Medical image processing method and device, computer and storage medium

Abstract

The invention provides a medical image processing method and device, a computer and a storage medium, and the method comprises the steps: carrying out the labeling of an image sample through an active learning strategy, and obtaining a labeled image sample; adding noise to the labeled image sample, calculating a judgment value of the labeled image sample to which the noise is added by using a selection formula, and optimizing the labeled image sample based on the judgment value to form a stable sample set; performing feature extraction on the stable sample set by using a multi-scale mixed cavity convolution module to obtain image features; a segmentation model is obtained based on image feature training, a medical image is obtained, the segmentation model is segmented based on the segmentation model, the segmentation model comprises a window-based converter, a feature integration mechanism and a linear set layer, the capture ability of the model to image details is improved, and the image segmentation efficiency is improved. Therefore, better performance is obtained in various visual tasks, and the segmentation precision is further improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image processing, and in particular relates to a medical image processing method, device, computer and storage medium. Background Art

[0002] Medical imaging is a key tool for diagnosis and treatment planning in the medical field, covering images obtained by a variety of imaging technologies. These include computed tomography (CT), positron emission tomography (PET), single photon emission tomography (SPECT), magnetic resonance imaging (MRI), and images generated by other medical imaging equipment. There are differences in the imaging principles and tissue characteristics of medical images. At the same time, they are affected by multiple factors such as noise, field offset effects, local body effects, and tissue movement, making medical images appear blurred and uneven compared to ordinary images.

[0003] The complexity of human anatomy and the differences between individuals further increase the difficulty of medical image processing. Medical image segmentation is the process of dividing an image into non-intersecting "connected" regions based on similarity features, which may include brightness, color, texture, area, shape, position, local statistical features or spectral features. During the segmentation process, pixels within the same region should show consistency or similarity, while there should be obvious differences between different regions, and pixels on the region boundaries should show discontinuity.

[0004] Existing medical image segmentation technologies mostly rely on a single modality, which limits the segmentation effect and accuracy. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a medical image processing method, device, computer and storage medium, which are used to solve the technical problems in the prior art.

[0006] On the one hand, the invention provides the following technical solution, a medical image processing method, the method comprising:

[0007] Annotate image samples using active learning strategies to obtain annotated image samples;

[0008] adding noise to the annotated image samples, calculating the judgment value of the annotated image samples with added noise using a selection formula, and optimizing the annotated image samples based on the judgment value to form a stable sample set;

[0009] Using a multi-scale mixed dilated convolution module to extract features from the stable sample set to obtain image features;

[0010] A segmentation model is obtained based on the image feature training, a medical image is acquired, and the segmentation model is segmented based on the segmentation model, wherein the segmentation model includes a window-based transformer, a feature integration mechanism and a linear aggregation layer.

[0011] Compared with the existing technology, the beneficial effects of the present application are: through active learning strategies and HDRFE modules, efficient and accurate segmentation of medical images is achieved; the segmentation model can effectively extract and integrate multi-scale features from the image, improve the model's ability to capture image details, thereby achieving better performance in various visual tasks, and thus improving segmentation accuracy.

[0012] Furthermore, the active learning strategy includes:

[0013] Randomly select some samples as the initial labeled sample set, and use the initial labeled sample set to train the active learning model to obtain a classifier;

[0014] Using the classifier to perform prediction evaluation on each sample in the unlabeled sample set, and evaluating the uncertainty or information gain of each sample in the unlabeled sample set;

[0015] Determining a sampling strategy, and using a mathematical model of the sampling strategy to calculate the potential value of each sample in the unlabeled sample set based on the uncertainty or the information gain;

[0016] According to the potential value, a plurality of most valuable samples are selected to form a sample set, the formed sample set is manually labeled, and the manually labeled sample set is added to the initial labeled sample set to obtain a labeled sample set;

[0017] Repeating the training of the active learning model according to the labeled sample set to obtain a classifier again;

[0018] Evaluate the performance of the recovered classifier on the validation set, adjust the sampling strategy according to the performance of the recovered classifier, and repeat the steps of using the classifier to perform predictive evaluation on each sample in the unlabeled sample set and evaluate the uncertainty or information gain of each sample in the unlabeled sample set.

