Identification processing method and system for medical image file

Abstract

The invention relates to the technical field of medical image file processing, in particular to an identification processing method and system for medical image files. According to the invention, typical occlusion characteristic data and target organization characteristic data are analyzed through the preset occlusion risk identification unit to obtain a boundary risk value, a space risk value, a signal risk value and a registration risk value, then the comprehensive occlusion risk value is obtained through comprehensive calculation, and the image occlusion state is judged based on the comprehensive occlusion risk value. Focus omission or misdiagnosis caused by shielding is avoided, and the image recognition accuracy is improved; according to the method, a feature completion unit is used for training and generating a structure completion mapping model by using historical occlusion data, structure completion is performed on an occlusion region image, and complete features of a target tissue are recovered; and meanwhile, the task scheduling management module divides execution queues based on clinical urgency and data complexity, optimizes calculation resource allocation, improves processing efficiency, and provides high-timeliness diagnostic images for doctors.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of medical image file processing, in particular to a recognition processing method and system for medical image files. BACKGROUND

[0002] Medical images have undergone years of technological development, gradually moving from morphological observation to precise analysis at the functional and molecular levels. With the development of AI, AI models are increasingly used in medical images, driving the intelligent development of medical images. The recognition and processing of medical image files have also evolved from early rule-based image processing to the use of generative AI to analyze and synthesize visual diagnostic criteria.

[0003] Existing recognition processing methods for medical image files often miss lesions or misdiagnose problems in chest image diagnosis due to occluded areas in the image. For example, ribs in chest images can obscure lung tissue and mediastinal structures, and bones in pelvic images can obscure organs. Although existing recognition models can improve automation, they may not fully extract lesion characteristics due to feature overlap between the occluding object and the target tissue in occluded scenarios. SUMMARY

[0004] The present application provides a recognition processing method and system for medical image files to solve the technical problems raised in the background.

[0005] The first aspect of the present application provides a recognition processing method for medical image files, comprising the following steps: Step 1: Collecting data from the medical image files to be processed using a pre-set data extraction module to obtain multi-modal image data, and performing step 2 on the multi-modal image data; Step 2: Inputting the multi-modal image data into a pre-set occlusion risk identification unit, analyzing the occlusion risk of the multi-modal image data to obtain occlusion state information, identifying the occlusion state information, and marking the corresponding multi-modal image data as incomplete optimization data when the occlusion state information corresponds to occlusion, and performing step 3 on the incomplete optimization data; when the occlusion state information corresponds to no occlusion, marking the corresponding multi-modal image data as no-occlusion multi-modal image, and performing step 4 on the no-occlusion multi-modal image; As a further improvement of the present application, the occlusion risk identification unit analyzes the occlusion risk of the multi-modal image data to obtain occlusion state information, which specifically includes: Obtaining typical occlusion feature data and target tissue feature data from the incomplete optimization data; The typical feature data includes occlusion boundary parameters and occlusion object space parameters corresponding to a typical occlusion region of the target tissue image region, and the boundary risk value and the space risk value are obtained by analyzing the occlusion boundary parameters and the occlusion object space parameters. The target tissue feature data includes modal signal response parameters and space registration parameters, and the signal risk value and the registration risk value are obtained by analyzing the modal signal response parameters and the space registration parameters. The boundary risk value, the space risk value, the signal risk value and the registration risk value are all normalized, and the numerical values are taken to obtain a preset weighted sum formula The comprehensive occlusion risk value Q is calculated; wherein, pz represents the registration risk value. represents the preset weight factor corresponding to the qth risk value; when the comprehensive occlusion risk value corresponding to the target tissue image region exceeds the preset comprehensive occlusion risk threshold, the occlusion state information is generated as occluded, otherwise, it corresponds to no occlusion.

[0006] Further, the boundary risk value and the space risk value are obtained by analyzing the occlusion boundary parameters and the occlusion object space parameters, which are specifically: Based on the occlusion boundary parameters, a preset unit length is obtained, the occlusion boundary is divided into a plurality of boundary segments according to the unit length, the number of boundary fracture points corresponding to each boundary segment is obtained, when the number of boundary fracture points exceeds the set upper limit of the number of fracture points, the corresponding boundary segment is recorded as a fracture boundary, the number of fracture boundaries is counted to obtain the fracture boundary number, and the fracture boundary number is compared with the total number of boundary segments to obtain the boundary fracture density; and the deviation of the occlusion boundary from the preset ideal smooth boundary is identified to obtain the boundary deviation degree; the occlusion boundary is input into the box dimension method to obtain the boundary fractal dimension, and the reference fractal dimension corresponding to the normal non-occluded region of each type of region is obtained, and the fractal dimension is valued as [1, 2]; The boundary fracture density and the boundary deviation degree are normalized, and the formula The boundary risk value bj is calculated; wherein, FX and FX0 represent the boundary fractal dimension and the reference fractal dimension, respectively. and represent the normalized values of the boundary fracture density and the boundary deviation degree, respectively.

