WO2021121159A1 - 基于神经网络的腰椎病变诊断结果输出系统以及方法 - Google Patents

基于神经网络的腰椎病变诊断结果输出系统以及方法 Download PDF

Info

Publication number
WO2021121159A1
WO2021121159A1 PCT/CN2020/135720 CN2020135720W WO2021121159A1 WO 2021121159 A1 WO2021121159 A1 WO 2021121159A1 CN 2020135720 W CN2020135720 W CN 2020135720W WO 2021121159 A1 WO2021121159 A1 WO 2021121159A1
Authority
WO
WIPO (PCT)
Prior art keywords
lesion
lumbar
prolapsed
intervertebral disc
marking line
Prior art date
Application number
PCT/CN2020/135720
Other languages
English (en)
French (fr)
Inventor
张芮溟
吴海萍
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2021121159A1 publication Critical patent/WO2021121159A1/zh

Links

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
    • G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
    • G16H50/20—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for computer-aided diagnosis, e.g. based on medical expert systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/05—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves 
    • A61B5/055—Detecting, measuring or recording for diagnosis by means of electric currents or magnetic fields; Measuring using microwaves or radio waves  involving electronic [EMR] or nuclear [NMR] magnetic resonance, e.g. magnetic resonance imaging
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/45—For evaluating or diagnosing the musculoskeletal system or teeth
    • A61B5/4538—Evaluating a particular part of the muscoloskeletal system or a particular medical condition
    • A61B5/4566—Evaluating the spine
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B5/00—Measuring for diagnostic purposes; Identification of persons
    • A61B5/72—Signal processing specially adapted for physiological signals or for diagnostic purposes
    • A61B5/7235—Details of waveform analysis
    • A61B5/7264—Classification of physiological signals or data, e.g. using neural networks, statistical classifiers, expert systems or fuzzy systems
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/02—Arrangements for diagnosis sequentially in different planes; Stereoscopic radiation diagnosis
    • A61B6/03—Computed tomography [CT]
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/50—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications
    • A61B6/505—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment specially adapted for specific body parts; specially adapted for specific clinical applications for diagnosis of bone
    • A—HUMAN NECESSITIES
    • A61—MEDICAL OR VETERINARY SCIENCE; HYGIENE
    • A61B—DIAGNOSIS; SURGERY; IDENTIFICATION
    • A61B6/00—Apparatus or devices for radiation diagnosis; Apparatus or devices for radiation diagnosis combined with radiation therapy equipment
    • A61B6/52—Devices using data or image processing specially adapted for radiation diagnosis
    • A61B6/5211—Devices using data or image processing specially adapted for radiation diagnosis involving processing of medical diagnostic data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/11—Region-based segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10081—Computed x-ray tomography [CT]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10116—X-ray image
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20084—Artificial neural networks [ANN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30008—Bone
    • G06T2207/30012—Spine; Backbone
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion

Definitions

  • This application relates to application fields such as intelligent decision-making based on the field of artificial intelligence technology, and in particular to a system and method for outputting diagnosis results of lumbar lesions based on a neural network.
  • Lumbar disc herniation is one of the most common diseases that cause low back and leg pain. Lumbar disc herniation includes signs of prolapse, herniation, bulging, and free disc. Lumbar disc herniation is extremely repetitive in imaging diagnosis. At present, there has not yet been a mature neural network-based image-aided diagnosis system for various diseases of lumbar intervertebral discs on the market. Early imaging analysis methods for lumbar intervertebral discs mainly use traditional imaging methods, such as edge extraction.
  • This application provides a neural network-based lumbar spine disease diagnosis result output system and method to solve the artificial interpretation of the lumbar spine disease diagnosis result output system based on neural network.
  • the workload of imaging doctors is large, the work repetition is high, and different doctors will meet There are their own descriptive terms, which leads to the problem of large descriptive errors in the diagnosis results.
  • a neural network-based lumbar spine disease diagnosis result output system including: an image acquisition device for acquiring lumbar scan sequence images; an image processing device for preprocessing the lumbar spine scan sequence images to obtain a disc layer scan image
  • the image processing device is also used to input the scan image of the intervertebral disc layer into a segmentation model based on a neural network for segmentation processing to segment lumbar intervertebral discs, prolapsed lesions, and the degree of disease for lumbar disc herniation classification Marking line;
  • the image processing device is also used to obtain a diagnosis result of the lumbar spine lesion based on the segmented lumbar intervertebral disc, the prolapsed lesion and the marking line of the lesion degree; the output device is used to output the diagnosis result of the lesion.
  • a method for outputting diagnosis results of lumbar spine lesions based on a neural network includes: preprocessing the lumbar scan sequence images to obtain a scan image of the disc layer; inputting the scan image of the disc layer into a segmentation model based on the neural network for processing The segmentation process is to segment the lumbar intervertebral disc, the prolapsed lesion, and the lesion degree marking line for the classification of lumbar intervertebral disc herniation; based on the segmented lumbar intervertebral disc, the prolapsed lesion and the lesion degree marking line, the lumbar lesion diagnosis result is obtained.
  • One or more readable storage media storing computer readable instructions, when the computer readable instructions are executed by one or more processors, the one or more processors execute the following steps: scan the lumbar spine The sequence image is preprocessed to obtain the intervertebral disc layer scan image; the intervertebral disc layer scan image is input into the segmentation model based on neural network for segmentation processing to segment the lumbar intervertebral disc, the prolapsed lesion and the classification of lumbar intervertebral disc herniation Marking line for the degree of lesions; based on the segmented lumbar intervertebral discs, prolapsed lesions, and marking lines for the degree of lesions, the diagnosis results of the lumbar spine are obtained.
  • the neural network-based diagnosis result output system for lumbar spine lesions proposed in this application includes an image acquisition device for acquiring lumbar scan sequence images; an image processing device for preprocessing the lumbar scan sequence images to obtain a disc layer scan image
  • the image processing device is also used to input the scan image of the intervertebral disc layer into a segmentation model based on a neural network for segmentation processing to segment lumbar intervertebral discs, prolapsed lesions, and the degree of disease for lumbar disc herniation classification Marking line;
  • the image processing device is also used to obtain the diagnosis result of the lumbar spine lesion based on the segmented lumbar intervertebral disc, the prolapsed lesion and the marking line of the lesion degree; the output device is used to output the diagnosis result of the lesion.
  • the neural network-based diagnosis result output system for lumbar spine lesions proposed in this application can automatically interpret the diagnosis results of lumbar intervertebral disc lesions, and provide a valuable reference for the selection of disease treatment options; in addition, the neural network-based diagnosis of lumbar spine lesions proposed in this application
  • the result output system also has the advantages of fast reading speed and higher accuracy of interpretation.
  • the unified output of the disease diagnosis results through the system can bypass the differences in the level of the same medical staff or the level of diagnosis and treatment in different regions, resulting in a relatively different description of the diagnosis results. The problem of large errors.
  • Fig. 1 is a schematic block diagram of a lumbar spine lesion diagnosis result output system based on a neural network in an embodiment of the present application;
  • FIG. 2 is a schematic diagram of the segmentation network structure of the preset Vnet model adopted in the neural network-based diagnosis result output system of lumbar spine lesions in an embodiment of the present application;
  • FIG. 3 is a schematic diagram of the first coding unit of the segmentation network structure of the preset Vnet model adopted in the neural network-based diagnosis result output system of lumbar spine lesions in an embodiment of the present application;
  • FIG. 4 is a schematic diagram of the fourth decoding unit of the segmentation network structure of the preset Vnet model adopted in the neural network-based diagnosis result output system of lumbar spine lesions in an embodiment of the present application;
  • FIG. 5 is a schematic diagram of two lesion degree marking lines used in a neural network-based diagnosis result output system for lumbar spine lesions in an embodiment of the present application;
  • FIG. 6 is a schematic diagram of four lesion location marking lines used in the neural network-based diagnosis result output system of lumbar spine lesions in an embodiment of the present application;
  • FIG. 7 is a schematic diagram of a nine-square grid used in the output system for diagnosis results of lumbar lesions based on a neural network in an embodiment of the present application;
  • Fig. 8 is a schematic diagram of different protrusions of lumbar intervertebral discs in the lumbar spine disease diagnosis result output system based on neural network in an embodiment of the present application.
  • an image acquisition device 10 an image processing device 20, and an output device 3 of a lumbar disease diagnosis result output system based on a neural network are provided, wherein the functions or functions of each device are as follows :
  • the image acquisition device 10 is used to acquire scan sequence images of the lumbar spine.
  • this solution can use a CT scanning device to obtain a CT scan sequence image of the lumbar spine, or use an X-ray scanning device to obtain a lumbar spine X-ray image, so as to obtain a lumbar spine scan sequence image. It is even possible to obtain medical images such as MRI of the lumbar spine, which is not specifically limited in this application, that is, the lumbar spine scan sequence image may be a lumbar spine CT scan sequence image, a lumbar spine X-ray scan sequence image or a lumbar spine magnetic resonance image.
  • the image processing device 20 is used for preprocessing the lumbar scan sequence image to obtain the interdiscal layer scan image.
  • the lumbar scan sequence image includes a positioning image and an axial scan image
  • the image processing device preprocesses the lumbar scan sequence image to obtain a disc layer scan image, which specifically refers to:
  • the lumbar spine positioning image of the lumbar spine scan sequence image is eliminated to obtain the axial scan image of the lumbar spine scan sequence.
  • the spiral scan image of the lumbar spine scan sequence is selected from the axial scan image.
  • the spiral scan image is reconstructed to form the disc layer scan image.
  • the lumbar spine positioning image refers to the image that reflects whether the lumbar intervertebral space is narrowed and the physiological curvature;
  • the axial scan image refers to the image that reflects whether the lumbar intervertebral disc is herniated and the lumbar intervertebral disc is abnormal.
  • the lumbar spine scan sequence may include lumbar spine positioning images and axial scan images, and the axial scan images may be spiral scans or non-helical scans.
  • the prolapsed lesions are mainly observed through the lumbar intervertebral disc images of the axial scan. Therefore, the positioning images are required. Filtering is performed to obtain the axial scan image after the positioning image is removed; then the spiral scan image in the axial scan image is selected, and the selected spiral scan image is reconstructed to form a disc layer scan image.
  • the spiral scan image is used to reconstruct the intervertebral disc layer scan image as the input of the subsequent segmentation model, which can make the segmented lumbar intervertebral disc, prolapsed lesion and the two lesion degree marking lines used for the MSU classification of lumbar disc herniation more accurate.
  • the image processing device 20 is also used to input the intervertebral disc layer scan image into a neural network-based segmentation model for segmentation processing, so as to segment the lumbar intervertebral disc, the prolapsed focus, and the degree of disease for the classification of lumbar intervertebral disc herniation Mark the line.
  • the classification of lumbar disc herniation is referred to as MSU classification.
  • the MSU classification of lumbar disc herniation is a method of classifying the degree of lesions shown in the cross-sectional image of the lumbar spine.
  • the marked line of the degree of lesions is the MSU classification.
  • the image processing device 20 is further used for:
  • the scan image of the interdiscal layer is adjusted to a horizontal position, so that the obtained marking line of the degree of lesion is a horizontal marking line.
  • the marked lines of the segmented lesions are not necessarily horizontal. If the patient’s scan body is skewed, then from the disc layer scan image, the vertebral mass, interstitial Both the disc and the horizontal marking line will rotate at a certain angle; if the scan image of the disc layer is not in a horizontal position, adjust the scan image of the disc layer to a horizontal position, so as to correct the scanned image to make the result visualized They all appear to be horizontal, so that the resulting marked line of the degree of the lesion is a horizontal marked line.
  • the scan image of the disc layer is in a horizontal position relative to the position of the scan image of the normal disc layer. If it is not in the horizontal position, the scan image of the disc layer is adjusted to the horizontal position, so that The visualized results seem to be level, which improves the reading speed.
  • the image processing device 20 is specifically further used for:
  • the pre-trained preset Vnet model is used to segment the scan image of the intervertebral disc layer to obtain a segmented image, wherein the segmented image includes the segmented lumbar intervertebral disc, the prolapsed lesion, and the degree of lesion labeling line.
  • the separation of medical images usually uses the Vnet model, and the pre-trained Vnet (U-Net Convolutional Networks for BiomedicalImage Segmentation) model.
  • the segmentation content includes the lumbar intervertebral disc, the prolapsed lesion and the marking line of the lesion degree used for the MSU classification of lumbar intervertebral disc herniation.
  • the pre-trained preset Vnet model is used to segment the scan image of the disc layer to obtain a segmented disc layer scan image, wherein the segmented disc layer scan image has been preliminarily divided Lumbar intervertebral disc, prolapsed lesion, and two pathological marking lines for MSU classification of lumbar intervertebral disc herniation.
  • the traditional Vnet model is more accurate for the segmentation of the intervertebral disc and the positioning of the two lines, but the segmentation effect for the prolapsed lesion is not good.
  • Lumbar disc herniation is mainly manifested as a series of subtle changes in the lower edge of the intervertebral disc.
  • the prolapse is manifested as a clear protrusion on the lower edge of the intervertebral disc, and the bulging is manifested as the overall outward expansion of the lower edge of the intervertebral disc.
  • the distribution area of the signs is very small compared to the original image.
  • the deep neural network has significant advantages for the extraction of structured features in the large receptive field. Therefore, the preset Vnet model adopts the segmentation network structure training proposed by this scheme. Formed, the segmentation network structure integrates the features of the breakout to improve the segmentation effect of fine-grained features.
  • the preset Vnet model adopts the following segmented network structure, including an up-sampling network and a down-sampling network connected in sequence, and the up-sampling network includes a first coding unit and a second coding unit connected in sequence down-sampling Unit, the third coding unit, the fourth coding unit, and the fifth coding unit; the down-sampling network includes a fourth decoding unit, a third decoding unit, a second decoding unit, and a first decoding unit that are sequentially connected up-sampling.
  • the output layer of the fifth coding unit is connected to the fourth decoding unit:
  • each of the first coding unit, the second coding unit, the third coding unit, and the fourth coding unit is connected to a corresponding decoding unit in the down-sampling network, and the output end of the third decoding unit is also Connected to the input end of the first decoding unit, so as to use the feature map output by the third decoding unit as the input of the first decoding unit;
  • Each encoding unit in the upsampling network includes an input layer, a convolutional layer, a pooling layer, and an output layer that are sequentially connected; each decoding unit in the downsampling network includes an input layer, a deconvolution layer, and a convolutional layer that are sequentially connected. Layer and output layer.
  • U-Net U-Net Convolutional Networks for BiomedicalImage Segmentation
  • U-Net is suitable for medical image segmentation and natural image generation.
  • U-net segmentation network can perform end-to-end segmentation of images, that is, the input is an image, and the output is also an image.
  • the segmentation network structure in this solution is similar to the traditional U-Net structure, and it is also divided into a down-sampling stage and an up-sampling stage, but the down-sampling stage is improved. See the following description for details.
  • each level of the down-sampling process is called a coding unit, which includes The first coding unit, the second coding unit, the third coding unit, the fourth coding unit, and the fifth coding unit; each level of the up-sampling process is represented by a decoding unit, which in turn includes the fourth decoding unit and the third decoding Unit, second decoding unit, first decoding unit.
  • the first coding unit includes an input layer, a convolutional layer, a pooling layer, and an output layer that are sequentially connected.
  • the function of the first coding unit is as follows: input 572 ⁇ 572 ⁇ 1 through the input layer
  • the single-channel image is transformed into 570 ⁇ 570 ⁇ 64 after 3 ⁇ 3 convolutional layer convolution, and then after 3 ⁇ 3 convolution, the image size is reduced by 2.
  • after convolution it becomes 568 ⁇ 568 ⁇ 64
  • after the convolution is completed, it will pass through the 2 ⁇ 2 pooling layer, and then the image size will be reduced by a factor of 284 ⁇ 284 ⁇ 64, and finally output to the next level unit through the output layer , That is, the subsequent second coding unit, third coding unit, fourth coding unit, and fifth coding unit.
  • the processing process of the next level unit is similar to the operation of the aforementioned first coding unit.
  • the fourth decoding unit is specifically as follows: First, the input image undergoes 2 ⁇ 2 up-sampling convolution, also known as deconvolution or transposed convolution, which can double the size of the image. The number of feature maps needs to be doubled to get 56 ⁇ 56 ⁇ 512 data.
  • the input of the subsequent third decoding unit is the 1024-channel data formed by superimposing the output of the corresponding coding unit, and then the valid convolution is performed twice with the input of 56 ⁇ 56 ⁇ 1024, and then up-sampling, superimposing, and convolution.
  • the subsequent The third decoding unit, the second decoding unit, and the first decoding unit will repeat the operations of the fourth decoding unit.
  • the convolution and other parameters involved are different. The details will not be described in detail. Please refer to Figure 3, and finally to 388 ⁇ 388 ⁇ 64 output.
  • the third decoding unit can extract better and strong expressions of the features of the prolapsed lesions.
  • some features of the prolapsed lesions will be gradually discarded, resulting in the first decoding unit
  • the segmentation effect of the output prolapsed lesion is not good. Therefore, in this solution, the output terminal of the third decoding unit is connected with the input terminal of the first decoding unit to use the feature map output by the third encoding unit as the first Input to the decoding unit.
  • the output terminal of the third coding unit is connected to the input terminal of the first decoding unit, so that the feature map output by the third coding unit is used as the input of the first decoding unit, which can be reserved.
  • the image processing device 20 is also used to obtain a diagnosis result of the lumbar spine lesion based on the segmented lumbar intervertebral disc, the prolapsed lesion, and the marking line of the lesion degree.
  • the lesion degree marking line includes multiple, and the image processing device 20 is used for:
  • the diagnosis result of lumbar disc herniation is obtained by judging the location of the prolapsed lesions on the grid map.
  • multiple lesion location marking lines perpendicular to the lesion degree marking line are drawn with the center of the spinal canal as the center to obtain a grid map of lumbar intervertebral disc herniation.
  • the grid map can accurately locate the lesion Position and judge the degree of lesions to make the interpretation more accurate and improve the efficiency of diagnosis and treatment.
  • the lesion degree marking line includes two, and the image processing device 20 is used for:
