Method, device and electronic device for determining puncture target point

Abstract

The present invention relates to the field of medical technology, and discloses a method, device and electronic device for determining a puncture target point. The method comprises: acquiring CT image data of a patient, and preprocessing the CT image data to obtain a bone structure map; segmenting the bone structure map to obtain a segmented bone structure; determining the intervertebral space width based on the centroid of the segmented bone structure; determining the puncture intervertebral space based on the intervertebral space width; inputting ultrasound image data and anatomical prior knowledge corresponding to the puncture intervertebral space into a trained ultrasound image recognition model, and outputting a label structure and its corresponding label; determining the puncture target point of epidural puncture based on the label structure and its corresponding label, identifying and prompting the label structure in the real-time ultrasound image during the epidural puncture process, and improving the efficiency and safety of epidural puncture.

Description

Technical Field

[0001] The present disclosure relates to the field of medical technology, and in particular to a method, device and electronic equipment for determining a puncture target point. Background Art

[0002] Epidural puncture is a common and important anesthesia technique, which is widely used in medical practice. This operation can effectively achieve the effects of anesthesia and analgesia by accurately injecting drugs between the dura mater and the periosteum of the spinal canal. Epidural puncture places extremely high demands on the doctor's operating skills and accuracy, because during the operation, the doctor needs to accurately locate and puncture, and must carefully avoid structures such as bone tissue to ensure the safety and effectiveness of the operation. However, traditional epidural puncture has many challenges, including difficulty in operation, high technical requirements for doctors, and high accuracy requirements, which can easily lead to complications and increased risks during the operation, thereby affecting the results of the operation and the safety of the patient.

[0003] At present, auxiliary technologies for epidural puncture are constantly developing. In clinical practice, epidural puncture is often assisted by preoperative chest and abdominal CT images and ultrasound imaging. However, although the traditional epidural puncture technology based on preoperative chest and abdominal CT images and ultrasound guidance has improved the accuracy and safety of the operation to a certain extent, it still has some limitations and disadvantages:

[0004] 1. Dependence on physician experience: Although preoperative chest and abdominal CT imaging and ultrasound guidance technology can provide image assistance, the interpretation of images and judgment of operation guidelines still rely on the physician's rich experience and skills, which is subjective and has certain limitations.

[0005] 2. High technical requirements for operation: Ultrasound-guided technology requires doctors to be proficient in ultrasound imaging principles and operation techniques, and the operation requirements are relatively high.

[0006] 3. Limited by real-time and precision: Ultrasonic imaging is affected by factors such as tissue structure and depth. There may be insufficient image clarity or low real-time performance, which affects the accuracy of the operation.

[0007] In view of this, this application is filed. Summary of the invention

[0008] In order to solve the above technical problems or at least partially solve the above technical problems, the embodiments of the present disclosure provide a method, device and electronic device for determining a puncture target point, which identify and prompt the label structure in the real-time ultrasound image during epidural puncture, thereby improving the efficiency and safety of epidural puncture.

[0009] In a first aspect, an embodiment of the present disclosure provides a method for determining a puncture target point, the method comprising:

[0010] Acquiring CT image data of a patient, and preprocessing the CT image data to obtain a bone structure map;

[0011] Segmenting the bone structure image to obtain segmented bone structure;

[0012] determining the intervertebral disc width based on the centroid of the segmented bony structure;

[0013] Based on the intervertebral disc width, determining the punctured intervertebral disc space;

[0014] Inputting the ultrasound image data corresponding to the punctured intervertebral space and the anatomical prior knowledge into the trained ultrasound image recognition model, and outputting the label structure and its corresponding label;

[0015] Based on the tag structure and its corresponding tag, a puncture target point of epidural puncture is determined.

[0016] In a second aspect, the present disclosure also provides a device for determining a puncture target point, the device comprising:

[0017] A first acquisition module is used to acquire CT image data of a patient and pre-process the CT image data to obtain a bone structure map;

[0018] A second acquisition module is used to segment the bone structure image to obtain segmented bone structure;

[0019] A first determination module, configured to determine the intervertebral disc width based on the centroid of the segmented bone structure;

[0020] A second determination module is used to determine the puncture intervertebral space based on the intervertebral space width;

[0021] An output module, used to input the ultrasound image data corresponding to the punctured intervertebral space and anatomical prior knowledge into a trained ultrasound image recognition model, and output a label structure and its corresponding label;

[0022] The third determination module is used to determine the puncture target point of the epidural puncture based on the label structure and its corresponding label.