[0007] A unique space coordinate system is established for the target tissue image region, the occlusion object region is divided into a plurality of triangular units based on the Delaunay triangulation method, the actual area corresponding to each triangular unit is obtained, the difference between the actual area and the preset ideal non-occluded area is calculated, and the absolute value of the difference is recorded as the area deviation value; the unit gravity coordinates are obtained based on the unique space coordinate system through the occlusion object space parameters, and the tissue center coordinates of the target tissue corresponding to the target tissue image region are obtained, and the maximum distance from the tissue center coordinates to the edge of the occlusion object is obtained, and the formula The unit weight corresponding to the kth triangular element is calculated ; wherein, respectively represent the element barycenter coordinates and the tissue center coordinates corresponding to the kth triangular element; corresponding to the maximum distance from the tissue center coordinates to the edge of the shield; The actual area and the area deviation value are normalized, and the normalized value and the unit weight are substituted into the preset weighted Delaunay triangulation deviation model The spatial risk value kj is calculated; wherein, respectively represent the actual area and the area deviation value normalized corresponding to the kth triangular element.

[0008] Further, the signal risk value and the registration risk value are obtained by analyzing the modal signal response parameters and the spatial registration parameters, which are specifically: According to the modal signal response parameters, the signal intensity of each type of modal signal corresponding to the same target tissue is obtained, the signal intensity corresponding to any two types of modal signals is calculated by difference to obtain the cross-modal signal intensity difference value, the standard signal intensity difference value of each type of target tissue corresponding to any two types of modal signals is obtained, and the cross-modal signal intensity difference value and the standard signal intensity difference value are calculated by ratio to obtain the cross-modal signal deviation value; the signal similarity is obtained by calculating the signal feature vectors under each type of modal by the cosine similarity calculation method, and the value of the signal similarity is marked as the modal consistency index; The cross-modal signal reference deviation value corresponding to the preset signal normal state is obtained, and the cross-modal signal deviation value, the modal consistency index, and the cross-modal signal reference deviation value are substituted into the preset distortion coefficient calculation formula The distortion coefficient J is calculated xh ; and the distortion coefficient and the cross-modal signal deviation value are substituted into the preset tanh function coupling formula The signal risk value is calculated; wherein, respectively represent the cross-modal signal deviation value and the cross-modal signal reference deviation value; represents the modal consistency index.

[0009] Further, the registration risk value is obtained by comprehensively analyzing the local feature point data and the modal Euclidean distance data, which is specifically: According to the local feature point data, a local gradient direction histogram of the target tissue is obtained, and a preset number of local feature points are extracted based on the local gradient direction histogram; a structure position corresponding to each local feature point is obtained, the structure position including a preset core region and an edge region of the target tissue, a structure importance coefficient is obtained according to the structure position of the local feature point, and a preset coefficient value range of the structure importance coefficient is divided into a high importance range and a low importance range; when the structure position corresponds to the core region, the value range of the structure importance coefficient corresponds to the high importance range; and when the structure position corresponds to the edge region, the value range of the structure importance coefficient corresponds to the low importance range. The modality to be registered and the preset reference modality of each local feature point are obtained, and the descriptor included angle on the local gradient direction histogram of the modality to be registered and the reference modality is obtained, and the descriptor included angle cosine value is obtained by a cosine function; the modality to be registered is defined as a target image modality that may exist occlusion, and the spatial deviation obtained by aligning with the reference modality, which directly reflects the position distortion caused by occlusion; the reference modality is defined as a reference image modality without occlusion, low noise and clear structure, and the corresponding spatial position and tissue contour are used to calibrate the spatial deviation of other modalities; the structure importance coefficient and the descriptor included angle cosine value are calculated to obtain a local feature consistency score.

[0010] The modality types corresponding to the multiple modalities participating in the spatial registration are obtained, each modality type is numbered to obtain a modality index number, a preset number of feature matching pairs corresponding to the modality index number are obtained, and the preset number of feature matching pairs is less than or equal to the preset number of feature points; based on the modality Euclidean distance data, the modality Euclidean distance of each feature matching pair corresponding to each modality index number is obtained, and the modality adaptive registration error value is obtained by registration error calculation on the modality Euclidean distance corresponding to each modality index number.

[0011] The local feature consistency score and the modality adaptive registration error value are substituted into a preset spatial registration parameter calculation formula to obtain a spatial registration value, and when the spatial registration value exceeds a preset standard registration value range, the part of the spatial registration value exceeding the standard registration value range is marked as a registration risk value.

