An image recognition-based medical image intelligent evaluation system

Abstract

The present application relates to the technical field of image recognition, and more particularly to a medical image intelligent evaluation system based on image recognition; the system comprises an image registration module, an image segmentation module, a preliminary fusion module, an image evaluation module, an optimization feedback module and an image output module. The present application firstly registers the CT image and the MRI image to ensure spatial alignment, then fuses the images by segmenting the region of interest, i.e. superimposes the bone contour in the CT image onto the MRI image. Since motion artifacts caused by patient movement during scanning may exist in the original CT image, the degree of coincidence and the blank degree of the bone contour and the edge of the anatomical structure of the MRI image are analyzed to feed back the optimization registration parameters or segmentation parameters, so as to improve the segmentation accuracy under the conditions of artifacts and low contrast during training of the segmentation network, and to maximize the preservation of the spectral and texture information of the two images, thereby ensuring the fusion effect.

Description

Technical Field

[0001] This invention relates to the field of image recognition technology, and in particular to an intelligent medical image evaluation system based on image recognition. Background Technology

[0002] Medical imaging technology plays a crucial role in clinical diagnosis and treatment planning, with CT (computed tomography) and MRI (magnetic resonance imaging) being two of the most commonly used imaging modalities. CT images have significant advantages in displaying bone structures, calcifications, and acute hemorrhages, while MRI images excel in soft tissue contrast, clearly displaying soft tissue structures such as gray matter, white matter, cerebrospinal fluid, and tumors. Therefore, effectively fusing CT and MRI images can provide physicians with more comprehensive and accurate diagnostic information.

[0003] While multimodal medical image fusion technology has made some progress, it still faces numerous challenges in practical applications. These include insufficient registration accuracy, a lack of diverse segmentation methods, and a lack of effective quality assessment mechanisms. Insufficient registration accuracy stems from existing methods often employing fixed marker density, which fails to adapt to varying image qualities and anatomical structures. When motion artifacts or noise interference exist in the image, fixed-parameter registration methods are prone to spatial alignment errors, leading to poor subsequent fusion results. Traditional segmentation methods typically employ uniform processing strategies, failing to fully consider the essential differences between healthy and diseased tissues. For lesion regions with blurred boundaries, conventional segmentation methods often struggle to accurately delineate the contours, while excessive complexity can reduce efficiency when processing healthy tissue. Existing fusion systems largely rely on manual visual evaluation of the fusion effect, lacking objective and quantitative evaluation metrics. Furthermore, issues such as edge mismatches and artifacts in the fused image are difficult to detect and automatically correct in a timely manner.

[0004] Chinese Patent Publication No. CN114782352A discloses an artificial intelligence-based medical image recognition system and method, including an image acquisition module, an image detection module, a normalization processing module, and a treatment evaluation module; the image acquisition module, image detection module, normalization processing module, and treatment evaluation module are connected sequentially. It is evident that existing intelligent medical image evaluation technologies lack an optimization feedback mechanism and suffer from motion artifacts and low contrast issues. This is because most systems employ open-loop processing, making it impossible to dynamically adjust processing parameters based on fusion quality. When fusion quality problems occur, manual intervention is required to readjust parameters, reducing the system's automation level and clinical application efficiency. In actual clinical environments, motion artifacts caused by patient movement and low contrast between lesions and normal tissues are common, severely impacting the accuracy of segmentation and fusion. Summary of the Invention

[0005] To address this issue, the present invention provides an intelligent medical image evaluation system based on image recognition, which overcomes the problem in the prior art of lacking intelligent feedback optimization of registration and segmentation processes based on the fusion effect of medical images, resulting in low fusion effect.

[0006] To achieve the above objectives, the present invention provides an intelligent medical image assessment system based on image recognition, comprising:

[0007] The image registration module is used to register the input original CT image and MRI image based on a preset marker density, and output the registered CT image and MRI image.

[0008] An image segmentation module, which is connected to the image registration module, is used to select the corresponding segmentation method according to the category of the registered CT image to segment the registered CT image and obtain several regions to be mosaicked.

[0009] A preliminary fusion module, which is connected to the image segmentation module, is used to mosaic any region to be mosaicked into the MRI image to form a mosaicked region, so as to generate a preliminary fusion image based on each mosaicked region.

