CN110599508B

CN110599508B - Artificial intelligence-based spine image processing method and related equipment

Info

Publication number: CN110599508B
Application number: CN201910706559.9A
Authority: CN
Inventors: 陶蓉; 吴海萍; 吕传峰
Original assignee: Ping An Technology Shenzhen Co Ltd
Current assignee: Ping An Technology Shenzhen Co Ltd
Priority date: 2019-08-01
Filing date: 2019-08-01
Publication date: 2023-10-27
Anticipated expiration: 2039-08-01
Also published as: WO2021017297A1; CN110599508A

Abstract

The invention relates to the field of artificial intelligence, and discloses a spine image processing method and related equipment based on artificial intelligence, which improve the segmentation accuracy of vertebrae and the specificity and sensitivity of sign recognition. The method comprises the following steps: acquiring an original spine image; preprocessing an original spine image to generate a target spine image; dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks; performing vertebra contour recognition and corner detection on a plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners; synthesizing the N bone block outlines, N4 vertebra corner points and the target spine image to generate a synthesized vertebra image; extracting a plurality of small images from the composite vertebra image, each small image including information of a target area; and identifying the plurality of small images through a preset sign classification model to generate an identification result.

Description

Artificial intelligence-based spine image processing method and related equipment

Technical Field

The invention relates to the field of artificial intelligence, in particular to a spine image processing method and related equipment based on artificial intelligence.

Background

The sagittal image of the spine has important clinical significance in evaluating the function of the spine and diagnosing the related diseases of the spine, and is specifically expressed in the following steps: 1. the spine sagittal data source is wide, and various scanning modes including X-ray, electronic computer tomography (computed tomography, CT), magnetic resonance imaging (magnetic resonance imaging, MRI) and the like provide sagittal images. 2. The sagittal image has wide application range in disease diagnosis, relates to various clinical related symptoms such as fracture, slippage, hyperosteogeny and the like, and covers lesions of a plurality of areas of sacrum, lumbar vertebra, thoracic vertebra and cervical vertebra. 3. Closely related to prognosis and quality of life of patients, parameters extracted from sagittal images are clinically used for quantitatively evaluating recovery of patients after spine surgery.

In the current market, the spine sagittal image analysis method uses traditional imaging methods, such as edge detection and the like, to position the edges, angles and other positions of vertebrae, and the method has low precision and large image quality, and particularly greatly reduces the success rate of identifying the boundary and points of a lesion area.

Disclosure of Invention

The invention provides an artificial intelligence-based spine image processing method and related equipment, which are used for respectively realizing vertebra segmentation and disease symptom classification through two different depth networks, so that the segmentation accuracy of vertebrae is improved, and the specificity and sensitivity of symptom recognition are improved.

A first aspect of an embodiment of the present invention provides a spine image processing method based on artificial intelligence, including: acquiring an original spine image, wherein the original spine image is a sagittal radiological image of a spine; preprocessing the original spine image to generate a target spine image; dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra; performing vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1; synthesizing the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image; extracting a plurality of small images from the synthesized vertebra image, each small image comprising information of a target area; and identifying the plurality of small images through a preset sign classification model to generate an identification result.

Optionally, in a first implementation manner of the first aspect of the embodiment of the present invention, the preprocessing the original spine image to generate a target spine image includes: processing the pixel size of the original spine image to obtain a processed first spine image; and carrying out parameter adjustment on the first spine image to generate a target spine image.

Optionally, in a second implementation manner of the first aspect of the embodiment of the present invention, the processing the pixel size of the original spine image to obtain a processed first spine image includes: performing black edge cutting treatment on the original spine image; cutting the spine image after the black edge is cut off; and adjusting the size of the cut spine image to obtain a processed first spine image.

Optionally, in a third implementation manner of the first aspect of the embodiment of the present invention, the performing parameter adjustment on the first spine image, generating a target spine image includes: determining the number of image channels of the first spine image; and adjusting the window width and the window level of the first spine image according to the image channel number to generate a target spine image.

Optionally, in a fourth implementation manner of the first aspect of the embodiment of the present invention, the segmenting each vertebra in the target spine image by using a preset segmentation model generates a plurality of vertebra masks, where each vertebra mask corresponds to a different vertebra, and the method includes: judging whether sacral vertebrae exist in the target spine image through a preset first segmentation model; if the sacral vertebrae exist in the target spinal column image, separating the sacral vertebrae to generate a sacral vertebrae mask, and marking the mask as a category I; judging whether a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image or not through a preset second segmentation model; if a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image, separating the fifth lumbar vertebra, generating a fifth lumbar vertebra mask, and marking the fifth lumbar vertebra mask as a category II; judging whether a thoracic vertebra, a first lumbar vertebra, a second lumbar vertebra, a third lumbar vertebra or a fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image through a preset third segmentation model; if the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image, separating the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which exist, generating a corresponding thoracic vertebra mask or a lumbar vertebra mask, and marking the thoracic vertebra mask or the lumbar vertebra mask as a category III; judging whether cervical vertebrae exist in the target spine image or not through a preset fourth segmentation model; if the cervical vertebrae exist in the target spine image, separating the cervical vertebrae to generate a cervical vertebrae mask, and marking the mask as a category IV.