[0023] In a third aspect, an embodiment of the present disclosure further provides an electronic device, comprising: one or more processors; a storage device for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining the puncture target point as described above.

[0024] The disclosed embodiment provides a method for determining a puncture target point, which obtains a bone structure map by acquiring CT image data of a patient and preprocessing the CT image data; segments the bone structure map based on an automatic segmentation algorithm to obtain a segmented bone structure; determines the intervertebral space width based on the centroid of the segmented bone structure; determines the puncture intervertebral space based on the intervertebral space width; reconstructs a three-dimensional image from a two-dimensional image, and uses an automatic measurement function to determine the intervertebral space most suitable for the puncture operation, thereby improving the efficiency of the doctor's puncture path planning before the operation, thereby improving the safety and success rate of the operation; then, the ultrasound image data and anatomical prior knowledge corresponding to the puncture intervertebral space are input into a trained ultrasound image recognition model, and a label structure and its corresponding label are output; based on the label structure and its corresponding label, the puncture target point of the epidural puncture is determined. The label structure in the real-time ultrasound image during the epidural puncture process is identified and prompted to improve the efficiency and safety of the epidural puncture. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and the originals and elements are not necessarily drawn to scale.

[0026] Figure 1 The present invention is a flowchart of a method for determining a puncture target point in an embodiment of the present invention.

[0027] Figure 2 It is a schematic diagram of an anatomical structure in an ultrasound image in an embodiment of the present disclosure.

[0028] Figure 3 It is a schematic diagram of another anatomical structure in an ultrasound image in an embodiment of the present disclosure.

[0029] Figure 4 Schematic diagram of the structure of a device for determining a puncture target point in an embodiment of the present disclosure.

[0030] Figure 5 It is a structural schematic diagram of an electronic device in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, which are instead provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0032] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0033] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0034] In view of the above problems, an embodiment of the present disclosure provides a method for determining a puncture target point.

[0035] Figure 1 1 is a flow chart of a method for determining a puncture target point in an embodiment of the present disclosure. The method can be executed by a device for determining a puncture target point, which can be implemented in software and / or hardware, and can be configured in an electronic device. Figure 1 As shown, the method may specifically include the following steps:

[0036] S110, acquiring CT image data of the patient, and preprocessing the CT image data to obtain a bone structure map;

[0037] Specifically, the CT image data of the patient's chest and abdomen are collected, and the corresponding DICOM format CT image data are analyzed to obtain image data with spatial position information; the detailed method for preprocessing the CT image data is as follows: the corresponding bone (bone) structure in each CT slice image data is marked and outlined along its outer edge on the cross section, and the annotation information in the coronal plane is paid attention to to ensure the accuracy of the bone edge annotation. All the outlined data are reconstructed in three dimensions through the existing neural network model, and each outlined CT image is superimposed and fused according to the spatial coordinate information to form a three-dimensional bone structure map.

[0038] S120, segmenting the bone structure image to obtain segmented bone structure.

[0039] Specifically, the bone structure is segmented using an existing automatic segmentation algorithm to obtain a segmented bone structure.

[0040] S130. Determine the intervertebral space width based on the centroid of the segmented bone structure.

[0041] It can be understood that the center of mass of the segmented bone structure in three-dimensional space is calculated, and the spatial straight line connecting the center of mass of two segmented bone structures is used as the intervertebral space width (the width of the intervertebral space).

[0042] S140. Determine the puncture intervertebral space based on the intervertebral space width.

[0043] Among them, by comparing all intervertebral space widths, an automatic screening algorithm can be used to obtain the intervertebral space most suitable for puncture, that is, the intervertebral space corresponding to the maximum intervertebral space width; or a preset threshold can be determined, and all intervertebral space widths are compared with the preset threshold, and the intervertebral space width exceeding the preset threshold is taken as the preferred intervertebral space suitable for puncture, and then the intervertebral space most suitable for puncture is determined among the preferred suitable intervertebral spaces for puncture.

[0044] It is understandable that the intervertebral space most suitable for puncture is the suitable puncture area recommended by the doctor.

