampling and alignment and similarity comparison and other identification work. As an example, the preset length threshold is 20. Of course, in actual applications, this value can also be set based on needs, and is not limited here.

[0105] In the embodiment of the present application, the format of the original image needs to be taken into consideration when reading the image, so as to automatically identify and successfully read the image and complete subsequent resampling, registration and other operations.

[0106] Reading 3D images: 3D medical images are typically in the DICOM format or the NII.GZ compressed format. Depending on the image format, the corresponding reading method can be used. For example, for DICOM-formatted 3D medical images, enter the image sequence path and use sitk.ImageSeriesReader() to read; for NII.GZ-formatted 3D medical images, enter the image file name and use sitk.ReadImage() to read.

[0107] It should be noted that, in practical applications, images with localizers, including those containing localizers, may interfere with the normal reading of medical images. For example, if a DICOM-formatted image sequence path includes a localizer, normal reading will fail. If this occurs, the system automatically switches to reading images with a localizer. Specifically, it obtains all file names in the image directory. A loop is used to read each file one by one, such as using pydicom.dcmread(). The sequence information is extracted. If the information contains 'LOCALIZER' (or other localizer value), it is identified as a localizer image, skipped, and the next image is read until all images are read. The resulting array is then converted to image format and combined using sitk.JoinSeries(). The resolution of the original image is set to the final image. This method eliminates the impact of the localizer image during image reading and facilitates successful medical image reading.

[0108] 2D images can be read using a DICOM viewer or a hospital's PACS (Picture Archiving and Communication System) client. DICOM formats can also be converted to common image formats, such as jpg and png, using common image viewing software. Medical images in common image formats can also be viewed using the operating system's built-in image viewer. Images can also be viewed through online image viewing platforms.

[0109] During the reading process, if the file is identified as damaged, the result of image file abnormality will be directly returned, indicating that the reading failed.

[0110] Based on the medical image fraud identification method introduced in the above embodiment, the present application also proposes a medical image fraud identification device. Figure 4 Figure 2 is a schematic diagram of the structure of the device. Figure 4 As shown, the medical image fraud identification device includes:

[0111] An image acquisition module 41 is configured to acquire a medical image pair to be analyzed, wherein the medical image pair to be analyzed includes two images of different sizes; wherein the smaller image is used as a fixed image during registration, and the larger image is used as a moving image during registration;

[0112] An image registration module 42 is configured to register the fixed image with the moving image, determine an image range from the moving image to be registered with the fixed image, and intercept the moving image using the image range to obtain a registered image corresponding to the size of the fixed image;

[0113] A similarity evaluation module 43 is used to evaluate the similarity between the registered image and the fixed image to obtain a similarity representation value;

[0114] The suspicious identification module 44 is configured to identify the medical image pair to be analyzed as a suspicious fraudulent image pair if the similarity representation value satisfies a similarity identification condition.

[0115] In an optional implementation, the image acquisition module 41 specifically includes:

[0116] A first acquisition unit is configured to acquire an initial medical image pair, wherein the initial medical image pair includes a first image of smaller size and a second image of larger size;

[0117] a sampling and processing unit, configured to perform sampling and processing on the first image and the second image using the same resolution to obtain a first sampled and processed image corresponding to the first image and a second sampled and processed image corresponding to the second image;

[0118] An image pair determination unit is used to treat the first sampled and processed image and the second sampled and processed image as a medical image pair to be analyzed; wherein the first sampled and processed image is used as a fixed image during registration, and the second sampled and processed image is used as a moving image during registration.

[0119] In an optional implementation, the sampling processing unit is specifically configured to:

[0120] acquiring a shape of the first image and a resolution of the first image;

[0121] Calculating a pseudo-sampling resolution based on the shape of the first image, the resolution of the first image, and the pseudo-sampled shape of the first image;

[0122] The first image is sampled and processed based on the pseudo-sampling resolution to obtain a first sampled and processed image; and the second image is sampled and processed based on the pseudo-sampling resolution to obtain a second sampled and processed image.

