Medical image registration method, system and equipment based on semi-supervised learning and medium
Through the medical image registration method based on semi-supervised learning, the registration model is trained using labeled data and non-articulated data, the problem of relying on manual labeling and strong supervised learning in the existing technology is solved, and efficient and accurate automatic registration of multimodal medical images is achieved.
Patent Information
- Application Number
- CN202411974240.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-12-30
- Publication Date
- 2025-04-25
AI Technical Summary
Existing medical imaging registration schemes rely on a large amount of manual annotation and strongly supervised learning, are time-consuming and labor-intensive, and are susceptible to data noise and uncertainty.
Using a medical image registration method based on semi-supervised learning, the medical image registration model is trained using labeled data and non-labeled data to realize automatic registration of multimodal medical images.
Without manual intervention and large amount of labeling data, automatic registration between data of different modal images is realized, and the generalization ability and robustness of the model is improved, making multimodal medical image registration more efficient, accurate and reliable.
Smart Images
Figure CN119887858A_ABST
Abstract
Description
Technical Field
[0001] The embodiments of the present invention relate to the technical field of medical image registration, and in particular to a medical image registration method, system, device and medium based on semi-supervised learning. Background Art
[0002] CT is computer tomography. It uses a series of X-ray images taken from different angles to reconstruct cross-sectional images using a computer. It has high spatial resolution and fast scanning speed, but the radiation dose is high and the contrast for soft tissue is low. MR is magnetic resonance imaging. It uses strong magnetic fields and radio frequency waves to make hydrogen nuclei in the body generate signals, which are reconstructed into images by computers. It can provide excellent soft tissue contrast and obtain multi-directional slice images, but the scanning time is long. PET is positron emission tomography. After the tracer containing radioactive isotopes is injected, the isotopes gather and release positrons in specific parts of the body, which annihilate with negative electrons in the body to produce gamma photons, which are captured by the detector to form images. It is mainly used for functional imaging to reflect the level of metabolic activity, but the spatial resolution is low and the cost is high. In the medical field, multimodal medical imaging (such as CT, MR, PET, etc.) can provide rich information for clinical diagnosis and treatment. However, due to the differences between different modalities, how to effectively integrate this information has become a challenge. Existing medical image registration schemes are still based on traditional manual registration or strongly supervised learning, which requires a lot of manual annotation, is time-consuming and labor-intensive, and is easily affected by data noise and uncertainty.
[0003] The above problems need to be solved urgently. Summary of the invention
[0004] In order to solve the related technical problems, the present invention provides a medical image registration method, system, device and medium based on semi-supervised learning to solve the problems mentioned in the above background technology part.
[0005] To achieve the above objectives, the embodiments of the present invention adopt the following technical solutions:
[0006] In a first aspect, an embodiment of the present invention provides a medical image registration method based on semi-supervised learning, the method comprising:
[0007] Receiving a multimodal medical image data acquisition instruction, and completing the multimodal medical image data acquisition according to the acquisition instruction;
[0008] Build a semi-supervised learning framework to train medical image registration models using labeled and unlabeled data;
[0009] The multimodal medical image data to be registered is input into the medical image registration model, and the multimodal medical image automatic registration is completed through the medical image registration model.
[0010] As an optional implementation manner, the receiving of a multimodal medical image data acquisition instruction and completing the multimodal medical image data acquisition according to the acquisition instruction further includes:
[0011] The multimodal medical image data is preprocessed; wherein the preprocessing includes but is not limited to denoising and standardization.
[0012] As an optional implementation, the construction of a semi-supervised learning framework includes:
[0013] Construct the registration network and evaluation network;
[0014] Generate a multimodal medical image registration result through the registration network;
[0015] The quality of the multimodal medical image registration result is evaluated by the evaluation network.
[0016] As an optional implementation, the evaluating the quality of the multimodal medical image registration result by the evaluation network includes:
[0017] Receiving the multimodal medical image registration result and the original image output by the registration network;
[0018] The multimodal medical image registration result and the original image are evaluated, and an evaluation score is output.