[0010] An image evaluation module, connected to the preliminary fusion module, is used to calculate the edge overlap and edge blanking based on the contour edge pixels of the region to be mosaicked, and to analyze the preliminary fused image based on the edge overlap and edge blanking to obtain diagnostic results.

[0011] The optimization feedback module, together with the image evaluation module, is used to optimize the segmentation method based on the edge overlap and the category of the registered CT image, and to correct the preset marker density based on the edge blankness, so as to generate a secondary fused image.

[0012] An image output module, which is connected to the image evaluation module and the optimization feedback module respectively, is used to use the preliminary fusion image as the final fusion image or the secondary fusion image as the final fusion image based on the diagnostic results.

[0013] Furthermore, the image segmentation module includes an image classification unit and an image segmentation unit;

[0014] The image classification unit is used to classify registered CT images into healthy images and disease images;

[0015] The image segmentation unit is used to segment the lesion image using a first segmentation method and to segment the healthy image using a second segmentation method.

[0016] Furthermore, the image segmentation unit includes a first segmentation subunit and a second segmentation subunit;

[0017] The first segmentation subunit is used to directly segment the skeletal region of the health image using a first segmentation method;

[0018] The second segmentation subunit is used to segment the lesion image in a second segmentation manner after the edge of the target segmentation contour is gradient processed.

[0019] Furthermore, the image evaluation module includes a first computing unit and a second computing unit;

[0020] The first calculation unit is used to calculate the ratio of the number of pixels whose contour edges of the region to be mosaicked coincide with the edges of the MRI image to the total number of pixels of the contour edges of the region to be mosaicked, and obtain the edge overlap degree.

[0021] The second calculation unit is used to calculate the ratio of the number of pixels in the MRI image whose anatomical structure edges are not covered by the area to be mosaicked to the total number of pixels in the MRI image's anatomical structure edges, thus obtaining the edge blanking degree.

[0022] Furthermore, the image evaluation module also includes a problem diagnosis unit and an evaluation unit;

[0023] The problem diagnosis unit is used to compare the edge overlap with the standard overlap, and to compare the edge blankness with the standard blankness.

[0024] The evaluation unit is used to obtain an oversegmentation diagnosis result based on the comparison between edge overlap and standard overlap, and to obtain a segmentation quality qualified result based on the comparison between edge blankness and standard blankness.

[0025] Furthermore, the optimization feedback module includes a segmentation parameter optimization unit;

[0026] The segmentation parameter optimization unit is used to generate a secondary fusion image in response to the oversegmentation diagnosis result, according to the optimization processing method.

[0027] Furthermore, the optimization feedback module also includes a registration verification unit;

[0028] The registration verification unit is used to verify the registration accuracy when the edge blankness is greater than the standard blankness, and to determine whether to reduce the preset marker point density based on the verification result in order to trigger the re-registration process.

[0029] Furthermore, the registration and verification unit includes a first error calculation subunit, a marking subunit, a second error calculation subunit, and a parameter correction subunit;

[0030] The first error calculation subunit is used to calculate the distance between the registered CT image and the MRI image at anatomical landmarks as the actual distance, and to calculate the difference between the actual distance and the standard distance to obtain the actual error;

[0031] The marking subunit is used to mark the corresponding anatomical landmark as an error anatomical point when the actual error is greater than the allowable error;

[0032] The second error calculation subunit is used to calculate the percentage of error anatomical points to total anatomical landmarks, and obtain the actual error rate;

[0033] The parameter correction subunit is used to reduce the preset marker density to the corrected marker density when the actual error rate is greater than the standard error rate.

[0034] Furthermore, the first segmentation method is,

[0035] Obtain the CT values ​​of all pixels within the lesion contour region in the registered CT image and store them as an initial lesion value list;

[0036] Calculate the average CT value of any region within the lesion contour to obtain the corresponding average lesion CT value;

[0037] Obtain the CT values ​​of all pixels in the normal tissue surrounding each lesion outline and store them as a background CT value list;

[0038] Calculate the average CT value of the region outside the lesion outline to obtain the average background CT value;

[0039] The difference between the average background CT value and the average lesion CT value is obtained to determine the actual CT value difference.