[0045] S150, inputting the ultrasound image data corresponding to the punctured intervertebral space and the anatomical prior knowledge into a trained ultrasound image recognition model, and outputting a label structure and its corresponding label.

[0046] The ultrasound image data corresponding to the punctured intervertebral space can be obtained from clinical data, and the anatomical prior knowledge refers to the constraints and settings of the positional relationship of the corresponding annotated anatomical structures in the ultrasound image, which assists in training, optimizes the final model training effect, and improves the overall recognition accuracy. Figure 2 As shown in the figure, in ultrasound image recognition, the posterior complex (dorsal complex) is always above the anterior complex (ventral complex), and the bone structure (spinous process) is always above the posterior complex in the image. At the same time, the bone structure (spinous process) often appears in a symmetrical structure. Through the prior knowledge of anatomy, the positional relationship of the identification tags is set, so as to customize the design of the ultrasound image recognition model, which greatly improves the accuracy of the ultrasound image recognition model in ultrasound image recognition.

[0047] It should be noted that the intervertebral space is the space or gap between two vertebrae. Before starting epidural puncture, the intervertebral space for puncture is selected based on the surgical level and the patient's spinal condition as shown by CT. The intervertebral space selected for puncture is then subjected to ultrasonic scanning to obtain ultrasonic image data corresponding to the punctured intervertebral space.

[0048] Before obtaining a trained ultrasound image recognition model, it is necessary to build and train the ultrasound image recognition model. Specifically, collect patient sample ultrasound data and sample clinical data, annotate the ultrasound image corresponding to the punctured intervertebral space, and mark the bone structure, anterior complex, posterior complex and other label structures in the ultrasound image. Different labels are used to distinguish them. The specific label classification is shown in Table 1 below:

[0049] ,

[0050] Designated professionals will conduct quality inspections on the annotated ultrasound images, and those that meet the requirements will be placed in the model training library for subsequent ultrasound image model training. Unqualified ultrasound images will be returned to the annotators for re-annotation.

[0051] It can be understood that the ultrasound image recognition model can be an improved UNet neural network, which is constructed by an input layer, an encoder, a bridge layer, a decoder, a structural constraint layer and an output layer, and introduces an attention mechanism (Attention Mechanism Module), a gain-aware module and prior knowledge of anatomical structure to improve the recognition accuracy of ultrasound images.

[0052] In one embodiment, the trained ultrasound image recognition model includes an input layer, an encoder, a bridge layer, a decoder, a structure constraint layer and an output layer; the step of inputting the ultrasound image data and anatomical prior knowledge into the trained ultrasound image recognition model and outputting the target structure and its corresponding label includes:

[0053] Inputting the ultrasound image data and anatomical prior knowledge into the input layer for normalization and position encoding to obtain anatomical ultrasound image features;

[0054] Inputting the anatomical ultrasound image feature into the encoder and performing gain through a gain sensing module in the encoder to obtain a coding gain feature;

[0055] The coding gain feature is input into the bridge layer, and the attention is increased through the attention module in the bridge layer to obtain the attention coding gain feature;

[0056] The attention encoding gain feature is input into the decoder and amplified by the gain perception module in the decoder to obtain the decoding gain feature;

[0057] The decoding gain features and anatomical prior knowledge are input into the transposed convolution layer and the structure constraint layer in the decoder, and the structure constraint decoding gain features are output;

[0058] The structure-constrained decoding gain features are input into the output layer, and the target structure and its corresponding label are output.

[0059] Specifically, before the ultrasound image data and anatomical prior data are input into the improved UNet neural network, the ultrasound image data is normalized to reduce the image differences under different ultrasound gain settings. Before inputting into the input layer of the improved UNet neural network, the anatomical prior knowledge needs to be positionally encoded, specifically the encoding of the anatomical position, to help the improved UNet neural network understand the relative positional relationship of different tissues and structures.

[0060] The annotated ultrasound image data and prior knowledge of anatomy are input into the ultrasound image model to train the ultrasound image model and obtain a trained ultrasound image model. After obtaining the trained ultrasound image model, multiple volunteers are selected and the expert team verifies the actual ultrasound image bone structure and anterior-posterior complex recognition on the volunteers. The expert team judges the accuracy, stability, latency, and ease of use of the ultrasound image model in bone structure and anterior-posterior complex recognition to evaluate whether the ultrasound image model is well trained.