[0019] As an optional implementation, the method of training a medical image registration model using labeled data and unlabeled data includes:
[0020] Prepare some labeled gold standard data;
[0021] Prepare some unlabeled input data;
[0022] training a registration network using the labeled gold standard data;
[0023] Input data is fed into the registration network, which outputs a deformation field;
[0024] Use the deformation field output by the registration network as the input of the evaluation network;
[0025] Use labeled gold standard data to supervise the training of the evaluation network, and use unlabeled input data to assist the training of the evaluation network;
[0026] Update the parameters of the evaluation network to minimize the loss function.
[0027] As an optional implementation, updating the parameters of the evaluation network to minimize the loss function includes:
[0028] The reconstruction error and regularization term are calculated, and the parameters of the evaluation network are updated to minimize the loss function.
[0029] In a second aspect, an embodiment of the present invention provides a medical image registration system based on semi-supervised learning, the system comprising:
[0030] A medical image data acquisition module, used to receive a multimodal medical image data acquisition instruction and complete the multimodal medical image data acquisition according to the acquisition instruction;
[0031] Semi-supervised learning module, which is used to build a semi-supervised learning framework and train the medical image registration model using labeled data and unlabeled data;
[0032] The multimodal medical image automatic registration module is used to input the multimodal medical image data to be registered into the medical image registration model, and complete the multimodal medical image automatic registration through the medical image registration model.
[0033] As an optional implementation, the medical image registration system based on semi-supervised learning further includes:
[0034] The multimodal medical image data preprocessing module is used to preprocess the multimodal medical image data; wherein the preprocessing includes but is not limited to denoising and standardization.
[0035] In a third aspect, an embodiment of the present invention provides an electronic device, comprising a processor and a memory connected to the processor, wherein program data is stored in the memory, and the processor calls the program data stored in the memory to execute the medical image registration method based on semi-supervised learning provided in the embodiment of the first aspect above.
[0036] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium, in which computer execution instructions are stored. When the computer execution instructions are executed by a processor, they are used to implement the medical image registration method based on semi-supervised learning as provided in the embodiment of the first aspect above.
[0037] The technical solution proposed in the embodiment of the present invention solves the shortcomings of the existing medical image registration solutions, receives a multimodal medical image data acquisition instruction, completes the multimodal medical image data acquisition according to the acquisition instruction; constructs a semi-supervised learning framework, and uses labeled data and unlabeled data to train a medical image registration model; inputs the multimodal medical image data to be registered into the medical image registration model, and completes the automatic registration of multimodal medical images through the medical image registration model. The technical solution proposed in the embodiment of the present invention does not require manual intervention and a large amount of labeled data, realizes the automatic registration between different modal image data, and is not limited to a specific modality combination; using a semi-supervised learning framework, unlabeled data can be effectively used, improving the generalization ability and robustness of the model, making the multimodal medical image registration more efficient, accurate, and reliable. BRIEF DESCRIPTION OF THE DRAWINGS
[0038] In order to more clearly illustrate and understand the technical solutions in the embodiments of the present invention, the following briefly introduces the background technology of the present invention and the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the contents of the embodiments of the present invention and these drawings without paying any creative work.
[0039] Figure 1 A flow chart of a medical image registration method based on semi-supervised learning provided by an embodiment of the present invention;
[0040] Figure 2 A structural block diagram of a medical image registration system based on semi-supervised learning provided in an embodiment of the present invention. DETAILED DESCRIPTION
[0041] In order to make the technical problems solved by the present invention, the technical solutions adopted and the technical effects achieved clearer, the technical solutions of the embodiments of the present invention will be further described in detail below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.