Optionally, in a fifth implementation manner of the first aspect of the embodiment of the present invention, performing vertebra contour recognition and corner detection on the plurality of vertebra masks by using a preset clustering algorithm to obtain N bone block contours and n×4 vertebra corners, where N is greater than or equal to 1, and includes: identifying bone block contours of N vertebrae through a preset fuzzy energy algorithm; obtaining M candidate points of each vertebra through a preset Harris corner detection algorithm, wherein M is greater than or equal to 4; dividing the M candidate points into P point clusters by a density-based clustering algorithm DBSCAN algorithm with noise; respectively calculating the central points of the P point clusters, and determining the P central points as P vertebra corner points; and removing redundant vertebral corner points or filling up missing vertebral corner points through a minimum circumscribed rectangle algorithm to obtain N4 vertebral corner points.

Optionally, in a sixth implementation manner of the first aspect of the embodiment of the present invention, the identifying, by a preset sign classification model, the plurality of small-block images includes: calling a preset depth residual error network model to identify the plurality of small images; isolating a disease patch image comprising a disease sign; the location of each disease patch image on the composite vertebral image is determined and a recognition result is output, including the vertebral center point, corner offset, and disc thickness.

A second aspect of an embodiment of the present invention provides an artificial intelligence-based spine image processing apparatus, including: the first acquisition unit is used for acquiring an original spine image, wherein the original spine image is a sagittal radial image of a spine; the preprocessing unit is used for preprocessing the original spine image to generate a target spine image; the segmentation unit is used for segmenting each vertebra in the target spine image through a preset segmentation model to generate a plurality of vertebra masks, and each vertebra mask corresponds to a different vertebra; the identification detection unit is used for carrying out vertebra outline identification and corner detection on the vertebra masks through a preset clustering algorithm to obtain N bone block outlines and N4 vertebra corner points, wherein N is greater than or equal to 1; the synthesis unit is used for synthesizing the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image; an extraction unit for extracting a plurality of small images from the synthesized vertebra image, each small image including information of a target area; and the identification generating unit is used for identifying the plurality of small images through a preset sign classification model and generating an identification result.

Optionally, in a first implementation manner of the second aspect of the embodiment of the present invention, the preprocessing unit includes: the processing module is used for processing the pixel size of the original spine image to obtain a processed first spine image; and the adjusting module is used for carrying out parameter adjustment on the first spine image and generating a target spine image.

Optionally, in a second implementation manner of the second aspect of the embodiment of the present invention, the processing module is specifically configured to: performing black edge cutting treatment on the original spine image; cutting the spine image after the black edge is cut off; and adjusting the size of the cut spine image to obtain a processed first spine image.

Optionally, in a third implementation manner of the second aspect of the embodiment of the present invention, the adjusting module is specifically configured to: determining the number of image channels of the first spine image; and adjusting the window width and the window level of the first spine image according to the image channel number to generate a target spine image.

Optionally, in a fourth implementation manner of the second aspect of the embodiment of the present invention, the dividing unit is specifically configured to: judging whether sacral vertebrae exist in the target spine image through a preset first segmentation model; if the sacral vertebrae exist in the target spinal column image, separating the sacral vertebrae to generate a sacral vertebrae mask, and marking the mask as a category I; judging whether a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image or not through a preset second segmentation model; if a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image, separating the fifth lumbar vertebra, generating a fifth lumbar vertebra mask, and marking the fifth lumbar vertebra mask as a category II; judging whether a thoracic vertebra, a first lumbar vertebra, a second lumbar vertebra, a third lumbar vertebra or a fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image through a preset third segmentation model; if the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image, separating the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which exist, generating a corresponding thoracic vertebra mask or a lumbar vertebra mask, and marking the thoracic vertebra mask or the lumbar vertebra mask as a category III; judging whether cervical vertebrae exist in the target spine image or not through a preset fourth segmentation model; if the cervical vertebrae exist in the target spine image, separating the cervical vertebrae to generate a cervical vertebrae mask, and marking the mask as a category IV.

Optionally, in a fifth implementation manner of the second aspect of the embodiment of the present invention, the identification detection unit is specifically configured to: identifying bone block contours of N vertebrae through a preset fuzzy energy algorithm; obtaining M candidate points of each vertebra through a preset Harris corner detection algorithm, wherein M is greater than or equal to 4; dividing the M candidate points into P point clusters by a density-based clustering algorithm DBSCAN algorithm with noise; respectively calculating the central points of the P point clusters, and determining the P central points as P vertebra corner points; and removing redundant vertebral corner points or filling up missing vertebral corner points through a minimum circumscribed rectangle algorithm to obtain N4 vertebral corner points.

Optionally, in a sixth implementation manner of the second aspect of the embodiment of the present invention, the identification generating unit is specifically configured to: calling a preset depth residual error network model to identify the plurality of small images; isolating a disease patch image comprising a disease sign; the location of each disease patch image on the composite vertebral image is determined and a recognition result is output, including the vertebral center point, corner offset, and disc thickness.

A third aspect of the embodiments of the present invention provides an artificial intelligence-based spine image processing apparatus, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the artificial intelligence-based spine image processing method according to any one of the foregoing embodiments when executing the computer program.

A fourth aspect of the embodiments of the present invention provides a computer-readable storage medium comprising instructions that, when executed on a computer, cause the computer to perform the steps of the artificial intelligence-based spine image processing method of any one of the embodiments described above.

