Knowledge Fusion Method and Device for Aneurysm Lesion Detection Neural Network

Abstract

The present invention relates to the technical field of aneurysm lesion detection, and discloses a knowledge fusion method and device for an aneurysm lesion detection neural network. The method includes: obtaining artery segmentation data, aneurysm region manual annotation data, and aneurysm knowledge data; respectively constructing a blood vessel - algorithm mapping matrix, an annotation - algorithm mapping matrix, and a knowledge - algorithm mapping matrix according to the three types of data; successively constructing a blood vessel atlas basic structure, updating the node information and edge weights of the blood vessel atlas basic structure to obtain an initial arterial data fusion graph; and obtaining an optimal arterial data fusion graph by optimizing the grid parameters of the three mapping matrices. By quantitatively characterizing and fusing arterial data and knowledge, the present invention can establish a three - dimensional arterial data fusion graph containing rich artery - related information, providing more reliable and comprehensive data support for the detection of aneurysm lesions.

Description

Technical Field

[0001] The present invention relates to the technical field of aneurysm lesion detection, and in particular to a knowledge fusion method and device for aneurysm lesion detection neural network. Background Art

[0002] Existing studies have focused on the use of three-dimensional morphological features of blood vessels and the spatial distribution of aneurysms. Studies have found that various prior knowledge of aneurysms and cerebral blood vessels can improve the performance of the detection network, and the effective representation forms of knowledge are very diverse. Although various methods have achieved improvements in detection performance, the theoretical basis for the optimal processing method of input data is still unclear. In the case that the optimal processing method of input data is still unclear, starting from the labeled data, the texture features of the aneurysm expansion area are used instead of the labeled input, which achieves an improvement of about 5% in sensitivity, indicating that the representation form of the labeled data also affects the detection performance. From the above research, it can be seen that in the input data stage of the aneurysm detection network, the processing of input data and label data is strongly correlated with the detection performance of the network model. Therefore, researchers have discovered a number of supervised data generation methods that can improve detection sensitivity. However, existing studies have the problem of one-sided description and rely heavily on manual observation and experience. The optimal processing path for supervised data is still unclear. How to find the optimal aneurysm knowledge fusion method and fusion method to construct supervised data is still an important problem that currently plagues researchers. Summary of the invention

[0003] In view of this, the present invention provides a knowledge fusion method and device for an aneurysm lesion detection neural network to solve the problem of knowledge fusion of supervision data and one-sidedness of knowledge fusion mode in knowledge fusion of aneurysm detection network.

[0004] In a first aspect, the present invention provides a knowledge fusion method for aneurysm lesion detection neural network, the method comprising:

[0005] Obtaining artery segmentation data, aneurysm region manual annotation data, and aneurysm knowledge data as input data;

[0006] The vessel-algorithm mapping matrix of artery segmentation data is constructed by using a combination of multiple algorithms included in the vascular image processing algorithm set, and the vascular feature data is obtained in combination with the grid decision mechanism, and the basic structure of the vascular atlas is constructed based on the vascular feature data;

[0007] The annotation-algorithm mapping matrix of the manually annotated data of the aneurysm region is constructed by using the permutations and combinations of various algorithms included in the vascular image processing algorithm set, and the feature data of the annotated region is obtained in combination with the arterial segmentation data, and the node information of the basic structure of the vascular atlas is updated according to the feature data of the annotated region;

[0008] Construct a knowledge-algorithm mapping matrix based on aneurysm knowledge data, and combine artery segmentation data to obtain knowledge quantization feature data. Update the edge weights of the basic structure of the vascular atlas according to the knowledge quantization feature data to obtain an initial arterial data fusion graph;

[0009] Use the aneurysm detection graph convolutional network to optimize the grid parameters of the vascular-algorithm mapping matrix, the annotation-algorithm mapping matrix, and the knowledge-algorithm mapping matrix, and optimize the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained.

[0010] The knowledge fusion method of the aneurysm lesion detection neural network provided by the embodiments of the present invention first constructs a vascular-algorithm mapping matrix of artery segmentation data to obtain vascular feature data and constructs the basic structure of the vascular atlas. Secondly, it constructs an annotation-algorithm mapping matrix of aneurysm region artificial annotation data to obtain annotation region feature data and updates the node information of the basic structure of the vascular atlas. Then, it constructs a knowledge-algorithm mapping matrix according to aneurysm knowledge data to obtain knowledge quantization feature data and updates the edge weights of the basic structure of the vascular atlas to obtain an initial arterial data fusion graph. Finally, the initial arterial data fusion graph is optimized by grid search for parameters to obtain the best arterial data fusion graph. The present invention realizes the quantitative representation of knowledge by constructing a knowledge-algorithm mapping matrix, and then performs knowledge fusion based on artery segmentation data and aneurysm region artificial annotation data to establish a three-dimensional arterial data fusion graph, which can contain rich artery-related information and provide more reliable and comprehensive data support for the detection of aneurysm lesions.

[0011] In an alternative embodiment, the artery segmentation data is three-dimensional image data; the aneurysm region artificial annotation data is binary three-dimensional image data; the aneurysm knowledge data is imaging description text data on artery course, partition, segmentation, and aneurysm diagnosis criteria.