[0042] Embodiment 1
[0043] Please refer to Figure 1 As stated, Figure 1 A flowchart of a medical image registration method based on semi-supervised learning provided in an embodiment of the present invention. As shown in the figure, a medical image registration method 100 based on semi-supervised learning in this embodiment includes:
[0044] S101. Receive a multimodal medical image data acquisition instruction, and complete the multimodal medical image data acquisition according to the acquisition instruction;
[0045] S102. Build a semi-supervised learning framework to train a medical image registration model using labeled and unlabeled data;
[0046] S103. Input the multimodal medical image data to be registered into the medical image registration model, and complete the automatic registration of multimodal medical images through the medical image registration model.
[0047] Exemplarily, the receiving of a multimodal medical image data acquisition instruction, and completing the multimodal medical image data acquisition according to the acquisition instruction, further includes:
[0048] The multimodal medical image data is preprocessed; wherein the preprocessing includes but is not limited to denoising and standardization.
[0049] Exemplarily, the construction of a semi-supervised learning framework includes:
[0050] Construct the registration network and evaluation network;
[0051] Generate a multimodal medical image registration result through the registration network;
[0052] The quality of the multimodal medical image registration result is evaluated by the evaluation network.
[0053] Exemplarily, the evaluating the quality of the multimodal medical image registration result by the evaluation network includes:
[0054] Receiving the multimodal medical image registration result and the original image output by the registration network;
[0055] The multimodal medical image registration result and the original image are evaluated, and an evaluation score is output.
[0056] Exemplarily, the method of training a medical image registration model using labeled data and unlabeled data includes:
[0057] Prepare some labeled gold standard data;
[0058] Prepare some unlabeled input data;
[0059] training a registration network using the labeled gold standard data;
[0060] Input data is fed into the registration network, which outputs a deformation field;
[0061] Use the deformation field output by the registration network as the input of the evaluation network;
[0062] Use labeled gold standard data to supervise the training of the evaluation network, and use unlabeled input data to assist the training of the evaluation network;
[0063] Update the parameters of the evaluation network to minimize the loss function.
[0064] Exemplarily, updating the parameters of the evaluation network to minimize the loss function includes:
[0065] The reconstruction error and regularization term are calculated, and the parameters of the evaluation network are updated to minimize the loss function.
[0066] The medical image registration method 100 based on semi-supervised learning proposed in this embodiment receives a multi-modal medical image data acquisition instruction, completes the multi-modal medical image data acquisition according to the acquisition instruction; constructs a semi-supervised learning framework, and uses labeled data and unlabeled data to train a medical image registration model; inputs the multi-modal medical image data to be registered into the medical image registration model, and completes the automatic registration of multi-modal medical images through the medical image registration model. The medical image registration method 100 based on semi-supervised learning proposed in this embodiment does not require manual intervention and a large amount of labeled data, realizes the automatic registration between different modal influence data, and is not limited to a specific modality combination; using a semi-supervised learning framework, unlabeled data can be effectively used, the generalization ability and robustness of the model are improved, and the multi-modal medical influence data registration is more efficient, accurate, and reliable.
[0067] Embodiment 2
[0068] The medical image registration method based on semi-supervised learning in this embodiment specifically includes:
[0069] S201. Receive a multimodal medical image data acquisition instruction, and complete the multimodal medical image data acquisition according to the acquisition instruction;
[0070] S202. Build a semi-supervised learning framework to train a medical image registration model using labeled and unlabeled data;
[0071] S203. Input the multimodal medical image data to be registered into the medical image registration model, and complete the automatic registration of multimodal medical images through the medical image registration model.
[0072] Exemplarily, the receiving of a multimodal medical image data acquisition instruction, and completing the multimodal medical image data acquisition according to the acquisition instruction, further includes:
[0073] The multimodal medical image data is preprocessed; wherein the preprocessing includes but is not limited to denoising and standardization.
[0074] Exemplarily, the construction of a semi-supervised learning framework includes:
[0075] Construct the registration network and evaluation network;
[0076] Generate a multimodal medical image registration result through the registration network;
[0077] The quality of the multimodal medical image registration result is evaluated by the evaluation network.