[0019] Furthermore, the selection formula includes:

[0020]

[0021] Among them, a is the number of labeled image samples with added noise, The confidence of the optimal class label in the jth labeled image sample with added noise for the i-th labeled image sample, D i is the judgment value.

[0022] Furthermore, the step of optimizing the labeled image samples based on the judgment value to form a stable sample set includes:

[0023] If the judgment value is greater than the threshold, the labeled image sample before the preset parameter is selected to form a stable sample set;

[0024] If the judgment value is less than the threshold, the labeled image sample after the preset parameters is selected to form a stable sample set.

[0025] Furthermore, in the step of optimizing the labeled image samples based on the judgment value to form a stable sample set, a stable formula is introduced to better select the labeled image samples;

[0026] The stability formula includes:

[0027]

[0028] Represents the detection box of the labeled image sample The stability score is the average stability score of all the annotated image samples with added noise. a represents the number of annotated image samples with added noise. Represents the detection box of the labeled image sample The detection box of the nth annotated image sample with added noise The intersection-over-union ratio is an indicator to measure the degree of overlap between two detection boxes. ′ ) represents the prediction box y of the labeled image sample with added noise and the prediction box y of the labeled image sample ′ The Euclidean distance between the center points of the image, x represents the diagonal distance of the minimum enclosing area (i.e., the minimum circumscribed rectangle) containing the noise sample prediction box and the original sample prediction box, and y represents the center point of the prediction box of the annotated image sample with added noise. ′ Represents the center point of the prediction box of the labeled image sample, Representing and labeling image sample detection boxes The nth annotated image sample detection box with the highest DIOU.

[0029] Furthermore, the step of extracting features from the stable sample set using a multi-scale mixed dilated convolution module to obtain image features includes:

[0030] Inputting the stable sample set into a convolutional network for preliminary processing to obtain a feature map;

[0031] Inputting the feature maps into multiple convolution paths respectively to obtain multiple features, and concatenating and integrating the multiple features to obtain feature information;

[0032] The feature information and the stable sample set are then subjected to feature extraction through a convolutional network to obtain image features of contextual information of different scales.

[0033] Furthermore, the segmentation model also includes an encoder, a bottleneck, a decoder and a redesigned multi-size skip connection.

[0034] In a second aspect, the invention provides the following technical solution: a medical image processing device, the device comprising:

[0035] A labeling module is used to label image samples using an active learning strategy to obtain labeled image samples;

[0036] an optimization module, configured to add noise to the annotated image samples, calculate the judgment value of the annotated image samples with added noise using a selection formula, and optimize the annotated image samples based on the judgment value to form a stable sample set;

[0037] An extraction module, used for extracting features from the stable sample set using a multi-scale mixed dilated convolution module to obtain image features;

[0038] A segmentation module is used to obtain a segmentation model based on the image feature training, acquire a medical image, and segment the segmentation model based on the segmentation model, wherein the segmentation model includes a window-based transformer, a feature integration mechanism and a linear aggregation layer.

[0039] In a third aspect, the invention provides the following technical solution: a computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the medical image processing method as described above when executing the computer program.

[0040] In a fourth aspect, the invention provides the following technical solution: a storage medium having a computer program stored thereon, wherein the computer program implements the medical image processing method as described above when executed by a processor. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0042] Figure 1 A flowchart of a medical image processing method provided by a first embodiment of the present invention;

[0043] Figure 2A schematic diagram of the structure of a segmentation model of a medical image processing method provided in the first embodiment of the present invention;

[0044] Figure 3 A structural block diagram of a medical image processing device provided by a second embodiment of the present invention;

[0045] Figure 4 A schematic diagram of the hardware structure of a computer provided in the third embodiment of the present invention.