[0061] Exemplarily, the input layer receives ultrasound image data and anatomical prior knowledge at the same time, and the ultrasound image model provides input information, wherein the ultrasound image data and prior knowledge of anatomy

[0062] Input layer definition:

[0063] ,

[0064] The encoder consists of convolutional layers and gain-aware module layers, each of which is followed by a gain-aware module layer to enhance the expressiveness of features. By extracting image features through the convolutional layer, the gain-aware module can adaptively adjust the response of the feature map and enhance the model's ability to capture important features.

[0065] The convolutional layer is represented as:

[0066] ,

[0067] in Indicates The convolutional layer output of layer, Indicates layer convolution kernel, represents the ultrasound image data and anatomical prior knowledge set of the input layer, Indicates The bias term of the layer.

[0068] The gain-aware module layer is expressed as:

[0069] ,

[0070] in, represents the coding gain characteristic, Indicates The convolutional layer output of layer, and represents the gain and offset parameters, represents the activation function, Represents a pooling operation.

[0071] The bridge layer further extracts features through a downsampling, which further reduces the spatial dimension of the data while increasing the expressiveness of the features. An attention layer is introduced to focus on important areas in the image.

[0072] The bridge hierarchy is represented as:

[0073] ,

[0074] in, represents the attention encoding gain feature, represents the coding gain characteristic, represents the strided convolution operation, is the step length, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the probability distribution function.

[0075] The decoder gradually restores the spatial resolution of the image through transposed convolutional layers and structure constraint layers.

[0076] The transposed convolutional layer is expressed as:

[0077] D l = RELU [ ( W l ) T * softmax ( Q l ( K l ) T d l ) V l + b l ] ,

[0078] in, represents the output features of the transposed convolutional layer, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the activation function, represents the probability distribution function, Indicates The bias term of the layer.

[0079] The structural constraint layer is expressed as:

[0080] ,

[0081] in, represents the structural constraint decoding gain feature, represents the output features of the transposed convolutional layer, represents the constraint strength factor, Indicates prior knowledge of anatomy.

[0082] The output layer is used to output the target structure and its corresponding label.

[0083] The output layer is represented as:

[0084] ,

[0085] in, represents the target structure, Indicates The structure of the layer constrains the decoding gain feature, Indicates The weight parameters of the output layer, Indicates upsampling.

[0086] It should be noted that the above content is the construction process of the ultrasound image model. It is also necessary to use the loss function optimization to construct the ultrasound image model to obtain the optimal model parameters of the ultrasound image model in order to obtain a trained ultrasound image model. The construction process of the ultrasound image model is the same as the process of the ultrasound image model, and the difference is the model parameters of the ultrasound image model.

[0087] The difference between the model prediction and the true label is quantified by the loss function. During the training process, the parameters of the model are adjusted by minimizing the loss function. Compared with the traditional mean square error (MSE) or cross entropy loss, the loss function in this invention pays more attention to the overlap between the predicted area and the true area, and is more robust to class imbalance and objects of different sizes.

[0088] Among them, the loss function is expressed as:

[0089] ,

[0090] in, represents the loss function, , represents ultrasound image data, Represents element-wise multiplication.

[0091] The model parameters of the ultrasound imaging model are optimized through the loss function, and the optimal model parameters corresponding to the target structure are substituted into the ultrasound imaging model to obtain a trained ultrasound imaging model. Then, the new ultrasound imaging data and anatomical prior knowledge are input into the trained ultrasound imaging model to predict the target structure and its corresponding label.

[0092] S160: Determine a puncture target point for epidural puncture based on the tag structure and its corresponding tag.

[0093] Based on the above embodiment, the tag structure includes a bone structure, an anterior complex and a posterior complex. For details, please refer to Figure 3 As shown in the figure, the bony structure includes vertebral lamina; the posterior complex is composed of dorsal dura mater, epidural space, and yellow ligament; the anterior complex is composed of ventral dura mater, posterior longitudinal ligament, and posterior edge of vertebral body. The puncture target point is located in the epidural space of the posterior complex. Therefore, the posterior complex can be determined by the label structure, and then the posterior complex can be used to determine the puncture target point, and the puncture can be guided in real time by ultrasound images.