[0123] In an optional implementation, the medical image fraud identification device further includes:

[0124] A reading module is used to read the medical images of each target object from the medical image directory and put them into the initial image list;

[0125] a combination module, configured to combine each medical image in the initial image list in pairs using a combination calculation method, and add the formed image combination pairs to the list of image pairs to be analyzed;

[0126] The first acquiring unit is specifically configured to:

[0127] An initial medical image pair is obtained from the list of image pairs to be analyzed.

[0128] In an optional implementation, the first acquiring unit is specifically configured to:

[0129] Obtain a pair of image combination pairs from the list of image pairs to be analyzed;

[0130] If the target object information fields of the two medical images in the acquired image combination pair are the same, filtering the image combination pair;

[0131] If the target object information fields of the two medical images in the acquired image combination pair are different, the image combination pair is regarded as an initial medical image pair.

[0132] In an optional implementation, the similarity evaluation module 43 includes:

[0133] a second acquiring unit, configured to acquire a pixel value array of the registered image and a pixel value array of the fixed image;

[0134] a first calculation unit, configured to calculate an average pixel value of the registered image according to the pixel value array of the registered image, and to calculate an average pixel value of the fixed image according to the pixel value array of the fixed image;

[0135] The second calculation unit is used to calculate the similarity representation value between the registered image and the fixed image based on the pixel value array of the registered image, the average pixel value of the registered image, the pixel value array of the fixed image and the average pixel value of the fixed image.

[0136] In an optional implementation, the second computing unit is specifically configured to:

[0137] Calculating, based on the pixel value array of the fixed image and the average pixel value of the fixed image, a first difference between the pixel value of each pixel point at each position of the fixed image and the average pixel value of the fixed image; and calculating, based on the pixel value array of the registered image and the average pixel value of the registered image, a second difference between the pixel value of each pixel point at each position of the registered image and the average pixel value of the registered image;

[0138] Based on the correspondence between the pixel positions of the fixed image and the registered image, obtaining a first product of the first difference and the second difference of the pixel points at each corresponding position, and summing the calculated first products to obtain a first summation result;

[0139] Calculating a first square value of each of the first differences and summing the calculated first square values ​​to obtain a second summation result; calculating a second square value of each of the second differences and summing the calculated second square values ​​to obtain a third summation result;

[0140] Calculating a second product of the second summation result and the third summation result;

[0141] The arithmetic square root of the second product is used as a denominator and the first summation result is used as a numerator to calculate a similarity representation value between the registered image and the fixed image.

[0142] Based on the medical image fraud identification method and apparatus described in the aforementioned embodiments, the present application also provides a medical image fraud identification device, which includes: a processor and a memory that are communicatively connected to each other;

[0143] The memory stores a computer program;

[0144] The processor is used to run the computer program to implement the medical image fraud identification method as described in the method embodiment.

[0145] In addition, the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is processed, it implements the steps of the medical image fraud identification method as described in the method embodiment.

[0146] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and equipment embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and equipment embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0147] The above is merely one specific embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for identifying medical image fraud, characterized in that: include: Acquire a pair of medical images to be analyzed, wherein the pair of medical images to be analyzed includes two images of different sizes; wherein the image with the smaller size is used as a fixed image during registration, and the image with the larger size is used as a moving image during registration; Registering the fixed image using the moving image, determining an image range from the moving image to be registered with the fixed image, and intercepting the moving image using the image range to obtain a registered image corresponding to the size of the fixed image; Evaluating the similarity between the registered image and the fixed image to obtain a similarity representation value; If the similarity representation value satisfies the similarity identification condition, the medical image pair to be analyzed is identified as a suspected fraudulent image pair.

2. The method according to claim 1, characterized in that The obtaining of the medical image pair to be analyzed includes: Acquire an initial medical image pair; the initial medical image pair includes a first image of smaller size and a second image of larger size; Sampling the first image and the second image using the same resolution to obtain a first sampled and processed image corresponding to the first image and a second sampled and processed image corresponding to the second image; The first sampled and processed image and the second sampled and processed image are used as a medical image pair to be analyzed; wherein the first sampled and processed image is used as a fixed image during registration, and the second sampled and processed image is used as a moving image during registration.