[0078] In this embodiment, the registration network is a deep learning network designed based on the U-Net structure to achieve automatic registration between different modalities. The registration network accepts multimodal medical image data as input and outputs a registration result.
[0079] Specifically, the registration network adopts a U-Net structure, which includes two main parts: an encoder and a decoder:
[0080] Among them, for the encoder:
[0081] Input: Two multimodal medical imaging data
[0082] Fixed image (fix): shape is (H x W x D x C)
[0083] Move image (move): shape is (H x W x D x C)
[0084] Layer structure:
[0085] Downsampling layer 1: convolution layer + batch normalization layer + ReLU activation function
[0086] Input: Two multimodal medical image data, shape is (H x W x D x C)
[0087] Output: shape is (H / 2x W / 2x D / 2x 32)
[0088] Downsampling layer 2: convolution layer + batch normalization layer + ReLU activation function
[0089] Input: Output of downsampling layer 1, shape (H / 2x W / 2x D / 2x 32)
[0090] Output: shape is (H / 4x W / 4x D / 4x 64)
[0091] Downsampling layer 3: convolution layer + batch normalization layer + ReLU activation function
[0092] Input: Output of downsampling layer 2, shape (H / 4x W / 4x D / 4x 64)
[0093] Output: shape is (H / 8x W / 8x D / 8x 128)
[0094] Downsampling layer 4: convolution layer + batch normalization layer + ReLU activation function
[0095] Input: Output of downsampling layer 3, shape (H / 8x W / 8x D / 8x 128)
[0096] Output: shape is (H / 16x W / 16x D / 16x 256)
[0097] Downsampling layer 5: convolution layer + batch normalization layer + ReLU activation function
[0098] Input: Output of downsampling layer 4, shape (H / 16x W / 16x D / 16x 256)
[0099] Output: shape is (H / 32x W / 32x D / 32x 512)
[0100] For the decoder:
[0101] Input: Output of encoder, shape is (H / 32x W / 32x D / 32x 512)
[0102] Layer structure:
[0103] Upsampling layer 1: transposed convolution layer + batch normalization layer + ReLU activation function
[0104] Input: Output of encoder, shape is (H / 32x W / 32x D / 32x 512)
[0105] Output: shape is (H / 16x W / 16x D / 16x 256)
[0106] Upsampling layer 2: transposed convolution layer + batch normalization layer + ReLU activation function
[0107] Input: Output of upsampling layer 1, shape (H / 16x W / 16x D / 16x 256)
[0108] Output: shape is (H / 8x W / 8x D / 8x 128)
[0109] Upsampling layer 3: transposed convolution layer + batch normalization layer + ReLU activation function
[0110] Input: Output of upsampling layer 2, shape (H / 8x W / 8x D / 8x 128)
[0111] Output: shape is (H / 4x W / 4x D / 4x 64)
[0112] Upsampling layer 4: transposed convolution layer + batch normalization layer + ReLU activation function
[0113] Input: Output of upsampling layer 3, shape (H / 4x W / 4x D / 4x 64)
[0114] Output: shape is (H / 2x W / 2x D / 2x 32)
[0115] Upsampling layer 5: transposed convolution layer + batch normalization layer + ReLU activation function
[0116] Input: Output of upsampling layer 4, shape (H / 2x W / 2x D / 2x 32)
[0117] Output: shape is (H x W x D x 3)
[0118] Output layer: Output the registration result, the shape is (H x W x D x C).
[0119] Exemplarily, the evaluating the quality of the multimodal medical image registration result by the evaluation network includes:
[0120] Receiving the multimodal medical image registration result and the original image output by the registration network;
[0121] The multimodal medical image registration result and the original image are evaluated, and an evaluation score is output.