[0040] Update each initial CT value of each lesion contour in the registered CT image using the corrected CT value;

[0041] Specifically, for each initial CT value, the sum of the differences between the initial CT value and the actual CT value is calculated to obtain the corrected CT value; the average lesion CT value is the ratio of the initial CT value of each pixel within the lesion contour to the number of pixels; the average background CT value is the ratio of the initial CT value of each pixel outside the lesion contour to the number of pixels.

[0042] Furthermore, the second segmentation method is as follows:

[0043] The CT values ​​of all pixels within the contour region of each brain tissue in the registered CT image are obtained and recorded as the first actual CT value.

[0044] The CT values ​​of all pixels outside the contour of each brain tissue in the registered CT image are obtained and recorded as the second actual CT value.

[0045] The average CT value corresponding to the pixel point whose first actual CT value is less than the first average contour CT value is obtained as the second average contour CT value.

[0046] The average CT value corresponding to the pixel point whose second actual CT value is less than the first average out-of-contour CT value is obtained as the second average out-of-contour CT value.

[0047] The average CT value corresponding to the pixel point whose first actual CT value is greater than the first average contour CT value is obtained as the third average contour CT value.

[0048] The average CT value corresponding to the pixel point whose second actual CT value is greater than the first average out-of-contour CT value is obtained as the third average out-of-contour CT value.

[0049] Calculate the square of the CT value within each average contour and the corresponding square of the CT value outside the average contour to obtain the real-time CT value difference.

[0050] The target CT value is obtained by adding the differences between the CT values ​​within each average contour and the corresponding actual CT values.

[0051] Adjust the original CT values ​​of the registered CT images to the corresponding target CT values.

[0052] Compared with the prior art, the beneficial effects of the present invention are that by first registering CT images and MRI images to ensure spatial alignment, and then by segmenting the region of interest and then fusing them, the bone contours in the CT image are superimposed on the MRI image. By analyzing the degree of overlap and blankness between the bone contours and the anatomical structure edges of the MRI image, it is determined whether the fused image is natural at the junction of bone and soft tissue, and whether there are obvious jagged edges or artifacts. The registration parameters or segmentation parameters are optimized to retain the spectral (such as density in CT) and texture (such as soft tissue contrast in MRI) information of the two images to the maximum extent, and to ensure the fusion effect.

[0053] Furthermore, when the edge blankness is greater than the standard blankness, it may be due to the image occlusion causing some bones to be missed when segmenting the CT image, or it may be due to insufficient registration accuracy. Therefore, it is determined whether to trigger the re-registration process based on the verification results to improve the registration accuracy and thus improve the fusion effect of CT and MRI images.

[0054] Furthermore, by calculating the distance between the registered CT image and MRI image at known anatomical landmarks, if multiple landmarks have large errors, it indicates that the registration accuracy is insufficient. In this case, the registration accuracy is improved by reducing the density of the preset markers. Attached Figure Description

[0055] Figure 1This is a schematic diagram of the structure of the image recognition-based intelligent medical image evaluation system according to an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the image segmentation module according to an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of the image evaluation module according to an embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram of the optimized feedback module in an embodiment of the present invention. Detailed Implementation

[0059] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0060] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0061] It should be noted that in the description of this invention, the terms "upper", "lower", "left", "right", "inner", "outer", etc., which indicate directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings. This is only for the convenience of description and is not intended to indicate or imply that the device or element must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, it should not be construed as a limitation of this invention.

[0062] Furthermore, it should be noted that, in the description of this invention, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this invention according to the specific circumstances.

[0063] Please see Figure 1 As shown, this is a structural schematic diagram of an image recognition-based intelligent medical image assessment system according to an embodiment of the present invention. The present invention provides an image recognition-based intelligent medical image assessment system, comprising:

[0064] The image registration module is used to register the input original CT image and MRI image based on a preset marker density, and output the registered CT image and MRI image.

[0065] An image segmentation module, which is connected to the image registration module, is used to select the corresponding segmentation method according to the category of the registered CT image to segment the registered CT image and obtain several regions to be mosaicked.

[0066] A preliminary fusion module, which is connected to the image segmentation module, is used to mosaic any region to be mosaicked into the MRI image to form a mosaicked region, so as to generate a preliminary fusion image based on each mosaicked region.

[0067] An image evaluation module, connected to the preliminary fusion module, is used to calculate the edge overlap and edge blanking based on the contour edge pixels of the region to be mosaicked, and to analyze the preliminary fused image based on the edge overlap and edge blanking to obtain diagnostic results.