3. The method according to claim 2, characterized in that The sampling and processing of the first image and the second image using the same resolution to obtain a first sampled and processed image corresponding to the first image and a second sampled and processed image corresponding to the second image includes: acquiring a shape of the first image and a resolution of the first image; Calculating a pseudo-sampling resolution based on the shape of the first image, the resolution of the first image, and the pseudo-sampled shape of the first image; The first image is sampled and processed based on the pseudo-sampling resolution to obtain a first sampled and processed image; and the second image is sampled and processed based on the pseudo-sampling resolution to obtain a second sampled and processed image.

4. The method according to claim 2, characterized in that Before acquiring the initial medical image pair, the method further includes: Read the medical images of each target object from the medical image directory and put them into the initial image list; For each medical image in the initial image list, use a combination calculation method to combine two of them, and add the formed image combination pairs to the list of image pairs to be analyzed; The obtaining of the initial medical image pair is specifically as follows: An initial medical image pair is obtained from the list of image pairs to be analyzed.

5. The method according to claim 4, characterized in that The obtaining of the initial medical image pair from the list of image pairs to be analyzed comprises: Obtain a pair of image combination pairs from the list of image pairs to be analyzed; If the target object information fields of the two medical images in the acquired image combination pair are the same, filtering the image combination pair; If the target object information fields of the two medical images in the acquired image combination pair are different, the image combination pair is regarded as an initial medical image pair.

6. The method according to any one of claims 1 to 5, characterized in that The evaluating the similarity between the registered image and the fixed image to obtain a similarity representation value includes: Obtaining a pixel value array of the registered image and a pixel value array of the fixed image; Calculating an average pixel value of the registered image according to the pixel value array of the registered image, and calculating an average pixel value of the fixed image according to the pixel value array of the fixed image; A similarity representation value between the registered image and the fixed image is calculated according to the pixel value array of the registered image, the average pixel value of the registered image, the pixel value array of the fixed image, and the average pixel value of the fixed image.

7. The method according to claim 6, characterized in that Calculating a similarity representation value between the registered image and the fixed image based on the pixel value array of the registered image, the average pixel value of the registered image, the pixel value array of the fixed image, and the average pixel value of the fixed image includes: Calculating, based on the pixel value array of the fixed image and the average pixel value of the fixed image, a first difference between the pixel value of each pixel point at each position of the fixed image and the average pixel value of the fixed image; and calculating, based on the pixel value array of the registered image and the average pixel value of the registered image, a second difference between the pixel value of each pixel point at each position of the registered image and the average pixel value of the registered image; Based on the correspondence between the pixel positions of the fixed image and the registered image, obtaining a first product of the first difference and the second difference of the pixel points at each corresponding position, and summing the calculated first products to obtain a first summation result; Calculating a first square value of each of the first differences and summing the calculated first square values ​​to obtain a second summation result; calculating a second square value of each of the second differences and summing the calculated second square values ​​to obtain a third summation result; Calculating a second product of the second summation result and the third summation result; The arithmetic square root of the second product is used as a denominator and the first summation result is used as a numerator to calculate a similarity representation value between the registered image and the fixed image.

8. A medical image fraud identification device, characterized in that: include: An image acquisition module is used to acquire a medical image pair to be analyzed, wherein the medical image pair to be analyzed includes two images of different sizes; wherein the smaller image is used as a fixed image during registration, and the larger image is used as a moving image during registration; an image registration module, configured to register the fixed image using the moving image, determine an image range from the moving image to be registered with the fixed image, and intercept the moving image using the image range to obtain a registered image corresponding to the size of the fixed image; A similarity evaluation module is used to evaluate the similarity between the registered image and the fixed image to obtain a similarity representation value; The suspicious identification module is configured to identify the medical image pair to be analyzed as a suspicious fraud image pair if the similarity representation value satisfies a similarity identification condition.

9. A medical image fraud identification device, characterized in that: include: a processor and memory communicatively connected to each other; The memory stores a computer program; The processor is configured to run the computer program to implement the medical image fraud identification method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that A computer program is stored, and when the computer program is processed, the steps of the medical image fraud identification method according to any one of claims 1 to 7 are implemented.

Images