In the technical scheme provided by the embodiment of the invention, an original spine image is obtained, wherein the original spine image is a sagittal radiological image of a spine; preprocessing an original spine image to generate a target spine image; dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra; performing vertebra contour recognition and corner detection on a plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1; synthesizing the N bone block outlines, N4 vertebra corner points and the target spine image to generate a synthesized vertebra image; extracting a plurality of small images from the composite vertebra image, each small image including information of a target area; and identifying the plurality of small images through a preset sign classification model to generate an identification result. According to the embodiment of the invention, the vertebrae segmentation and disease symptom classification are respectively realized through two different depth networks, the segmentation accuracy of vertebrae is improved, and the specificity and sensitivity of symptom recognition are improved.

Drawings

FIG. 1 is a schematic diagram of an embodiment of a spine image processing method based on artificial intelligence according to an embodiment of the present invention;

FIG. 2 is a schematic diagram of another embodiment of an artificial intelligence-based spine image processing method according to an embodiment of the present invention;

FIG. 3 is a schematic view of an embodiment of an artificial intelligence-based spine image processing apparatus according to an embodiment of the present invention;

FIG. 4 is a schematic view of another embodiment of an artificial intelligence based spine image processing device according to an embodiment of the present invention;

FIG. 5 is a schematic diagram of an embodiment of an artificial intelligence-based spine image processing apparatus according to the present invention.

Detailed Description

In order to enable those skilled in the art to better understand the present invention, embodiments of the present invention will be described below with reference to the accompanying drawings.

The terms "first," "second," "third," "fourth" and the like in the description and in the claims and in the above drawings, if any, are used for distinguishing between similar objects and not necessarily for describing a particular sequential or chronological order. It is to be understood that the data so used may be interchanged where appropriate such that the embodiments described herein may be implemented in other sequences than those illustrated or otherwise described herein. Furthermore, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of steps or elements is not necessarily limited to those steps or elements expressly listed or inherent to such process, method, article, or apparatus.

Referring to fig. 1, a flowchart of an artificial intelligence-based spine image processing method according to an embodiment of the present invention specifically includes:

101. an original spine image is obtained, wherein the original spine image is a sagittal radiological image of the spine.

The server acquires an original spine image, wherein the original spine image is a sagittal radiological image of the spine. The sagittal plane is the plane of the anatomy that divides the body into left and right planes, and is parallel to this plane of anatomy. In this position, the sagittal position (Median sagittal section). The sagittal plane is relative to the coronal plane and the horizontal plane, wherein the coronal plane refers to a longitudinal section which divides the human body into front and rear parts in the left-right direction, and the longitudinal section is mutually perpendicular to the sagittal plane and the horizontal plane; the horizontal plane (also called a transverse plane) is a plane parallel to the ground plane and dividing the human body into an upper part and a lower part, and the plane is perpendicular to the coronal plane and the sagittal plane.

It will be appreciated that the implementation subject of the present invention may be an artificial intelligence-based spine image processing device, and may also be a terminal or a server, which is not limited herein. The embodiment of the invention is described by taking a server as an execution main body as an example.

102. Preprocessing the original spine image to generate a target spine image.

The server preprocesses the original spine image to generate a target spine image. Specifically, the server processes the pixel size of the original spine image to obtain a processed first spine image; and the server carries out parameter adjustment on the first spine image to generate a target spine image.

The server processes the pixel size of the original spine image, and the obtained first spine image after processing specifically comprises: cutting off black edges of the original spine image; cutting the spine image after the black edge is cut off; and adjusting the size of the cut spine image to obtain a processed first spine image. For example, assume that the input file size of the model is n×n pixels.

The server performs parameter adjustment on the first spine image, and generating the target spine image specifically includes: determining the number of image channels of the first spine image; and adjusting the window width and the window level of the first spine image according to the number of the image channels to generate a target spine image.

103. And dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra.

The server segments each vertebra in the target spine image through a preset segmentation model to generate a plurality of vertebra masks, and each vertebra mask corresponds to a different vertebra. Each vertebra mask corresponds to a mask type label, different colors or numbers or letters can be used for distinguishing the type labels, and the segmentation model can be a mask nn model.

Wherein, for vertebrae at different positions, such as sacral vertebrae, lumbar vertebrae, thoracic vertebrae, etc., different model training strategies are adopted in advance for training, so that the separation model can accurately separate vertebrae at different positions. The segmentation model is a neural network model, and the training process of the model is the prior art, and is not repeated here.

It should be noted that the model training strategy includes: 1. training a segmentation model by taking sacral vertebrae and other vertebrae as different categories; 2. since the fifth lumbar vertebra is adjacent to the sacral vertebrae and the remaining vertebrae are adjacent to the vertebrae, the fifth lumbar vertebra is also considered as a separate category; 3. taking 4-1 lumbar vertebrae and thoracic vertebrae as one category; 4. cervical vertebrae are taken as a category.

For example, each vertebra appearing on the target spine image is separately segmented, assuming that the sacrum 1, lumbar vertebrae 5-1, and thoracic vertebrae 12-11 appear on one image. The method comprises the following steps of: the mask labels of the sacrum 1 are of a first type (red), the mask labels of the lumbar vertebrae 5 are of a second type (green), the mask labels of the lumbar vertebrae 4-1 and the thoracic vertebrae 12-11 are of a third type (blue), the vertebra masks of different labels are combined into one output, and different vertebrae are distinguished through different labels.