[0012] The present invention performs diverse feature quantization expressions on artery segmentation data, aneurysm region artificial annotation data, and aneurysm knowledge data, and constructs a map with the same structure, different node information, and different edge weights with the arterial centerline as the skeleton. Among them, the vascular atlas takes into account both the global and local aspects, the annotation atlas focuses on the local, and the knowledge text atlas focuses on multi-level expression, and finally obtains a fusion graph containing rich artery information.

[0013] In an alternative embodiment, the vascular image processing algorithm set includes: a preprocessing algorithm and a feature extraction algorithm; the preprocessing algorithm includes: a histogram equalization algorithm and a threshold segmentation algorithm, and the omics preprocessing operations adopted by each algorithm include wavelet transform, LoG transform, and gradient transform; the feature extraction algorithm includes: a principal component analysis algorithm, a Hessian matrix filtering algorithm, and a Sobel edge operator algorithm, and the omics feature extraction filters adopted by each algorithm include gray-level co-occurrence matrix features, gray-level size zone matrix features, and gray-level run-length matrix features.

[0014] In an alternative embodiment, a method for constructing a blood vessel-algorithm mapping matrix and an annotation-algorithm mapping matrix by arranging and combining a variety of algorithms included in the vascular image processing algorithm set includes: using the preprocessing algorithm as the horizontal axis and the feature extraction algorithm as the vertical axis; sequentially selecting one from each of the two types of algorithms to form a combination of algorithm processing; based on a grid decision-making mechanism, using arterial segmentation data or aneurysm region manual annotation data as input, constructing a parameter search table for the arrangement and combination, and respectively obtaining the blood vessel-algorithm mapping matrix and the annotation-algorithm mapping matrix.

[0015] The present invention processes arterial segmentation data and aneurysm region manual annotation data through a variety of preprocessing algorithms and feature extraction algorithms, quantifies and characterizes each data through the arrangement and combination of algorithms, obtains the mapping relationship between blood vessel data or annotation data and algorithms, so as to be able to extract corresponding feature data, ensure the reliability of the fusion map construction, and thus improve the detection effect of aneurysm lesions.

[0016] In an alternative embodiment, the process of constructing the basic structure of the blood vessel atlas according to the blood vessel feature data includes: extracting the centerline of the blood vessel tree according to the arterial segmentation data; using the centerline as the skeleton, sampling preset structural points in the blood vessel feature data to obtain nodes representing preset blood vessel regions; using the path of the centerline as the edge, and constructing the basic information of the blood vessel atlas according to the connection relationship of different nodes through the edge.

[0017] The present invention constructs an atlas with the same structure, different node information, and different edge weights with the arterial centerline as the skeleton, can include various information in the atlas in the optimal representation manner, constructs supervised data through the arterial data fusion map containing rich information, and thus performs the detection of aneurysm lesions, which can improve the detection effect of aneurysm lesions.

[0018] In an alternative embodiment, the process of constructing a knowledge-algorithm mapping matrix according to the aneurysm knowledge data includes: extracting the first preset number of text information entries according to the aneurysm knowledge data; selecting the second preset number of quantification representation algorithms; establishing a first preset number * second preset number of knowledge-algorithm mapping matrix between the text information entries and the knowledge quantification representation algorithms.

[0019] By matching the text information entries of arterial knowledge data with the corresponding quantization representation algorithms, the present invention can form relatively independent quantization feature data for each knowledge point, realize the expression of the knowledge level, no longer rely on manual observation and experience, and improve the reliability of knowledge representation and fusion.

[0020] In an optional implementation manner, the process of optimizing the grid parameters of the blood vessel-algorithm mapping matrix, annotation-algorithm mapping matrix, and knowledge-algorithm mapping matrix by using an aneurysm detection graph convolutional network includes: based on the blood vessel-algorithm mapping matrix, annotation-algorithm mapping matrix, and knowledge-algorithm mapping matrix, respectively select a preset number of points; take the central line of the blood vessel tree as the skeleton, and construct node information and edge weights according to the selected points; input the node information and edge weights into the graph convolutional network for grid parameter searching; perform iterative optimization according to the change of the aneurysm detection performance until the aneurysm detection performance reaches a local maximum value, or obtain the global maximum value after traversing all combinations of points.

[0021] After constructing the initial arterial data fusion graph, the present invention performs optimal parameter searching for each mapping matrix through iterative optimization. After the final iteration ends, it can generate optimal blood vessel preprocessing data and optimal annotation preprocessing data, and can also optimize the initial arterial data fusion graph to obtain the optimal arterial data fusion graph, so as to realize rich optimal knowledge representation and data fusion, which is helpful for the detection of aneurysm lesions.