[0122] In this embodiment, the evaluation network includes:
[0123] Input: deformation field output by the registration network
[0124] Goal: Reconstruct the original input image (including fix image and move image)
[0125] Autoencoder structure
[0126] For the encoder:
[0127] Input layer: receives the deformation field output by the registration network
[0128] Hidden layer 1: convolutional layer, using ReLU activation function
[0129] Hidden layer 2: convolutional layer, using ReLU activation function
[0130] Output layer: fully connected layer, the output dimension is the same as the original input image
[0131] For the decoder:
[0132] Input layer: receives the low-dimensional representation of the encoder output
[0133] Hidden layer 1: Deconvolution layer, using ReLU activation function
[0134] Hidden layer 2: Deconvolution layer, using ReLU activation function
[0135] Output layer: A fully connected layer that outputs the original input image.
[0136] In this embodiment, for the loss function
[0137] Reconstruction Error: Contrastive Loss or Mean Squared Error. Regularization Term: L1 or L2 regularization to prevent overfitting.
[0138] Exemplarily, the method of training a medical image registration model using labeled data and unlabeled data includes:
[0139] The registration network and the evaluation network are trained together using labeled data and unlabeled data, and the quality of the registration results is continuously improved by alternately optimizing the two networks. Specifically,
[0140] Prepare some labeled gold standard data, including fix images, move images and gold standard deformation fields;
[0141] Prepare some unlabeled input data, including fix images and move images;
[0142] training a registration network using the labeled gold standard data;
[0143] Inputting input data into a registration network, and the registration network outputs a deformation field; wherein the input data includes a fix image and a move image;
[0144] Using the deformation field output by the registration network as the input of the evaluation network; in this embodiment, the goal of the evaluation network is to reconstruct the original input image, wherein the original input image includes a fix image and a move image;
[0145] Use labeled gold standard data to supervise the training of the evaluation network, and use unlabeled input data to assist the training of the evaluation network;
[0146] Update the parameters of the evaluation network to minimize the loss function.
[0147] Perform iterative training and repeat the above steps until the model converges.
[0148] In this embodiment, updating the parameters of the evaluation network to minimize the loss function includes:
[0149] The reconstruction error and regularization term are calculated, and the parameters of the evaluation network are updated to minimize the loss function.
[0150] It can be seen that the medical image registration method based on semi-supervised learning proposed in this embodiment can use labeled gold standard data and unlabeled input data to train the evaluation network and the registration network at the same time; the evaluation network can learn to reconstruct the original input image from the deformation field, and the registration network can predict the deformation field. In this embodiment, the performance of the method is evaluated using a test data set, including indicators such as precision and recall.
[0151] The medical image registration method based on semi-supervised learning proposed in this embodiment receives a multimodal medical image data acquisition instruction, completes the multimodal medical image data acquisition according to the acquisition instruction; constructs a semi-supervised learning framework, and uses labeled data and unlabeled data to train a medical image registration model; inputs the multimodal medical image data to be registered into the medical image registration model, and completes the automatic registration of multimodal medical images through the medical image registration model. The medical image registration method based on semi-supervised learning proposed in this embodiment does not require manual intervention and a large amount of labeled data, realizes the automatic registration between different modal influence data, and is not limited to a specific modality combination; using a semi-supervised learning framework, unlabeled data can be effectively used, which improves the generalization ability and robustness of the model, making the multimodal medical influence data registration more efficient, accurate, and reliable.
[0152] Embodiment 3
[0153] like Figure 2 As shown, Figure 2 This is a block diagram of a medical image registration system based on semi-supervised learning provided by an embodiment of the present invention. As shown in the figure, the medical image registration system 200 based on semi-supervised learning in this embodiment adopts the medical image registration method based on semi-supervised learning proposed in any one of Embodiment 1 or Embodiment 2, and the system includes:
[0154] The medical image data acquisition module 201 is used to receive a multi-modal medical image data acquisition instruction and complete the multi-modal medical image data acquisition according to the acquisition instruction;
[0155] A semi-supervised learning module 202, for constructing a semi-supervised learning framework and training a medical image registration model using labeled data and unlabeled data;
[0156] The multimodal medical image automatic registration module 203 is used to input the multimodal medical image data to be registered into the medical image registration model, and complete the multimodal medical image automatic registration through the medical image registration model.
