A method for constructing a case image resource library for a smart medical system

Abstract

The present invention relates to the technical field of case image storage management, and specifically to a method for constructing a case image resource library of a smart medical system. The present invention first obtains all case images, and further classifies and obtains all image groups; constructs a spatial image model by sorting and stacking similar case images, and obtains a sub-image model formed by stacking similar local tissue areas in each case image, and determines its importance by analyzing the texture structure of the sub-image model, thereby reducing the workload of analyzing the importance of the local tissue area in each case image separately, thereby improving efficiency; and based on the importance, obtains the dimensionality reduction result of the local tissue area of ​​each case image, while retaining the important lesion features, reduces the information redundancy of the non-important tissue area features as much as possible, improves the compression effect of the case image, and thus improves the construction effect of the case image resource library.

Description

Technical Field

[0001] The present invention relates to the technical field of case image storage management, and in particular to a method for constructing a case image resource library of a smart medical system. Background Art

[0002] The smart medical system is a medical service system based on cloud computing, big data, artificial intelligence and other technologies. The case image resource library is an important module in the smart medical system. It integrates a large amount of patient case image data, realizes the storage and sharing of medical cases, and promotes the development and progress of the medical industry. By classifying case image data into multiple levels according to imaging attributes, acquisition sites and lesion characteristics, each case image data of the same category is compressed and stored in the same node of the resource library, thereby building a case image resource library for the management and retrieval of case image data.

[0003] Traditional image compression methods include using the principal component analysis algorithm (PCA) to reduce the dimension of each image to achieve the compression purpose. However, different local areas in case images have different importance for disease research. Directly using traditional methods to extract the principal components of the entire case image may cause feature distortion of important lesion areas in the case image or feature redundancy of normal tissue background areas, resulting in poor compression of the case image, which in turn affects the construction of the case image resource library. Summary of the invention

[0004] In order to solve the technical problem that the existing technology has poor compression effect on case images, which leads to poor construction effect of case image resource library, the purpose of the present invention is to provide a method for constructing a case image resource library of a smart medical system. The technical solution adopted is as follows:

[0005] Acquire all case images; classify all case images according to the imaging properties, acquisition sites and lesion characteristics of the case images, and acquire all image groups;

[0006] In each of the image groups, all the case images are sorted and stacked according to the similarities between different case images, and all spatial image models are constructed, wherein each model unit in the spatial image model corresponds to a pixel point in the case image; in each of the spatial image models, each of the spatial image models is divided according to the grayscale similarity of the model units to obtain all sub-image models;

[0007] According to the grayscale distribution of all model units in each sub-image model, combined with the shape information of each sub-image model and the changes after morphological processing, the importance of each sub-image model is obtained; in each of the spatial image models, a dimensionality reduction analysis is performed on each sub-image model, and combined with the importance of each sub-image model, a dimensionality reduction case image of each case image is obtained;

[0008] All dimension-reduced case images in each of the image groups are stored in the corresponding group nodes in the resource library structure model to construct a case image resource library.

[0009] Furthermore, the method for obtaining the image group includes:

[0010] Dividing all the case images according to the imaging attributes of the case images to obtain all primary image groups;

[0011] Dividing all the case images in the first-level image group according to the acquisition site of the case images to obtain all the second-level image groups;

[0012] All the case images in the secondary image group are divided according to the lesion characteristics of the case images to obtain all image subgroups.

[0013] Furthermore, the method for acquiring the spatial image model includes:

[0014] In each of the image groups, all of the case images are sorted and segmented according to the similarity between each of the case images and all other case images, to obtain all sequence sub-segments;

[0015] The pixels in each of the case images are converted into cubic pixels, the model units corresponding to all the pixels are obtained, and the stereo grayscale model of each of the case images is acquired; the stereo grayscale models corresponding to all the case images in each of the sequence sub-segments are sorted and stacked layer by layer to construct the spatial image model corresponding to each of the sequence sub-segments.

[0016] Furthermore, the method for obtaining the sequence sub-segments includes:

[0017] In each of the image groups, the average of the Hamming distances of the hash values ​​between each of the case images and all other case images is used as a sorting parameter for each of the case images;

[0018] Taking the case image corresponding to the maximum sorting parameter as one endpoint of the image sequence, taking the case image corresponding to the minimum sorting parameter as the other endpoint of the image sequence, and obtaining the image sequence of each image group based on a state compression dynamic programming algorithm;

[0019] In each of the image sequences, it is determined whether the Hamming distance of the hash values ​​between adjacent case images exceeds a preset threshold. If it exceeds the preset threshold, the image sequence is segmented to obtain all sequence sub-segments of each of the image sequences.

[0020] Furthermore, the method for obtaining the sub-image model includes:

[0021] In each of the spatial image models, a preset first number of seed units are selected from all model units, where the preset first number is the same as the number of grayscale values ​​of all model units; based on a regional connectivity algorithm and grayscale differences between the model units, a sub-image model of each of the seed units is obtained;

[0022] If there is a model unit without a corresponding sub-image model, the model unit without a corresponding sub-image model is used as a unit to be clustered, and a preset second number of new seed units are screened out from all the units to be clustered, and the preset second number is the same as the number of grayscale values ​​of all the units to be clustered; in all the units to be clustered in each of the spatial image models, a sub-image model of each new seed unit is obtained based on the regional connectivity algorithm and the grayscale difference between the units to be clustered;

[0023] The iteration is continued until all model units in each of the spatial image models have corresponding sub-image models, thereby obtaining all sub-image models in each of the spatial image models.