[0046] The embodiments of the present invention will be further described below with reference to the accompanying drawings. DETAILED DESCRIPTION

[0047] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout are the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the embodiments of the present invention, and should not be construed as limiting the present invention.

[0048] In the description of the embodiments of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside" and "outside" etc., indicating orientations or positional relationships, are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the embodiments of the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.

[0049] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the embodiments of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0050] Embodiment 1

[0051] In the first embodiment of the present invention, see Figure 1 , a medical image processing method, comprising the following steps S01 to S04:

[0052] S01, using active learning strategy to label image samples to obtain labeled image samples;

[0053] The active learning strategy includes:

[0054] S11, randomly selecting some samples as an initial labeled sample set, and using the initial labeled sample set to train an active learning model to obtain a classifier;

[0055] S12, using the classifier to perform prediction evaluation on each sample in the unlabeled sample set, and evaluating the uncertainty or information gain of each sample in the unlabeled sample set;

[0056] S13, determining a sampling strategy, and using a mathematical model of the sampling strategy to calculate the potential value of each sample in the unlabeled sample set based on the uncertainty or the information gain;

[0057] S14, selecting a plurality of most valuable samples according to the potential value to form a sample set, manually annotating the formed sample set, and adding the manually annotated sample set to the initial annotated sample set to obtain an annotated sample set;

[0058] S15, repeatedly training the active learning model according to the labeled sample set to obtain a classifier again;

[0059] S16, evaluate the performance of the recovered classifier on the validation set, adjust the sampling strategy according to the performance of the recovered classifier, and repeat the steps of using the classifier to perform predictive evaluation on each sample in the unlabeled sample set, and evaluate the uncertainty or information gain of each sample in the unlabeled sample set.

[0060] It is worth noting that active learning strategies can effectively guide sample selection and labeling to improve learning efficiency and model performance.

[0061] S02, adding noise to the annotated image sample, calculating the judgment value of the annotated image sample with added noise by using a selection formula, optimizing the annotated image sample based on the judgment value, and forming a stable sample set;

[0062] Specifically, the selection formula includes:

[0063]

[0064] Among them, a is the number of labeled image samples with added noise, The confidence of the optimal class label in the jth labeled image sample with added noise for the i-th labeled image sample, D i is the judgment value.

[0065] Specifically, the step of optimizing the labeled image samples based on the judgment value to form a stable sample set includes:

[0066] If the judgment value is greater than the threshold, the labeled image sample before the preset parameter is selected to form a stable sample set;

[0067] If the judgment value is less than the threshold, the labeled image sample after the preset parameters is selected to form a stable sample set.

[0068] In this embodiment, the threshold is 0.5. i >0.5 sets parameter (preset parameter) b and selects the first b% of noise samples (annotated image samples with added noise); otherwise, selects the last b% of noise samples.

[0069] It is worth mentioning that in the process of active learning, we use carefully designed strategies to optimize the stability of samples. First, when facing those noise samples with high confidence, the core of the strategy is to identify and remove those extreme noise samples. These samples may deviate from the normal distribution of data due to abnormalities or mislabeling. By removing these samples, we can reduce their negative impact on model training, thereby improving the stability of samples.

[0070] Next, when dealing with low-confidence noisy samples, the strategy focuses on removing those samples that may actually perform well despite low confidence. This approach is intended to prevent the model from over-relying on samples that may contain misleading information and avoid introducing instability into the model's decision-making.

[0071] In addition, our strategy has the ability of dynamic balance and can find the balance point among noise samples with different confidence levels. Whether it is enhancing stability by removing extreme samples in high confidence cases or maintaining stability by removing misleading samples in low confidence cases, the strategy can respond flexibly.