[0012] Step 3: Obtain the to-be-completed optimization data and input the to-be-completed optimization data into a preset feature completion unit, the feature completion unit performs completion analysis on the to-be-completed optimization data to obtain completed image data, and step 4 is performed on the completed image data; As a further improvement of the present application, the feature completion unit performs completion analysis on the to-be-completed optimization data to obtain completed image data, which specifically comprises: the feature completion unit obtains historical occlusion data, trains a structure completion mapping model based on the historical occlusion data, identifies the to-be-completed optimization data to obtain an occlusion risk area, inputs the multi-modal image data corresponding to the occlusion risk area into the structure completion mapping model for matching, and generates the completed image data.

[0013] Step 4: receiving unoccluded multi-modal image and completed image data, and obtaining corresponding medical diagnostic image based on the unoccluded multi-modal image and the completed image data, and delivering the medical diagnostic image to the corresponding physician work terminal.

[0014] The second aspect of the present application provides a medical image file recognition processing system, comprising: The task data extraction module is used for extracting the medical image files corresponding to each patient to obtain corresponding image task packages.

[0015] The task scheduling management module is used for obtaining the image task packages of each patient, and inputting each image task package into a preset task scheduler, and the task scheduler performs scheduling analysis on each image task package to obtain a corresponding execution label, and performs recognition processing and execution on the image task package based on the execution label to obtain image data.

[0016] The task scheduler performs scheduling analysis on each image task package to obtain a corresponding execution label, which specifically comprises: obtaining image clinical information and image size data based on the image task package; obtaining an examination description word from the image clinical information, dividing an examination description word library into a high emergency word library and a low emergency word library, matching the current corresponding examination description word with the examination description word library to obtain a high emergency word number, taking the high emergency word number as a clinical emergency value, when the clinical emergency value is greater than 0, generating a high emergency clinical state, when the clinical emergency value corresponds to 0, generating a low emergency clinical state; obtaining an image matrix and an image slice number based on the image size data, dividing the image matrix into a high matrix interval and a low matrix interval based on a preset matrix division value, and similarly, dividing the image slice number into a high slice interval and a low slice interval based on a preset slice division value; when the current image size data corresponds to the high matrix interval and the high slice interval, generating a data size signal as a high complexity signal; otherwise, generating a data size signal as a low complexity signal; generating execution queue information based on the clinical state and the size signal corresponding to each image task package, that is, when the clinical state corresponds to a high emergency and a high complexity signal, generating the execution queue information as a priority calculation queue; when the clinical state corresponds to a high emergency and a low complexity signal, generating the execution queue information as a real-time processing queue; When the clinical state corresponds to a low-emergency, low-complexity signal, the execution queue information is generated as a batch queue; An execution label is generated based on the execution queue information corresponding to the image task package.

[0017] The recognition result output module is configured to obtain image data corresponding to the recognition processing execution, obtain patient information corresponding to each image data, and output the image data to a patient information terminal based on the patient information.

[0018] Compared with the prior art, the technical scheme provided by the present application has the beneficial effects that: 1. The preset occlusion risk identification unit analyzes the typical occlusion feature data and the target tissue feature data to obtain a boundary risk value, a spatial risk value, a signal risk value and a registration risk value, and then obtains a comprehensive occlusion risk value through comprehensive calculation, and determines the image occlusion state based on the comprehensive occlusion risk value, thereby avoiding lesion omission or misdiagnosis caused by occlusion and improving the image recognition accuracy.

[0019] 2. The feature completion unit uses historical occlusion data to train a structure completion mapping model to complete the structure of the occlusion area image and restore the complete features of the target tissue, and the task scheduling management module divides the execution queue based on the clinical emergency degree and data complexity, optimizes the allocation of computing resources, improves the processing efficiency, and provides high-time-efficiency diagnostic images for doctors. DETAILED DESCRIPTION

[0020] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description, and the following drawings are not deliberately scaled according to the actual size, and the focus is to show the main idea of the present application.

[0021] Figure 1 A method flowchart of a medical image file recognition processing method of the present application; Figure 2 A principle block diagram of a medical image file recognition processing system of the present application. DETAILED DESCRIPTION

[0022] The technical solutions in the embodiments of the present application will be described clearly and completely in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0023] For the convenience of understanding, the specific process of the embodiments of the present application will be described below, please refer to Figure 1In an embodiment of the present application, one embodiment of a medical image file recognition processing method comprises: Step 1. Multi-modal medical image data input: The data extraction module collects multi-modal image data from the medical image file to be processed, and executes step 2 on the multi-modal image data. The medical image file includes but is not limited to CT, MRI, and PET-CT image data.