[0024] Furthermore, the method for obtaining the importance includes:

[0025] In each of the sub-image models, according to the quantity distribution of the module units at each gray value, the grayscale stability of each of the sub-image models is obtained;

[0026] According to the quantitative difference between the total number of corresponding module units of the sub-image model after three-dimensional expansion and three-dimensional erosion, the degree of morphological change of each sub-image model is obtained; the quantitative difference is positively correlated with the degree of morphological change;

[0027] In each of the sub-image models, according to the shape differences between the single-layer modules formed by the module units corresponding to different case images, the inter-layer difference parameters of each of the sub-image models are obtained, and the shape difference is positively correlated with the inter-layer difference parameters;

[0028] The importance of the corresponding sub-image model is obtained according to the grayscale stability, the morphological change degree and the inter-layer difference parameter; the grayscale stability is negatively correlated with the importance, and the morphological change degree and the inter-layer difference parameter are both positively correlated with the importance.

[0029] Furthermore, the grayscale stability acquisition method includes:

[0030] In each of the sub-image models, the total number of module units at each gray value is counted; the total number of gray values ​​where the total number of units is 0 is used as a first stable parameter; and the difference between the total number of units at all adjacent gray values ​​is accumulated as a second stable parameter;

[0031] The grayscale stability is obtained according to the first stability parameter and the second stability parameter; the first stability parameter and the second stability parameter are both positively correlated with the grayscale stability.

[0032] Furthermore, the method for obtaining the inter-layer difference parameter includes:

[0033] In each of the sub-image models, all edge modules in the single-layer module composed of the module units corresponding to each of the case images are obtained; in each of the spatial image models, any module unit is used as a reference unit; in each of the single-layer modules, the edge module closest to the reference unit is used as a starting point to obtain the edge chain codes of all edge modules in each of the single-layer modules;

[0034] Based on the multiple sequence comparison algorithm, a distance matrix of all the edge chain codes in each sub-image model is obtained, and the mean of all matrix elements in the distance matrix is ​​used as a first difference parameter; a second difference parameter is obtained according to the length difference of all the edge chain codes, and the length difference is positively correlated with the second difference parameter;

[0035] An inter-layer difference parameter is obtained according to the first difference parameter and the second difference parameter; the first difference parameter and the second difference parameter are both positively correlated with the inter-layer difference parameter.

[0036] Furthermore, the method for obtaining the reduced-dimensional case image includes:

[0037] In the stereo grayscale model corresponding to each case image of each of the spatial image models, all module units belonging to the same sub-image model constitute a sub-stereo grayscale model as a model to be reduced in dimension, and principal component analysis is performed on each of the models to be reduced in dimension based on the grayscale information of the module units to obtain all eigenvalues ​​and corresponding eigenvectors of each model to be reduced in dimension;

[0038] According to the importance of the sub-image model to which each model to be reduced in dimensionality belongs, the eigenvalue of each model to be reduced in dimensionality and the retention result of the corresponding eigenvector are determined; based on the retention result, a reconstructed model of each model to be reduced in dimensionality is obtained; in the stereo grayscale model corresponding to each case image, the splicing results of all the reconstructed models are used as the reduced dimensionality case image of each case image.

[0039] Furthermore, the method for determining the retention result includes:

[0040] In each model to be reduced in dimensionality, all eigenvalues ​​are sorted in descending order to obtain an eigenvalue sequence; if the importance of the sub-image model to which each model to be reduced in dimensionality belongs is in a preset first interval, then the eigenvalues ​​corresponding to the preset first number of eigenvalues ​​in the eigenvalue sequence are retained; if the importance of the sub-image model to which each model to be reduced in dimensionality belongs is in a preset second interval, then the eigenvalues ​​corresponding to the preset second number of eigenvalues ​​in the eigenvalue sequence are retained; if the importance of the sub-image model to which each model to be reduced in dimensionality belongs is in a preset third interval, then the eigenvalues ​​corresponding to the preset third number of eigenvalues ​​in the eigenvalue sequence are retained;

[0041] The left endpoint of the preset first interval is greater than the right endpoint of the preset second interval, and the left endpoint of the preset second interval is greater than the right endpoint of the preset third interval; the preset first quantity is greater than the preset second quantity, and the preset second quantity is greater than the preset third quantity.

[0042] The present invention has the following beneficial effects:

[0043] The present invention first obtains and classifies all case images, obtains all image groups, and provides data preparation for the establishment of a resource library; in each image group, all case images are sorted and stacked according to the similarities between different case images, and all spatial image models are constructed; the construction of spatial image models uses the idea of ​​computer tomography (CT) to improve the efficiency of subsequent analysis of the importance of local areas in each case image; in each spatial image model, each spatial image model is divided according to the grayscale similarity of the model units, and all sub-image models are obtained. The sub-image model can be regarded as the stacking result of local areas in multiple case images, so as to facilitate the subsequent unified analysis of similar groups. The importance of tissue area; according to the grayscale distribution of all model units in each sub-image model, combined with the shape information of each sub-image model and the changes after morphological processing, the importance of each sub-image model is obtained, and the importance of the sub-image model reflects the importance of the local area in each case image, so as to effectively reduce the dimension and compress it later; in each spatial image model, each sub-image model is subjected to dimensionality reduction analysis, and the dimension reduction case image of each case image is obtained in combination with the importance of each sub-image model; all dimension reduction case images in each image group are stored in the corresponding group node in the resource library structure model to obtain a constructed case image resource library. The present invention constructs a spatial image model by sorting and stacking similar case images, and determines its importance by analyzing the texture structure of the sub-image model, thereby reducing the workload of analyzing the importance of the local tissue area in each case image separately, thereby improving efficiency; and based on the importance, the dimension reduction result of the local tissue area of ​​each case image is obtained, while retaining the important lesion features, the information redundancy of the non-important tissue area features is reduced as much as possible, and the compression effect of the case image is improved, thereby improving the construction effect of the case image resource library. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0045] Figure 1 A flowchart of a method for constructing a case image resource library of a smart medical system provided by one embodiment of the present invention;

[0046] Figure 2 A schematic diagram of classification results of an image group provided by an embodiment of the present invention;

[0047] Figure 3A schematic diagram of stacking case images layer by layer in a sequence sub-segment provided by an embodiment of the present invention;

[0048] Figure 4 A schematic diagram of a corresponding spatial image model in a sequence sub-segment provided by an embodiment of the present invention;

[0049] Figure 5 A flow chart of a method for obtaining importance provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0050] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the method for constructing a case image resource library of a smart medical system proposed by the present invention, its specific implementation method, structure, features and effects, in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0051] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0052] The following is a detailed description of a method for constructing a case image resource library for a smart medical system provided by the present invention with reference to the accompanying drawings.