104. And performing vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1.

And the server performs vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1.

For example, a preset fuzzy energy algorithm is first used to identify the bone block contours of multiple vertebrae, and an approximately rectangular vertebra generates four clusters of points, but in practice, the number of clusters of points may be greater than or equal to four due to morphological changes of the vertebrae or uneven edges of the segmented mask. The harris corner detection (Harris Coner Detection) algorithm can only provide the position information of the points, but cannot provide the information of the point clusters, i.e. cannot output which points belong to the same point cluster. Therefore, in the Harris corner detection algorithm, a wider threshold is given, so that N point clusters (N is not necessarily equal to 4) are generated for each vertebra, each cluster comprises 20-40 points, and 100-200 points are obtained in total. And dividing 100-200 into N point clusters by using a density-based clustering algorithm (density-based spatial clustering of applications with noise, DBSCAN) algorithm with noise. Taking the central points of N point clusters as corner points.

It will be appreciated that for most vertebrae (non-sacral vertebrae) n=4, in a few cases, when N is not equal to 4, a minimum circumscribed rectangular algorithm is used to reject the redundant vertebral corner or fill the missing vertebral corner, and four vertebral corner points are output. To improve the efficiency of the algorithm, the four outputted vertebral corner points are ordered in a counterclockwise direction.

It should be noted that, the server detects the corner point on the mask, and the key point on the mask is not interfered by the image ghost and the low pixel, so that the detected point is mapped from the mask to the corresponding position of the target spine image more accurately.

105. And synthesizing the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image.

And the server synthesizes the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image. Wherein each vertebral mask is superimposed over a corresponding vertebra in the image of the target spine.

The color areas in the generated synthetic vertebra images are masks output by the segmentation model, different colors represent different types of labels, the gray areas are target spine images, and the gray areas are background images. The model outputs only the mask and the class corresponding to the mask, and the generated mask is drawn on the target spine image for convenience of display.

106. A plurality of tile images are extracted from the composite vertebral image, each tile image including information of a target region.

The server extracts a plurality of small images from the composite vertebral image, each small image including information of the target region. The target area may include an intervertebral disc, vertebrae, sacral vertebrae, vertebral corner points, biforaries, and the like. Depending on the detection target, this patch image may comprise 1 to 2 vertebrae, intervertebral discs, or sacral vertebrae (different patch images are extracted depending on the different detection targets). For example, on a composite vertebra image, assuming that N bone block contours and n×4 corner points are obtained by segmentation, N image patches including 1 vertebra can be extracted, (N-1) patch images including two adjacent vertebrae, n×4 patch images including vertebra corner points, (N-1) patch images including intervertebral disc, 1 sacral patch image (only in the presence of sacral vertebrae).

107. And identifying the plurality of small images through a preset sign classification model to generate an identification result.

The server identifies a plurality of small images through a preset sign classification model, and generates an identification result. The symptom classification model may be a deep residual network (deep residual network, resNet) model, among others. The server separates out the small image including the disease symptom, outputs the position of the disease small image in the synthesized vertebra image, and outputs the recognition result. The server calculates parameters such as the center point, the angular point offset, the disc thickness and the like of the bone block according to the position information of the edge and the angular point of the bone block, and can further calculate parameters such as the vertebra thickness, the thoracic vertebra kyphosis curvature, the lumbar lordosis curvature, the sacrum inclination angle, the sagittal plane axial distance and the like so as to assist doctors in quantitative analysis.

According to the embodiment of the invention, the vertebrae segmentation and disease symptom classification are respectively realized through two different depth networks, the segmentation accuracy of vertebrae is improved, and the specificity and sensitivity of symptom recognition are improved.

Referring to fig. 2, another flowchart of an artificial intelligence-based spine image processing method according to an embodiment of the present invention specifically includes:

201. an original spine image is obtained, wherein the original spine image is a sagittal radiological image of the spine.

202. And processing the pixel size of the original spine image to obtain a processed first spine image.

And the server processes the pixel size of the original spine image to obtain a processed first spine image. Specifically, the server performs black edge cutting treatment on the original spine image; the server cuts the spine image after the black edge cutting treatment; and the server adjusts the size of the cut spine image to obtain a processed first spine image. For example, assume that the input file size of the model is n×n pixels.

For example, when the size of the original spine image (i.e., DICOM file) is 512×512 pixels, the server first checks whether the image has a frame, if the frame is found to have a thickness of 77 pixels around the image, the server first cuts out the frame, the image size is changed to 435×435 pixels, and then the server scales the image after cutting out the frame to n×n pixels. For another example, when the size of the input original spine image is 888×678 pixels, the server first checks whether the image has a frame, if no frame is found around the image, the server cuts the image to a size of 678×678 pixels, and then the server scales the cut image to a size of n×n pixels.

The original spine image is subjected to consistency adjustment, the obtained adjusted image spine occupies the main body of the target spine image, the image is stretched into uniform N by N size in equal proportion, and no frame exists.

203. And carrying out parameter adjustment on the first spine image to generate a target spine image.

And the server carries out parameter adjustment on the first spine image to generate a target spine image. Specifically, the server determines the number of image channels of the first spine image; and the server adjusts the window width and the window level of the first spine image according to the number of the image channels to generate a target spine image.