[0022] In a second aspect, the present invention provides a knowledge fusion device for an aneurysm lesion detection neural network, including:

[0023] A data acquisition module, configured to acquire arterial segmentation data, aneurysm region manual annotation data, and aneurysm knowledge data as input data;

[0024] A map construction module, configured to construct a blood vessel-algorithm mapping matrix of arterial segmentation data by arranging and combining multiple algorithms included in the blood vessel image processing algorithm set, and obtain blood vessel feature data in combination with a grid decision mechanism, and construct a basic structure of the blood vessel map according to the blood vessel feature data;

[0025] A first map update module, configured to construct an annotation-algorithm mapping matrix of aneurysm region manual annotation data by arranging and combining multiple algorithms included in the blood vessel image processing algorithm set, and obtain annotation region feature data in combination with arterial segmentation data, and update the node information of the basic structure of the blood vessel map according to the annotation region feature data;

[0026] A second map update module, configured to construct a knowledge-algorithm mapping matrix according to aneurysm knowledge data, and obtain knowledge quantization feature data in combination with arterial segmentation data, and update the edge weights of the basic structure of the blood vessel map according to the knowledge quantization feature data to obtain an initial arterial data fusion graph;

[0027] The atlas optimization module is used to optimize the grid parameters of the blood vessel - algorithm mapping matrix, the annotation - algorithm mapping matrix, and the knowledge - algorithm mapping matrix by using the aneurysm detection graph convolutional network, and optimize the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained.

[0028] In a third aspect, the present invention provides a computer device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to execute the knowledge fusion method of the aneurysm lesion detection neural network in the first aspect or any corresponding embodiment thereof.

[0029] In a fourth aspect, the present invention provides a computer - readable storage medium, on which computer instructions are stored. The computer instructions are used to cause a computer to execute the knowledge fusion method of the aneurysm lesion detection neural network in the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0031] Figure 1 is a schematic flowchart of the knowledge fusion method of the aneurysm lesion detection neural network according to an embodiment of the present invention;

[0032] Figure 2 is a schematic flowchart of the construction process of the initial arterial data fusion graph of the knowledge fusion method of the aneurysm lesion detection neural network according to an embodiment of the present invention;

[0033] Figure 3 is a schematic flowchart of the knowledge fusion method of another aneurysm lesion detection neural network according to an embodiment of the present invention;

[0034] Figure 4 is a schematic flowchart of the construction process of the best arterial data fusion graph of the knowledge fusion method of the aneurysm lesion detection neural network according to an embodiment of the present invention;

[0035] Figure 5 is a structural block diagram of the knowledge fusion device of the aneurysm lesion detection neural network according to an embodiment of the present invention;

[0036] Figure 6It is a schematic diagram of the hardware structure of the computer device according to an embodiment of the present invention. Detailed implementation manners

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0038] The embodiments of the present invention are applicable to scenarios for intelligent detection of aneurysm lesions.

[0039] The embodiments of the present invention provide a knowledge fusion method for an aneurysm lesion detection neural network, which quantifies and fuses various types of aneurysm-related data to provide rich data for aneurysm lesion detection.

[0040] According to an embodiment of the present invention, an embodiment of a knowledge fusion method for an aneurysm lesion detection neural network is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0041] In this embodiment, a knowledge fusion method for an aneurysm lesion detection neural network is provided, which can be used in the above-mentioned mobile terminals, such as computers, lesion detectors, etc. Figure 1 It is a flowchart of the knowledge fusion method for an aneurysm lesion detection neural network according to an embodiment of the present invention, as Figure 1 shown, and the process includes the following steps:

[0042] Step S101, obtain artery segmentation data, aneurysm region manual annotation data, and aneurysm knowledge data as input data.

[0043] Specifically, in the embodiments of the present invention, when performing intelligent detection of aneurysm lesions on a certain part of the human body after obtaining the angiography (TOF-MRA) of this part, rich relevant data is required as a basis. Taking the detection of cerebral aneurysm lesions as an example, but not limited thereto, three-dimensional image data of cerebral arteries is selected as artery segmentation data, and the three-dimensional image data is binarized and used as aneurysm region manual annotation data, and the text data of the course, partition, segmentation of cerebral arteries, and the imaging description of aneurysm diagnosis criteria is used as aneurysm knowledge data.

[0044] Step S102: Construct a blood vessel - algorithm mapping matrix for arterial segmentation data by arranging and combining multiple algorithms included in the blood vessel image processing algorithm set, and obtain blood vessel feature data in combination with a grid - based decision - making mechanism. Then, construct the basic structure of the blood vessel atlas according to the blood vessel feature data.

[0045] Specifically, in the embodiments of the present invention, a blood vessel image processing algorithm set is prepared in advance to quantitatively characterize arterial segmentation data, including a pre - processing algorithm and a feature extraction algorithm. Among them, the pre - processing algorithm includes: histogram equalization algorithm and threshold segmentation algorithm, and the omics pre - processing operations adopted by each algorithm include wavelet transform, LoG transform, and gradient transform; the feature extraction algorithm includes: principal component analysis algorithm, Hessian matrix filtering algorithm, Sobel edge operator algorithm, and the omics feature extraction filters adopted by each algorithm include gray - level co - occurrence matrix feature, gray - level size - region matrix feature, and gray - level run - length matrix feature.

[0046] In some optional embodiments, a blood vessel - algorithm mapping matrix for arterial segmentation data is constructed by arranging and combining multiple algorithms included in the blood vessel image processing algorithm set. As Figure 2 shown, in combination with the grid - based decision - making mechanism, taking arterial segmentation data as input, after analysis by the blood vessel - algorithm mapping matrix, multiple pieces of blood vessel feature data with different pre - processing - feature extraction are obtained, and then the basic structure of the blood vessel atlas is constructed according to the blood vessel feature data. Among them, the grid - based decision - making mechanism is to use the points traversed or randomly selected from the blood vessel - algorithm mapping matrix as the processing and analysis method.