[0094] The method for determining the puncture target point provided in this embodiment reconstructs a three-dimensional image from a two-dimensional image, and uses the automatic measurement function to determine the intervertebral space most suitable for the puncture operation. This improves the efficiency of the doctor's puncture path planning before the operation, thereby improving the safety and success rate of the operation. Then, the trained ultrasound image recognition model is used to identify the label structure, and the label structure in the real-time ultrasound image during the epidural puncture process is identified and prompted, thereby improving the efficiency and safety of epidural puncture.

[0095] In one embodiment, the method further comprises:

[0096] The puncture path is determined according to the predetermined needle entry point and the puncture target point.

[0097] Specifically, the puncture path is manually annotated on the acquired ultrasound image, mainly marking the needle entry point and the puncture target point, and the puncture guide line is a straight line between the needle entry point and the puncture target point.

[0098] Secondly, the ultrasound image data that marks the relative position relationship between the needle entry point, puncture target point and label structure is used as training data. The training data is input into the unsupervised learning machine learning algorithm for algorithm training and learning, and the needle entry point and puncture target point are automatically output. Then the straight line between the needle entry point and the puncture target point is determined as the puncture path.

[0099] This application uses a large amount of labeled data and an unsupervised learning machine learning algorithm to perform deep learning on the simulated puncture path and automatically plan the puncture path. The puncture path can provide a reasonable needle entry point, avoid dangerous and unnecessary structures, accurately identify the target point, and improve the success rate, safety and efficiency of epidural puncture.

[0100] Figure 4 FIG. 1 is a schematic diagram of a device for determining a puncture target point in an embodiment of the present disclosure. Figure 4 As shown: The device includes:

[0101] The first acquisition module 410 is used to acquire CT image data of the patient and pre-process the CT image data to obtain a bone structure map;

[0102] A second acquisition module 420 segments the bone structure image to obtain segmented bone structures;

[0103] A first determination module 430 is used to determine the intervertebral space width based on the centroid of the segmented bone structure;

[0104] A second determination module 440 is used to determine the puncture intervertebral space based on the intervertebral space width;

[0105] An output module 450 is used to input the ultrasound image data corresponding to the punctured intervertebral space and the anatomical prior knowledge into a trained ultrasound image recognition model, and output a label structure and its corresponding label;

[0106] The third determination module 460 is used to determine the puncture target point of the epidural puncture based on the tag structure and its corresponding tag.

[0107] Optionally, the output module 450 is further configured to:

[0108] Inputting the ultrasound image data and anatomical prior knowledge into the input layer for normalization and position encoding to obtain anatomical ultrasound image features;

[0109] Inputting the anatomical ultrasound image feature into the encoder and performing gain through a gain sensing module in the encoder to obtain a coding gain feature;

[0110] The coding gain feature is input into the bridge layer, and the attention is increased through the attention module in the bridge layer to obtain the attention coding gain feature;

[0111] The attention encoding gain feature is input into the decoder and amplified by the gain perception module in the decoder to obtain the decoding gain feature;

[0112] The decoding gain features and anatomical prior knowledge are input into the transposed convolution layer and the structure constraint layer in the decoder, and the structure constraint decoding gain features are output;

[0113] The structure-constrained decoding gain features are input into the output layer, and the target structure and its corresponding label are output.

[0114] Among them, the expression of the gain perception module is:

[0115] ,

[0116] in, represents the coding gain characteristic, and represents the gain and offset parameters, represents the activation function, Represents a pooling operation.

[0117] The bridge hierarchy is expressed as:

[0118] ,

[0119] in, represents the attention encoding gain feature, represents the coding gain characteristic, represents the strided convolution operation, is the step length, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the probability distribution function.

[0120] The expression of the transposed convolutional layer is:

[0121] D l = RELU [ ( W l ) T * softmax ( Q l ( K l ) T d l ) V l + b l ] ,

[0122] in, represents the output features of the transposed convolutional layer, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the activation function, represents the probability distribution function, Indicates The bias term of the layer.

[0123] The expression of the structural constraint layer is:

[0124] ,

[0125] in, represents the structural constraint decoding gain feature, represents the output features of the transposed convolutional layer, represents the constraint strength factor, Indicates prior knowledge of anatomy.