  • the diagnosis result of lumbar disc herniation is obtained by judging that the prolapsed lesions are distributed in the position of the Jiugongge.
  • post-processing is performed based on the segmentation results of the U-Net segmentation network, and four lesion location labeling lines perpendicular to the two lesion degree labeling lines are drawn with the center of the spinal canal as the center, according to the location and degree of lumbar disc herniation , Divide different areas, so as to obtain the nine-square grid map of lumbar intervertebral disc herniation; determine the location of the prolapsed lesions in the nine-square grid map to obtain the diagnosis result of the lumbar intervertebral disc herniation.
  • the nine-square grid chart makes the interpretation more accurate and improves the efficiency of diagnosis and treatment.
  • the image processing device 20 is used to:
  • first and second pathological degree marking lines between the lumbar intervertebral disc and the small appendages, wherein the first and second pathological degree marking lines are arranged sequentially from top to bottom;
  • the first, second, third, and fourth lesion location labeling lines are lesion location labeling lines arranged in order from left to right, and the first lesion location labeling line is located on the edge of one of the facet joints, the The fourth lesion location marking line is located on the edge of the facet joint on the other side, the second lesion location marking line is located between the center of the spinal canal and the edge of the facet joint on the one side, and the third lesion The position marking line is located between the center of the spinal canal and the edge of the facet joint on the other side.
  • the nine-square grid chart is used to describe the degree of lesions, and the first and second marking lines for the degree of lesions are drawn between the lumbar intervertebral disc and the small appendages.
  • the extent of lesions is determined by the first and second lines.
  • the three graded lesion degrees obtained by dividing the two lesion degree marking lines.
  • the two dashed lines from top to bottom indicate the first lesion degree marking line and the second lesion degree marking line respectively.
  • the Jiugongge is used to describe the location of the lesion.
  • the dotted lines from the left to the right represent the first, second, third, and fourth lesion location marking lines, respectively.
  • the first, second, third, and fourth lesion location marking lines are drawn with the protruding part being bounded by the midline of the spinal canal and the edges of the articular joints on both sides.
  • the nine-square grid chart is drawn according to the first and second lesion degree marking lines and the first, second, third, and fourth lesion location marking lines, and at the same time, the lesion degree marking line and the lesion location marking line vertical.
  • the first and the second marking lines are drawn with the center of the spinal canal as the center and the edges of the facet joints on both sides as the boundary.
  • Two, three, and four marking lines for the location of the lesion to obtain the nine-square grid, which can quickly determine the diagnosis of lumbar intervertebral disc herniation through the nine-square grid, which has the advantages of fast reading speed and higher interpretation accuracy, which greatly relieves medical staff Burden, improve the efficiency of diagnosis and treatment.
  • the image processing device 20 is used to:
  • the corresponding surgical strategy is determined according to the degree of lesion of the lumbar intervertebral disc herniation and the location area of the prolapsed lesion.
  • the nine-square grid is used to describe the degree of lesions. See Figures 5 and 7.
  • the first and second lesion degree marking lines are drawn between the lumbar intervertebral disc and the vertebral appendages, and the first and second lesion degrees are marked according to the first and second lesions.
  • Level 1, 2, and 3 for example, level 1 is the level of the upper facet joint, level 2 is the facet joint space, and level 3 is the level of the lower facet joint.
  • the three grades of lesions are mild, moderate, and severe, and mild, moderate, and severe represent the mild, moderate, and severe lumbar spine lesions, respectively.
  • the lumbar intervertebral disc herniation is a mild lesion, that is, the lumbar intervertebral disc herniation is grade 1 "upper articular process level”, then it is judged that the Lumbar disc herniation is a mild disease. If it is determined that the prolapsed lesion protrudes from the first lesion degree marking line and does not protrude from the second lesion degree marking line, that is, the lumbar disc herniation is grade 2 "articular joint space”, then The lumbar disc herniation is judged to be a moderate disease.
  • the lumbar intervertebral disc herniation is grade 3 "inferior articular process level"
  • the lumbar intervertebral disc herniation is a severe disease.
  • the first, second, third, and fourth lesion location marking lines are drawn with the protruding part being bounded by the midline of the spinal canal and the edges of the facet joints on both sides, and the first, second, third, and fourth lesion locations are marked according to the first, second, third, and fourth lesion locations.
  • the line is divided into three areas, A, B, and C.
  • the three areas A, B, and C respectively represent the first location area, the second location area, and the third location area.
  • area A is the central area of the spinal canal and area B
  • the C area is the articular process, and the recess area is measured.
  • the areas divided by the four lesion location marking lines from left to right are the third location area, the second location area, the first location area, the second location area, and the third location area. If it is determined that the prolapsed lesion is between the second lesion location marking line and the third lesion location marking line, that is, the prolapsed lesion is located in the "central area of the spinal canal" in area A, then it is determined that the prolapsed lesion is located in the first area. Location area.
  • the prolapsed lesion is between the first lesion location marking line and the second lesion location marking line, or it is determined that the prolapsed lesion is between the third lesion location marking line and the fourth lesion location
  • the prolapsed lesion is located in the area B "areas beyond the center of the spinal canal and within the spinal canal"
  • the prolapsed lesion is located in the second location area. If it is determined that the prolapsed lesion protrudes beyond the first lesion location marking line, or protrudes beyond the fourth lesion location marking line, that is, the prolapsed lesion is located in the C area "articular process, measuring recess area", then it is judged The prolapsed lesion is located in the third location area.
  • the corresponding surgical strategy is determined according to the extent of the lumbar disc herniation and the location area of the prolapsed lesion to obtain the diagnosis result of the lumbar disc herniation.
  • the diagnosis result can be 1-A, 1 -C, 2-B, 3-B, etc., the specific application is not limited.
  • the extent of lumbar disc herniation is judged by the degree of protrusion of the prolapsed lesion
  • the location area of the prolapsed lesion is judged by the location of the prolapsed lesion
  • the extent of the lumbar intervertebral disc herniation and the extent of the prolapsed lesion The location area can quickly determine the diagnosis result of lumbar intervertebral disc herniation, which has the advantages of fast reading speed and higher accuracy of interpretation, which greatly relieves the burden of medical staff and improves the efficiency of diagnosis and treatment.
  • the output device 30 is used to output lesion diagnosis results.
  • the result of the lesion diagnosis can be output, thereby assisting the doctor in making a more accurate diagnosis.
  • This solution uses the display device to receive the lesion diagnosis result output by the output device, so as to realize the visualization of the lesion diagnosis result. It should be emphasized that, in order to further ensure the privacy and safety of the above-mentioned lesion diagnosis results, the above-mentioned lesion diagnosis results can also be stored in a node of a blockchain.
  • the neural network-based diagnosis result output system of lumbar spine disease proposed in this application includes an image acquisition device for acquiring lumbar spine scan sequence images; an image processing device for performing lumbar scan sequence images. Preprocessing to obtain a scan image of the intervertebral disc layer; the image processing device is also used for inputting the scan image of the intervertebral disc layer into a segmentation model based on a neural network for segmentation processing to segment the lumbar intervertebral disc, prolapsed lesions, and for Lumbar disc herniation classification line of the lesion degree; the image processing device is also used to obtain the diagnosis result of the lumbar spine lesion based on the segmented lumbar intervertebral disc, the prolapsed lesion and the lesion degree label line; the output device is used to output the lesion diagnosis result.
  • the neural network-based diagnosis result output system of lumbar spine disease proposed in this application can automatically determine whether the lumbar intervertebral disc is herniated and the degree of herniation, which provides a valuable reference for the selection of disease treatment options; in addition, the neural network-based lumbar spine disease proposed in this application
  • the diagnostic result output system also has the advantages of fast reading speed and higher accuracy of interpretation, which reduces the difference in the level of medical staff or the level of diagnosis and treatment in different regions, greatly reduces the burden on medical staff, and improves the efficiency of diagnosis and treatment.
  • a method for outputting diagnosis results of lumbar spine lesions is applied to a system for outputting diagnosis results of lumbar spine lesions, which will be described in detail below in conjunction with examples.
  • the method for outputting diagnosis results of lumbar spine lesions includes the following steps: preprocessing the lumbar scan sequence image to obtain a disc layer scan image; and inputting the disc layer scan image into a segmentation based on a neural network Segmentation is performed in the model to segment lumbar intervertebral discs, prolapsed lesions, and lesion degree marking lines for the classification of lumbar disc herniation; based on the segmented lumbar intervertebral disc, prolapsed lesions and lesion degree marking lines, the diagnosis results of lumbar spine lesions are obtained.
  • the preprocessing of the lumbar scan sequence image to obtain the interdiscal layer scan image includes the following steps: removing the lumbar spine positioning image of the lumbar scan sequence image to obtain the lumbar scan sequence An axial scan image; select a spiral scan image of the lumbar scan sequence from the axial scan image; reconstruct the spiral scan image to form the interdiscal layer scan image.
  • the said scanning image of the intervertebral disc layer is input into a segmentation model based on a neural network for segmentation processing, so as to segment the lumbar intervertebral disc, the prolapsed lesion, and the lesion degree marking line for the classification of lumbar intervertebral disc herniation , Including the following steps: segmenting the scan image of the intervertebral disc layer through a preset trained Vnet model to obtain segmented images, the segmented images including the segmented lumbar intervertebral disc, prolapsed lesions, and The extent of the lesion is marked with a line.
  • the preset Vnet model adopts the following segmented network structure: including an up-sampling network and a down-sampling network connected in sequence, and the up-sampling network includes a first coding unit and a second coding unit connected in sequence down-sampling. Unit, the third coding unit, the fourth coding unit, and the fifth coding unit; the down-sampling network includes a fourth decoding unit, a third decoding unit, a second decoding unit, and a first decoding unit that are sequentially connected up-sampling.
  • the output layer of the fifth coding unit is connected to the fourth decoding unit; wherein, each of the first coding unit, the second coding unit, the third coding unit, and the fourth coding unit is connected to the down-sampling network
  • the corresponding decoding unit is connected, and the output end of the third decoding unit is also connected to the input end of the first decoding unit, so as to use the feature map output by the third decoding unit as the input of the first decoding unit;
  • Each encoding unit in the up-sampling network includes an input layer, a convolutional layer, a pooling layer, and an output layer that are sequentially connected;
  • each decoding unit in the down-sampling network includes an input layer, a deconvolution layer, a convolution layer, and Output layer.
  • the method further includes the following step: judging that the scan image of the intervertebral disc is relative to the normal disc Whether the position of the layer scan image is in a horizontal position; if it is not in a horizontal position, adjust the layer scan image of the disc to a horizontal position, so that the obtained marking line of the degree of lesion is a horizontal marking line.
  • the lesion degree marking line includes multiple, after the segmentation of the lumbar intervertebral disc, the prolapsed lesion, and the lesion degree marking line for the classification of lumbar intervertebral disc herniation, the following step is further included: In the center, draw multiple marking lines of the lesion position perpendicular to the marking line of the extent of the lesion to obtain a grid map of lumbar intervertebral disc herniation; determine the location of the prolapsed lesions on the grid map to obtain lumbar disc herniation The result of the diagnosis.
  • the lesion degree marking line includes two, after the segmentation of the lumbar intervertebral disc, the prolapsed lesion, and the lesion degree marking line for the classification of lumbar disc herniation, the method further includes the following steps: With the center of the spinal canal as the center, draw four marking lines of the lesion position perpendicular to the marking line of the extent of the lesion to obtain the nine-square grid map of lumbar disc herniation; determine the location of the prolapsed lesions on the nine-square grid map to obtain the lumbar intervertebral disc Outstanding diagnosis results.
  • the drawing four lesion location marking lines perpendicular to the lesion degree marking line with the center of the spinal canal as the center to obtain a nine-square grid map of lumbar intervertebral disc herniation includes the following steps: The first and second pathological degree marking lines are drawn between the intervertebral disc and the small appendages, wherein the first and second pathological degree marking lines are arranged sequentially from top to bottom; with the center of the spinal canal as the center, Draw the first, second, third, and fourth lesion location marking lines on the edges of the articular joints on both sides to obtain the nine-square grid map; wherein the first, second, third, and fourth lesion location marking lines are from the left
  • the second lesion location marking line is located between the center of the spinal
  • the method further includes the following steps: if it is determined that the prolapsed lesion does not protrude from the first lesion degree marking line, determining that the lumbar intervertebral disc protrusion is a mild lesion If it is determined that the prolapsed lesion protrudes from the first lesion degree marking line and does not protrude from the second lesion degree marking line, it is determined that the lumbar disc herniation is a moderate lesion; if the prolapsed lesion is determined If it protrudes beyond the second marking line of the degree of lesion, it is determined that the lumbar intervertebral disc herniation is a severe lesion; if it is determined that the prolapsed lesion is between the second lesion location marking line and the third lesion location marking line, Then it is determined that the prolapsed lesion is located in the first location area; if it is determined that the prolapsed lesion is between the first lesion location marking line and the second le
  • the prolapsed lesion is located in the second location area; if it is determined that the prolapsed lesion protrudes beyond the first lesion location marking line, or protrudes from the first lesion location marking line. If the four lesion locations are outside the marked line, it is determined that the prolapsed lesion is located in the third location area; the corresponding surgical strategy is determined according to the extent of the lesion of the lumbar intervertebral disc herniation and the location area of the prolapsed lesion.
  • the lumbar spine scan sequence image is a lumbar spine CT scan sequence image, a lumbar spine X-ray scan sequence image, or a lumbar spine magnetic resonance image.
  • the various modules in the above-mentioned lumbar disease diagnosis result output system can be implemented in whole or in part by software, hardware and a combination thereof.
  • the above-mentioned modules may be embedded in the form of hardware or independent of the processor in the computer equipment, or may be stored in the memory of the computer equipment in the form of software, so that the processor can call and execute the operations corresponding to the above-mentioned modules.
  • one or more readable storage media storing computer readable instructions are provided.
  • the readable storage medium stores computer readable instructions.
  • the readable storage media provided in this embodiment include non-transitory A volatile readable storage medium and a volatile readable storage medium.
  • the one or more processors implement the following steps: pre-processing the lumbar scan sequence images Process to obtain a scan image of the intervertebral disc layer; input the scan image of the intervertebral disc layer into a segmentation model based on a neural network for segmentation processing to segment the lumbar intervertebral disc, prolapsed lesions, and mark the degree of disease for the classification of lumbar intervertebral disc herniation Line; based on the segmented lumbar intervertebral disc, the prolapsed lesion and the marked line of the extent of the lesion, the diagnosis result of the lumbar spine lesion is obtained.
  • Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
  • Volatile memory may include random access memory (RAM) or external cache memory.
  • RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous chain Channel (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.
  • the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.