[0024] For example, the patient is a suspected lung cancer patient. The data extraction module retrieves the patient's chest CT image modal data (such as 1mm slice thickness, 512x512 matrix, containing lung structure details), lung MRI image modal data, and PET-CT metabolic image modal data from the hospital PACS system.

[0025] Step 2. Occlusion area identification and segmentation: input the multi-modal image data into the preset occlusion risk identification unit, the occlusion risk identification unit analyzes the multi-modal image data to obtain occlusion state information, identifies the occlusion state information, and when the occlusion state information corresponds to occlusion, marks the corresponding multi-modal image data as incomplete optimization data, and executes step 3 on the incomplete optimization data; when the occlusion state information corresponds to no occlusion, mark the corresponding multi-modal image data as no-occlusion multi-modal image, and execute step 4 on the no-occlusion multi-modal image.

[0026] The specific execution content of the occlusion risk identification unit is: A1: Obtain typical occlusion feature data and target tissue feature data from the incomplete optimization data; A2: The typical feature data includes the occlusion boundary parameter and the occlusion object space parameter of the typical occlusion area corresponding to the target tissue image area; the typical occlusion area is, for example, the rib-lung tissue and the bone-organ corresponding to the chest and the pelvic cavity; A21: Based on the occlusion boundary parameter, divide the occlusion boundary into multiple boundary segments according to the preset unit length, obtain the number of boundary fracture points corresponding to each boundary segment, when the number of boundary fracture points exceeds the set upper limit of the number of fracture points, mark the corresponding boundary segment as a fracture boundary, count the number of fracture boundaries to obtain the fracture boundary number, calculate the boundary fracture density by comparing the fracture boundary number with the total number of boundary segments; and identify the boundary deviation degree by comparing the occlusion boundary with the preset ideal smooth boundary; input the occlusion boundary into the box dimension method to obtain the boundary fractal dimension, and obtain the reference fractal dimension corresponding to the normal non-occlusion area corresponding to each type of area. The fractal dimension is in the range of [1, 2], that is, the closer the value is to 2, the more complex the boundary is; Normalize the boundary fracture density and boundary deviation degree, and use the formula The boundary risk value bj is calculated, wherein FX and FX0 represent the boundary fractal dimension and the reference fractal dimension, respectively. respectively represent the normalized values of the boundary fracture density and the boundary deviation degree.

[0027] A22: A unique spatial coordinate system is established for the target tissue image region, the occlusion region is divided into multiple triangular units based on the Delaunay triangulation method, the actual area corresponding to each triangular unit is obtained, the actual area is calculated from the occlusion pixels and voxels corresponding to the target tissue image, the difference between the actual area and the preset ideal unoccluded area is calculated, and the absolute value of the difference is recorded as an area deviation value; the unit gravity coordinates are obtained from the spatial parameters of the occlusion through the occlusion space parameter acquisition unit based on the unique spatial coordinate system, and are recorded as , and the tissue center coordinates of the target tissue corresponding to the target tissue image region are recorded as The maximum distance from the tissue center coordinates to the edge of the occlusion is obtained, and the formula is used to calculate the unit weight of the kth triangular unit ; wherein, respectively represent the unit gravity coordinates and the tissue center coordinates of the kth triangular unit. correspond to the maximum distance from the tissue center coordinates to the edge of the occlusion. The normalized values of the actual area and the area deviation value are taken, and the normalized values and the unit weight are substituted into the preset weighted Delaunay triangulation deviation model to calculate the spatial risk value kj; wherein, respectively represent the normalized values of the actual area and the area deviation value of the kth triangular unit.