[0072] Finally, the adaptability of the active learning strategy is a highlight. It can dynamically adjust according to the characteristics of the data set and the feedback of the model, making the learning process more flexible and able to optimize accordingly for different data distributions and noise levels. Through this optimization method, the active learning strategy can more effectively utilize limited annotation resources and improve the generalization ability and robustness of the model, especially when dealing with noisy data.

[0073] Specifically, in the step of optimizing the labeled image samples based on the judgment value to form a stable sample set, a stable formula is introduced to better select the labeled image samples;

[0074] The stability formula includes:

[0075]

[0076] Represents the detection box of the labeled image sample The stability score is the average stability score of all the annotated image samples with added noise. a represents the number of annotated image samples with added noise. Represents the detection box of the labeled image sample The detection box of the nth annotated image sample with added noise The intersection-over-union ratio is an indicator to measure the degree of overlap between two detection boxes. ′ ) represents the prediction box y of the labeled image sample with added noise and the prediction box y of the labeled image sample ′ The Euclidean distance between the center points of the image, x represents the diagonal distance of the minimum enclosing area (i.e., the minimum circumscribed rectangle) containing the noise sample prediction box and the original sample prediction box, and y represents the center point of the prediction box of the annotated image sample with added noise. ′ Represents the center point of the prediction box of the labeled image sample, Representing and labeling image sample detection boxes The nth annotated image sample detection box with the highest DIOU.

[0077] It is worth noting that the introduction of the DIOU indicator through the stable formula can significantly improve the effect of active learning strategies in processing noisy medical imaging data, allowing the model to learn and generalize more effectively. The DIOU indicator enables the model to more accurately identify those noise samples that are closer to the original samples in space and have a higher degree of overlap, so as to select these samples for annotation, thereby improving the quality of sample selection.

[0078] S03, using a multi-scale mixed dilated convolution module to extract features from the stable sample set to obtain image features;

[0079] Specifically, the step of extracting features from the stable sample set using a multi-scale mixed dilated convolution module to obtain image features includes:

[0080] Inputting the stable sample set into a convolutional network for preliminary processing to obtain a feature map;

[0081] Inputting the feature maps into multiple convolution paths respectively to obtain multiple features, and concatenating and integrating the multiple features to obtain feature information;

[0082] The feature information and the stable sample set are then subjected to feature extraction through a convolutional network to obtain image features of contextual information of different scales.

[0083] In this embodiment, the multi-scale hybrid dilated convolution module is a hybrid dilated residual feature extraction (HDRFE) module.

[0084] The design and workflow of the HDRFE module are as follows:

[0085] Multi-scale mixed dilated convolution:

[0086] The HDRFE module uses multi-scale hybrid dilated convolution technology, which contains three parallel convolution paths to capture features of different scales:

[0087] Path 1: Use standard convolution, 3×3 convolution with dilation rate r=1.

[0088] Path 2: Combines two dilated convolutions with dilation rates of r=1 and r=3 respectively.

[0089] Path 3: Three dilated convolutions are connected in series, with dilation rates of r=1, r=3, and r=5 respectively.

[0090] Feature extraction: The input feature map is first processed through a 1×1 convolutional layer to adjust the number of channels or perform feature transformation.

[0091] Then, the feature maps are extracted through the above three mixed atrous convolution paths respectively.

[0092] Feature Fusion:

[0093] The feature maps extracted from the three paths are concatenated in the channel dimension to integrate feature information of different scales.

[0094] Further feature extraction:

[0095] The concatenated feature map passes through a 1×1 convolutional layer again to further extract and integrate features and obtain contextual information of different scales.

[0096] Residual Connection:

[0097] Finally, the features output by the HDRFE module are added to the original input features through the residual structure. This residual connection helps alleviate the gradient vanishing problem in deep networks while promoting the flow of information, thereby improving the performance of the model.

[0098] Output:

[0099] The feature map processed by the HDRFE module contains rich multi-scale contextual information, providing richer feature representation for subsequent network layers.

[0100] Through this design, the HDRFE module can effectively extract and integrate multi-scale features from images, improve the model's ability to capture image details, and thus achieve better performance in various visual tasks.