Landscapes

  • Health & Medical Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Medical Informatics (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Public Health (AREA)
  • Biophysics (AREA)
  • Pathology (AREA)
  • Heart & Thoracic Surgery (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Veterinary Medicine (AREA)
  • Surgery (AREA)
  • Animal Behavior & Ethology (AREA)
  • Radiology & Medical Imaging (AREA)
  • Theoretical Computer Science (AREA)
  • High Energy & Nuclear Physics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Optics & Photonics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Dentistry (AREA)
  • General Engineering & Computer Science (AREA)
  • Software Systems (AREA)
  • Computational Linguistics (AREA)
  • Computing Systems (AREA)
  • Orthopedic Medicine & Surgery (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Quality & Reliability (AREA)
  • Physical Education & Sports Medicine (AREA)
  • Databases & Information Systems (AREA)
  • Epidemiology (AREA)
  • Rheumatology (AREA)
  • Fuzzy Systems (AREA)
  • Physiology (AREA)

Abstract

一种人工智能技术领域的智能决策应用领域，其中，腰椎病变诊断结果输出系统，包括：图像获取装置，用于获取腰椎扫描序列图像；图像处理装置，用于对腰椎扫描序列图像进行预处理，得到间盘层扫描图像；图像处理装置，还用于将间盘层扫描图像输入基于神经网络的分割模型中进行分割处理；图像处理装置，还用于基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果；输出装置，用于输出病变诊断结果。基于神经网络的腰椎病变诊断结果输出系统可以自动判断腰椎的病变诊断结果；另外还具有判读精确程度更高的优点。所述病变诊断结果存储于区块链中。可以提高诊疗效率。

Description

基于神经网络的腰椎病变诊断结果输出系统以及方法
本申请要求于2020年9月04日提交中国专利局、申请号为202010922602.8，申请名称为“基于神经网络的腰椎病变诊断结果输出系统以及方法”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及基于人工智能技术领域的智能决策等应用领域，尤其涉及一种基于神经网络的腰椎病变诊断结果输出系统以及方法。
 
背景技术
腰椎间盘突出症是引起腰腿痛的最常见疾病之一，腰椎间盘突出症包含脱出、突出、膨出、间盘游离等征象，腰椎间盘突出症在影像诊断工作中重复性极高。目前市场上尚未出现成熟落地的针对腰椎间盘上各类病变的基于神经网络的影像辅助诊断系统，早期针对腰椎间盘的影像学分析方法主要是采用传统图像学方法，如边缘提取等。
 
技术问题
申请人意识到，这类传统图像学方法精度低，且腰椎间盘突出的辅助检查主要为影像学检查，而影像学报告目前均为人工判读，影像医生的工作量大，工作重复度高。同时阅片的主观性比较强，虽然诊断结果有一定的撰写规范，但不同的医生会有其各自的描述用语，导致诊断结果存在较大的描述误差。
 
技术解决方案
本申请提供一种基于神经网络的腰椎病变诊断结果输出系统以及方法，以解决基于神经网络的腰椎病变诊断结果输出系统的人工判读，影像医生的工作量大，工作重复度高以及不同的医生会有其各自的描述用语，导致诊断结果存在较大的描述误差的问题。
一种基于神经网络的腰椎病变诊断结果输出系统，包括：图像获取装置，用于获取腰椎扫描序列图像；图像处理装置，用于对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；所述图像处理装置，还用于将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；所述图像处理装置，还用于基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果；输出装置，用于输出所述病变诊断结果。
一种基于神经网络的腰椎病变诊断结果输出方法，包括：对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
一个或多个存储有计算机可读指令的可读存储介质，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
 
有益效果
本申请提出的基于神经网络的腰椎病变诊断结果输出系统包括图像获取装置，用于获取腰椎扫描序列图像；图像处理装置，用于对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；所述图像处理装置，还用于将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；所述图像处理装置，还用于基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果；输出装置，用于输出病变诊断结果。本申请提出的基于神经网络的腰椎病变诊断结果输出系统可以自动判读腰椎间盘的病变诊断结果，为疾病治疗方案的选择提供有价值的参考依据；另外，本申请提出的基于神经网络的腰椎病变诊断结果输出系统还具有阅片速度快、判读精确程度更高的优点，通过该系统统一输出的病变诊断结果，可以绕开同医务人员水平或不同地区诊疗水平的差异，导致诊断结果的描述存在较大误差的问题。
本申请的一个或多个实施例的细节在下面的附图和描述中提出，本申请的其他特征和优点将从说明书、附图以及权利要求变得明显。
 
附图说明
为了更清楚地说明本申请的技术方案，下面将对本申请的描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动性的前提下，还可以根据这些附图获得其他的附图。
图1是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统的原理框图；
图2是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中采用的预设Vnet模型的分割网络结构示意图；
图3是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中采用的预设Vnet模型的分割网络结构的第一编码单元的示意图；
图4是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中采用的预设Vnet模型的分割网络结构的第四解码单元的示意图；
图5是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中采用的两条病变程度标注线的示意图；
图6是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中采用的四条病变位置标注线的示意图；
图7是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中采用的九宫格图的示意图；
图8是本申请一实施例中基于神经网络的腰椎病变诊断结果输出系统中腰椎间盘不同的突出情况的示意图。
 