[0053] See also Figure 1 , which shows a flowchart of a method for constructing a case image resource library of a smart medical system provided by an embodiment of the present invention, specifically comprising:

[0054] Step S1, obtaining all case images; classifying all case images according to the imaging properties, acquisition sites and lesion characteristics of the case images to obtain all image groups.

[0055] In one embodiment of the present invention, case images collected by medical imaging departments of major hospitals are collected to provide data support for the subsequent construction of a case image resource library; then all case images are classified to facilitate subsequent unified processing and storage.

[0056] Preferably, in one embodiment of the present invention, considering that the imaging properties, acquisition sites and lesion characteristics of case images are inconsistent, all case images can be preliminarily classified to obtain all image groups; the method for obtaining the image groups includes:

[0057] All case images are divided according to the imaging attributes of the case images to obtain all primary image groups; the imaging attributes at least include X-ray, CT, B-ultrasound and MRI;

[0058] In each primary image group, all case images in the primary image group are divided according to the acquisition site of the case images to obtain all secondary image groups; the acquisition site includes at least the head, chest and limbs;

[0059] In each secondary image group, all case images within the secondary image group are divided according to the lesion characteristics of the case images to obtain all image subgroups; the lesion characteristics at least include defects, cysts and stones.

[0060] See also Figure 2 , which shows a schematic diagram of the classification results of an image group provided by an embodiment of the present invention; in other embodiments, the implementer may also set other classification criteria such as age, gender, etc. to classify all case images to obtain the final image group.

[0061] It should be noted that the case images are grayscale images when they are collected, but there may be cases where the case images are in color, in which case they need to be further converted to grayscale.

[0062] It should be noted that, since the imaging properties and acquisition angles of each case image may be different, the scales of case images in the same image group are different. Therefore, it is necessary to use the Scale-Invariant Feature Transform (SIFT) algorithm to perform feature matching on all case images in each image group, and then calculate the scale transformation matrix through the matching key points, so as to scale all case images in each image group to a uniform scale for subsequent analysis and processing; grayscale, SIFT algorithm and image scaling are already existing technologies well known to technicians in this field and will not be elaborated here; all case images in each image group used for subsequent analysis and processing are images scaled to the same scale.

[0063] Step S2: In each image group, all case images are sorted and stacked according to the similarities between different case images to construct all spatial image models. Each model unit in the spatial image model corresponds to a pixel point in the case image. In each spatial image model, each spatial image model is divided according to the grayscale similarity of the model units to obtain all sub-image models.

[0064] In order to effectively reduce the dimensionality of each case image, the embodiment of the present invention needs to analyze and evaluate the importance of different local areas in each case image, so as to retain the principal components according to the importance, so that the dimensionality reduction result can retain the important lesion characteristics while reducing the information redundancy of non-important tissue area characteristics as much as possible, thereby improving the compression effect of the case image, so as to subsequently improve the construction effect of the case image resource library.

[0065] Taking into account that the importance of local areas is analyzed for the case images in each imaging group, the calculation amount is large and the efficiency is low; and considering that the case images in each imaging group have been preliminarily classified and have similar imaging features, similar case images are stacked to construct a spatial imaging model, so that the three-dimensional feature areas corresponding to similar feature areas can be obtained, such as three-dimensional lesion areas, three-dimensional normal tissue areas, etc., and then the importance of the corresponding local planar area in each case image can be obtained by analyzing the importance of each three-dimensional area; the construction of the spatial imaging model uses the idea of ​​computed tomography (CT) to improve the efficiency of analyzing the importance of the corresponding local planar area in each case image.

[0066] Therefore, the embodiment of the present invention will first sort and stack all case images in each image group according to the similarities between different case images, and construct all spatial image models, in which each model unit corresponds to a pixel point in the case image.

[0067] Preferably, in one embodiment of the present invention, considering that the lesion features in each image group are similar, but the severity or specific location of the lesion is different, it is necessary to first sort and segment all case images in the image group before stacking and constructing the spatial image model; therefore, the method for obtaining the spatial image model includes:

[0068] In each image group, all case images are sorted and segmented according to the similarity between each case image and all other case images, and all sequence subsegments are obtained;

[0069] The pixels in each case image are converted into cubic pixels, and the model units corresponding to all pixels are obtained to obtain the stereo grayscale model of each case image; the stereo grayscale models corresponding to all case images in each sequence sub-segment are sorted and stacked layer by layer to construct the spatial image model corresponding to each sequence sub-segment.

[0070] Among them, considering that the hash value of the image is also called a digital fingerprint, it can help describe the structural characteristics of the image; considering that the sorting and stacking of case images should follow the order from the most representative to the most unique, so that the spatial image model can transition naturally during the construction process to avoid poor construction effect; considering that there may be large differences between adjacent case images in the image sequence, the image sequence can be divided and truncated, and then multiple spatial image models can be constructed to eliminate the situation where the entire spatial image model is poorly constructed; therefore, the method for obtaining the sequence sub-segments includes:

[0071] In each image group, the mean Hamming distance of the hash values ​​between each case image and all other case images is used as the ranking parameter of each case image;

[0072] The case image corresponding to the maximum sorting parameter is used as one end point of the image sequence, and the case image corresponding to the minimum sorting parameter is used as the other end point of the image sequence, and the image sequence of each image group is obtained based on the state compression dynamic programming algorithm;

[0073] In each image sequence, it is determined whether the Hamming distance of the hash values ​​between adjacent case images exceeds a preset threshold. If it exceeds the preset threshold, the image sequence is segmented to obtain all sequence sub-segments of each image sequence.