For example, the server first determines the number of image channels of the first spine image; if the number of the image channels of the first spine image is 1, the first spine image is represented as a gray scale image; if the number of image channels of the first spine image is 3, the first spine image is represented as an RGB image. It is understood that the medical image is generally a single-channel image, i.e. the number of image channels is 1, and will not be described herein. The following description will take the number of channels as 1 as an example.

For example, assume that the gray value range of the first spine image is: -638-904. The server firstly calculates a gray level histogram of the first spine image, then calculates the area from the histogram to the coordinate axis, takes the proportion of the envelope area to the total area as a threshold (comprising an area upper limit threshold and an area lower limit threshold), and intercepts the image in the target gray level interval. Assuming that the area upper threshold is 0.01, the area lower threshold is 0.6, the area of the histogram after the upper threshold and the lower threshold are removed is cut, for example, the gray level of the interval of 209-685 gray level histogram gray level value is cut. The specific calculation formula is as follows: window width of CT image= (685-209)/2; window level of CT image = 209+ 685)/2; and uniformly stretching the intercepted gray value to a 0-255 interval to realize window width and window level adjustment.

204. And dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra.

The server segments each vertebra in the target spine image through a preset segmentation model to generate a plurality of vertebra masks, and each vertebra mask corresponds to a different vertebra. Each vertebra mask corresponds to a mask type label, different colors or numbers or letters can be used for distinguishing the type labels, and the segmentation model can be a mask nn model. Wherein, for vertebrae at different positions, such as sacral vertebrae, lumbar vertebrae, thoracic vertebrae, etc., different model training strategies are adopted in advance for training, so that the separation model can accurately separate vertebrae at different positions. The segmentation model is a neural network model, and the training process of the model is the prior art, and is not repeated here.

It should be noted that the model training strategy includes: 1. training a segmentation model by taking sacral vertebrae and other vertebrae as different categories; 2. since the fifth lumbar vertebra is adjacent to the sacral vertebrae and the remaining vertebrae are adjacent to the vertebrae, the fifth lumbar vertebra is also considered as a separate category; 3. taking 4-1 lumbar vertebrae and thoracic vertebrae as one category; 4. cervical vertebrae are taken as a category. For example, each vertebra appearing on the target spine image is separately segmented, assuming that the sacrum 1, lumbar vertebrae 5-1, and thoracic vertebrae 12-11 appear on one image. The method comprises the following steps of: the mask labels of the sacrum 1 are of a first type (red), the mask labels of the lumbar vertebrae 5 are of a second type (green), the mask labels of the lumbar vertebrae 4-1 and the thoracic vertebrae 12-11 are of a third type (blue), the vertebra masks of different labels are combined into one output, and different vertebrae are distinguished through different labels.

Specifically, the server judges whether sacral vertebrae exist in the target spine image through a preset first segmentation model; if the sacral vertebrae exist in the target spinal column image, the server separates the sacral vertebrae, generates a sacral vertebrae mask, and marks the mask as a category I; the server judges whether a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image or not through a preset second segmentation model; if a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image, the server separates the fifth lumbar vertebra, generates a fifth lumbar vertebra mask and marks the fifth lumbar vertebra mask as a category II; the server judges whether thoracic vertebrae, a first lumbar vertebra, a second lumbar vertebra, a third lumbar vertebra or a fourth lumbar vertebra which are sequentially connected with the thoracic vertebrae exist in the target spine image through a preset third segmentation model; if the thoracic vertebrae, the first lumbar vertebrae, the second lumbar vertebrae, the third lumbar vertebrae or the fourth lumbar vertebrae which are sequentially connected with the thoracic vertebrae exist in the target spine image, the server separates the existing thoracic vertebrae, the first lumbar vertebrae, the second lumbar vertebrae, the third lumbar vertebrae or the fourth lumbar vertebrae, generates corresponding thoracic vertebrae masks or lumbar vertebrae masks, and marks the corresponding thoracic vertebrae masks or lumbar vertebrae masks as a category III; the server judges whether cervical vertebrae exist in the target spine image or not through a preset fourth segmentation model; if the cervical vertebrae exist in the target spine image, the server separates the cervical vertebrae to generate a cervical vertebrae mask, and the mask is marked as category four.

205. And performing vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1.

And the server performs vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1. Specifically, the server identifies bone block contours of N vertebrae through a preset fuzzy energy algorithm; the server acquires M candidate points of each vertebra through a preset Harris corner detection algorithm, wherein M is greater than or equal to 4; the server divides M candidate points into P point clusters through a density-based clustering algorithm DBSCAN algorithm with noise; the server calculates the central points of the P point clusters respectively, and the P central points are determined to be P vertebra corner points; the server eliminates redundant vertebra corner points or fills up missing vertebra corner points through a minimum circumscribed rectangle algorithm to obtain N4 vertebra corner points.

206. And synthesizing the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image.

207. A plurality of tile images are extracted from the composite vertebral image, each tile image including information of a target region.

208. And identifying the plurality of small images through a preset sign classification model to generate an identification result.

Before the preset symptom classification model is called, the server can use the extracted corresponding small image to train to obtain the symptom classification model in combination with the gold standard of the clinical spinal disease symptom database marked by doctors.