[0047] Step S103: Construct a annotation - algorithm mapping matrix for aneurysm region manually annotated data by arranging and combining multiple algorithms included in the blood vessel image processing algorithm set, and obtain annotation region feature data in combination with arterial segmentation data. Then, update the node information of the basic structure of the blood vessel atlas according to the annotation region feature data.

[0048] Specifically, in the embodiments of the present invention, a annotation - algorithm mapping matrix for aneurysm region manually annotated data is constructed by arranging and combining multiple algorithms included in the blood vessel image processing algorithm set. As Figure 2 shown, in combination with the grid - based decision - making mechanism, taking arterial segmentation data and aneurysm region manually annotated data as input, after analysis by the blood vessel - algorithm mapping matrix, multiple pieces of annotation region feature data with different pre - processing - feature extraction are obtained, and then the minority node information of the corresponding region in the basic structure of the blood vessel atlas is updated according to the annotation region feature data.

[0049] Step S104: Construct a knowledge - algorithm mapping matrix according to aneurysm knowledge data, and obtain knowledge - quantified feature data in combination with arterial segmentation data. Then, update the edge weights of the basic structure of the blood vessel atlas according to the knowledge - quantified feature data to obtain an initial arterial data fusion graph.

[0050] Specifically, in the embodiments of the present invention, a knowledge-algorithm mapping matrix is constructed based on aneurysm knowledge data. Then, based on the vascular segmentation data, each knowledge point forms relatively independent quantitative feature data. According to the correspondence between the knowledge level and the vascular structure, the edge weights of the basic structure of the vascular atlas are updated based on the quantitative features.

[0051] In some alternative embodiments, as Figure 2 shown, through the above generation of the basic structure of the vascular atlas, update of the node information of the lesion area, and update of the edge weights of the whole graph under the guidance of knowledge, an initial arterial data fusion graph is finally formed.

[0052] Step S105: Use the aneurysm detection graph convolutional network to optimize the grid parameters of the vascular-algorithm mapping matrix, the annotation-algorithm mapping matrix, and the knowledge-algorithm mapping matrix, and optimize the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained.

[0053] Specifically, in the embodiments of the present invention, taking the initial arterial data fusion graph as the input, using the aneurysm detection graph convolutional network, with aneurysm detection as the task, under the condition of the same number of training steps, optimize the grid parameters of the vascular-algorithm mapping matrix, the annotation-algorithm mapping matrix, and the knowledge-algorithm mapping matrix. In the iterative optimization process, use the aneurysm detection performance of the graph convolutional network as the feedback index. The iterative optimization method adopts the class hierarchical clustering method, only as an example, not limited thereto. Classify each group of mapping relationships based on the feedback index ranking, and optimize the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained. In addition, after the final iteration ends, the optimal vascular preprocessing data and the optimal annotation preprocessing data can be generated, and these two types of data can be used in the aneurysm detection method.

[0054] The knowledge fusion method of the aneurysm lesion detection neural network provided in this embodiment first obtains vascular feature data by constructing a vascular-algorithm mapping matrix of arterial segmentation data and constructs the basic structure of the vascular atlas. Secondly, it obtains the feature data of the labeled area by constructing a labeling-algorithm mapping matrix of the artificially labeled data of the aneurysm area and updates the node information of the basic structure of the vascular atlas. Then, it constructs a knowledge-algorithm mapping matrix according to the aneurysm knowledge data, obtains the knowledge quantization feature data, and updates the edge weights of the basic structure of the vascular atlas to obtain the initial arterial data fusion graph. Finally, the initial arterial data fusion graph is optimized by grid search for parameters to obtain the optimal arterial data fusion graph. The present invention realizes the quantitative representation of knowledge by constructing a knowledge-algorithm mapping matrix, and then performs knowledge fusion based on the arterial segmentation data and the artificially labeled data of the aneurysm area to establish a three-dimensional arterial data fusion graph, which can contain rich artery-related information and provide more reliable and comprehensive data support for the detection of aneurysm lesions.

[0055] In this embodiment, a knowledge fusion method of an aneurysm lesion detection neural network is provided, which can be used in the above-mentioned mobile terminals, such as computers, lesion detectors, etc. Figure 3 It is a flowchart of the knowledge fusion method of the aneurysm lesion detection neural network according to an embodiment of the present invention, as Figure 3 shown, and this process includes the following steps:

[0056] Step S301, obtain arterial segmentation data, artificially labeled data of the aneurysm area, and aneurysm knowledge data as input data. For details, please refer to Figure 1 Step S101 of the embodiment shown, which will not be elaborated here.

[0057] Step S302, use various algorithms included in the vascular image processing algorithm set to construct a vascular-algorithm mapping matrix of arterial segmentation data, and combine the grid decision mechanism to obtain vascular feature data, and construct the basic structure of the vascular atlas according to the vascular feature data.