[0157] Exemplarily, the medical image registration system 200 based on semi-supervised learning further includes:
[0158] The multimodal medical image data preprocessing module 204 is used to preprocess the multimodal medical image data; wherein the preprocessing includes but is not limited to denoising and standardization.
[0159] In this embodiment, the construction of a semi-supervised learning framework includes:
[0160] Construct the registration network and evaluation network;
[0161] Generate a multimodal medical image registration result through the registration network;
[0162] The quality of the multimodal medical image registration result is evaluated by the evaluation network.
[0163] In this embodiment, the registration network is a deep learning network designed based on the U-Net structure to achieve automatic registration between different modalities. The registration network accepts multimodal medical image data as input and outputs a registration result.
[0164] Specifically, the registration network adopts a U-Net structure, which includes two main parts: an encoder and a decoder:
[0165] Among them, for the encoder:
[0166] Input: Two multimodal medical imaging data
[0167] Fixed image (fix): shape is (H x W x D x C)
[0168] Move image (move): shape is (H x W x D x C)
[0169] Layer structure:
[0170] Downsampling layer 1: convolution layer + batch normalization layer + ReLU activation function
[0171] Input: Two multimodal medical image data, shape is (H x W x D x C)
[0172] Output: shape is (H / 2x W / 2x D / 2x 32)
[0173] Downsampling layer 2: convolution layer + batch normalization layer + ReLU activation function
[0174] Input: Output of downsampling layer 1, shape (H / 2x W / 2x D / 2x 32)
[0175] Output: shape is (H / 4x W / 4x D / 4x 64)
[0176] Downsampling layer 3: convolution layer + batch normalization layer + ReLU activation function
[0177] Input: Output of downsampling layer 2, shape (H / 4x W / 4x D / 4x 64)
[0178] Output: shape is (H / 8x W / 8x D / 8x 128)
[0179] Downsampling layer 4: convolution layer + batch normalization layer + ReLU activation function
[0180] Input: Output of downsampling layer 3, shape (H / 8x W / 8x D / 8x 128)
[0181] Output: shape is (H / 16x W / 16x D / 16x 256)
[0182] Downsampling layer 5: convolution layer + batch normalization layer + ReLU activation function
[0183] Input: Output of downsampling layer 4, shape (H / 16x W / 16x D / 16x 256)
[0184] Output: shape is (H / 32x W / 32x D / 32x 512)
[0185] For the decoder:
[0186] Input: Output of encoder, shape is (H / 32x W / 32x D / 32x 512)
[0187] Layer structure:
[0188] Upsampling layer 1: transposed convolution layer + batch normalization layer + ReLU activation function
[0189] Input: Output of encoder, shape is (H / 32x W / 32x D / 32x 512)
[0190] Output: shape is (H / 16x W / 16x D / 16x 256)
[0191] Upsampling layer 2: transposed convolution layer + batch normalization layer + ReLU activation function
[0192] Input: Output of upsampling layer 1, shape (H / 16x W / 16x D / 16x 256)
[0193] Output: shape is (H / 8x W / 8x D / 8x 128)
[0194] Upsampling layer 3: transposed convolution layer + batch normalization layer + ReLU activation function
[0195] Input: Output of upsampling layer 2, shape (H / 8x W / 8x D / 8x 128)
[0196] Output: shape is (H / 4x W / 4x D / 4x 64)
[0197] Upsampling layer 4: transposed convolution layer + batch normalization layer + ReLU activation function
[0198] Input: Output of upsampling layer 3, shape (H / 4x W / 4x D / 4x 64)
[0199] Output: shape is (H / 2x W / 2x D / 2x 32)
[0200] Upsampling layer 5: transposed convolution layer + batch normalization layer + ReLU activation function
[0201] Input: Output of upsampling layer 4, shape (H / 2x W / 2x D / 2x 32)
[0202] Output: shape is (H x W x D x 3)
[0203] Output layer: Output the registration result, the shape is (H x W x D x C).