[0074] As an example, firstly, based on the average hash algorithm, the 64-bit hash value of each case image in each image group is obtained, and then the average of the Hamming distances between the hash values ​​of each case image and all other case images is calculated to obtain the sorting parameters of each case image, and the case image corresponding to the minimum sorting parameter is used as the sequence starting point of the image sequence, and the case image corresponding to the maximum sorting parameter is used as the sequence end point, and then the sorting method of all case images in each image group is planned based on the state compression dynamic programming algorithm, so that the Hamming distance between all adjacent case images is minimized, thereby obtaining the image sequence of each image group;

[0075] Then, in each image sequence, if the Hamming distance of the hash values ​​between adjacent case images exceeds a preset threshold value, such as 50, the image sequence is disconnected in the corresponding adjacent case images to obtain all sequence sub-segments, and the preset threshold value can also be set by oneself; for example, there is an image sequence of a certain image group {A, B, C, D, E, F, G, H, I}, where the Hamming distance of the hash value between DE and GH exceeds the preset threshold value, then all the sequence sub-segments of the image group are {A, B, C, D}, {E, F, G}, {H, I};

[0076] Then, in each sequence sub-segment, the pixels in each case image are converted from plane to three-dimensional to obtain the model unit corresponding to each pixel, thereby obtaining the three-dimensional grayscale model of each case image; the three-dimensional grayscale models corresponding to all case images in each sequence sub-segment are sorted and stacked layer by layer according to the sorting order of the corresponding case images in the sequence sub-segment, thereby constructing the spatial image model corresponding to each sequence sub-segment.

[0077] See also Figure 3 and Figure 4 , which shows a schematic diagram of stacking case images layer by layer in a sequence sub-segment provided by an embodiment of the present invention, and a schematic diagram of a corresponding spatial image model in a sequence sub-segment provided by an embodiment of the present invention; Figure 3 In the example, case images of size L×L are stacked layer by layer along the Z axis, where the bottom case image corresponds to the end point of the sequence subsegment, and the Figure 4 The spatial image model of the case image corresponding to the stereo grayscale model after layer-by-layer stacking, Figure 4 Each small cube corresponds to a pixel, which is also a model unit. The grayscale value of the pixel is the grayscale value of the model unit. Each layer of model units corresponds to a case image.

[0078] In another embodiment of the present invention, the implementer may also use the Euclidean distance between the grayscale histograms corresponding to the case images to replace the Hamming distance of the hash values ​​between the case images, to obtain the sorting parameters of each case image, thereby screening out the starting point and the end point of the image sequence, and then plan the sorting method of all case images in each image group based on the state compression dynamic programming algorithm to obtain the image sequence of each image group; then, the image sequence is segmented using the judgment method based on the preset threshold to obtain all sequence sub-segments; then, in each sequence sub-segment, each pixel point in each case image is assigned a spatial coordinate label, such as (i, p, q) the pixel point in the pth row and qth column in the ith case image, mapped to the spatial rectangular coordinate system, thereby constructing a structure such as Figure 4 In the spatial image model shown, each module unit corresponds to a pixel point.

[0079] In other embodiments of the present invention, the implementer may also use the SSIM structural similarity index between case images to replace the Hamming distance of the hash values ​​between case images, and then construct a spatial image model. It should be noted that the average hash algorithm, Hamming distance, state compression dynamic programming algorithm, grayscale histogram and SSIM structural similarity index are all prior arts well known to those skilled in the art and will not be described in detail here.

[0080] After obtaining all the spatial image models constructed by stacking case images, each spatial image model can be divided according to the grayscale similarity of the model units in each spatial image model to obtain all sub-image models; the sub-image model roughly represents the stacking results of the corresponding local similar tissue areas in each case image in the sequence sub-segment, such as normal organ tissue or lesion tissue, so as to subsequently uniformly analyze the importance of similar tissue areas, thereby obtaining the importance of each local area of ​​each case image in the sequence sub-segment and improving the efficiency of dimensionality reduction in subsequent analysis.

[0081] Preferably, in one embodiment of the present invention, considering that the grayscale expressions of the module units corresponding to the same tissue region should be similar, the module units can be subjected to regional connectivity detection in combination with the grayscale information of the module units, and the module units with similar grayscales are clustered and analyzed to obtain all sub-image models; therefore, the method for obtaining the sub-image model includes:

[0082] In each spatial image model, a preset first number of seed units is selected from all model units, where the preset first number is the same as the number of grayscale values ​​of all model units; based on the regional connectivity algorithm and the grayscale difference between the model units, a sub-image model of each seed unit is obtained;

[0083] If there is a model unit without a corresponding sub-image model, the model unit without a corresponding sub-image model is used as a unit to be clustered, and a preset second number of new seed units are screened out from all the units to be clustered, and the preset second number is the same as the number of grayscale values ​​of all the units to be clustered; among all the units to be clustered in each spatial image model, the sub-image model of each new seed unit is obtained based on the regional connectivity algorithm and the grayscale difference between the units to be clustered;

[0084] The iteration is continued until all model units in each spatial image model have their corresponding sub-image models, and all sub-image models in each spatial image model are obtained.