[0028] For example, the scene of a patient's pelvic image in which the bladder is occluded by bone is obtained, and a spatial coordinate system is established with the center of the bladder as the origin. The bone occlusion region is divided into 50 triangular units using the Delaunay triangulation method. The actual area of a certain triangular unit is measured to be 8 mm², the preset ideal unoccluded area is 10 mm², and the area deviation value is 2 mm². The unit gravity coordinates are (20, 15, 5), the bladder center coordinates are (0, 0, 0), the maximum distance from the tissue center to the edge of the occlusion is 30 mm, the unit weight is calculated to be 0.25 by substituting the formula, and the spatial risk value is calculated to be 0.32 by combining the normalized area (i.e. 0.8) and the deviation value (i.e. 0.2) using the formula A3: The target tissue feature data includes modal signal response parameters and spatial registration parameters. A31: obtaining signal intensities of each type of modality signal corresponding to the same target tissue according to the modal signal response parameters, performing difference calculation on the signal intensities corresponding to any two types of modality signals to obtain a cross-modality signal intensity difference, obtaining a standard signal intensity difference corresponding to any two types of modality signals of each type of target tissue, performing ratio calculation on the cross-modality signal intensity difference and the standard signal intensity difference to obtain a cross-modality signal deviation value; calculating the signal similarity by calculating the signal feature vectors under each type of modality by using the cosine similarity calculation method, and marking the numerical value of the signal similarity as a modality consistency index; obtaining a cross-modality signal reference deviation value corresponding to a preset signal normal state, substituting the cross-modality signal deviation value, the modality consistency index, and the cross-modality signal reference deviation value into a preset distortion coefficient calculation formula to obtain a distortion coefficient J xh ; and substituting the distortion coefficient and the cross-modality signal deviation value into a preset tanh function coupling formula to obtain a signal risk value; wherein, respectively represent the cross-modality signal deviation value and the cross-modality signal reference deviation value; represents the modality consistency index, and the lower the modality consistency index, the worse the signal consistency, which leads to distortion amplification. The meaning is that the cross-modality signal deviation value deviates from the cross-modality signal reference deviation value by an increased deviation, the modality consistency decreases, the corresponding sub-risk is higher, and the tanh function limits the value range of the signal risk value to [0, 1], and the square term is used to amplify extreme deviation.

[0029] A32: obtaining local feature point data and modality Euclidean distance data based on spatial registration parameters; obtaining a local gradient direction histogram of the target tissue according to the local feature point data, extracting a local feature point with a preset feature point number based on the local gradient direction histogram, obtaining a structure position corresponding to each local feature point, the structure position including a core region and an edge region of the target tissue, obtaining a structure importance coefficient according to the structure position of the local feature point, dividing a coefficient value range of the structure importance coefficient into a high importance range and a low importance range, when the structure position corresponds to the core region, the value range of the structure importance coefficient corresponds to the high importance range; when the structure position corresponds to the edge region, the value range of the structure importance coefficient corresponds to the low importance range; The modality to be registered and the preset reference modality of each local feature point are obtained, and the descriptor angle on the local gradient direction histogram of the modality to be registered and the reference modality is obtained, and the descriptor angle cosine value is calculated by the cosine function; the modality to be registered is defined as the target image modality that may exist occlusion (such as metal artifact, body fluid occlusion, foreign body occlusion), and the spatial deviation obtained by aligning with the reference modality directly reflects the position distortion caused by the occlusion; the reference modality is defined as the reference image modality without occlusion, low noise and clear structure, and the corresponding spatial position and tissue contour are used to calibrate the spatial deviation of other modalities; the local feature consistency score is calculated by the structure importance coefficient and the descriptor angle cosine value.

[0030] The modality types corresponding to the multi-modality participating in the spatial registration are obtained, each modality type is numbered to obtain a modality index number, a feature matching pair corresponding to the modality index number and a preset feature matching pair number are obtained, and the preset feature matching pair number is less than or equal to the preset feature point number; the modality Euclidean distance of each feature matching pair corresponding to each modality index number is obtained based on the modality Euclidean distance data, and the modality adaptive registration error value is calculated by the modality Euclidean distance corresponding to each modality index number.

[0031] The local feature consistency score and the modality adaptive registration error value are substituted into the preset spatial registration parameter calculation formula to calculate the spatial registration value, and when the spatial registration value exceeds the preset standard registration value interval, the part of the spatial registration value exceeding the standard registration value interval is marked as a registration risk value.

[0032] The formula used in the above calculation includes: ; ; ; Wherein, respectively represent the local feature consistency score, the modality adaptive registration error value and the spatial registration value; E represents the preset feature point number, e represents the local feature point, e=1, 2, E; respectively represent the structure importance coefficient and the descriptor angle cosine value; B represents the total number of modality index numbers, b represents the modality index number, b=1, 2, E, for example, b=1 represents CT, and b=2 represents MRI; represents the preset weight factor corresponding to the bth modality, which increases with the increase of the signal-to-noise ratio corresponding to the modality; represents the modality Euclidean distance of the a th feature matching pair in the b th modality, a=1, 2, A; MAD represents the median absolute deviation, which is used to measure the dispersion degree of the residual set formed by the modality Euclidean distance corresponding to each modality, and the smaller the value, the more concentrated the matching quality is; the larger the value, the more abnormal values.

[0033] A4: normalizing the boundary risk value, the spatial risk value, the signal risk value and the registration risk value, and taking the numerical value thereof to obtain a preset weighted sum formula to obtain a comprehensive occlusion risk value Q; wherein, pz represents the registration risk value; indicates a preset weight factor corresponding to the qth risk value; when the comprehensive occlusion risk value corresponding to the target tissue image region exceeds a preset comprehensive occlusion risk threshold, the occlusion state information is generated as occlusion, otherwise, it corresponds to no occlusion.