According to the embodiment of the invention, based on the deep neural network, the multi-mode multi-size spine medical image is input, the consistency of different types of images is enhanced through preprocessing, two different deep networks are used for respectively realizing vertebra segmentation and disease symptom classification, multiple symptom identification of the vertebra is further performed on the basis of vertebra segmentation, the segmentation accuracy of the vertebra is improved, and the specificity and sensitivity of the symptom identification are improved. The spine image processing method based on artificial intelligence in the embodiment of the present invention is described above, and the spine image processing device based on artificial intelligence in the embodiment of the present invention is described below, referring to fig. 3, one embodiment of the spine image processing device based on artificial intelligence in the embodiment of the present invention includes:

a first acquiring unit 301, configured to acquire an original spine image, where the original spine image is a sagittal radiological image of a spine;

a preprocessing unit 302, configured to perform preprocessing on the original spine image to generate a target spine image;

The segmentation unit 303 is configured to segment each vertebra in the target spine image through a preset segmentation model, and generate a plurality of vertebra masks, where each vertebra mask corresponds to a different vertebra;

the recognition detection unit 304 is configured to perform vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and n×4 vertebra corners, where N is greater than or equal to 1;

a synthesizing unit 305, configured to synthesize the N bone block contours, the N x 4 vertebra corner points, and the target spine image, and generate a synthesized vertebra image;

an extracting unit 306 for extracting a plurality of small images from the synthesized vertebra image, each small image including information of a target area;

the recognition generating unit 307 is configured to identify the plurality of small images through a preset sign classification model, and generate a recognition result.

Referring to fig. 4, another embodiment of the spinal image processing apparatus based on artificial intelligence according to the present invention includes:

Optionally, the preprocessing unit 302 includes:

The processing module 3021 is configured to process the pixel size of the original spine image to obtain a processed first spine image;

the adjustment module 3022 is configured to perform parameter adjustment on the first spine image to generate a target spine image.

Optionally, the processing module 3021 is specifically configured to:

performing black edge cutting treatment on the original spine image; cutting the spine image after the black edge is cut off; and adjusting the size of the cut spine image to obtain a processed first spine image. Optionally, the adjusting module 3022 is specifically configured to:

determining the number of image channels of the first spine image; and adjusting the window width and the window level of the first spine image according to the image channel number to generate a target spine image.

Optionally, the dividing unit 303 is specifically configured to:

judging whether sacral vertebrae exist in the target spine image through a preset first segmentation model; if the sacral vertebrae exist in the target spinal column image, separating the sacral vertebrae to generate a sacral vertebrae mask, and marking the mask as a category I; judging whether a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image or not through a preset second segmentation model; if a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image, separating the fifth lumbar vertebra, generating a fifth lumbar vertebra mask, and marking the fifth lumbar vertebra mask as a category II; judging whether a thoracic vertebra, a first lumbar vertebra, a second lumbar vertebra, a third lumbar vertebra or a fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image through a preset third segmentation model; if the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image, separating the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which exist, generating a corresponding thoracic vertebra mask or a lumbar vertebra mask, and marking the thoracic vertebra mask or the lumbar vertebra mask as a category III; judging whether cervical vertebrae exist in the target spine image or not through a preset fourth segmentation model; if the cervical vertebrae exist in the target spine image, separating the cervical vertebrae to generate a cervical vertebrae mask, and marking the mask as a category IV.

Optionally, the identification detection unit 304 is specifically configured to:

identifying bone block contours of N vertebrae through a preset fuzzy energy algorithm; obtaining M candidate points of each vertebra through a preset Harris corner detection algorithm, wherein M is greater than or equal to 4; dividing the M candidate points into P point clusters by a density-based clustering algorithm DBSCAN algorithm with noise; respectively calculating the central points of the P point clusters, and determining the P central points as P vertebra corner points; and removing redundant vertebral corner points or filling up missing vertebral corner points through a minimum circumscribed rectangle algorithm to obtain N4 vertebral corner points.

Optionally, the identification generating unit 307 is specifically configured to:

calling a preset depth residual error network model to identify the plurality of small images; isolating a disease patch image comprising a disease sign; the location of each disease patch image on the composite vertebral image is determined and a recognition result is output, including the vertebral center point, corner offset, and disc thickness.

According to the embodiment of the invention, an original spine image is obtained, wherein the original spine image is a sagittal radial image of a spine; processing the pixel size of the original spine image to obtain a processed first spine image; parameter adjustment is carried out on the first spine image, and a target spine image is generated; dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra; performing vertebra contour recognition and corner detection on a plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1; synthesizing the N bone block outlines, N4 vertebra corner points and the target spine image to generate a synthesized vertebra image; extracting a plurality of small images from the composite vertebra image, each small image including information of a target area; and identifying the plurality of small images through a preset sign classification model to generate an identification result. According to the embodiment of the invention, based on the deep neural network, the multi-mode multi-size spine medical image is input, the consistency of different types of images is enhanced through preprocessing, two different deep networks are used for respectively realizing vertebra segmentation and disease symptom classification, multiple symptom identification of the vertebra is further performed on the basis of vertebra segmentation, the segmentation accuracy of the vertebra is improved, and the specificity and sensitivity of the symptom identification are improved.