[0101] It is worth noting that combining the HDRFE (Hybrid Dilated Residual Feature Extraction) module with the active learning strategy can bring the following potential effects:

[0102] Improve the efficiency of feature extraction:

[0103] The HDRFE module is able to capture features of different scales through multi-scale mixed dilated convolutions, which helps the active learning strategy to more effectively identify and select information-rich samples for annotation.

[0104] Enhance the generalization ability of the model:

[0105] Hybrid atrous convolution can provide richer contextual information, and combined with active learning strategies, it can ensure that the model is trained on limited labeled data while maintaining good generalization capabilities for new data.

[0106] Optimizing the sample selection process:

[0107] The active learning strategy can intelligently select the most representative and informative samples for labeling based on the features provided by the HDRFE module, reducing unnecessary labeling work and improving labeling efficiency.

[0108] Improve model performance:

[0109] The HDRFE module can provide high-quality feature representation, which helps the active learning strategy select more valuable samples in each iteration, thereby improving model performance.

[0110] Reduce the reliance on large amounts of labeled data:

[0111] The active learning strategy combined with the HDRFE module can reduce the need for a large amount of labeled data and achieve rapid convergence by selectively labeling samples that are most helpful for model learning.

[0112] Improve robustness to noise and outliers:

[0113] The HDRFE module can alleviate the gradient vanishing problem through the residual structure and improve the robustness of the model to noise and outliers, which is especially important for the uncertainty and data incompleteness problems that may be encountered in active learning.

[0114] To achieve finer segmentation:

[0115] In tasks such as medical image segmentation, the HDRFE module can provide more refined boundary features, which, combined with active learning strategies, can further improve the segmentation accuracy.

[0116] Through this combination, the HDRFE module and the active learning strategy can complement each other and jointly drive the model to achieve better performance under limited resources.

[0117] S04, obtaining a segmentation model based on the image feature training, acquiring a medical image, and segmenting the segmentation model based on the segmentation model, wherein the segmentation model includes a window-based transformer, a feature integration mechanism, and a linear aggregation layer.

[0118] Specifically, the segmentation model also includes an encoder, a bottleneck, a decoder and a redesigned multi-size skip connection.

[0119] See also Figure 2 , the segmentation model is SWTRU (Symmetric U-type Network with ResidualConnections and U-Net-like structure).

[0120] SWTRU network structure consists of:

[0121] Encoder part:

[0122] Medical images are input into the classic convolutional neural network (CNN).

[0123] The network structure consists of multiple 3x3 convolutional layers, each of which is followed by a ReLU activation function.

[0124] Each convolutional layer is followed by a 2x2 max pooling operation to achieve 2x downsampling.

[0125] In each downsampling step, the number of feature channels is doubled to capture richer feature information.

[0126] Bottleneck part:

[0127] The feature map is evenly divided into non-overlapping small blocks at the bottleneck, and the size of each module is Where R is the initial block size.

[0128] Full-scale skip connections:

[0129] The encoder uses a redesigned full-scale skip connections to perform full-scale feature fusion, which helps preserve more spatial information during decoding.

[0130] Decoder part:

[0131] The decoder contains the S2Win Transformer block (Star Window Transformer block), which is a window-based transformer for processing feature maps.

[0132] The FFIM (Filtering Feature Integration Mechanism) block is used to reduce the number of parameters while integrating shallow refinement information and deep semantic information.

[0133] The linear aggregation layer is the last layer of the decoder and consists of three parts: convolution, patch merging, and linear projection. It is used to integrate feature maps and restore the resolution to the input resolution.

[0134] Output layer:

[0135] The purpose of the output layer is to integrate the feature maps and restore the resolution to the input resolution H×W×Class to output pixel-level segmentation predictions.

[0136] Advantages of SWTRU networks:

[0137] Feature fusion: Through full-scale skip connections, SWTRU can effectively fuse features from the encoder during the decoding process to improve segmentation accuracy.