[0126] The expression of the output layer is:

[0127] ,

[0128] in, represents the target structure, Indicates The structure of the layer constrains the decoding gain feature, Indicates The weight parameters of the output layer, Indicates upsampling.

[0129] Optionally, the device further comprises a path module, and the path module is used to determine the puncture path according to a predetermined needle entry point and the target point.

[0130] The device for determining the puncture target point provided in the embodiment of the present disclosure can execute the steps in the method for determining the puncture target point provided in the method embodiment of the present disclosure, and the execution steps and beneficial effects are not repeated here.

[0131] Figure 5 Schematic diagram of the structure of an electronic device in the embodiment of the present disclosure. Figure 5, which shows a structural schematic diagram of an electronic device 500 suitable for implementing the embodiments of the present disclosure. Figure 5 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0132] like Figure 5 As shown, the electronic device 500 may include a processing device 501, a ROM 502, a RAM 503, a bus 504, an input / output (I / O) interface 505, an input device 506, an output device 507, a storage device 508, and a communication device 509. The processing device (e.g., a central processing unit, a graphics processor, etc.) 501 can perform various appropriate actions and processes according to the program stored in the read-only memory (ROM) 502 or the program loaded from the storage device 508 to the random access memory (RAM) 503 to implement the method for determining the puncture target point of the embodiment described in the present disclosure. In the RAM 503, various programs and data required for the operation of the electronic device 500 are also stored. The processing device 501, the ROM 502, and the RAM 503 are connected to each other through the bus 504. The input / output (I / O) interface 505 is also connected to the bus 504.

[0133] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the method shown in the flowchart, thereby implementing the method for determining the puncture target point as described above. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 509, or installed from the storage device 508, or installed from the ROM 502. When the computer program is executed by the processing device 501, the above functions defined in the method of the embodiment of the present disclosure are executed.

[0134] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. Computer readable signal media may also be any computer readable medium other than computer readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0135] The computer-readable medium may be included in the electronic device; or it may exist independently without being assembled into the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains the patient's CT image data, and pre-processes the CT image data to obtain a bone structure map; segments the bone structure map to obtain a segmented bone structure; determines the intervertebral space width based on the centroid of the segmented bone structure; determines the puncture intervertebral space based on the intervertebral space width; inputs the ultrasound image data and anatomical prior knowledge corresponding to the puncture intervertebral space into a trained ultrasound image recognition model, and outputs a label structure and its corresponding label; determines the puncture target point of the epidural puncture based on the label structure and its corresponding label.

[0136] Optionally, when the above one or more programs are executed by the electronic device, the electronic device may also execute other steps described in the above embodiments.

[0137] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0138] The above description is only a preferred embodiment of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of disclosure involved in the present disclosure is not limited to the technical solutions formed by a specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above disclosed concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present disclosure (but not limited to) by each other to form a technical solution.

Claims

1. A method for determining a puncture target point, characterized in that: The method for determining the puncture target point comprises: Acquiring CT image data of a patient, and preprocessing the CT image data to obtain a bone structure map; Segmenting the bone structure image to obtain segmented bone structure; determining the intervertebral disc width based on the centroid of the segmented bony structure; Based on the intervertebral disc width, determining the punctured intervertebral disc space; Inputting the ultrasound image data corresponding to the punctured intervertebral space and the anatomical prior knowledge into the trained ultrasound image recognition model, and outputting the label structure and its corresponding label; Based on the label structure and its corresponding label, the puncture target point of the epidural puncture is determined, The trained ultrasound image recognition model includes an input layer, an encoder, a bridge layer, a decoder, a structure constraint layer and an output layer; the ultrasound image data and anatomical prior knowledge are input into the trained ultrasound image recognition model, and the target structure and its corresponding label are output, including: Inputting the ultrasound image data and anatomical prior knowledge into the input layer for normalization and position encoding to obtain anatomical ultrasound image features; Inputting the anatomical ultrasound image feature into the encoder and performing gain through a gain sensing module in the encoder to obtain a coding gain feature; The coding gain feature is input into the bridge layer, and the attention is increased through the attention module in the bridge layer to obtain the attention coding gain feature; The attention encoding gain feature is input into the decoder and amplified by the gain perception module in the decoder to obtain the decoding gain feature; The decoding gain features and anatomical prior knowledge are input into the transposed convolution layer and the structure constraint layer in the decoder, and the structure constraint decoding gain features are output; The structural constraint decoding gain feature is input into the output layer, and the target structure and its corresponding label are output. Among them, the expression of the gain perception module is: ， in, represents the coding gain characteristic, and represents the gain and offset parameters, represents the activation function, represents the pooling operation, Indicates The convolutional layer output of layer, The bridge hierarchy is expressed as: ， in, represents the attention encoding gain feature, represents the coding gain characteristic, represents the strided convolution operation, is the step length, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the probability distribution function, Among them, the expression of the transposed convolutional layer is: ， in, represents the output features of the transposed convolutional layer, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the activation function, represents the probability distribution function, Indicates The bias term of the layer, Wherein, the expression of the structural constraint layer is: ， in, represents the structural constraint decoding gain feature, represents the output features of the transposed convolutional layer, represents the constraint strength factor, represents prior knowledge of anatomy, Among them, the expression of the output layer is: ， in, represents the target structure, Indicates The structure of the layer constrains the decoding gain feature, Indicates The weight parameters of the output layer, Indicates upsampling.