Fig. 3 to 4 above describe the artificial intelligence-based spine image processing apparatus in the embodiment of the present invention in detail from the point of view of modularized functional entities, and the artificial intelligence-based spine image processing device in the embodiment of the present invention is described in detail from the point of view of hardware processing.

Fig. 5 is a schematic diagram of an artificial intelligence-based spine image processing apparatus 500 according to an embodiment of the present invention, where the artificial intelligence-based spine image processing apparatus 500 may have a relatively large difference due to different configurations or performances, and may include one or more processors (central processing units, CPU) 501 (e.g., one or more processors) and a memory 509, and one or more storage media 508 (e.g., one or more mass storage devices) storing application programs 507 or data 506. Wherein the memory 509 and storage medium 508 may be transitory or persistent storage. The program stored on the storage medium 508 may include one or more modules (not shown), each of which may include a series of instruction operations for an artificial intelligence-based spinal image processing device. Still further, the processor 501 may be configured to communicate with the storage medium 508 and execute a series of instruction operations in the storage medium 508 on the artificial intelligence based spine image processing device 500.

The artificial intelligence based spine image processing device 500 may also include one or more power supplies 502, one or more wired or wireless network interfaces 503, one or more input/output interfaces 504, and/or one or more operating systems 505, such as Windows Serve, mac OS X, unix, linux, freeBSD, etc. It will be appreciated by those skilled in the art that the artificial intelligence based spinal imaging processing device configuration illustrated in FIG. 5 is not limiting of the artificial intelligence based spinal imaging processing device and may include more or fewer components than illustrated, or may combine certain components, or a different arrangement of components. The processor 501 may perform the functions of the first acquisition unit 301, the preprocessing unit 302, the segmentation unit 303, the identification detection unit 304, the synthesis unit 305, the extraction unit 306, and the identification generation unit 307 in the above-described embodiments.

The following describes the components of the artificial intelligence-based spine image processing apparatus in detail with reference to fig. 5:

the processor 501 is a control center of the artificial intelligence-based spine image processing apparatus, and may perform processing according to a set artificial intelligence-based spine image processing method. The processor 501 utilizes various interfaces and lines to connect the various parts of the overall artificial intelligence based spinal imaging processing device to perform various functions and processing data of the artificial intelligence based spinal imaging processing device by running or executing software programs and/or modules stored in the memory 509 and invoking data stored in the memory 509 to thereby effect vertebrae segmentation and disease symptom classification. The storage medium 508 and the memory 509 are both carriers for storing data, and in the embodiment of the present invention, the storage medium 508 may refer to an internal memory with a small storage capacity but a fast speed, and the memory 509 may be an external memory with a large storage capacity but a slow storage speed.

The memory 509 may be used to store software programs and modules that the processor 501 performs various functional applications and data processing of the artificial intelligence-based spine image processing device 500 by running the software programs and modules stored in the memory 509. The memory 509 may mainly include a memory program area and a memory data area, where the memory program area may store an operating system, an application program required for at least one function (such as dividing each vertebra in the target spine image by a preset division model to generate a plurality of vertebra masks, each vertebra mask corresponding to a different vertebra), and so on; the stored data area may store data created from use of an artificial intelligence based spine image processing device (e.g., synthetic vertebra images, etc.), and so forth. In addition, the memory 509 may include high-speed random access memory, and may also include non-volatile memory, such as at least one magnetic disk storage device, flash memory device, or other volatile solid-state storage device. The artificial intelligence based spine image processing method program and received data streams provided in embodiments of the present invention are stored in memory and when needed, processor 501 recalls from memory 509.

When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present invention, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a computer network, or other programmable apparatus. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, twisted pair), or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer readable storage medium may be any available medium that can be stored by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a magnetic tape), an optical medium (e.g., an optical disk), or a semiconductor medium (e.g., a Solid State Disk (SSD)), or the like.

It will be clear to those skilled in the art that, for convenience and brevity of description, specific working procedures of the above-described systems, apparatuses and units may refer to corresponding procedures in the foregoing method embodiments, which are not repeated herein.

In the several embodiments provided in the present invention, it should be understood that the disclosed systems, devices, and methods may be implemented in other manners. For example, the apparatus embodiments described above are merely illustrative, e.g., the division of the units is merely a logical function division, and there may be additional divisions when actually implemented, e.g., multiple units or components may be combined or integrated into another system, or some features may be omitted or not performed. Alternatively, the coupling or direct coupling or communication connection shown or discussed with each other may be an indirect coupling or communication connection via some interfaces, devices or units, which may be in electrical, mechanical or other form.

The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in the embodiment of the present invention may be integrated in one processing unit, or each unit may exist alone physically, or two or more units may be integrated in one unit. The integrated units may be implemented in hardware or in software functional units.

The integrated units, if implemented in the form of software functional units and sold or used as stand-alone products, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied essentially or in part or all of the technical solution or in part in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a read-only memory (ROM), a random access memory (random access memory, RAM), a magnetic disk, or an optical disk, or other various media capable of storing program codes.