[0138] Parameter efficiency: The design of the FFIM block reduces the number of parameters of the model while maintaining feature richness.

[0139] Flexibility: The S2Win Transformer block provides fine-grained processing of local features while capturing global contextual information through a window mechanism.

[0140] The SWTRU network structure is designed to combine the precise segmentation capability of U-Net and the global feature learning capability of Transformer to improve the performance of medical image segmentation tasks.

[0141] In summary, a medical image processing method uses an active learning strategy to intelligently annotate image samples, and effectively improves learning efficiency and model performance by randomly selecting initial samples, training classifiers, evaluating uncertainty or information gain, calculating potential value, selecting the most valuable samples for manual annotation, and repeating training and performance evaluation; the step of forming a stable sample set, adding noise to the annotated image samples, and calculating the judgment value by selecting a formula to optimize the samples to form a stable sample set. The stable sample set is feature extracted by a multi-scale mixed dilated convolution module, which captures features of different scales through multi-scale mixed dilated convolution technology and improves model performance through residual connection. A segmentation model is used to achieve high-precision medical image segmentation. It can be seen that the multi-scale mixed dilated convolution module provides fine boundary features, combined with active learning strategies, to improve segmentation accuracy, thereby improving the segmentation accuracy of the segmentation model.

[0142] The present invention proposes a medical image processing method, and designs a SWTRU network structure by combining the advantages of U-Net and Transformer. The network realizes efficient and accurate segmentation of medical images through active learning strategy and HDRFE module. This method is particularly suitable for multimodal medical image segmentation and has important clinical significance. Through this structure, the SWTRU network can effectively extract and integrate multi-scale features from images, improve the model's ability to capture image details, and thus achieve better performance in various visual tasks.

[0143] Embodiment 2

[0144] like Figure 3 As shown, in a second embodiment of the present invention, a medical image processing device is provided, the device comprising:

[0145] A labeling module 10 is used to label image samples using an active learning strategy to obtain labeled image samples;

[0146] An optimization module 20, configured to add noise to the annotated image sample, calculate a judgment value of the annotated image sample with added noise using a selection formula, and optimize the annotated image sample based on the judgment value to form a stable sample set;

[0147] An extraction module 30, configured to extract features from the stable sample set using a multi-scale mixed dilated convolution module to obtain image features;

[0148] The segmentation module 40 is used to obtain a segmentation model based on the image feature training, acquire a medical image, and segment the segmentation model based on the segmentation model, wherein the segmentation model includes a window-based transformer, a feature integration mechanism and a linear aggregation layer.

[0149] The medical image processing device provided in the embodiment of the present invention has the same implementation principle and technical effects as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference may be made to the corresponding contents in the aforementioned method embodiment.

[0150] Embodiment 3

[0151] like Figure 4 As shown, in the third embodiment of the present invention, the embodiment of the present invention provides the following technical solution: a computer, comprising a memory 202, a processor 201, and a computer program stored in the memory 202 and executable on the processor 201, wherein the processor 201 implements the medical image processing method as described above when executing the computer program.

[0152] Specifically, the processor 201 may include a central processing unit (CPU), or an application specific integrated circuit (ASIC), or may be configured to implement one or more integrated circuits of the embodiments of the present application.