[0058] Specifically, the above step S302 includes:

[0059] Step S3021, taking the preprocessing algorithm as the horizontal axis and the feature extraction algorithm as the vertical axis, select one from each of the two types of algorithms in turn to form a combination of algorithm processing, construct a parameter search table for permutation and combination, and based on the grid decision mechanism, use the arterial segmentation data or the artificially labeled data of the aneurysm area as input to obtain the vascular-algorithm mapping matrix and the labeling-algorithm mapping matrix respectively.

[0060] Specifically, in the embodiments of the present invention, a grid-based decision-making mechanism is adopted. Taking two algorithms as the horizontal and vertical axes respectively, a grid-shaped parameter search table is formed. Each time, one algorithm is selected from each of the two types of algorithms to form a combination for algorithm processing. All algorithms are traversed according to the grid, and the algorithms are permuted and combined. Taking arterial segmentation data as the input, a blood vessel-algorithm mapping matrix is constructed; taking arterial segmentation data and manually annotated data of the aneurysm region as the input, a annotation-algorithm mapping matrix is constructed. Through the permutation and combination of algorithms, each data is quantitatively characterized, and the mapping relationship between blood vessel data or annotation data and algorithms is obtained, so that the corresponding feature data can be extracted, the reliability of the fusion graph construction is ensured, and the detection effect of the aneurysm lesion is improved.

[0061] Step S3022: Extract the centerline of the blood vessel tree according to the arterial segmentation data. Taking the centerline as the skeleton, sample the preset structural points in the blood vessel feature data to obtain the nodes representing the preset blood vessel regions. Taking the path of the centerline as the edge, and construct the basic information of the blood vessel atlas according to the connection relationship of different nodes through the edges.

[0062] Specifically, in the embodiments of the present invention, the centerline of the blood vessel tree is extracted by using the cerebral arterial segmentation data. Sampling is performed with the centerline as the skeleton. The sampling result uses the SLIC superpixel segmentation algorithm based on Euclidean distance and gray difference to automatically segment the adjacent pixels and cluster them as the nodes of the graph. Taking the path along the centerline as the edge, and generating the adjacency relationship of the graph according to the connection relationship of different nodes through the edges, to construct the basic structure of the blood vessel atlas, where the node information comes from the sampling of specific structural points of each feature data.

[0063] The present invention constructs an atlas with the same structure, different node information, and different edge weights with the arterial centerline as the skeleton, which can include various information in the atlas in the optimal representation manner. By constructing supervised data through the arterial data fusion graph containing rich information, the detection of aneurysm lesions is carried out, and the detection effect of aneurysm lesions can be improved.

[0064] Step S303: Use the permutation and combination of multiple algorithms included in the blood vessel image processing algorithm set to construct an annotation-algorithm mapping matrix for the manually annotated data of the aneurysm region, and combine the arterial segmentation data to obtain the feature data of the annotation region. Update the node information of the basic structure of the blood vessel atlas according to the feature data of the annotation region. For details, please refer to Figure 1 Step S103 of the embodiment shown, which will not be elaborated here.

[0065] Step S304: Construct a knowledge-algorithm mapping matrix according to the aneurysm knowledge data, and combine the arterial segmentation data to obtain the knowledge quantization feature data. Update the edge weights of the basic structure of the blood vessel atlas according to the knowledge quantization feature data to obtain the initial arterial data fusion graph.

[0066] Specifically, the above step S304 includes:

[0067] Step S3041: Extract a first preset number of text information entries according to the aneurysm knowledge data.

[0068] Specifically, in the embodiment of the present invention, extract the text information entries in the aneurysm knowledge data. Taking the aneurysm knowledge data as: "An aneurysm refers to a local outward bulge of all three layers of the arterial wall, often occurring at the arterial branch. Approximately 85% is located in the anterior circulation, 30% - 40% originates from the anterior cerebral artery or the anterior communicating artery, 30% is located in the posterior communicating artery, 20% - 30% is located in the branches of the middle cerebral artery, and 5% - 10% is located in the internal carotid artery. Approximately 15% originates from the posterior circulation, including the basilar apex, the superior cerebellar artery, and the posterior inferior cerebellar artery. On CT, an aneurysm appears as a well-defined, slightly hyperdense, round lesion, and sometimes there may be surrounding calcification" as an example, the extracted text information entries are as follows:

[0069] 1. All three layers of the arterial wall bulge locally outward

[0070] 2. Often occurs at the arterial branch

[0071] 3. Approximately 85% is located in the anterior circulation

[0072] 4. 30% - 40% originates from the anterior cerebral artery or the anterior communicating artery

[0073] 5. 30% is located in the posterior communicating artery

[0074] 6. 20% - 30% is located in the branches of the middle cerebral artery

[0075] 7. 5% - 10% is located in the internal carotid artery

[0076] 8. Approximately 15% originates from the posterior circulation, including the basilar apex, the superior cerebellar artery, and the posterior inferior cerebellar artery

[0077] 9. Appears as a well-defined, slightly hyperdense, round lesion

[0078] 10. Sometimes there may be surrounding calcification

[0079] Step S3042: Select a second preset number of quantization representation algorithms.