[0068] The optimization feedback module, together with the image evaluation module, is used to optimize the segmentation method based on the edge overlap and the category of the registered CT image, and to correct the preset marker density based on the edge blankness, so as to generate a secondary fused image.

[0069] An image output module, which is connected to the image evaluation module and the optimization feedback module respectively, is used to use the preliminary fusion image as the final fusion image or the secondary fusion image as the final fusion image based on the diagnostic results.

[0070] In this embodiment, the soft tissue background is extracted from the original MRI image and used as the MRI image. The CT and MRI images are registered to ensure spatial alignment. Then, the region of interest is segmented and fused, i.e., the bone contour from the CT image is superimposed onto the MRI image. By analyzing the degree of overlap and gaps between the bone contour and the anatomical structure edges of the MRI image, the naturalness of the fused image at the bone-soft tissue interface is determined, and whether there are obvious jagged edges or artifacts. The registration or segmentation parameters are then optimized to maximize the preservation of spectral and texture information of both images and ensure a good fusion effect.

[0071] In the registration process of this embodiment, based on the preset initial marker density, the MRI image is used as a reference and remains fixed. The original CT image needs to be transformed to match the fixed MRI image, that is, stable anatomical landmarks such as the sella turcica and anterior commissure are automatically extracted to form a feature point set. Subsequently, through optimization algorithms, such as minimizing the mean square error, an optimal spatial transformation model is calculated, starting from rigid body or affine transformation, and this transformation is applied to the entire CT image. Interpolation technology is used to generate a registered CT image that is spatially precisely aligned with the MRI image. At the same time, the system uses the high-density bone information in the registered CT image, such as through threshold segmentation, to create a binary mask that highlights the bone region, and applies it to the original CT image, thereby generating a registered CT image that retains only the key bone structure and filters out redundant soft tissue information. Finally, the registered CT image and the original MRI image as the background reference are output, which together provide accurate and clean input data for the subsequent image segmentation, fusion, and intelligent evaluation modules. Through the subsequent evaluation feedback module, the marker density can also be dynamically optimized, achieving an adaptive balance between registration accuracy and computational efficiency.

[0072] During the fusion process, each organ or tissue is first identified, and then the information of different modalities is correlated. The registered CT and MRI images are then precisely segmented. For CT images, structures that are clearly visible on CT, such as bones (skull) and calcifications, are segmented. For MRI images, soft tissues such as gray matter, white matter, cerebrospinal fluid, and tumors are segmented. If the images are precisely registered, the outline of the "skull" segmented from CT and the outline of the "brain parenchyma" segmented from MRI are naturally aligned in space. However, since motion artifacts caused by patient movement during the scanning process may exist in the original CT images, the edge overlap and edge blankness are analyzed to improve the segmentation accuracy under artifact and low contrast conditions when training the segmentation network, thereby improving the fusion effect.

[0073] Specifically, the image segmentation module includes an image classification unit and an image segmentation unit;

[0074] The image classification unit is used to classify registered CT images into healthy images and disease images;

[0075] The image segmentation unit is used to segment the lesion image using a first segmentation method and to segment the healthy image using a second segmentation method.

[0076] Specifically, the image segmentation unit includes a first segmentation subunit and a second segmentation subunit;

[0077] The first segmentation subunit is used to directly segment the skeletal region of the health image using a first segmentation method;

[0078] The second segmentation subunit is used to segment the lesion image in a second segmentation manner after the edge of the target segmentation contour is gradient processed.

[0079] Specifically, the process of applying a gradient to the edges of the target segmentation contour of the lesion image is as follows:

[0080] Radial gradient transition processing is applied to each gradient transition layer at the edge of the target segmentation contour;

[0081] The determination of the number of radial gradient transition layers includes: in the target segmentation contour edge region of the lesion image, obtaining the contour curve segment as the initial radial gradient transition layer, obtaining each pixel point constituting the contour curve segment as the edge center point, calculating the CT value difference between the adjacent inner and outer layer pixels and the edge center point from each edge center point, obtaining several real-time CT value differences, if there is a real-time CT value difference greater than the preset CT value difference, then selecting the layer where the pixel point is located as the radial gradient transition layer, and accumulating the initial number of radial gradient transition layers;

[0082] The radial gradient transition process involves adjusting the CT value of each gradient transition layer using a stacking algorithm.