[0085] As an example, in each spatial image model, the gray levels of all model units are first counted, and then a preset first number of seed units are screened out from all model units. The preset first number is consistent with the number of gray value types of all model units. For example, if there are 20 gray value types in all model units, the number of seed units is set to 20, that is, one module unit is selected from each gray value type as a seed unit; then, in a 3×3×3 spatial neighborhood centered on each seed unit, it is determined whether all neighborhood module units meet the connected growth conditions, such as whether the gray difference between the neighborhood module unit and the seed unit is less than 8. If it is less than 8, spatial region connected growth is performed, and then each module unit of the connected growth in the spatial neighborhood is used as the center for the next connected growth, until no module unit can be added. At this point, the spatial connected growth result of each seed unit can be obtained, that is, the corresponding sub-image model;

[0086] Since the connected growth results of the initially screened seed units may not be sufficient to fill the entire spatial image model, it is necessary to further judge and screen out all model units in the spatial image model that have no corresponding sub-image models, and then select new seed units based on the screening from all model units that have no corresponding sub-image models, and perform spatial connected growth to obtain sub-image models corresponding to all new seed units; wherein the preset second number is consistent with the number of grayscale value types of all model units that have no corresponding sub-image models after the first spatial connected growth; the iteration is continued until all model units have sub-image models, and all sub-image models in each spatial image model are obtained.

[0087] In other embodiments of the present invention, implementers may also replace the idea of ​​spatial connected growth with the idea of ​​clustering to obtain all sub-image models, which and the regional connected algorithm are well-known technologies and will not be described in detail here.

[0088] Step S3, according to the grayscale distribution of all model units in each sub-image model, combined with the shape information of each sub-image model and the changes after morphological processing, obtain the importance of each sub-image model; in each spatial image model, perform dimensionality reduction analysis on each sub-image model, and obtain the reduced dimensionality case image of each case image combined with the importance of each sub-image model.

[0089] Taking into account that the more complex the texture features of a local area in a case image, the more likely it is to correspond to a key observation area for medical research such as a lesion tissue, while the more simple the texture features, the more likely it is to be a normal tissue area; the importance of the local area with complex texture features is relatively high, and the importance of the sub-image model corresponding to the superposition of many case images is also relatively high; and considering that the grayscale distribution and morphological structure can reflect the complexity of the texture structure, the embodiment of the present invention obtains the importance of each sub-image model based on the grayscale distribution of all model units in each sub-image model, combined with the shape information of each sub-image model and the changes after morphological processing; the importance of the sub-image model indirectly reflects the importance of the corresponding local tissue area in each case image, so as to facilitate subsequent effective dimensionality reduction and compression.

[0090] Preferably, in one embodiment of the present invention, the method for obtaining the importance includes:

[0091] See also Figure 5 , which shows a flow chart of a method for obtaining importance provided by an embodiment of the present invention, specifically comprising:

[0092] Step S501 , in each sub-image model, according to the quantity distribution of module units at each gray value, the gray stability of each sub-image model is obtained.

[0093] In a preferred embodiment of the present invention, considering that the more extreme and clustered the distribution of the number of module units at each grayscale value in each sub-image model is, the closer the grayscale values ​​of the module units in the sub-image model are, and the simpler the grayscale distribution is; therefore, the method for obtaining grayscale stability includes:

[0094] In each sub-image model, the total number of module units at each grayscale value is counted; the total number of grayscale values ​​with a total number of units of 0 is taken as the first stability parameter; the difference between the total number of units at all adjacent grayscale values ​​is accumulated as the second stability parameter; the grayscale stability is obtained according to the first stability parameter and the second stability parameter; the first stability parameter and the second stability parameter are both positively correlated with the grayscale stability.

[0095] As an example, the grayscale stability is calculated as: ;in For the Grayscale stability of each sub-image model; For the The first stable parameter of the sub-image model; is the sequence number of the gray value; For the The total number of cells of the module unit under gray value; For the The total number of cells of the module unit under gray value; is the absolute value symbol.

[0096] In the grayscale stability calculation formula, the greater the total number of grayscale values ​​with a total number of units of 0, that is, the greater the first stability parameter, the more the grayscale values ​​of the module units in the sub-image model tend to be similar, and the greater the grayscale stability; at the same time, the greater the difference between the total number of units under all adjacent grayscales, the greater the possibility of a cliff-like difference in the total number of units between adjacent grayscale values, which indirectly reflects that the grayscale of the module units in the sub-image model tends to be similar, and the greater the grayscale stability. Multiply the first stability parameter and the second stability parameter to obtain the grayscale stability. In other examples, the implementer can also use basic mathematical operations such as addition or weighted summation or related mapping methods to combine the two, which will not be repeated here.

[0097] Step S502, obtaining the degree of morphological change of each sub-image model according to the quantitative difference between the total number of module units corresponding to the sub-image model after three-dimensional expansion and three-dimensional erosion; the quantitative difference is positively correlated with the degree of morphological change.

[0098] Considering that after the morphological operation on the sub-image model, if the difference between the before and after changes is large, the degree of morphological change is also greater; it usually means that the texture and shape changes in the sub-image model are more complex, containing more details and features, and its importance is greater; therefore, the degree of morphological change indirectly reflects the importance of the sub-image model.

[0099] As an example, the total number of module units of the sub-image model after three-dimensional expansion is used as the numerator, the total number of module units after three-dimensional erosion plus a preset normal number 1 is used as the denominator, and the ratio of the numerator to the denominator is used as the degree of morphological change; the preset normal number 1 is set to avoid the disappearance of the sub-image model after three-dimensional erosion, which makes the fraction meaningless; the larger the fraction ratio, the greater the degree of morphological change; in other examples, the absolute value of the difference between the total number of module units of the sub-image model after three-dimensional expansion and three-dimensional erosion can also be directly used as the degree of morphological change.

[0100] It should be noted that three-dimensional erosion and three-dimensional expansion are both existing technologies and will not be described in detail here.

[0101] Step S503, in each sub-image model, according to the shape difference between the single-layer modules formed by the module units corresponding to different case images, the inter-layer difference parameter of each sub-image model is obtained, and the shape difference is positively correlated with the inter-layer difference parameter.