具体实施方式
下面将结合本申请中的附图，对本申请中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本申请一部分实施例，而不是全部的实施例。基于本申请中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本申请保护的范围。
在一实施例中，如图1所示，提供一种基于神经网络的腰椎病变诊断结果输出系统图像获取装置10、图像处理装置20和输出装置3，其中，各装置的功能或作用如下所示：
图像获取装置10，用于获取腰椎扫描序列图像。
可理解地，本方案可以利用CT扫描装置获取腰椎CT扫描序列图像、或利用X光扫描装置获取腰椎X光图像，从而得到腰椎扫描序列图像。甚至还可以获取腰椎的MRI等医学图像，具体本申请不做限定，也就是说，所述腰椎扫描序列图像可以为腰椎CT扫描序列图像、腰椎X光扫描序列图像或腰椎磁共振图像。
图像处理装置20，用于对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像。
在一实施例中，所述腰椎扫描序列图像包括定位像和轴扫图像，图像处理装置对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像，具体指的是：
剔除所述腰椎扫描序列图像的腰椎定位像，以获取所述腰椎扫描序列的轴扫图像。
从所述轴扫图像中选取所述腰椎扫描序列的螺旋扫描图像。
将所述螺旋扫描图像重建形成所述间盘层扫描图像。
可理解地，腰椎定位像是指反映腰椎间隙有无变窄以及生理曲度的图像；轴扫图像是指反映腰椎间盘是否突出以及腰椎间盘四周异常的图像。
腰椎扫描序列中可能包含腰椎定位像以及轴扫图像，而轴扫图像中可能是螺旋扫描，也可能是非螺旋扫描，脱出病灶主要是通过轴扫的间腰椎间盘图像来观察，因此需要对定位像进行过滤去除，从而得到去除定位像后的轴扫图像；再选取所述轴扫图像中的螺旋扫描图像，将选取的螺旋扫描图像进行重建组成间盘层扫描图像。
其中，通过螺旋扫描图像进行重建组成间盘层扫描图像作为后续分割模型的输入，能够使得分割出的出腰椎间盘、脱出病灶以及用于腰椎间盘突出症MSU分型的两条病变程度标注线更精确。
所述图像处理装置20，还用于将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线。
可理解地，腰椎间盘突出症分型简称MSU分型，腰椎间盘突出症MSU分型是一种对腰椎横断面图像显示的病变程度进行分型的方法，该病变程度标注线为MSU分型中对腰椎病变程度进行划分的参考线。
在一实施例中，所述图像处理装置20，还用于：
在分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，判断所述间盘层扫描图像相对于正常间盘层扫描图像位置是否处于水平位置。
若不是处于水平位置，则将所述间盘层扫描图像调整至水平位置，以使得到的所述病变程度标注线为水平的标注线。
可理解地，由于获取的腰椎扫描序列图像所限制，使得分割出的病变程度标注线不一定是水平的，如果患者的扫描体歪了，那么从间盘层扫描图像上看，椎块、间盘、横向标注线都会有一定角度的旋转；若所述间盘层扫描图像不是处于水平位置，则将所述间盘层扫描图像调整至水平位置，从而将扫描图像转正，以使可视化的结果看起来都是水平的，以使得到的所述病变程度标注线为水平的标注线。
在本实施例中，判断所述间盘层扫描图像相对于正常间盘层扫描图像位置是否处于水平位置，若不是处于水平位置，则将所述间盘层扫描图像调整至水平位置，以使可视化的结果看起来都是水平的，进而提高阅片速度快。
在一实施例中，所述图像处理装置20具体还用于：
通过预设训练好的预设Vnet模型以对所述间盘层扫描图像进行分割处理，以得到分割图像，其中，所述分割图像包括已分割出的所述腰椎间盘、脱出病灶以及病变程度标注线。
可理解地，医学图像的分隔通常会使用Vnet模型，预设训练好的Vnet（U-Net Convolutional Networks for BiomedicalImage Segmentation）模型。基于神经网络的分割模型，分割内容包括腰椎间盘、脱出病灶以及用于腰椎间盘突出症MSU分型的病变程度标注线。
本方案中，通过预设训练好的预设Vnet模型以对所述间盘层扫描图像进行分割处理，得到分割后的间盘层扫描图像，其中，分割后的间盘层扫描图像已初步分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症MSU分型的两条病变程度标注线，传统的Vnet模型对于椎间盘的分割以及两条线的定位比较精准，但是对于脱出病灶的分割效果不佳，腰椎间盘脱出症主要表现为间盘下边缘的一系列形态上的细微变化，如脱出表现为间盘下边缘上的明显突起，膨出表现为间盘下边缘整体的向外扩张。征象的分布区域相对于原图来说比例非常小，深度神经网络对于大感受野下结构化特征的提取具有显著的优势，因此，所述预设Vnet模型采用本方案所提出的分割网络结构训练形成，分割网络结构中融合了间盘脱出特征，以提升细粒度特征的分割效果。
在一实施例中，所述预设Vnet模型采用如下分割网络结构，包括依次连接的上采样网络和下采样网络，所述上采样网络包括依次向下采样连接的第一编码单元、第二编码单元、第三编码单元、第四编码单元和第五编码单元；所述下采样网络包括依次向上采样连接的第四解码单元、第三解码单元、第二解码单元、第一解码单元，所述第五编码单元的输出层连接于所述第四解码单元：
其中，所述第一编码单元、第二编码单元、第三编码单元和第四编码单元中各编码单元与所述下采样网络中对应的解码单元连接，所述第三解码单元的输出端还与所述第一解码单元的输入端连接，以将所述第三解码单元输出的特征图作为第一解码单元的输入；
所述上采样网络中各编码单元包括依次连接的输入层、卷积层、池化层和输出层；所述下采样网络中各解码单元包括依次连接的输入层、反卷积层、卷积层和输出层。
可理解地，前述提到的分割网络结构，是对U-Net（U-Net Convolutional Networks for BiomedicalImage Segmentation）进行改进的分割网络结构，U-Net适用于医学图像分割、自然图像生成，U-net分割网络可以对图像进行端到端的分割，即输入是一幅图像，输出也是一幅图像，本方案中的分割网络结构与传统的U-Net结构类似，也分为下采样阶段和上采样阶段，但对下采样阶段做了改进，具体详见下文描述。
请参见图2，从左上方开始，沿着U型左侧上方到右侧上方，本方案的提出的分割网络的完整路径，向下采样过程中的每一层级过程称为编码单元，依次包括第一编码单元、第二编码单元、第三编码单元、第四编码单元和第五编码单元；将向上采样过程中的每一层级过程用解码单元表示，依次包括第四解码单元、第三解码单元、第二解码单元、第一解码单元。例如，请参见图3，第一编码单元包括依次连接的输入层、卷积层、池化层和输出层，该第一编码单元的功能如下所示：通过输入层输入572×572×1的单通道图像，经过3×3卷积层卷积后，转变为570×570×64，再经过3×3的卷积后，图像尺寸缩小了2，同理，再卷积后就变成了568×568×64，在卷积结束后，会经过2×2池化层，然后，图像尺寸就缩小了一倍，变成了284×284×64，最后通过输出层输出至下一层级单元，也即后续的第二编码单元、第三编码单元、第四编码单元及第五编码单元，下一层级单元的处理过程与前述第一编码单元的操作类似，其中的卷积参数和池化参数所有不同，可参阅图3所示，具体不再详述。到第五编码单元，也就是U型网络结构的最下方时，此时的第五编码单元输出的feature maps尺寸为28×28×1024；然后开始右侧的第四解码单元。请参见图4，第四解码单元具体如下：首先输入图像经过2×2上采样卷积，也称为反卷积或转置卷积，可以将图像尺寸扩大一倍，那么相对应的，也需要将feature map数量缩小一倍，得到56×56×512的数据。后续第三解码单元的输入是叠加相应编码单元的输出形成的1024通道的数据，然后，以56×56×1024的输入进行两次valid卷积，然后再上采样、叠加、卷积，后续的第三解码单元、第二解码单元、第一解码单元将重复上述第四解码单元的操作，其中涉及的卷积等参数不同，具体不再详述，可参阅图3，最后到388×388×64的输出。
经过大量实验，第三解码单元能提取较好强的脱出病灶特征表达，而在第二解码单元和第一解码单元筛选和融合过程中，脱出病灶的部分特征会逐渐被丢弃，导致第一解码单元输出的脱出病灶分割效果不佳，因此，本方案将所述第三解码单元的输出端与所述第一解码单元的输入端连接，以将所述第三编码单元输出的特征图作为第一解码单元的输入。
在本实施例中，将所述第三编码单元的输出端与所述第一解码单元的输入端连接，以将所述第三编码单元输出的特征图作为第一解码单元的输入，可保留第三编码单元获取到的较好的脱出病灶特征表达，从而提高脱出病灶分割效果。
所述图像处理装置20，还用于基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
在一实施例中，所述病变程度标注线包括多条，所述图像处理装置20用于：
以椎管中央为中心绘制多条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的宫格图；
通过判断脱出病灶分布在所述宫格图的位置，以获得腰椎间盘突出的诊断结果。
可理解地，以椎管中央为中心绘制多条与所述病变程度标注线垂直的病变位置标注线，根据腰椎间盘突出所在位置和突出程度，以划分不同区域，从而获得腰椎间盘突出症的宫格图。再通过判断脱出病灶分布在所述宫格图的位置，以获得腰椎间盘突出的诊断结果。
在本实施例中，通过以椎管中央为中心绘制多条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的宫格图，通过宫格图可以精准定位病灶的位置和判断病变程度，以使得判读精确程度更高，提高诊疗效率。
在一实施例中，所述病变程度标注线包括两条，所述图像处理装置20用于：
以椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图；
通过判断脱出病灶分布在所述九宫格图的位置，以获得腰椎间盘突出的诊断结果。
可理解地，基于U-Net分割网络的分割结果进行后处理，以椎管中央为中心绘制与两条所述病变程度标注线垂直的四条病变位置标注线，根据腰椎间盘突出所在位置和突出程度，划分不同区域，从而获得腰椎间盘突出症分的九宫格图；通过判断脱出病灶分布在所述九宫格图的位置，以获得腰椎间盘突出的诊断结果。
在本实施例中，通过以椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图，通过判断脱出病灶分布在所述九宫格图的位置，以获得腰椎间盘突出的诊断结果，以九宫格图使得判读精确程度更高，提高诊疗效率。
在一实施例中，所述图像处理装置20用于：
在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，其中，所述第一、二条病变程度标注线从上往下依次排列的病变程度标注线；
以所述椎管中央为中心，两侧的关节突关节边缘为界分别绘制第一、二、三、四条病变位置标注线，以获得所述九宫格图；
其中，所述第一、二、三、四条病变位置标注线为从左往右依次排列的病变位置标注线，所述第一条病变位置标注线位于其中一侧的关节突关节边缘，所述第四条病变位置标注线位于其中另一侧的关节突关节边缘，所述第二条病变位置标注线位于椎管中央与所述一侧的关节突关节边缘之间，所述第三条病变位置标注线位于椎管中央与所述另一侧的关节突关节边缘之间。
可理解地，所述九宫格图用来描述病变程度，在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，其中，参见图5及图7，病变程度由第一、二条病变程度标注线分割得到的三个等次的病变程度，图5中，由上往下两条虚线分别表示第一条病变程度标注线和第二条病变程度标注线。
可理解地，参见图6及图7，九宫格图用来描述病灶的位置，图6和7中，从左侧至右侧的虚线分别表示所述第一、二、三、四条病变位置标注线，所述第一、二、三、四条病变位置标注线以突出部位以椎管中线、两侧关节突关节边缘为界绘制。
所述九宫格图根据所述第一、二条病变程度标注线以及所述第一、二、三、四条病变位置标注线绘制而成，同时，所述病变程度标注线与所述条病变位置标注线垂直。
在本实施例中，通过在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，以所述椎管中央为中心，两侧的关节突关节边缘为界分别绘制第一、二、三、四条病变位置标注线，以获得所述九宫格图，通过九宫格图快速判断出腰椎间盘突出症的诊断结果，具有阅片速度快、判读精确程度更高的优点，大大缓解医务人员负担，提高诊疗效率。