2. The method for determining the puncture target point according to claim 1, characterized in that: The method for determining the puncture target point also includes: The puncture path is determined according to the predetermined needle entry point and the puncture target point.

3. A device for determining a puncture target point, characterized in that: include: A first acquisition module is used to acquire CT image data of a patient and pre-process the CT image data to obtain a bone structure map; A second acquisition module is used to segment the bone structure image to obtain segmented bone structure; A first determination module, configured to determine the intervertebral disc width based on the centroid of the segmented bone structure; A second determination module is used to determine the puncture intervertebral space based on the intervertebral space width; An output module, used to input the ultrasound image data corresponding to the punctured intervertebral space and anatomical prior knowledge into a trained ultrasound image recognition model, and output a label structure and its corresponding label; The third determination module is used to determine the puncture target point of the epidural puncture based on the label structure and its corresponding label, The trained ultrasound image recognition model includes an input layer, an encoder, a bridge layer, a decoder, a structure constraint layer and an output layer; the ultrasound image data and anatomical prior knowledge are input into the trained ultrasound image recognition model, and the target structure and its corresponding label are output, including: Inputting the ultrasound image data and anatomical prior knowledge into the input layer for normalization and position encoding to obtain anatomical ultrasound image features; Inputting the anatomical ultrasound image feature into the encoder and performing gain through a gain sensing module in the encoder to obtain a coding gain feature; The coding gain feature is input into the bridge layer, and the attention is increased through the attention module in the bridge layer to obtain the attention coding gain feature; The attention encoding gain feature is input into the decoder and amplified by the gain perception module in the decoder to obtain the decoding gain feature; The decoding gain features and anatomical prior knowledge are input into the transposed convolution layer and the structure constraint layer in the decoder, and the structure constraint decoding gain features are output; The structural constraint decoding gain feature is input into the output layer, and the target structure and its corresponding label are output. Among them, the expression of the gain perception module is: ， in, represents the coding gain characteristic, and represents the gain and offset parameters, represents the activation function, represents the pooling operation, Indicates The convolutional layer output of layer, The bridge hierarchy is expressed as: ， in, represents the attention encoding gain feature, represents the coding gain characteristic, represents the strided convolution operation, is the step length, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the probability distribution function, Among them, the expression of the transposed convolutional layer is: ， in, represents the output features of the transposed convolutional layer, Indicates layer convolution kernel, Indicates The query matrix in the layer attention mechanism, Indicates The key matrix in the layer attention mechanism, express The key-value matrix in the layer attention mechanism, T represents the transposed matrix, express The dimension of the keys in the layer attention mechanism, represents the activation function, represents the probability distribution function, Indicates The bias term of the layer, Wherein, the expression of the structural constraint layer is: ， in, represents the structural constraint decoding gain feature, represents the output features of the transposed convolutional layer, represents the constraint strength factor, represents prior knowledge of anatomy, Among them, the expression of the output layer is: ， in, represents the target structure, Indicates The structure of the layer constrains the decoding gain feature, Indicates The weight parameters of the output layer, Indicates upsampling.

4. An electronic device, characterized in that: The electronic device comprises: one or more processors; A storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method for determining a puncture target point according to any one of claims 1-2.

Images