[0034] Step 3. Occlusion multi-view feature completion: obtaining the to-be-completed optimization data and inputting the to-be-completed optimization data into a preset feature completion unit, the feature completion unit performs completion analysis on the to-be-completed optimization data to obtain completed image data, and executes step 4 on the completed image data, which is: the feature completion unit obtains historical occlusion data, trains a generation structure completion mapping model based on the historical occlusion data, identifies the to-be-completed optimization data to obtain an occlusion risk region, inputs the multi-modal image data corresponding to the occlusion risk region into the structure completion mapping model for matching, and generates the completed image data.

[0035] Step 4. Medical image output: receiving the non-occlusion multi-modal image and the completed image data, obtaining the corresponding medical diagnostic image based on the non-occlusion multi-modal image and the completed image data, and delivering the medical diagnostic image to the corresponding physician work end.

[0036] Please refer to Figure 2 The application also provides a medical image file recognition processing system, which comprises a task data extraction module, a task scheduling management module and a recognition result output module.

[0037] The task data extraction module extracts the medical image files corresponding to each patient to obtain corresponding image task packages; the image task package is the image clinical information and image scale data of each patient; for example, the task data extraction module extracts the chest enhanced CT image file of the patient from the hospital PACS system, and simultaneously obtains the examination description words (such as emergency chest pain and suspected pulmonary embolism) in the clinical information and the image scale data (such as matrix 512x512 and slice number 300) in the electronic medical record; the image task package can also include patient ID and examination time.

[0038] The task scheduling management module obtains the image task packages of each patient, and inputs each image task package into a preset task scheduler; the task scheduler analyzes each image task package to obtain a corresponding execution label, and performs recognition processing and execution on the image task package based on the execution label to obtain image data.

[0039] The task scheduler, which specifically analyzes and schedules, is: Obtaining image clinical information and image size data based on the image task package; According to the image clinical information, the examination description word is obtained, the examination description word library is divided into a high emergency word library and a low emergency word library, the high emergency word corresponding to the high emergency word library includes but is not limited to emergency, aortic dissection, pulmonary embolism, and brain perfusion, the current corresponding examination description word is matched with the examination description word library to obtain the number of high emergency words, and the number of high emergency words is recorded as a clinical emergency value; when the clinical emergency value is greater than 0, a clinical state of high emergency is generated; when the clinical emergency value corresponds to 0, a clinical state of low emergency is generated.

[0040] Based on the image size data, the image matrix and the number of image slices are obtained, the image matrix is divided into a high matrix interval and a low matrix interval based on a preset matrix division value, and similarly, the number of image slices is divided into a high slice interval and a low slice interval based on a preset slice division value; when the current image size data corresponds to the high matrix interval and the high slice interval, a data size signal of high complexity signal is generated; otherwise, a data size signal of low complexity signal is generated.

[0041] Based on the clinical state and the size signal corresponding to each image task package, execution queue information is generated, that is, when the clinical state corresponds to high emergency and high complexity signal, the execution queue information is generated as a priority calculation queue, which corresponds to high priority and is allocated with a large amount of computing resources; when the clinical state corresponds to high emergency and low complexity signal, the execution queue information is generated as a real-time processing queue, which corresponds to the highest priority and is processed immediately; when the clinical state corresponds to low emergency and low complexity signal, the execution queue information is generated as a batch queue, which corresponds to low priority and is processed when the system is idle; and an execution label is generated based on the execution queue information corresponding to the image task package.

[0042] The recognition result output module obtains the image data corresponding to the recognition processing execution, obtains the patient information corresponding to each image data, and outputs the image data to the patient information end based on the patient information. The patient information end includes a patient personal information end and a patient corresponding doctor work end.

[0043] The above-described embodiments are only used to illustrate the technical solutions of the present application, but not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features thereof; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for recognizing and processing medical image files, characterized in that, include: Step 1: Collect multimodal image data from the medical image file to be processed using the preset data extraction module, and then perform Step 2 on the multimodal image data; Step 2: Input the multimodal image data into the preset occlusion risk identification unit. The occlusion risk identification unit performs occlusion risk analysis on the multimodal image data to obtain occlusion status information. The occlusion status information is identified. When the occlusion status information corresponds to occlusion, the corresponding multimodal image data is marked as data to be completed and optimized, and Step 3 is executed on the data to be completed and optimized. When the occlusion status information corresponds to no occlusion, the corresponding multimodal image data is marked as an occluded multimodal image, and step 4 is performed on the occluded multimodal image. Step 3: Obtain the data to be completed and optimized and input the data to be completed and optimized into the preset feature completion unit. The feature completion unit performs completion analysis on the data to be completed and optimized to obtain the completed image data, and then performs step 4 on the completed image data. Step 4: Receive unobstructed multimodal images and completed image data, obtain corresponding medical diagnostic images based on the unobstructed multimodal images and completed image data, and send the medical diagnostic images to the corresponding physician's workstation.