在一实施例中，所述图像处理装置20用于：
若确定所述脱出病灶未凸出于第一条病变程度标注线，则判断所述腰椎间盘突出为轻度病变；
若确定所述脱出病灶凸出于所述第一条病变程度标注线且未凸出于第二条病变程度标注线，则判断所述腰椎间盘突出为中度病变；
若确定所述脱出病灶凸出于第二条病变程度标注线外，则判断所述腰椎间盘突出为重度病变；
若确定所述脱出病灶在所述第二条病变位置标注线与第三条病变位置标注线之间，则判断所述脱出病灶位于第一位置区域；
若确定所述脱出病灶在所述第一条病变位置标注线与第二条病变位置标注线之间，或者，确定所述脱出病灶在第三条病变位置标注线与第四条病变位置标注线之间，则判断所述脱出病灶位于第二位置区域；
若确定所述脱出病灶凸出于第一条病变位置标注线外，或凸出于第四条病变位置标注线外，则判断所述脱出病灶位于第三位置区域；
根据腰椎间盘突出的病变程度和所述脱出病灶的位置区域确定对应的手术策略。
可理解地，九宫格图用来描述病变程度，参见图5及图7，在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，根据第一、二条病变程度标注线分为1、2、3级程度，例如1级程度为上关节突水平，2级程度为关节突关节间隙，3级程度为下关节突水平。其中，三个等次的病变程度分别为轻度、中度和重度，轻度、中度和重度分别表示腰椎病变的轻、中、重。若确定所述脱出病灶未凸出于第一条病变程度标注线，则判断所述腰椎间盘突出为轻度病变，也即腰椎间盘突出为1级程度“上关节突水平”，则判断所述腰椎间盘突出为轻度病变。若确定所述脱出病灶凸出于所述第一条病变程度标注线且未凸出于所述第二条病变程度标注线，也即腰椎间盘突出为2级程度“关节突关节间隙”，则判断所述腰椎间盘突出为中度病变。若确定所述脱出病灶凸出于第二条病变程度标注线外，也即腰椎间盘突出为3级程度“下关节突水平”，则判断所述腰椎间盘突出为重度病变。
可理解地，以突出部位以椎管中线、两侧关节突关节边缘为界绘制所述第一、二、三、四条病变位置标注线，根据所述第一、二、三、四条病变位置标注线分为A、B、C三个区域，其中A、B、C三个区域分别表示第一位置区域、第二位置区域以及第三位置区域，例如，A区域为椎管中央区域，B区域为超过椎管中央，且在椎管内的区域，C区域为关节突，测隐窝区域。四条病变位置标注线分割的区域从左到右位置为第三位置区域、第二位置区域、第一位置区域、第二位置区域以及第三位置区域。若确定所述脱出病灶在所述第二条病变位置标注线与第三条病变位置标注线之间，也即脱出病灶位于A区域“椎管中央区域”，则判断所述脱出病灶位于第一位置区域。若确定所述脱出病灶在所述第一条病变位置标注线与第二条病变位置标注线之间，或者，确定所述脱出病灶在第三条病变位置标注线与所述第四条病变位置标注线之间，也即脱出病灶位于B区域“超过椎管中央，且在椎管内的区域”，则判断所述脱出病灶位于第二位置区域。若确定所述脱出病灶凸出于第一条病变位置标注线外，或凸出于第四条病变位置标注线外，也即脱出病灶位于C区域“关节突，测隐窝区域”，则判断所述脱出病灶位于第三位置区域。
可理解地，参见图7和图8，根据腰椎间盘突出的病变程度和所述脱出病灶的位置区域确定对应的手术策略，以获得腰椎间盘突出的诊断结果，诊断结果可以为1-A、1-C、2-B、3-B等，具体本申请不做限定，基于九宫格图的诊断结果的手术策略：诊断结果为1级程度的可以不考虑手术，1-AB可能对神经节压迫，可考虑需要手术或者保守；2级程度需要手术，尤其是2-B、2-AB需要手术，2-A类型如果症状较轻，可考虑保守治疗；3级程度，多数需要考虑手术。
在本实施例中，通过脱出病灶的凸出程度来判断腰椎间盘突出的病变程度，再通过脱出病灶所在位置来判断脱出病灶所处的位置区域，再通过腰椎间盘突出的病变程度以及脱出病灶的位置区域快速判断出腰椎间盘突出症的诊断结果，具有阅片速度快、判读精确程度更高的优点，大大缓解医务人员负担，提高诊疗效率。
输出装置30，用于输出病变诊断结果。
可理解地，通过判断脱出病灶分布在所述腰椎间盘突出症MSU分型的九宫格图的位置，通过观察脱出病灶在九宫格图的位置，以输出病变诊断结果，从而辅助医生做较为准确的诊断。
本方案通过显示装置来接收输出装置输出的病变诊断结果，以使病变诊断结果实现可视化。需要强调的是，为进一步保证上述病变诊断结果的私密和安全性，上述病变诊断结果还可以存储于一区块链的节点中。
在图1对应的实施例中，本申请提出的基于神经网络的腰椎病变诊断结果输出系统包括图像获取装置，用于获取腰椎扫描序列图像；图像处理装置，用于对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；所述图像处理装置，还用于将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；所述图像处理装置，还用于基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果；输出装置，用于输出病变诊断结果。本申请提出的基于神经网络的腰椎病变诊断结果输出系统可以自动判读腰椎间盘是否突出以及突出程度，为疾病治疗方案的选择提供有价值的参考依据；另外，本申请提出的基于神经网络的腰椎病变诊断结果输出系统还具有阅片速度快、判读精确程度更高的优点，缩小了不同医务人员水平或不同地区诊疗水平的差异，大大缓解医务人员负担，提高诊疗效率。
一种腰椎病变诊断结果输出方法应用于腰椎病变诊断结果输出系统中，下面结合示例进行具体描述。
在一实施例中，所述腰椎病变诊断结果输出方法包括如下步骤：对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
在一实施例中，所述对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像，包括以下步骤：剔除所述腰椎扫描序列图像的腰椎定位像，以获取所述腰椎扫描序列的轴扫图像；从所述轴扫图像中选取所述腰椎扫描序列的螺旋扫描图像；将所述螺旋扫描图像重建形成所述间盘层扫描图像。
在一实施例中，所述将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线，包括以下步骤：通过预设训练好的预设Vnet模型以对所述间盘层扫描图像进行分割处理，以得到分割图像，所述分割图像包括已分割出的所述腰椎间盘、脱出病灶以及病变程度标注线。
在一实施例中，所述预设Vnet模型采用如下分割网络结构：包括依次连接的上采样网络和下采样网络，所述上采样网络包括依次向下采样连接的第一编码单元、第二编码单元、第三编码单元、第四编码单元和第五编码单元；所述下采样网络包括依次向上采样连接的第四解码单元、第三解码单元、第二解码单元、第一解码单元，所述第五编码单元的输出层连接于所述第四解码单元；其中，所述第一编码单元、第二编码单元、第三编码单元和第四编码单元中各编码单元与所述下采样网络中对应的解码单元连接，所述第三解码单元的输出端还与所述第一解码单元的输入端连接，以将所述第三解码单元输出的特征图作为第一解码单元的输入；所述上采样网络中各编码单元包括依次连接的输入层、卷积层、池化层和输出层；所述下采样网络中各解码单元包括依次连接的输入层、反卷积层、卷积层和输出层。
在一实施例中，在所述分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，还包括以下步骤：判断所述间盘层扫描图像相对于正常间盘层扫描图像位置是否处于水平位置；若不是处于水平位置，则将所述间盘层扫描图像调整至水平位置，以使得到的所述病变程度标注线为水平的标注线。
在一实施例中，所述病变程度标注线包括多条，在所述分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，还包括以下步骤：以椎管中央为中心绘制多条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的宫格图；通过判断脱出病灶分布在所述宫格图的位置，以获得腰椎间盘突出的诊断结果。
在一实施例中，所述病变程度标注线包括两条，在所述分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，还包括以下步骤：以所述椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图；通过判断脱出病灶分布在所述九宫格图的位置，以获得所述腰椎间盘突出的诊断结果。
在一实施例中，所述以所述椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图，包括以下步骤：在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，其中，所述第一、二条病变程度标注线从上往下依次排列的病变程度标注线；以所述椎管中央为中心，两侧的关节突关节边缘为界分别绘制第一、二、三、四条病变位置标注线，以获得所述九宫格图；其中，所述第一、二、三、四条病变位置标注线为从左往右依次排列的病变位置标注线，所述第一条病变位置标注线位于其中一侧的关节突关节边缘，所述第四条病变位置标注线位于其中另一侧的关节突关节边缘，所述第二条病变位置标注线位于椎管中央与所述一侧的关节突关节边缘之间，所述第三条病变位置标注线位于椎管中央与所述另一侧的关节突关节边缘之间。
在一实施例中，在所述获得所述九宫格图之后，还包括以下步骤：若确定所述脱出病灶未凸出于第一条病变程度标注线，则判断所述腰椎间盘突出为轻度病变；若确定所述脱出病灶凸出于所述第一条病变程度标注线且未凸出于第二条病变程度标注线，则判断所述腰椎间盘突出为中度病变；若确定所述脱出病灶凸出于第二条病变程度标注线外，则判断所述腰椎间盘突出为重度病变；若确定所述脱出病灶在所述第二条病变位置标注线与第三条病变位置标注线之间，则判断所述脱出病灶位于第一位置区域；若确定所述脱出病灶在所述第一条病变位置标注线与第二条病变位置标注线之间，或者，确定所述脱出病灶在第三条病变位置标注线与第四条病变位置标注线之间，则判断所述脱出病灶位于第二位置区域；若确定所述脱出病灶凸出于第一条病变位置标注线外，或凸出于第四条病变位置标注线外，则判断所述脱出病灶位于第三位置区域；根据腰椎间盘突出的病变程度和所述脱出病灶的位置区域确定对应的手术策略。
在一实施例中，所述腰椎扫描序列图像为腰椎CT扫描序列图像、腰椎X光扫描序列图像或腰椎磁共振图像。
关于腰椎病变诊断结果输出系统的具体限定可以参见上文中对于腰椎病变诊断结果输出方法的限定，在此不再赘述。上述腰椎病变诊断结果输出系统中的各个模块可全部或部分通过软件、硬件及其组合来实现。上述各模块可以硬件形式内嵌于或独立于计算机设备中的处理器中，也可以以软件形式存储于计算机设备中的存储器中，以便于处理器调用执行以上各个模块对应的操作。
在一个实施例中，提供了一个或多个存储有计算机可读指令的可读存储介质，该可读存储介质上存储有计算机可读指令，本实施例所提供的可读存储介质包括非易失性可读存储介质和易失性可读存储介质，该计算机可读指令被一个或多个处理器执行时，使得一个或多个处理器实现以下步骤：对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，所述的计算机可读指令可存储于一非易失性计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，本申请所提供的各实施例中所使用的对存储器、存储、数据库或其它介质的任何引用，均可包括非易失性和/或易失性存储器。非易失性存储器可包括只读存储器（ROM）、可编程ROM（PROM）、电可编程ROM（EPROM）、电可擦除可编程ROM（EEPROM）或闪存。易失性存储器可包括随机存取存储器（RAM）或者外部高速缓冲存储器。作为说明而非局限，RAM以多种形式可得，诸如静态RAM（SRAM）、动态RAM（DRAM）、同步DRAM（SDRAM）、双数据率SDRAM（DDRSDRAM）、增强型SDRAM（ESDRAM）、同步链路（Synchlink） DRAM（SLDRAM）、存储器总线（Rambus）直接RAM（RDRAM）、直接存储器总线动态RAM（DRDRAM）、以及存储器总线动态RAM（RDRAM）等。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
所属领域的技术人员可以清楚地了解到，为了描述的方便和简洁，仅以上述各功能单元、模块的划分进行举例说明，实际应用中，可以根据需要而将上述功能分配由不同的功能单元、模块完成，即将所述装置的内部结构划分成不同的功能单元或模块，以完成以上描述的全部或者部分功能。
以上所述实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的精神和范围，均应包含在本申请的保护范围之内。
 