[0083] The overlay algorithm employs a hybrid overlay mode, adjusting the normalized CT value of the previous layer based on the normalized CT value of the current layer. Specifically, if the normalized CT value of the current layer is less than 0.5, the normalized CT value of the previous layer is multiplied by 2 to enhance the contrast of the CT value of the previous layer. If the normalized CT value of the current layer is greater than or equal to 0.5, the normalized CT value of the previous layer is multiplied by the sum of the two normalized CT values ​​minus the product of the two normalized CT values, thereby reducing the contrast of the CT value of the previous layer. This calculation is performed layer by layer to achieve a natural and gradual transition of lesion boundaries.

[0084] The CT value normalization process in this embodiment includes obtaining the minimum and maximum CT values ​​within the target segmentation contour region in the lesion image, linearly mapping the actual CT values ​​of each pixel to the [0,1] interval based on the minimum and maximum CT values, and obtaining the corresponding normalized CT values. The preset CT value difference is a key threshold parameter used to determine whether to extend the radial gradient transition layer. It determines the sensitivity and accuracy of the lesion boundary gradient processing. In CT images, the gray value is the HU value, and the preset CT value difference represents the difference in HU values. The setting value of the preset CT value difference is related to the lesion type and is dynamically adjusted according to the lesion type. A smaller preset CT value difference is used for invasive lesions, and a larger preset CT value difference is used for lesions with clear boundaries. For example, for invasive lesions, due to their blurred boundaries, the preset CT value difference is set between 8 and 12. For lesions with clear boundaries such as meningiomas, schwannomas, and pituitary adenomas, it is set between 18 and 25. When the lesion type is unknown, it is set to 15.

[0085] Achieving a natural transition through contour gradation aligns with the biological characteristics of diseased tissue, better captures areas with uncertain boundaries, provides a more natural boundary basis for subsequent segmentation operations, and reduces missed segmentation.

[0086] See Figure 3 As shown, it is a schematic diagram of the image evaluation module in an embodiment of the present invention;

[0087] Specifically, the image evaluation module includes a first computing unit and a second computing unit;

[0088] The first calculation unit is used to calculate the ratio of the number of pixels whose contour edges of the region to be mosaicked coincide with the edges of the MRI image to the total number of pixels of the contour edges of the region to be mosaicked, and obtain the edge overlap degree.

[0089] The second calculation unit is used to calculate the ratio of the number of pixels in the MRI image whose anatomical structure edges are not covered by the area to be mosaicked to the total number of pixels in the MRI image's anatomical structure edges, thus obtaining the edge blanking degree.

[0090] In this embodiment, the edges of the segmented regions in the registered CT image, i.e. the regions to be mosaicked, are extracted. The edge overlap is calculated by counting the number of pixels in the MRI image where the edge of the region to be mosaicked falls and the total number of edge pixels in the region to be mosaicked. The contour edges are extracted from the binary mask, and the boundaries are extracted using morphological gradients. The edge blanking is calculated by extracting the anatomical structure edges in the MRI image, counting the pixels where the anatomical structure edges fall outside the edge of the region to be mosaicked, and counting the total number of edge pixels in the MRI image. The region to be mosaicked is the skeletal region in the CT image, with a standard overlap of 0.9 and a standard blanking of 0.1.

[0091] In this embodiment, the edge overlap degree and the standard overlap degree are compared:

[0092] If the edge overlap is greater than the standard overlap, an oversegmentation diagnosis result is obtained;

[0093] If the edge overlap is less than or equal to the standard overlap, compare the edge blankness with the standard blankness:

[0094] If the edge blankness is greater than the standard blankness, the registration accuracy is verified, and the re-registration process is determined based on the verification results.

[0095] If the edge blankness is less than or equal to the standard blankness, the segmentation quality is acceptable.

[0096] When the edge blankness is greater than the standard blankness, it may be due to image occlusion causing some bones to be missed during CT image segmentation, or it may be due to insufficient registration accuracy. Based on the verification results, it is determined whether to trigger the re-registration process to improve the registration accuracy, thereby improving the fusion effect of CT and MRI images.