[0080] Specifically, in the embodiment of the present invention, select 7 quantization representation algorithms, but not limited thereto:

[0081] 1. canny operator

[0082] 2. sobel operator

[0083] 3. Density mean

[0084] 4. Surface curvature

[0085] 5. Vascular Zoning

[0086] 6. Density Variance

[0087] 7. Vascular Segmentation

[0088] Step S3043: Establish a knowledge - algorithm mapping matrix of the first preset quantity * the second preset quantity between the text information entries and the knowledge quantification representation algorithms.

[0089] Specifically, in the embodiment of the present invention, a 7 * 10 mapping matrix can be established according to 10 text information entries and 7 knowledge quantification representation algorithms, as described below:

[0090] 1. For text entry 1, algorithm entry 4 is 1;

[0091] 2. For text entry 2, algorithm entries 4, 5, and 7 are 1;

[0092] 3. For text 3, algorithm entries 5 and 7 are 1;

[0093] 4. The same as above for texts 4, 5, 6, 7, and 8;

[0094] 5. For text 9, algorithm entries 2, 3, and 6 are 1;

[0095] 6. For text 10, algorithm entries 1 and 7 are 1.

[0096] The above description is shown in the following table:

[0097] Step S305: Use the aneurysm detection graph convolutional network to optimize the grid parameters of the vascular - algorithm mapping matrix, annotation - algorithm mapping matrix, and knowledge - algorithm mapping matrix, and optimize the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained.

[0098] Specifically, the above - mentioned step S305 includes:

[0099] Step S3051: Based on the vascular - algorithm mapping matrix, annotation - algorithm mapping matrix, and knowledge - algorithm mapping matrix, select a preset number of points respectively.

[0100] Step S3052: Take the vascular tree centerline as the skeleton, construct node information and edge weights according to the selected points, and input the node information and edge weights into the graph convolutional network for grid - based parameter searching;

[0101] Step S3053: Perform iterative optimization according to the change of aneurysm detection performance until the aneurysm detection performance reaches a local maximum value, or a global maximum value is obtained after traversing all combinations of points.

[0102] Specifically, in the embodiments of the present invention, as Figure 4 shown, after constructing the initial arterial data fusion graph, the optimal parameters of each mapping matrix are obtained through iterative optimization. After the final iteration ends, the optimal vascular preprocessing data and the optimal annotation preprocessing data can be generated, and the initial arterial data fusion graph can also be optimized to obtain the optimal arterial data fusion graph, thereby realizing rich optimal knowledge representation and data fusion, which is helpful for the detection of aneurysm lesions.

[0103] The knowledge fusion method of the aneurysm lesion detection neural network provided in this embodiment first constructs a blood vessel - algorithm mapping matrix of arterial segmentation data to obtain blood vessel feature data and constructs the basic structure of the blood vessel atlas. Secondly, it constructs an annotation - algorithm mapping matrix of the manually annotated data in the aneurysm region to obtain the feature data of the annotated region and updates the node information of the basic structure of the blood vessel atlas. Then, it constructs a knowledge - algorithm mapping matrix according to the aneurysm knowledge data to obtain the knowledge quantization feature data and updates the edge weights of the basic structure of the blood vessel atlas to obtain the initial arterial data fusion graph. Finally, it performs grid search parameter optimization on the initial arterial data fusion graph to obtain the optimal arterial data fusion graph. The present invention realizes the quantitative representation of knowledge by constructing a knowledge - algorithm mapping matrix, and then performs knowledge fusion based on the arterial segmentation data and the manually annotated data in the aneurysm region to establish a three - dimensional arterial data fusion graph, which can contain rich arterial - related information and provide more reliable and comprehensive data support for the detection of aneurysm lesions.

[0104] In this embodiment, a knowledge fusion device for an aneurysm lesion detection neural network is also provided. This device is used to implement the above - mentioned embodiments and preferred implementation manners, and those that have been described will not be repeated. As used hereinafter, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, implementation in hardware, or a combination of software and hardware is also possible and contemplated.

[0105] This embodiment provides a knowledge fusion device for an aneurysm lesion detection neural network, as Figure 5 shown, including:

[0106] A data acquisition module 501, configured to acquire arterial segmentation data, manually annotated data in the aneurysm region, and aneurysm knowledge data as input data.

[0107] An atlas construction module 502, configured to construct a blood vessel - algorithm mapping matrix of arterial segmentation data by arranging and combining various algorithms included in the blood vessel image processing algorithm set, obtain blood vessel feature data in combination with a grid - based decision - making mechanism, and construct the basic structure of the blood vessel atlas according to the blood vessel feature data;

[0108] The first atlas update module 503 is configured to construct an annotation-algorithm mapping matrix of the aneurysm region artificial annotation data by arranging and combining multiple algorithms included in the vascular image processing algorithm set, obtain the annotation region feature data in combination with the artery segmentation data, and update the node information of the vascular atlas basic structure according to the annotation region feature data;

[0109] The second atlas update module 504 is configured to construct a knowledge-algorithm mapping matrix according to the aneurysm knowledge data, obtain the knowledge quantization feature data in combination with the artery segmentation data, and update the edge weights of the vascular atlas basic structure according to the knowledge quantization feature data to obtain an initial artery data fusion graph;

[0110] The atlas optimization module 505 is configured to optimize the grid parameters of the vascular-algorithm mapping matrix, the annotation-algorithm mapping matrix, and the knowledge-algorithm mapping matrix by using the aneurysm detection graph convolutional network, and optimize the graph node information and edge weights of the initial artery data fusion graph according to the parameter optimization results until the optimal artery data fusion graph is obtained.