2. The method for recognizing and processing medical image files according to claim 1, characterized in that, The occlusion risk identification unit performs occlusion risk analysis on multimodal image data to obtain occlusion status information. Its specific execution content is as follows: Typical occlusion feature data and target tissue feature data are obtained from the data to be completed and optimized. Typical feature data include occlusion boundary parameters and occlusion object spatial parameters of typical occlusion areas corresponding to the target tissue image area. The boundary risk value and spatial risk value are obtained by analyzing the occlusion boundary parameters and occlusion object spatial parameters. The target organization characteristic data includes modal signal response parameters and spatial registration parameters. The signal risk value and registration risk value are obtained by analyzing the modal signal response parameters and spatial registration parameters. The boundary risk value, spatial risk value, signal risk value and registration risk value are all normalized, and their values ​​are used to calculate the comprehensive occlusion risk value using a preset weighted summation formula. When the overall occlusion risk value of the target tissue image area exceeds the preset overall occlusion risk threshold, the occlusion status information is generated as occluded; otherwise, it corresponds to unoccluded.

3. The method for recognizing and processing medical image files according to claim 2, characterized in that, The analysis of the occlusion boundary parameters and the spatial parameters of the occlusion object yields the boundary risk value and the spatial risk value, specifically as follows: Based on the occlusion boundary parameters, a preset unit length is obtained. The occlusion boundary is divided into multiple boundary segments according to the unit length. The number of boundary break points corresponding to each boundary segment is obtained. When the number of boundary break points exceeds the set upper limit of the number of break points, the corresponding boundary segment is recorded as a break boundary. The number of break boundaries is counted to obtain the number of break boundaries. The ratio of the number of break boundaries to the total number of boundary segments is calculated to obtain the boundary break density. Furthermore, the deviation between the occlusion boundary and the preset ideal smooth boundary is identified to obtain the boundary deviation degree. The occlusion boundary is input into the box dimension method to obtain the boundary fractal dimension, and the benchmark fractal dimension corresponding to the normal unoccluded area for each type of region is obtained. The boundary break density and boundary deviation degree are normalized and comprehensively calculated using a comprehensive formula to obtain the boundary risk value. A unique spatial coordinate system is established for the target tissue image region. Based on the Delaunay triangulation method, the occlusion region is subdivided into multiple triangular units. The actual area of ​​each triangular unit is obtained, and the difference between the actual area and the preset ideal unobstructed area is calculated. The absolute value of the difference is recorded as the area deviation value. Based on the unique spatial coordinate system, the centroid coordinates of the unit are obtained through the spatial parameters of the occlusion, as well as the coordinates of the tissue center corresponding to the target tissue image region. The maximum distance from the tissue center coordinates to the edge of the occlusion is also obtained. The unit weight corresponding to each triangular unit is calculated using a preset unit weight calculation formula. The normalized values ​​of the actual area and the area deviation value are taken, and the normalized values ​​and unit weights are substituted into a preset weighted Delaunay triangulation deviation model to calculate the spatial risk value.

4. The method for recognizing and processing medical image files according to claim 3, characterized in that, The analysis of modal signal response parameters and spatial registration parameters yields signal risk values ​​and registration risk values, specifically as follows: Based on the modal signal response parameters, the signal strengths of various modal signals corresponding to the same target tissue are obtained. The difference between the signal strengths of any two modal signals is calculated to obtain the cross-modal signal strength difference. The standard signal strength difference between any two modal signals corresponding to various target tissues is obtained. The cross-modal signal strength difference is calculated by the ratio of the cross-modal signal strength difference to the standard signal strength difference. The signal similarity is calculated by the cosine similarity calculation method for the signal feature vectors under various modes. The numerical value of the signal similarity is marked as the modal consistency index. Obtain the cross-modal signal reference deviation value corresponding to the normal state of the preset signal; substitute the cross-modal signal deviation value, modal consistency index and cross-modal signal reference deviation value into the preset distortion coefficient calculation formula to calculate the distortion coefficient; and substitute the distortion coefficient and cross-modal signal deviation value into the preset tanh function coupling formula to calculate the signal risk value. Local feature point data and modal Euclidean distance data are obtained based on spatial registration parameters, and registration risk values ​​are obtained by comprehensive analysis of the local feature point data and modal Euclidean distance data.