[0097] Specifically, the image evaluation module further includes a problem diagnosis unit and an evaluation unit;

[0098] The problem diagnosis unit is used to compare the edge overlap with the standard overlap, and to compare the edge blankness with the standard blankness.

[0099] The evaluation unit is used to obtain an oversegmentation diagnosis result based on the comparison between edge overlap and standard overlap, and to obtain a segmentation quality qualified result based on the comparison between edge blankness and standard blankness.

[0100] See Figure 4 As shown, it is a schematic diagram of the structure of the optimization feedback module in an embodiment of the present invention;

[0101] Specifically, the optimization feedback module includes a segmentation parameter optimization unit;

[0102] The segmentation parameter optimization unit is used to generate a secondary fusion image in response to the oversegmentation diagnosis result, according to the optimization processing method;

[0103] The optimization process involves increasing the preset CT value difference by a step size of 5HU, re-performing the gradient process and the second segmentation, until the overlap is within an acceptable range, and then adjusting the upper limit of the preset CT value difference to the maximum number of iterations.

[0104] In this embodiment, the maximum number of iterations is 5. The overlap is within an acceptable range, meaning that the edge overlap is less than or equal to the standard overlap. If the maximum number of iterations is reached and the edge overlap is still greater than the standard overlap, an early warning will be issued.

[0105] When oversegmentation is detected, it means that soft tissue near the bone has also been segmented. In this case, the contour edge of the bone region is enhanced to reduce the segmented area and make the edge fit the actual bone better. The segmentation result is adjusted according to the diagnosis result, and then the adjusted bone is inlaid into the MRI image to form the final fused image.

[0106] Specifically, the optimization feedback module also includes a registration verification unit;

[0107] The registration verification unit is used to verify the registration accuracy when the edge blankness is greater than the standard blankness, and to determine whether to reduce the preset marker point density based on the verification result in order to trigger the re-registration process.

[0108] Specifically, the registration and verification unit includes a first error calculation subunit, a marking subunit, a second error calculation subunit, and a parameter correction subunit;

[0109] The first error calculation subunit is used to calculate the distance between the registered CT image and the MRI image at anatomical landmarks as the actual distance, and to calculate the difference between the actual distance and the standard distance to obtain the actual error;

[0110] The marking subunit is used to mark the corresponding anatomical landmark as an error anatomical point when the actual error is greater than the allowable error;

[0111] The second error calculation subunit is used to calculate the percentage of error anatomical points to total anatomical landmarks, and obtain the actual error rate;

[0112] The parameter correction subunit is used to reduce the preset marker density to the corrected marker density when the actual error rate is greater than the standard error rate.

[0113] Wherein, the corrected marker density = max{minimum marker density, preset marker density × [1 - (actual error rate - standard error rate) / actual error rate]}.

[0114] In this embodiment, the minimum marker density is 4 pixels, and the preset marker density is set between 16 and 32 pixels, preferably 16 pixels. When the registration error rate exceeds the standard error rate, the preset marker density is corrected according to the formula. For example, if the preset marker density is 16, the actual error rate is 20%, and the standard error rate is 10%, then the correction value is: max{4, 16×[1-(20%-10%) / 20%]}=8.

[0115] The registration accuracy is verified when the edge blankness is greater than the standard blankness. This involves calculating the distance between the registered CT and MRI images at known anatomical landmarks. If multiple landmarks have large errors, the registration accuracy is insufficient. The registration accuracy is improved by reducing the density of the preset markers. When the actual error rate is less than or equal to the standard error rate, the registration is considered accurate. In this case, some bones may be missed when segmenting the CT image due to image occlusion. In this case, manual verification is performed before the image is put into the process.

[0116] Specifically, the first segmentation method is,

[0117] Obtain the CT values ​​of all pixels within the lesion contour region in the registered CT image and store them as an initial lesion value list;

[0118] Calculate the average CT value of any region within the lesion contour to obtain the corresponding average lesion CT value;

[0119] Obtain the CT values ​​of all pixels in the normal tissue surrounding each lesion outline and store them as a background CT value list;

[0120] Calculate the average CT value of the region outside the lesion outline to obtain the average background CT value;

[0121] The difference between the average background CT value and the average lesion CT value is obtained to determine the actual CT value difference.