[0111] The further function descriptions of the above-mentioned various modules and units are the same as those in the corresponding embodiments above, and will not be repeated here.

[0112] The knowledge fusion device of the aneurysm lesion detection neural network in this embodiment is presented in the form of functional units. Here, the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and a memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0113] The embodiment of the present invention also provides a computer device having the above Figure 6 knowledge fusion device of the aneurysm lesion detection neural network shown.

[0114] Please refer to Figure 6 , Figure 6 which is a schematic structural diagram of a computer device provided by an alternative embodiment of the present invention. As shown in Figure 6As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including a high-speed interface and a low-speed interface. Each component communicates with each other using different buses and can be installed on a common motherboard or installed in other ways as needed. The processor can process instructions executed within the computer device, including instructions stored in the memory or on the memory to display graphical information of the GUI on an external input / output device (such as a display device coupled to the interface). In some alternative embodiments, if needed, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a set of blade servers, or a multi-processor system). Figure 6 In the figure, one processor 10 is taken as an example.

[0115] The processor 10 can be a central processing unit, a network processor, or a combination thereof. Among them, the processor 10 can further include a hardware chip. The above hardware chip can be an application-specific integrated circuit, a programmable logic device, or a combination thereof. The above programmable logic device can be a complex programmable logic device, a field programmable gate array, a generic array logic, or any combination thereof.

[0116] Among them, the memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiments.

[0117] The memory 20 can include a program storage area and a data storage area. Among them, the program storage area can store an operating system and application programs required for at least one function; the data storage area can store data created according to the use of the computer device, etc. In addition, the memory 20 can include a high-speed random access memory, and can also include a non-transitory memory, such as at least one disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some alternative embodiments, the memory 20 can optionally include a memory remotely set relative to the processor 10, and these remote memories can be connected to the computer device through a network. Examples of the above network include but are not limited to the Internet, an enterprise intranet, a local area network, a mobile communication network, and combinations thereof.

[0118] The memory 20 can include a volatile memory, such as a random access memory; the memory can also include a non-volatile memory, such as a flash memory, a hard disk, or a solid-state drive; the memory 20 can also include a combination of the above types of memories.

[0119] The computer device further includes a communication interface 30 for the computer device to communicate with other devices or communication networks.

[0120] Embodiments of the present invention also provide a computer-readable storage medium. The methods according to the embodiments of the present invention can be implemented in hardware, firmware, or be implemented as computer code that can be recorded on a storage medium, or be implemented as computer code that is originally stored in a remote storage medium or a non-transitory machine-readable storage medium and downloaded through a network and will be stored in a local storage medium, so that the methods described herein can be stored as such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only memory, a random access memory, a flash memory, a hard disk, or a solid-state drive, etc.; further, the storage medium can also include a combination of the above-mentioned types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code, and when the software or computer code is accessed and executed by the computer, the processor, or the hardware, the methods shown in the above embodiments are implemented.

[0121] Although embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A knowledge fusion method for an aneurysm lesion detection neural network, characterized in that The method includes: Obtaining arterial segmentation data, manually labeled aneurysm region data, and aneurysm knowledge data as input data; Using various algorithms included in the vascular image processing algorithm set to construct a vascular-algorithm mapping matrix for the arterial segmentation data, and combining a grid-based decision-making mechanism to obtain vascular feature data, and constructing a basic structure of the vascular atlas according to the vascular feature data; The process of constructing the basic structure of the vascular atlas according to the vascular feature data includes: extracting the centerline of the vascular tree from the arterial segmentation data; using the centerline as a skeleton, sampling preset structural points in the vascular feature data to obtain nodes representing preset vascular regions; using the path of the centerline as edges, and constructing basic information of the vascular atlas according to the connection relationships of different nodes through the edges; Using various algorithms included in the vascular image processing algorithm set to construct a labeling-algorithm mapping matrix for the manually labeled aneurysm region data, and combining the arterial segmentation data to obtain labeled region feature data, and updating the node information of the basic structure of the vascular atlas according to the labeled region feature data; Constructing a knowledge-algorithm mapping matrix according to the aneurysm knowledge data, and combining the arterial segmentation data to obtain knowledge quantization feature data, and updating the edge weights of the basic structure of the vascular atlas according to the knowledge quantization feature data to obtain an initial arterial data fusion graph; Using an aneurysm detection graph convolutional network to optimize the grid parameters of the vascular-algorithm mapping matrix, labeling-algorithm mapping matrix, and knowledge-algorithm mapping matrix, and optimizing the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained; The process of using the aneurysm detection graph convolutional network to optimize the grid parameters of the vascular-algorithm mapping matrix, labeling-algorithm mapping matrix, and knowledge-algorithm mapping matrix includes: based on the vascular-algorithm mapping matrix, labeling-algorithm mapping matrix, and knowledge-algorithm mapping matrix, selecting a preset number of points respectively; using the centerline of the vascular tree as a skeleton, constructing node information and edge weights according to the selected points; inputting the node information and edge weights into the graph convolutional network for grid-based parameter searching; performing iterative optimization according to the change of the aneurysm detection performance until the aneurysm detection performance reaches a local maximum value, or obtaining a global maximum value after traversing all combinations of points.