5. The method for recognizing and processing medical image files according to claim 4, characterized in that, The registration risk value is obtained by comprehensively analyzing local feature point data and modal Euclidean distance data, specifically as follows: The local gradient direction histogram of the target tissue is obtained based on the local feature point data. A preset number of local feature points are extracted based on the local gradient direction histogram. The structural position corresponding to each local feature point is obtained. The structural position includes the preset core region and edge region of the target tissue. The structural importance coefficient is obtained based on the structural position of the local feature point. The preset value range of the structural importance coefficient is divided into a high importance range and a low importance range. When the structural position corresponds to the core region, the value range of its structural importance coefficient corresponds to the high importance range. When the structural location corresponds to the edge region, the range of its structural importance coefficient corresponds to the low importance range; The model obtains the mode to be registered and the preset reference mode for each local feature point, and obtains the descriptor angle between the mode to be registered and the reference mode on the local gradient direction histogram. The cosine value of the descriptor angle is calculated by the cosine function. The local feature consistency score is obtained by calculating the structural importance coefficient and the cosine value of the descriptor angle. The modality types corresponding to the multimodalities participating in spatial registration are obtained. Each modality type is numbered to obtain a modality index number. The number of feature matching pairs corresponding to the modality index number is obtained, and the number of preset feature matching pairs is less than or equal to the number of preset feature points. Based on the modal Euclidean distance data, the modal Euclidean distance of each feature matching pair corresponding to each modality index number is obtained. The registration error is calculated by performing registration error calculation on the modal Euclidean distance corresponding to each modality index number to obtain the modal adaptive registration error value. The spatial registration value is calculated by substituting the local feature consistency score and the modal adaptive registration error value into the preset spatial registration parameter calculation formula. When the spatial registration value exceeds the preset standard registration value range, the part of the spatial registration value that exceeds the standard registration value range is marked as the registration risk value.

6. The method for recognizing and processing medical image files according to claim 5, characterized in that, The feature completion unit performs completion analysis on the data to be completed and optimized to obtain the completed image data, specifically: features The completion unit acquires historical occlusion data, trains a generative structural completion mapping model based on the historical occlusion data, identifies occlusion risk areas in the data to be completed and optimized, inputs the multimodal image data corresponding to the occlusion risk areas into the structural completion mapping model for matching, and generates completed image data.

7. A medical image file recognition and processing system, characterized in that, It includes a task data extraction module, a task scheduling and management module, and a recognition result output module, so that the medical image file recognition and processing system can perform the medical image file recognition and processing method as described in any one of claims 1-6.

8. A medical image file recognition and processing system according to claim 7, characterized in that, The task data extraction module is used to extract the medical image files corresponding to each patient to obtain the corresponding image task package. The task scheduling and management module is used to obtain the image task packages of each patient and input each image task package into the preset task scheduler. The task scheduler performs scheduling analysis on each image task package to obtain the corresponding execution number. Based on the execution number, the image task package is identified, processed and executed to obtain image data. The recognition result output module is used to obtain the image data corresponding to the recognition processing, obtain the patient information corresponding to each image data, and output the image data to the patient information terminal based on the patient information.

9. A medical image file recognition and processing system according to claim 8, characterized in that, The task scheduler performs scheduling analysis on each image task package to obtain the corresponding execution label, specifically as follows: Acquire clinical information and image size data from imaging task packages; Based on the clinical information of the imaging, the examination descriptor words are obtained. The examination descriptor word library is divided into a high-urgency word library and a low-urgency word library. The current corresponding examination descriptor word is matched with the examination descriptor word library to obtain the number of high-urgency words. The number of high-urgency words is recorded as the clinical urgency value. When the clinical urgency value is greater than 0, the clinical status is generated as high-urgency. When the corresponding clinical urgency value is 0, the clinical status is generated as low urgency; Based on the image size data, the image matrix and the number of image slices are obtained. The image matrix is ​​divided into a high matrix interval and a low matrix interval based on a preset matrix partition value. Similarly, the number of image slices is divided into a high slice interval and a low slice interval based on a preset slice partition value. When the current image size data corresponds to a high matrix interval and a high slice interval, the generated data size signal is a high complexity signal; otherwise, the generated data size signal is a low complexity signal. Execution queue information is generated based on the clinical status and scale signals corresponding to each imaging task package. That is, when the clinical status corresponds to a high urgency and high complexity signal, the execution queue information is generated as the priority calculation queue. When the clinical status corresponds to a high urgency and low complexity signal, the generated execution queue information is a real-time processing queue. When the clinical status corresponds to a low urgency and low complexity signal, the generated execution queue information is a batch queue. An execution label is generated based on the execution queue information corresponding to the image task package.

Images