[0122] Update each initial CT value of each lesion contour in the registered CT image using the corrected CT value;

[0123] Specifically, for each initial CT value, the sum of the differences between the initial CT value and the actual CT value is calculated to obtain the corrected CT value; the average lesion CT value is the ratio of the initial CT value of each pixel within the lesion contour to the number of pixels; the average background CT value is the ratio of the initial CT value of each pixel outside the lesion contour to the number of pixels.

[0124] Specifically, the second segmentation method is as follows:

[0125] The CT values ​​of all pixels within the contour region of each brain tissue in the registered CT image are obtained and recorded as the first actual CT value.

[0126] The CT values ​​of all pixels outside the contour of each brain tissue in the registered CT image are obtained and recorded as the second actual CT value.

[0127] The average CT value corresponding to the pixel point whose first actual CT value is less than the first average contour CT value is obtained as the second average contour CT value.

[0128] The average CT value corresponding to the pixel point whose second actual CT value is less than the first average out-of-contour CT value is obtained as the second average out-of-contour CT value.

[0129] The average CT value corresponding to the pixel point whose first actual CT value is greater than the first average contour CT value is obtained as the third average contour CT value.

[0130] The average CT value corresponding to the pixel point whose second actual CT value is greater than the first average out-of-contour CT value is obtained as the third average out-of-contour CT value.

[0131] Calculate the square of the CT value within each average contour and the corresponding square of the CT value outside the average contour to obtain the real-time CT value difference.

[0132] The target CT value is obtained by adding the differences between the CT values ​​within each average contour and the corresponding actual CT values.

[0133] Adjust the original CT values ​​of the registered CT images to the corresponding target CT values.

[0134] The technical solution of the present invention has been described above with reference to the preferred embodiments shown in the accompanying drawings. However, it will be readily understood by those skilled in the art that the scope of protection of the present invention is obviously not limited to these specific embodiments. Without departing from the principles of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will all fall within the scope of protection of the present invention.