[0204] Exemplarily, the evaluating the quality of the multimodal medical image registration result by the evaluation network includes:
[0205] Receiving the multimodal medical image registration result and the original image output by the registration network;
[0206] The multimodal medical image registration result and the original image are evaluated, and an evaluation score is output.
[0207] In this embodiment, the evaluation network includes:
[0208] Input: deformation field output by the registration network
[0209] Goal: Reconstruct the original input image (including fix image and move image)
[0210] Autoencoder structure
[0211] For the encoder:
[0212] Input layer: receives the deformation field output by the registration network
[0213] Hidden layer 1: convolutional layer, using ReLU activation function
[0214] Hidden layer 2: convolutional layer, using ReLU activation function
[0215] Output layer: fully connected layer, the output dimension is the same as the original input image
[0216] For the decoder:
[0217] Input layer: receives the low-dimensional representation of the encoder output
[0218] Hidden layer 1: Deconvolution layer, using ReLU activation function
[0219] Hidden layer 2: Deconvolution layer, using ReLU activation function
[0220] Output layer: A fully connected layer that outputs the original input image.
[0221] In this embodiment, for the loss function
[0222] Reconstruction Error: Contrastive Loss or Mean Squared Error. Regularization Term: L1 or L2 regularization to prevent overfitting.
[0223] Exemplarily, the semi-supervised learning module 202 uses the labeled data and the unlabeled data to train the medical image registration model, including:
[0224] The registration network and the evaluation network are trained together using labeled data and unlabeled data, and the quality of the registration results is continuously improved by alternately optimizing the two networks. Specifically,
[0225] Prepare some labeled gold standard data, including fix images, move images and gold standard deformation fields;
[0226] Prepare some unlabeled input data, including fix images and move images;
[0227] training a registration network using the labeled gold standard data;
[0228] Inputting input data into a registration network, and the registration network outputs a deformation field; wherein the input data includes a fix image and a move image;
[0229] Using the deformation field output by the registration network as the input of the evaluation network; in this embodiment, the goal of the evaluation network is to reconstruct the original input image, wherein the original input image includes a fix image and a move image;
[0230] Use labeled gold standard data to supervise the training of the evaluation network, and use unlabeled input data to assist the training of the evaluation network;
[0231] Update the parameters of the evaluation network to minimize the loss function.
[0232] Perform iterative training and repeat the above steps until the model converges.
[0233] In this embodiment, updating the parameters of the evaluation network to minimize the loss function includes:
[0234] The reconstruction error and regularization term are calculated, and the parameters of the evaluation network are updated to minimize the loss function.
[0235] It can be seen that the medical image registration system based on semi-supervised learning proposed in this embodiment can use labeled gold standard data and unlabeled input data to train the evaluation network and the registration network at the same time; the evaluation network can learn to reconstruct the original input image from the deformation field, and the registration network can predict the deformation field. In this embodiment, the performance of the method is evaluated using a test data set, including indicators such as precision and recall.
[0236] The medical image registration system based on semi-supervised learning proposed in this embodiment receives a multi-modal medical image data acquisition instruction, completes the multi-modal medical image data acquisition according to the acquisition instruction; constructs a semi-supervised learning framework, and uses labeled data and unlabeled data to train a medical image registration model; inputs the multi-modal medical image data to be registered into the medical image registration model, and completes the automatic registration of multi-modal medical images through the medical image registration model. The medical image registration system based on semi-supervised learning proposed in this embodiment does not require manual intervention and a large amount of labeled data, realizes the automatic registration between different modal influence data, and is not limited to a specific modality combination; using a semi-supervised learning framework, unlabeled data can be effectively used, which improves the generalization ability and robustness of the model, making the multi-modal medical influence data registration more efficient, accurate, and reliable.