[0102] Considering that the sub-image model can be approximately regarded as the stacking result of local tissue areas of multiple case images, the greater the difference in local tissue areas between case images corresponding to different layers in the stacking process, the less likely it is that it corresponds to a normal tissue area, and the more likely it is that it is a lesion tissue area that needs to be studied in depth; therefore, the importance can be evaluated based on the difference in local tissue areas between case images corresponding to different layers in the sub-image model, that is, the difference in shape between single-layer modules composed of module units corresponding to different case images; single-layer modules composed of module units corresponding to different case images correspond to local tissue areas corresponding to different case images. The greater the difference in shape between different layers, the larger the inter-layer difference parameter, and the greater the importance of the sub-image model.

[0103] In a preferred embodiment of the present invention, considering that the edge shape of a region roughly reflects the shape of a local region, and the chain code can quantify the edge shape, it is convenient to measure the shape difference; considering that the greater the difference between edge chain codes of different layers, the greater the difference between layers; therefore, the method for obtaining the inter-layer difference parameter includes:

[0104] In each sub-image model, all edge modules in the single-layer module composed of the module units corresponding to each case image are obtained; in each spatial image model, any module unit is used as a reference unit; in each single-layer module, the edge module closest to the reference unit is used as the starting point to obtain the edge chain codes of all edge modules in each single-layer module;

[0105] Based on the multiple sequence comparison algorithm, the distance matrix of all edge chain codes in each sub-image model is obtained, and the mean of all matrix elements in the distance matrix is ​​used as the first difference parameter; the second difference parameter is obtained according to the length difference of all edge chain codes, and the length difference is positively correlated with the second difference parameter;

[0106] The inter-layer difference parameter is obtained according to the first difference parameter and the second difference parameter; the first difference parameter and the second difference parameter are both positively correlated with the inter-layer difference parameter.

[0107] As an example, in each sub-image model, all the outermost unit modules in the single-layer module composed of the module units corresponding to each case image are taken as edge modules, and then the coordinate origin of the spatial image model is taken as the reference unit. In each single-layer module corresponding to the case image, the edge module with the closest Euclidean distance to the reference unit is taken as the starting point, and then all the edge modules of each single-layer module are chain-code traced to obtain the edge chain code corresponding to each single-layer module;

[0108] Then, based on the multiple sequence comparison algorithm, the distance matrix of all edge chain codes in each sub-image model is obtained. Each matrix element in the distance matrix reflects the chain code difference between different edge chain codes, which can more clearly evaluate the difference between the local tissue regions corresponding to different case images. Then, when constructing the distance matrix, the number of vacancies that each edge chain code needs to be aligned and filled is evaluated, that is, the difference between the chain code length of each edge chain code and the maximum chain code length. Finally, the inter-layer difference parameter is obtained based on the calculation formula of the inter-layer difference parameter.

[0109] The calculation formula of the interlayer difference parameter is: ;in, For the The inter-layer difference parameters of the sub-image model; For the The sub-image model corresponds to the sequence number of the matrix element in the distance matrix; For the The total number of matrix elements in the distance matrix corresponding to each sub-image model; For the The sub-image model corresponds to the first matrix elements; is the first difference parameter; For the The serial number of the edge chain code in the sub-image model; For the The total number of edge chain codes in each sub-image model; For the In the sub-image model The number of vacancies in the edge chain code; is the second difference parameter.

[0110] In the calculation formula of the inter-layer difference parameter, the larger the first difference parameter and the second difference parameter, the greater the difference in shape of the single-layer modules between different layers; the first difference parameter and the second difference parameter are multiplied and combined to obtain the inter-layer difference parameter; in other examples, the implementer may also use basic mathematical operations such as addition or weighted summation or related mapping methods to combine the two, which will not be repeated here.

[0111] Step S504, obtaining the importance of the corresponding sub-image model according to the grayscale stability, the degree of morphological change and the inter-layer difference parameter; the grayscale stability is negatively correlated with the importance, and the degree of morphological change and the inter-layer difference parameter are both positively correlated with the importance.

[0112] As an example, the importance is calculated as: ;in, For the Grayscale stability of each sub-image model; For the Grayscale stability of each sub-image model; For the The inter-layer difference parameters of the sub-image model; For the The degree of morphological change of each sub-image model; is the standard normalization function; It is an exponential function with the natural constant e as its base.

[0113] In this example, linear normalization is specifically used for normalization, and other normalization methods may also be used, which will not be described in detail here; implementers may also use other basic mathematical operations or related mapping methods to combine the three, which will not be described in detail here.

[0114] At this point, the importance of each sub-image model is obtained, that is, the importance of the local tissue area corresponding to the single-layer module in each case image constituting the sub-image model is obtained.

[0115] The embodiment of the present invention further performs dimensionality reduction analysis on each sub-image model in each spatial image model, and obtains a dimensionality reduction case image of each case image in combination with the importance of each sub-image model. The importance can be combined to retain the principal components in different local tissue regions in each case image, so that the dimensionality reduction result can reduce the information redundancy of non-important tissue region features as much as possible while retaining important lesion features, thereby improving the compression effect of the case image.

[0116] Preferably, in one embodiment of the present invention, the method for obtaining the reduced-dimensional case image includes:

[0117] In the stereo grayscale model corresponding to each case image in each spatial image model, all module units belonging to the same sub-image model constitute a sub-stereo grayscale model as a model to be reduced in dimension, and principal component analysis is performed on each model to be reduced in dimension based on the grayscale information of the module units to obtain all eigenvalues ​​and corresponding eigenvectors of each model to be reduced in dimension;

[0118] According to the importance of the sub-image model to which each model to be reduced belongs, the retention results of the eigenvalues ​​and corresponding eigenvectors of each model to be reduced are determined; based on the retention results, the reconstructed model of each model to be reduced is obtained; in the stereo grayscale model corresponding to each case image, the splicing results of all the reconstructed models are used as the reduced dimensionality case image of each case image.