Claims

1. An image recognition-based intelligent medical image evaluation system, characterized by, The application relates to a medical image processing system, comprising: an image registration module, configured to register inputted original CT images and MRI images based on a preset marker point density, and output registered CT images and MRI images; the preset marker point density is set as 16 pixels; an image segmentation module, connected with the image registration module, configured to select a corresponding segmentation mode according to the category of the registered CT images, to segment the registered CT images, and to obtain a plurality of to-be-inlaid regions; a preliminary fusion module, connected with the image segmentation module, configured to inlay any to-be-inlaid region into the MRI images to form an inlaid region, and to generate a preliminary fusion image based on the inlaid region; an image evaluation module, connected with the preliminary fusion module, configured to calculate an edge coincidence degree and an edge blankness degree according to the contour edge pixels of the to-be-inlaid region, and to analyze the preliminary fusion image according to the edge coincidence degree and the edge blankness degree, to obtain a diagnosis result; an optimization feedback module, connected with the image evaluation module, configured to optimize the segmentation mode based on the edge coincidence degree and the category of the registered CT images, and to correct the preset marker point density based on the edge blankness degree, to generate a secondary fusion image; an image output module, connected with the image evaluation module and the optimization feedback module respectively, configured to take the preliminary fusion image as a final fusion image or take the secondary fusion image as the final fusion image according to the diagnosis result. 2.The image recognition based medical image intelligent evaluation system according to claim 1, characterized in that, The image segmentation module comprises an image classification unit and an image segmentation unit; the image classification unit is configured to divide the registered CT images into healthy images and lesion images; the image segmentation unit is configured to segment the lesion images in a first segmentation mode and to segment the healthy images in a second segmentation mode. 3.The image recognition based medical image intelligent evaluation system according to claim 2, characterized in that, The image segmentation unit comprises a first segmentation subunit and a second segmentation subunit; the first segmentation subunit is configured to directly segment the bone region of the healthy images in the first segmentation mode; the second segmentation subunit is configured to segment the lesion images in the second segmentation mode after performing gradient processing on the edge of a target segmentation contour of the lesion images. 4.The image recognition based medical image intelligent evaluation system according to claim 1, wherein, The image evaluation module comprises a first calculation unit and a second calculation unit; the first calculation unit is configured to calculate the ratio of the number of pixels of the edge coincidence of the contour edge of the to-be-inlaid region and the total number of pixels of the contour edge of the to-be-inlaid region, to obtain the edge coincidence degree; the second calculation unit is configured to calculate the ratio of the number of pixels of the edge blankness of the anatomical structure of the MRI images which is not covered by the to-be-inlaid region and the total number of pixels of the anatomical structure edge of the MRI images, to obtain the edge blankness degree. 5.The image recognition based medical image intelligent evaluation system according to claim 4, characterized in that, The image evaluation module further comprises a problem diagnosis unit and an evaluation unit; the problem diagnosis unit is configured to compare the edge coincidence degree with a standard coincidence degree, and to compare the edge blankness degree with a standard blankness degree; the evaluation unit is configured to obtain an over-segmentation diagnosis result according to the comparison result of the edge coincidence degree and the standard coincidence degree, and to obtain a segmentation quality qualified result according to the comparison result of the edge blankness degree and the standard blankness degree. 6.The image recognition based intelligent medical image assessment system according to claim 1, wherein, The optimization feedback module comprises a segmentation parameter optimization unit. The segmentation parameter optimization unit is configured to generate a secondary fusion image according to an optimization processing mode in response to an over-segmentation diagnosis result. 7.The image recognition based intelligent medical image assessment system according to claim 1, wherein, The optimization feedback module further comprises a registration verification unit; The registration verification unit is configured to verify the registration accuracy when the edge blankness is greater than the standard blankness, and determine whether to reduce the preset marker point density according to a verification result to trigger a re-registration process. 8.The image recognition based medical image intelligent assessment system according to claim 7, characterized in that, The registration verification unit comprises a first error calculation subunit, a marker subunit, a second error calculation subunit, and a parameter correction subunit. The first error calculation subunit is configured to calculate the distance between the registration CT image and the MRI image at the anatomical landmark point as an actual distance, and calculate the difference between the actual distance and the standard distance to obtain an actual error. The marker subunit is configured to mark the corresponding anatomical landmark point as an error anatomical point when the actual error is greater than the allowable error. The second error calculation subunit is configured to calculate the percentage of error anatomical points in total anatomical landmark points to obtain an actual error rate. The parameter correction subunit is configured to reduce the preset marker point density to a corrected marker point density when the actual error rate is greater than a standard error rate. 9.The image recognition based medical image intelligent assessment system according to claim 2, wherein, The first segmentation mode is, Obtain the CT values of all pixel points in each lesion contour region in the registration CT image and store them as an initial lesion value list; Calculate the average CT value of any of the lesion contour regions to obtain a corresponding average lesion CT value; Obtain the CT values of all pixel points of the surrounding normal tissue outside each lesion contour and store them as a background CT value list; Calculate the average CT value of the region outside the lesion contour to obtain an average background CT value; Obtain the difference between the average background CT value and the average lesion CT value to obtain an actual CT value difference; Update each initial CT value of each lesion contour in the registration CT image using the corrected CT value; Wherein, for each initial CT value, the sum of the initial CT value and the actual CT value difference is calculated to obtain the corrected CT value; the average lesion CT value is the ratio of the initial CT value of each pixel point in the lesion contour to the number of pixel points; and the average background CT value is the ratio of the initial CT value of each pixel point outside the lesion contour to the number of pixel points. 10.The image recognition based medical image intelligent evaluation system according to claim 2, wherein, The second segmentation mode is, Obtain the CT values of all pixel points in each brain tissue contour region in the registration CT image, denoted as a first actual CT value; Obtain the CT values of all pixel points in each brain tissue contour region outside the registration CT image, denoted as a second actual CT value; Obtain the average CT value corresponding to the pixel point whose first actual CT value is less than the first average contour CT value as a second average contour CT value; Obtain the average CT value corresponding to the pixel point whose second actual CT value is less than the first average contour CT value as a second average contour CT value; Obtain the average CT value corresponding to the pixel point whose first actual CT value is greater than the first average contour CT value as a third average contour CT value; Obtain the average CT value corresponding to the pixel point whose second actual CT value is greater than the first average contour CT value as a third average contour CT value; Calculate the square of each average contour CT value and the corresponding average contour CT value to obtain the corresponding real-time CT value difference; The target CT value is obtained by adding the CT value in each average profile to the corresponding actual CT value difference; Each original CT value of the registered CT image is adjusted to the corresponding target CT value.

Images