Claims (20)

  1. 一种基于神经网络的腰椎病变诊断结果输出系统，其中，包括：
    图像获取装置，用于获取腰椎扫描序列图像；
    图像处理装置，用于对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；
    所述图像处理装置，还用于将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；
    所述图像处理装置，还用于基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果；
    输出装置，用于输出所述病变诊断结果。
  2. 如权利要求1所述的腰椎病变诊断结果输出系统，其中，所述图像处理装置具体用于：
    剔除所述腰椎扫描序列图像的腰椎定位像，以获取所述腰椎扫描序列的轴扫图像；
    从所述轴扫图像中选取所述腰椎扫描序列的螺旋扫描图像；
    将所述螺旋扫描图像重建形成所述间盘层扫描图像。
  3. 如权利要求2所述的腰椎病变诊断结果输出系统，其中，所述图像处理装置具体还用于：
    通过预设训练好的预设Vnet模型以对所述间盘层扫描图像进行分割处理，以得到分割图像，所述分割图像包括已分割出的所述腰椎间盘、脱出病灶以及病变程度标注线。
  4. 如权利要求3所述的腰椎病变诊断结果输出系统，其中，所述预设Vnet模型采用如下分割网络结构：包括依次连接的上采样网络和下采样网络，所述上采样网络包括依次向下采样连接的第一编码单元、第二编码单元、第三编码单元、第四编码单元和第五编码单元；所述下采样网络包括依次向上采样连接的第四解码单元、第三解码单元、第二解码单元、第一解码单元，所述第五编码单元的输出层连接于所述第四解码单元；
    其中，所述第一编码单元、第二编码单元、第三编码单元和第四编码单元中各编码单元与所述下采样网络中对应的解码单元连接，所述第三解码单元的输出端还与所述第一解码单元的输入端连接，以将所述第三解码单元输出的特征图作为第一解码单元的输入；
    所述上采样网络中各编码单元包括依次连接的输入层、卷积层、池化层和输出层；所述下采样网络中各解码单元包括依次连接的输入层、反卷积层、卷积层和输出层。
  5. 如权利要求1-4任一项所述的腰椎病变诊断结果输出系统，其中，所述图像处理装置，还用于：
    在分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，判断所述间盘层扫描图像相对于正常间盘层扫描图像位置是否处于水平位置；
    若不是处于水平位置，则将所述间盘层扫描图像调整至水平位置，以使得到的所述病变程度标注线为水平的标注线。
  6. 如权利要求5所述的腰椎病变诊断结果输出系统，其中，所述病变程度标注线包括多条，所述图像处理装置用于：
    以椎管中央为中心绘制多条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的宫格图；
    通过判断脱出病灶分布在所述宫格图的位置，以获得腰椎间盘突出的诊断结果。
  7. 如权利要求6所述的腰椎病变诊断结果输出系统，其中，所述病变程度标注线包括两条，所述图像处理装置用于：
    以所述椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图；
    通过判断脱出病灶分布在所述九宫格图的位置，以获得所述腰椎间盘突出的诊断结果。
  8. 如权利要求7所述的腰椎病变诊断结果输出系统，其中，所述图像处理装置用于：
    在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，其中，所述第一、二条病变程度标注线从上往下依次排列的病变程度标注线；
    以所述椎管中央为中心，两侧的关节突关节边缘为界分别绘制第一、二、三、四条病变位置标注线，以获得所述九宫格图；
    其中，所述第一、二、三、四条病变位置标注线为从左往右依次排列的病变位置标注线，所述第一条病变位置标注线位于其中一侧的关节突关节边缘，所述第四条病变位置标注线位于其中另一侧的关节突关节边缘，所述第二条病变位置标注线位于椎管中央与所述一侧的关节突关节边缘之间，所述第三条病变位置标注线位于椎管中央与所述另一侧的关节突关节边缘之间。
  9. 如权利要求8所述的腰椎病变诊断结果输出系统，其中，所述图像处理装置用于：
    若确定所述脱出病灶未凸出于第一条病变程度标注线，则判断所述腰椎间盘突出为轻度病变；
    若确定所述脱出病灶凸出于所述第一条病变程度标注线且未凸出于第二条病变程度标注线，则判断所述腰椎间盘突出为中度病变；
    若确定所述脱出病灶凸出于第二条病变程度标注线外，则判断所述腰椎间盘突出为重度病变；
    若确定所述脱出病灶在所述第二条病变位置标注线与第三条病变位置标注线之间，则判断所述脱出病灶位于第一位置区域；
    若确定所述脱出病灶在所述第一条病变位置标注线与第二条病变位置标注线之间，或者，确定所述脱出病灶在第三条病变位置标注线与第四条病变位置标注线之间，则判断所述脱出病灶位于第二位置区域；
    若确定所述脱出病灶凸出于第一条病变位置标注线外，或凸出于第四条病变位置标注线外，则判断所述脱出病灶位于第三位置区域；
    根据腰椎间盘突出的病变程度和所述脱出病灶的位置区域确定对应的手术策略。
  10. 如权利要求1-4任一项所述的腰椎病变诊断结果输出系统，其中，所述腰椎扫描序列图像为腰椎CT扫描序列图像、腰椎X光扫描序列图像或腰椎磁共振图像。
  11. 一种基于神经网络的腰椎病变诊断结果输出方法，其中，包括：
    对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；
    将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；
    基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
  12. 一个或多个存储有计算机可读指令的可读存储介质，其中，所述计算机可读指令被一个或多个处理器执行时，使得所述一个或多个处理器执行如下步骤：
    对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像；
    将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线；
    基于分割出的腰椎间盘、脱出病灶以及病变程度标注线获得腰椎的病变诊断结果。
  13. 如权利要求12所述的可读存储介质，其中，所述对所述腰椎扫描序列图像进行预处理，得到间盘层扫描图像，包括以下步骤：
    剔除所述腰椎扫描序列图像的腰椎定位像，以获取所述腰椎扫描序列的轴扫图像；
    从所述轴扫图像中选取所述腰椎扫描序列的螺旋扫描图像；
    将所述螺旋扫描图像重建形成所述间盘层扫描图像。
  14. 如权利要求13所述的可读存储介质，其中，所述将所述间盘层扫描图像输入基于神经网络的分割模型中进行分割处理，以分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线，包括以下步骤：
    通过预设训练好的预设Vnet模型以对所述间盘层扫描图像进行分割处理，以得到分割图像，所述分割图像包括已分割出的所述腰椎间盘、脱出病灶以及病变程度标注线。
  15. 如权利要求14所述的可读存储介质，其中，所述预设Vnet模型采用如下分割网络结构：包括依次连接的上采样网络和下采样网络，所述上采样网络包括依次向下采样连接的第一编码单元、第二编码单元、第三编码单元、第四编码单元和第五编码单元；所述下采样网络包括依次向上采样连接的第四解码单元、第三解码单元、第二解码单元、第一解码单元，所述第五编码单元的输出层连接于所述第四解码单元；
    其中，所述第一编码单元、第二编码单元、第三编码单元和第四编码单元中各编码单元与所述下采样网络中对应的解码单元连接，所述第三解码单元的输出端还与所述第一解码单元的输入端连接，以将所述第三解码单元输出的特征图作为第一解码单元的输入；
    所述上采样网络中各编码单元包括依次连接的输入层、卷积层、池化层和输出层；所述下采样网络中各解码单元包括依次连接的输入层、反卷积层、卷积层和输出层。
  16. 如权利要求12-15任一项所述的可读存储介质，其中，在所述分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，还包括以下步骤：
    判断所述间盘层扫描图像相对于正常间盘层扫描图像位置是否处于水平位置；
    若不是处于水平位置，则将所述间盘层扫描图像调整至水平位置，以使得到的所述病变程度标注线为水平的标注线。
  17. 如权利要求16所述的可读存储介质，其中，所述病变程度标注线包括多条，在所述分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，还包括以下步骤：
    以椎管中央为中心绘制多条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的宫格图；
    通过判断脱出病灶分布在所述宫格图的位置，以获得腰椎间盘突出的诊断结果。
  18. 如权利要求17所述的可读存储介质，其中，所述病变程度标注线包括两条，在所述分割出腰椎间盘、脱出病灶以及用于腰椎间盘突出症分型的病变程度标注线之后，还包括以下步骤：
    以所述椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图；
    通过判断脱出病灶分布在所述九宫格图的位置，以获得所述腰椎间盘突出的诊断结果。
  19. 如权利要求18所述的可读存储介质，其中，所述以所述椎管中央为中心绘制四条与所述病变程度标注线垂直的病变位置标注线，以获得腰椎间盘突出症的九宫格图，包括以下步骤：
    在所述腰椎间盘与椎小附件之间绘制第一、二条病变程度标注线，其中，所述第一、二条病变程度标注线从上往下依次排列的病变程度标注线；
    以所述椎管中央为中心，两侧的关节突关节边缘为界分别绘制第一、二、三、四条病变位置标注线，以获得所述九宫格图；
    其中，所述第一、二、三、四条病变位置标注线为从左往右依次排列的病变位置标注线，所述第一条病变位置标注线位于其中一侧的关节突关节边缘，所述第四条病变位置标注线位于其中另一侧的关节突关节边缘，所述第二条病变位置标注线位于椎管中央与所述一侧的关节突关节边缘之间，所述第三条病变位置标注线位于椎管中央与所述另一侧的关节突关节边缘之间。
  20. 如权利要求19所述的可读存储介质，其中，在所述获得所述九宫格图之后，还包括以下步骤：
    若确定所述脱出病灶未凸出于第一条病变程度标注线，则判断所述腰椎间盘突出为轻度病变；
    若确定所述脱出病灶凸出于所述第一条病变程度标注线且未凸出于第二条病变程度标注线，则判断所述腰椎间盘突出为中度病变；
    若确定所述脱出病灶凸出于第二条病变程度标注线外，则判断所述腰椎间盘突出为重度病变；
    若确定所述脱出病灶在所述第二条病变位置标注线与第三条病变位置标注线之间，则判断所述脱出病灶位于第一位置区域；
    若确定所述脱出病灶在所述第一条病变位置标注线与第二条病变位置标注线之间，或者，确定所述脱出病灶在第三条病变位置标注线与第四条病变位置标注线之间，则判断所述脱出病灶位于第二位置区域；
    若确定所述脱出病灶凸出于第一条病变位置标注线外，或凸出于第四条病变位置标注线外，则判断所述脱出病灶位于第三位置区域；
    根据腰椎间盘突出的病变程度和所述脱出病灶的位置区域确定对应的手术策略。
     