[0237] Embodiment 4
[0238] This embodiment provides an electronic device, which includes a processor and a memory connected to the processor, wherein program data is stored in the memory, and the processor calls the program data stored in the memory to execute the medical image registration method based on semi-supervised learning as proposed in any of the above-mentioned embodiments one or two.
[0239] Embodiment 5
[0240] This embodiment provides a computer-readable storage medium, in which computer-executable instructions are stored. When the computer-executable instructions are executed by a processor, they are used to implement the medical image registration method based on semi-supervised learning proposed in any of the above-mentioned embodiments 1 or 2.
[0241] It should be noted that the above-mentioned readable storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The readable storage medium can be any available medium that can be accessed by a general or special-purpose computer.
[0242] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.
Claims
1. A medical image registration method based on semi-supervised learning, characterized in that: include: Receiving a multimodal medical image data acquisition instruction, and completing the multimodal medical image data acquisition according to the acquisition instruction; Build a semi-supervised learning framework to train medical image registration models using labeled and unlabeled data; The multimodal medical image data to be registered is input into the medical image registration model, and the multimodal medical image automatic registration is completed through the medical image registration model.
2. The medical image registration method based on semi-supervised learning according to claim 1, characterized in that: The method further comprises: receiving a multimodal medical image data acquisition instruction, and completing the multimodal medical image data acquisition according to the acquisition instruction, and then further comprising: The multimodal medical image data is preprocessed; wherein the preprocessing includes but is not limited to denoising and standardization.
3. The medical image registration method based on semi-supervised learning according to claim 1, characterized in that: The construction of the semi-supervised learning framework includes: Construct the registration network and evaluation network; Generate a multimodal medical image registration result through the registration network; The quality of the multimodal medical image registration result is evaluated by the evaluation network.
4. The medical image registration method based on semi-supervised learning according to claim 3, characterized in that: The evaluating the quality of the multimodal medical image registration result by the evaluation network includes: Receiving the multimodal medical image registration result and the original image output by the registration network; The multimodal medical image registration result and the original image are evaluated, and an evaluation score is output.
5. The medical image registration method based on semi-supervised learning according to claim 1, characterized in that: The method of training the medical image registration model using labeled data and unlabeled data includes: Prepare some labeled gold standard data; Prepare some unlabeled input data; training a registration network using the labeled gold standard data; Input data is fed into the registration network, which outputs a deformation field; Use the deformation field output by the registration network as the input of the evaluation network; Use labeled gold standard data to supervise the training of the evaluation network, and use unlabeled input data to assist the training of the evaluation network; Update the parameters of the evaluation network to minimize the loss function.
6. The medical image registration method based on semi-supervised learning according to claim 5, characterized in that: The updating of the parameters of the evaluation network to minimize the loss function includes: The reconstruction error and regularization term are calculated, and the parameters of the evaluation network are updated to minimize the loss function.
7. A medical image registration system based on semi-supervised learning, characterized in that: include: A medical image data acquisition module, used to receive a multimodal medical image data acquisition instruction and complete the multimodal medical image data acquisition according to the acquisition instruction; Semi-supervised learning module, which is used to build a semi-supervised learning framework and train the medical image registration model using labeled data and unlabeled data; The multimodal medical image automatic registration module is used to input the multimodal medical image data to be registered into the medical image registration model, and complete the multimodal medical image automatic registration through the medical image registration model.
8. The medical image registration system based on semi-supervised learning according to claim 7, characterized in that: Also includes: The multimodal medical image data preprocessing module is used to preprocess the multimodal medical image data; wherein the preprocessing includes but is not limited to denoising and standardization.
9. An electronic device, characterized in that: The electronic device includes a processor and a memory connected to the processor, wherein the memory stores program data, and the processor calls the program data stored in the memory to execute the medical image registration method based on semi-supervised learning as described in claim 1.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a processor, they are used to implement the medical image registration method based on semi-supervised learning as claimed in claim 1.