[0119] The methods for determining the retention results include:

[0120] In each model to be reduced in dimensionality, all eigenvalues ​​are sorted in descending order to obtain an eigenvalue sequence; if the importance of the sub-image model to which each model to be reduced in dimensionality is within a preset first interval, the eigenvalues ​​corresponding to the preset first number in the eigenvalue sequence are retained; if the importance of the sub-image model to which each model to be reduced in dimensionality is within a preset second interval, the eigenvalues ​​corresponding to the preset second number in the eigenvalue sequence are retained; if the importance of the sub-image model to which each model to be reduced in dimensionality is within a preset third interval, the eigenvalues ​​corresponding to the preset third number in the eigenvalue sequence are retained; the left end point of the preset first interval is greater than the right end point of the preset second interval, and the left end point of the preset second interval is greater than the right end point of the preset third interval; the preset first number is greater than the preset second number, and the preset second number is greater than the preset third number.

[0121] As an example, first obtain the model to be reduced in dimension in the stereo gray model corresponding to each case image in each spatial image model, and the model to be reduced is the local tissue area corresponding to the single-layer module in each case image constituting the sub-image model; it should be noted that the model to be reduced may be an irregular shape, so it is also necessary to obtain its minimum circumscribed cube, and the gray value corresponding to the newly added module unit in the minimum circumscribed rectangle is 0, and the models to be reduced mentioned later are all minimum circumscribed cubes; then obtain the local covariance matrix in the model to be reduced based on the gray information, so as to use the principal component analysis algorithm to obtain all the principal components in each model to be reduced, that is, the eigenvectors, and obtain the eigenvalues ​​corresponding to each eigenvector;

[0122] Among them, the preset first interval is (0.7, 1], the preset second interval is (0.5, 0.7], the preset third interval is [0, 0.5], the preset first number is the first 50, the preset second number is the first 20, and the preset third number is the first 10; that is, for each model to be reduced in dimension, when its importance is greater than 0.7, the eigenvalues ​​corresponding to the eigenvectors of the top 50 are retained; when its importance is in (0.5, 0.7], the eigenvalues ​​corresponding to the eigenvectors of the top 20 are retained; when its importance is less than or equal to 0.5, the eigenvalues ​​corresponding to the eigenvectors of the top 10 are retained; the smaller the importance, the fewer the principal components, i.e., the eigenvectors, are retained, thereby reducing information redundancy; on the contrary, the greater the importance, the more the principal components, i.e., the eigenvectors, are retained, thereby retaining texture details;

[0123] The dimension reduction retention result of each model to be reduced is used as the reconstructed model, and the splicing result of all the reconstructed models in the stereo grayscale model corresponding to each case image is used as the reduced dimension case image of each case image.

[0124] It should be noted that performing principal component analysis on an image is an existing technology well known to those skilled in the art and will not be elaborated here. The implementer may also set the eigenvalues ​​of each model to be reduced and the retention method of the corresponding eigenvector, such as the setting of the preset interval and the preset number, which will not be elaborated here.

[0125] Step S4, storing all dimension-reduced case images in each image group into the corresponding group node in the resource library structure model to construct a case image resource library.

[0126] In one embodiment of the present invention, all the reduced-dimensional case images in each image group are stored in the corresponding group node of the resource library structure model, that is, all the reduced-dimensional case images in the same image group are stored in the same node for subsequent retrieval and sharing, so that a constructed case image resource library can be obtained. It should be noted that the construction and application of the resource library model is a prior art well known to those skilled in the art and will not be described in detail.

[0127] In summary, the present invention first obtains all case images, and then classifies and obtains all image groups; sorts and stacks similar case images in each image group to construct a spatial image model, and obtains a sub-image model formed by stacking similar local tissue areas in each case image, and determines its importance by analyzing the texture structure of the sub-image model, thereby reducing the workload of separately analyzing the importance of the local tissue area in each case image, thereby improving efficiency; and based on the importance, obtains the dimensionality reduction result of the local tissue area of ​​each case image, while retaining the important lesion features, reduces the information redundancy of the non-important tissue area features as much as possible, improves the compression effect of the case image, and thus improves the construction effect of the case image resource library.

[0128] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0129] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A method for constructing a case image resource library of a smart medical system, characterized in that: The method comprises: Acquire all case images; classify all case images according to the imaging properties, acquisition sites and lesion characteristics of the case images, and acquire all image groups; In each of the image groups, all the case images are sorted and stacked according to the similarities between different case images, and all spatial image models are constructed, wherein each model unit in the spatial image model corresponds to a pixel point in the case image; in each of the spatial image models, each of the spatial image models is divided according to the grayscale similarity of the model units to obtain all sub-image models; According to the grayscale distribution of all model units in each sub-image model, combined with the shape information of each sub-image model and the changes after morphological processing, the importance of each sub-image model is obtained; in each of the spatial image models, a dimensionality reduction analysis is performed on each sub-image model, and combined with the importance of each sub-image model, a dimensionality reduction case image of each case image is obtained; All dimension-reduced case images in each of the image groups are stored in the corresponding group nodes in the resource library structure model to construct a case image resource library; The method for acquiring the spatial image model includes: In each of the image groups, all of the case images are sorted and segmented according to the similarity between each of the case images and all other case images, to obtain all sequence sub-segments; Convert the pixels in each case image into cubic pixels, obtain the model units corresponding to all the pixels, and obtain the stereo grayscale model of each case image; sort and stack the stereo grayscale models corresponding to all the case images in each sequence sub-segment layer by layer to construct the spatial image model corresponding to each sequence sub-segment; The method for obtaining the sub-image model includes: In each of the spatial image models, a preset first number of seed units are selected from all model units, where the preset first number is the same as the number of grayscale values ​​of all model units; based on a regional connectivity algorithm and grayscale differences between the model units, a sub-image model of each of the seed units is obtained; If there is a model unit without a corresponding sub-image model, the model unit without a corresponding sub-image model is used as a unit to be clustered, and a preset second number of new seed units are screened out from all the units to be clustered, and the preset second number is the same as the number of grayscale values ​​of all the units to be clustered; in all the units to be clustered in each of the spatial image models, a sub-image model of each new seed unit is obtained based on the regional connectivity algorithm and the grayscale difference between the units to be clustered; Iterate continuously until all model units in each of the spatial image models have corresponding sub-image models, and then obtain all sub-image models in each of the spatial image models; The method for obtaining the importance degree includes: In each of the sub-image models, according to the quantity distribution of the module units at each gray value, the grayscale stability of each of the sub-image models is obtained; According to the quantitative difference between the total number of corresponding module units of the sub-image model after three-dimensional expansion and three-dimensional erosion, the degree of morphological change of each sub-image model is obtained; the quantitative difference is positively correlated with the degree of morphological change; In each of the sub-image models, according to the shape differences between the single-layer modules formed by the module units corresponding to different case images, the inter-layer difference parameters of each of the sub-image models are obtained, and the shape difference is positively correlated with the inter-layer difference parameters; The importance of the corresponding sub-image model is obtained according to the grayscale stability, the morphological change degree and the inter-layer difference parameter; the grayscale stability is negatively correlated with the importance, and the morphological change degree and the inter-layer difference parameter are both positively correlated with the importance.