2. The method according to claim 1, wherein The arterial segmentation data is three-dimensional image data; The manually labeled aneurysm region data is binary three-dimensional image data; The aneurysm knowledge data is imaging description text data on arterial course, zoning, segmentation, and aneurysm diagnosis criteria.

3. The method according to claim 1, characterized in that, The vascular image processing algorithm set includes: a preprocessing algorithm and a feature extraction algorithm; The preprocessing algorithm includes: a histogram equalization algorithm and a threshold segmentation algorithm, and the omics preprocessing operations adopted by each algorithm include wavelet transform, LoG transform, and gradient transform; The feature extraction algorithms include: principal component analysis algorithm, Hessian matrix filtering algorithm, and Sobel edge operator algorithm. The omics feature extraction filters used in each algorithm include gray-level co-occurrence matrix features, gray-level size zone matrix features, and gray-level run-length matrix features.

4. The method according to claim 3, wherein The process of constructing a blood vessel-algorithm mapping matrix and an annotation-algorithm mapping matrix by arranging and combining various algorithms included in the blood vessel image processing algorithm set includes: Taking the preprocessing algorithm as the horizontal axis and the feature extraction algorithm as the vertical axis; Selecting one algorithm from each of the two types of algorithms in sequence to form a combination for algorithm processing, and constructing a parameter search table for permutation and combination; Based on the grid decision-making mechanism, using the artery segmentation data or the artificial annotation data of the aneurysm region as input, obtaining the blood vessel-algorithm mapping matrix and the annotation-algorithm mapping matrix respectively.

5. The method according to claim 1, wherein The process of constructing a knowledge-algorithm mapping matrix according to the aneurysm knowledge data includes: Extracting a first preset number of text information entries according to the aneurysm knowledge data; Selecting a second preset number of quantization representation algorithms; Establishing a first preset number * second preset number of knowledge-algorithm mapping matrices between the text information entries and the knowledge quantization representation algorithms.

6. A knowledge fusion device for an aneurysm lesion detection neural network, characterized in that, The device includes: A data acquisition module for acquiring artery segmentation data, artificial annotation data of the aneurysm region, and aneurysm knowledge data as input data; A map construction module for constructing a blood vessel-algorithm mapping matrix of the artery segmentation data by arranging and combining various algorithms included in the blood vessel image processing algorithm set, and obtaining blood vessel feature data in combination with the grid decision-making mechanism, and constructing a basic structure of the blood vessel map according to the blood vessel feature data. The process of constructing the basic structure of the blood vessel map according to the blood vessel feature data includes: extracting the centerline of the blood vessel tree according to the artery segmentation data; using the centerline as the skeleton, sampling preset structural points in the blood vessel feature data to obtain nodes representing preset blood vessel regions; using the path of the centerline as the edge, and constructing basic information of the blood vessel map according to the connection relationship of different nodes through the edge. A first map update module for constructing an annotation-algorithm mapping matrix of the artificial annotation data of the aneurysm region by arranging and combining various algorithms included in the blood vessel image processing algorithm set, and obtaining annotation region feature data in combination with the artery segmentation data, and updating the node information of the basic structure of the blood vessel map according to the annotation region feature data; A second map update module for constructing a knowledge-algorithm mapping matrix according to the aneurysm knowledge data, and obtaining knowledge quantization feature data in combination with the artery segmentation data, and updating the edge weights of the basic structure of the blood vessel map according to the knowledge quantization feature data to obtain an initial arterial data fusion map; The atlas optimization module is used to optimize the grid parameters of the blood vessel - algorithm mapping matrix, annotation - algorithm mapping matrix, and knowledge - algorithm mapping matrix by using the aneurysm detection graph convolutional network, and optimize the graph node information and edge weights of the initial arterial data fusion graph according to the parameter optimization results until the best arterial data fusion graph is obtained; the process of using the aneurysm detection graph convolutional network to optimize the grid parameters of the blood vessel - algorithm mapping matrix, annotation - algorithm mapping matrix, and knowledge - algorithm mapping matrix includes: based on the blood vessel - algorithm mapping matrix, annotation - algorithm mapping matrix, and knowledge - algorithm mapping matrix, respectively select a preset number of points; taking the center line of the blood vessel tree as the skeleton, construct node information and edge weights according to the selected points; input the node information and edge weights into the graph convolutional network for grid parameter searching; perform iterative optimization according to the change of aneurysm detection performance until the aneurysm detection performance reaches a local maximum value, or a global maximum value is obtained after traversing all combinations of points.

7. A computer device, characterized in that, Comprising: A memory and a processor, the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to execute the knowledge fusion method of the aneurysm lesion detection neural network according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, Computer instructions are stored on the computer - readable storage medium, and the computer instructions are used to cause a computer to execute the knowledge fusion method of the aneurysm lesion detection neural network according to any one of claims 1 to 5.

Images