     
PCT/CN2020/135720 2020-09-04 2020-12-11 基于神经网络的腰椎病变诊断结果输出系统以及方法 WO2021121159A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010922602.8 2020-09-04
CN202010922602.8A CN112071422B (zh) 2020-09-04 2020-09-04 基于神经网络的腰椎病变诊断系统

Publications (1)

Publication Number Publication Date
WO2021121159A1 true WO2021121159A1 (zh) 2021-06-24

Family

ID=73665544

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/135720 WO2021121159A1 (zh) 2020-09-04 2020-12-11 基于神经网络的腰椎病变诊断结果输出系统以及方法

Country Status (2)

Country Link
CN (1) CN112071422B (zh)
WO (1) WO2021121159A1 (zh)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113240661B (zh) * 2021-05-31 2023-09-26 平安科技（深圳）有限公司 基于深度学习的腰椎骨分析方法、装置、设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050033139A1 (en) * 2002-07-09 2005-02-10 Deus Technologies, Llc Adaptive segmentation of anatomic regions in medical images with fuzzy clustering
CN108399616A (zh) * 2018-02-11 2018-08-14 西安理工大学 基于深度残差网络的骨科病病变分类分级方法
CN109903269A (zh) * 2019-01-24 2019-06-18 刘星宇 确定脊柱横断面图像的异常类型的方法及计算设备
CN109903268A (zh) * 2019-01-24 2019-06-18 刘星宇 确定脊柱图像集的异常类型的方法及计算设备
CN110448270A (zh) * 2018-08-01 2019-11-15 冯世庆 人工智能腰椎间盘突出诊断及分型系统

Family Cites Families (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108288506A (zh) * 2018-01-23 2018-07-17 雨声智能科技（上海）有限公司 一种基于人工智能技术的癌症病理辅助诊断方法
CN108711187B (zh) * 2018-08-03 2021-12-31 华侨大学 配准融合ct和mri信号建立人体腰椎三维仿真模型的方法
CN109389603B (zh) * 2018-09-10 2021-09-24 北京大学 一种基于预加重策略的全自动腰椎图像分割方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20050033139A1 (en) * 2002-07-09 2005-02-10 Deus Technologies, Llc Adaptive segmentation of anatomic regions in medical images with fuzzy clustering
CN108399616A (zh) * 2018-02-11 2018-08-14 西安理工大学 基于深度残差网络的骨科病病变分类分级方法
CN110448270A (zh) * 2018-08-01 2019-11-15 冯世庆 人工智能腰椎间盘突出诊断及分型系统
CN109903269A (zh) * 2019-01-24 2019-06-18 刘星宇 确定脊柱横断面图像的异常类型的方法及计算设备
CN109903268A (zh) * 2019-01-24 2019-06-18 刘星宇 确定脊柱图像集的异常类型的方法及计算设备

Also Published As

Publication number Publication date
CN112071422B (zh) 2024-05-03
CN112071422A (zh) 2020-12-11

Similar Documents

Publication Publication Date Title
CN111047572B (zh) 一种基于Mask RCNN的医学图像中脊柱自动定位方法
US10068671B2 (en) Methods and systems for producing an implant
CN113506334B (zh) 基于深度学习的多模态医学图像融合方法及系统
CN109166105A (zh) 人工智能医学影像的肿瘤恶性风险分层辅助诊断系统
WO2024001140A1 (zh) 一种椎体亚区域分割方法、装置及存储介质
CN111340825A (zh) 一种纵膈淋巴结分割模型的生成方法及系统
CN111951219B (zh) 基于眼眶ct图像的甲状腺眼病筛查方法、系统及设备
CN110852991B (zh) 一种基于3d结合2d的脊椎骨检测方法
WO2021121159A1 (zh) 基于神经网络的腰椎病变诊断结果输出系统以及方法
Tran et al. MBNet: A multi-task deep neural network for semantic segmentation and lumbar vertebra inspection on X-ray images
CN110738633A (zh) 一种机体组织的三维图像处理方法及相关设备
Qin et al. Residual block-based multi-label classification and localization network with integral regression for vertebrae labeling
CN115953416A (zh) 基于深度学习的膝骨关节核磁共振图像自动分割方法
CN111918611A (zh) 胸部x线图像的异常显示控制方法、异常显示控制程序、异常显示控制装置以及服务器装置
CN116797519A (zh) 脑胶质瘤分割和三维可视化模型训练方法及系统
Sha et al. A robust segmentation method based on improved U-Net
CN114596286A (zh) 图像分割方法、系统、装置和存储介质
CN115205238A (zh) 骨折辅助检测方法及系统、计算机设备及可读存储介质
Cong et al. Domain-Distance Adapted Super-Resolution Reconstruction of Low-Field MR Brain Images
JP3996555B2 (ja) 画像処理装置、画像処理方法、及び当該画像処理をコンピュータに実行させるプログラムを格納する記録媒体
CN112509080A (zh) 一种基于迁移学习的颅内血管模拟三维模型的建立方法
Hu et al. Multirigid registration of MR and CT images of the cervical spine
CN117078612B (zh) 一种基于cbct图像的快速三维剂量验证方法及装置
Preetha et al. Computer aided diagnostic system for automatic cervical disc herniation classification
Qiao et al. Coarse-Fine View Attention Alignment-Based GAN for CT Reconstruction from Biplanar X-Rays

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20903019

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20903019

Country of ref document: EP

Kind code of ref document: A1