2. The method for constructing a case image resource library of a smart medical system according to claim 1, characterized in that: The method for obtaining the image group includes: Dividing all the case images according to the imaging attributes of the case images to obtain all primary image groups; Dividing all the case images in the first-level image group according to the acquisition site of the case images to obtain all the second-level image groups; All the case images in the secondary image group are divided according to the lesion characteristics of the case images to obtain all image subgroups.

3. The method for constructing a case image resource library of a smart medical system according to claim 1, characterized in that: The method for obtaining the sequence sub-segments includes: In each of the image groups, the average of the Hamming distances of the hash values ​​between each of the case images and all other case images is used as a sorting parameter for each of the case images; Taking the case image corresponding to the maximum sorting parameter as one endpoint of the image sequence, taking the case image corresponding to the minimum sorting parameter as the other endpoint of the image sequence, and obtaining the image sequence of each image group based on a state compression dynamic programming algorithm; In each of the image sequences, it is determined whether the Hamming distance of the hash values ​​between adjacent case images exceeds a preset threshold. If it exceeds the preset threshold, the image sequence is segmented to obtain all sequence sub-segments of each of the image sequences.

4. The method for constructing a case image resource library of a smart medical system according to claim 1, characterized in that: The grayscale stability acquisition method includes: In each of the sub-image models, the total number of module units at each gray value is counted; the total number of gray values ​​where the total number of units is 0 is used as a first stable parameter; and the difference between the total number of units at all adjacent gray values ​​is accumulated as a second stable parameter; The grayscale stability is obtained according to the first stability parameter and the second stability parameter; the first stability parameter and the second stability parameter are both positively correlated with the grayscale stability.

5. The method for constructing a case image resource library of a smart medical system according to claim 1, characterized in that: The method for obtaining the inter-layer difference parameter includes: In each of the sub-image models, all edge modules in the single-layer module composed of the module units corresponding to each of the case images are obtained; in each of the spatial image models, any module unit is used as a reference unit; in each of the single-layer modules, the edge module closest to the reference unit is used as a starting point to obtain the edge chain codes of all edge modules in each of the single-layer modules; Based on the multiple sequence comparison algorithm, a distance matrix of all the edge chain codes in each sub-image model is obtained, and the mean of all matrix elements in the distance matrix is ​​used as a first difference parameter; a second difference parameter is obtained according to the length difference of all the edge chain codes, and the length difference is positively correlated with the second difference parameter; An inter-layer difference parameter is obtained according to the first difference parameter and the second difference parameter; the first difference parameter and the second difference parameter are both positively correlated with the inter-layer difference parameter.

6. The method for constructing a case image resource library of a smart medical system according to claim 1, characterized in that: The method for obtaining the reduced-dimensional case image comprises: In the stereo grayscale model corresponding to each case image of each of the spatial image models, all module units belonging to the same sub-image model constitute a sub-stereo grayscale model as a model to be reduced in dimension, and principal component analysis is performed on each of the models to be reduced in dimension based on the grayscale information of the module units to obtain all eigenvalues ​​and corresponding eigenvectors of each model to be reduced in dimension; According to the importance of the sub-image model to which each model to be reduced in dimensionality belongs, the eigenvalue of each model to be reduced in dimensionality and the retention result of the corresponding eigenvector are determined; based on the retention result, a reconstructed model of each model to be reduced in dimensionality is obtained; in the stereo grayscale model corresponding to each case image, the splicing results of all the reconstructed models are used as the reduced dimensionality case image of each case image.

7. The method for constructing a case image resource library of a smart medical system according to claim 6, characterized in that: The method for determining the retention result includes: In each model to be reduced in dimensionality, all eigenvalues ​​are sorted in descending order to obtain an eigenvalue sequence; if the importance of the sub-image model to which each model to be reduced in dimensionality belongs is in a preset first interval, then the eigenvalues ​​corresponding to the preset first number of eigenvalues ​​in the eigenvalue sequence are retained; if the importance of the sub-image model to which each model to be reduced in dimensionality belongs is in a preset second interval, then the eigenvalues ​​corresponding to the preset second number of eigenvalues ​​in the eigenvalue sequence are retained; if the importance of the sub-image model to which each model to be reduced in dimensionality belongs is in a preset third interval, then the eigenvalues ​​corresponding to the preset third number of eigenvalues ​​in the eigenvalue sequence are retained; The left endpoint of the preset first interval is greater than the right endpoint of the preset second interval, and the left endpoint of the preset second interval is greater than the right endpoint of the preset third interval; the preset first quantity is greater than the preset second quantity, and the preset second quantity is greater than the preset third quantity.

Images