The above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An artificial intelligence-based spine image processing method is characterized by comprising the following steps:

acquiring an original spine image, wherein the original spine image is a sagittal radiological image of a spine;

preprocessing the original spine image to generate a target spine image;

dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra;

performing vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N4 vertebra corners, wherein N is greater than or equal to 1;

Synthesizing the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image;

extracting a plurality of small images from the synthesized vertebra image, each small image comprising information of a target area;

identifying the small images through a preset sign classification model to generate an identification result;

dividing each vertebra in the target spine image through a preset division model to generate a plurality of vertebra masks, wherein each vertebra mask corresponds to a different vertebra, and the method comprises the following steps:

judging whether sacral vertebrae exist in the target spine image through a preset first segmentation model;

if the sacral vertebrae exist in the target spinal column image, separating the sacral vertebrae to generate a sacral vertebrae mask, and marking the mask as a category I;

judging whether a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image or not through a preset second segmentation model;

if a fifth lumbar vertebra adjacent to the sacral vertebra exists in the target spine image, separating the fifth lumbar vertebra, generating a fifth lumbar vertebra mask, and marking the fifth lumbar vertebra mask as a category II;

judging whether a thoracic vertebra, a first lumbar vertebra, a second lumbar vertebra, a third lumbar vertebra or a fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image through a preset third segmentation model;

If the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which are sequentially connected with the thoracic vertebra exist in the target spine image, separating the thoracic vertebra, the first lumbar vertebra, the second lumbar vertebra, the third lumbar vertebra or the fourth lumbar vertebra which exist, generating a corresponding thoracic vertebra mask or a lumbar vertebra mask, and marking the thoracic vertebra mask or the lumbar vertebra mask as a category III;

judging whether cervical vertebrae exist in the target spine image or not through a preset fourth segmentation model;

if the cervical vertebrae exist in the target spine image, separating the cervical vertebrae to generate a cervical vertebrae mask, and marking the mask as a category IV;

performing vertebra contour recognition and corner detection on the plurality of vertebra masks through a preset clustering algorithm to obtain N bone block contours and N x 4 vertebra corners, wherein N is greater than or equal to 1, and the method comprises the following steps:

identifying bone block contours of N vertebrae through a preset fuzzy energy algorithm;

obtaining M candidate points of each vertebra through a preset Harris corner detection algorithm, wherein M is greater than or equal to 4;

dividing the M candidate points into P point clusters by a density-based clustering algorithm DBSCAN algorithm with noise;

respectively calculating the central points of the P point clusters, and determining the P central points as P vertebra corner points;

And removing redundant vertebral corner points or filling up missing vertebral corner points through a minimum circumscribed rectangle algorithm to obtain N4 vertebral corner points.

2. The artificial intelligence based spine image processing method of claim 1 wherein preprocessing the original spine image to generate a target spine image comprises:

processing the pixel size of the original spine image to obtain a processed first spine image;

and carrying out parameter adjustment on the first spine image to generate a target spine image.

3. The artificial intelligence based spine image processing method of claim 2 wherein the processing the pixel size of the original spine image to obtain a processed first spine image comprises:

performing black edge cutting treatment on the original spine image;

cutting the spine image after the black edge is cut off;

and adjusting the size of the cut spine image to obtain a processed first spine image.

4. The artificial intelligence based spine image processing method of claim 2 wherein the performing parameter adjustment on the first spine image to generate a target spine image comprises:

Determining the number of image channels of the first spine image;

and adjusting the window width and the window level of the first spine image according to the image channel number to generate a target spine image.

5. The artificial intelligence based spine image processing method of any one of claims 1-4 wherein the identifying the plurality of small images by a preset symptom classification model, generating an identification result comprises:

calling a preset depth residual error network model to identify the plurality of small images;

isolating a disease patch image comprising a disease sign;

the location of each disease patch image on the composite vertebral image is determined and a recognition result is output, including the vertebral center point, corner offset, and disc thickness.

6. An artificial intelligence-based spine image processing device, comprising:

the first acquisition unit is used for acquiring an original spine image, wherein the original spine image is a sagittal radial image of a spine;

the preprocessing unit is used for preprocessing the original spine image to generate a target spine image;

the segmentation unit is used for segmenting each vertebra in the target spine image through a preset segmentation model to generate a plurality of vertebra masks, and each vertebra mask corresponds to a different vertebra;

The identification detection unit is used for carrying out vertebra outline identification and corner detection on the vertebra masks through a preset clustering algorithm to obtain N bone block outlines and N4 vertebra corner points, wherein N is greater than or equal to 1;

the synthesis unit is used for synthesizing the N bone block outlines, the N4 vertebra corner points and the target spine image to generate a synthesized vertebra image;

an extraction unit for extracting a plurality of small images from the synthesized vertebra image, each small image including information of a target area;

the identification generation unit is used for identifying the small images through a preset sign classification model to generate an identification result;

the dividing unit is specifically configured to:

the identification detection unit is specifically configured to:

7. The artificial intelligence based spine image processing device of claim 6 wherein the preprocessing unit comprises:

the processing module is used for processing the pixel size of the original spine image to obtain a processed first spine image;

and the adjusting module is used for carrying out parameter adjustment on the first spine image and generating a target spine image.

8. The artificial intelligence based spine image processing apparatus of claim 6 or claim 7 wherein the identification generation unit is specifically configured to:

isolating a disease patch image comprising a disease sign;

9. An artificial intelligence based spine image processing apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor implementing the artificial intelligence based spine image processing method of any one of claims 1-5 when executing the computer program.

10. A computer readable storage medium comprising instructions which, when executed on a computer, cause the computer to perform the artificial intelligence based spine image processing method of any one of claims 1-5.