[0153] Among them, the memory 202 may include a large capacity memory for data or instructions. By way of example and not limitation, the memory 202 may include a hard disk drive (HDD), a floppy disk drive, a solid state drive (SSD), a flash memory, an optical disk, a magneto-optical disk, a magnetic tape, or a universal serial bus (USB) drive, or a combination of two or more of these. Where appropriate, the memory 202 may include a removable or non-removable (or fixed) medium. Where appropriate, the memory 202 may be inside or outside a data processing device. In a specific embodiment, the memory 202 is a non-volatile memory. In a specific embodiment, the memory 202 includes a read-only memory (ROM) and a random access memory (RAM). Where appropriate, the ROM may be a mask-programmed ROM, a programmable ROM (Programmable Read-Only Memory, PROM for short), an erasable PROM (ErasableProgrammable Read-Only Memory, EPROM for short), an electrically erasable PROM (Electrically ErasableProgrammable Read-Only Memory, EEPROM for short), an electrically alterable ROM (ElectricallyAlterable Read-Only Memory, EAROM for short) or a flash memory (FLASH) or a combination of two or more of these. Under appropriate circumstances, the RAM can be a static random access memory (SRAM) or a dynamic random access memory (DRAM), wherein the DRAM can be a fast page mode dynamic random access memory (FPMDRAM), an extended data output dynamic random access memory (EDODRAM), a synchronous dynamic random access memory (SDRAM), etc.

[0154] The memory 202 may be used to store or cache various data files that need to be processed and / or used for communication, as well as possible computer program instructions executed by the processor 201 .

[0155] The processor 201 implements the above-mentioned medical image processing method by reading and executing the computer program instructions stored in the memory 202 .

[0156] In some embodiments, the computer may further include a communication interface 203 and a bus 200. Figure 4 As shown, the processor 201, the memory 202, and the communication interface 203 are connected via a bus 200 and communicate with each other.

[0157] The communication interface 203 is used to implement communication between the modules, devices, units and / or equipment in the embodiment of the present application. The communication interface 203 can also implement data communication with other components such as: external devices, image / data acquisition equipment, databases, external storage, and image / data processing workstations.

[0158] The bus 200 includes hardware, software or both, and couples the components of the computer to each other. The bus 200 includes but is not limited to at least one of the following: a data bus, an address bus, a control bus, an expansion bus, and a local bus. By way of example and not limitation, bus 200 may include an Accelerated Graphics Port (AGP) or other graphics bus, an Extended Industry Standard Architecture (EISA) bus, a Front Side Bus (FSB), a Hyper Transport (HT) interconnect, an Industry Standard Architecture (ISA) bus, an InfiniBand interconnect, a Low Pin Count (LPC) bus, a memory bus, a Micro Channel Architecture (MCA) bus, a Peripheral Component Interconnect (PCI) bus, a PCI-Express (PCI-X) bus, a Serial Advanced Technology Attachment (SATA) bus, a Video Electronics Standards Association Local Bus (VLB) bus, or other suitable buses or a combination of two or more of these. Where appropriate, bus 200 may include one or more buses. Although embodiments of the present application describe and illustrate a particular bus, the present application contemplates any suitable bus or interconnect.

[0159] Embodiment 4

[0160] In a fourth embodiment of the present invention, in combination with the above-mentioned medical image processing method, the embodiment of the present invention provides the following technical solution, a storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the above-mentioned medical image processing method is implemented.

[0161] Those skilled in the art will appreciate that the logic and / or steps described in the flowchart as data or in other ways herein, for example, can be considered as a sequenced data table of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution device, apparatus or device (such as a computer-based device, a device including a processor, or other device that can fetch instructions from an instruction execution device, apparatus or device and execute instructions), or in combination with these instruction execution devices, apparatuses or devices. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution device, apparatus or device, or in combination with these instruction execution devices, apparatuses or devices.

[0162] More specific examples of computer-readable media (non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and editable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0163] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution device. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0164] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0165] The above-mentioned embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the patent of the present application shall be subject to the attached claims.

Claims

1. A medical image processing method, characterized in that: The method comprises: Annotate image samples using active learning strategies to obtain annotated image samples; adding noise to the annotated image samples, calculating the judgment value of the annotated image samples with added noise using a selection formula, and optimizing the annotated image samples based on the judgment value to form a stable sample set; Using a multi-scale mixed dilated convolution module to extract features from the stable sample set to obtain image features; A segmentation model is obtained based on the image feature training, a medical image is acquired, and the segmentation model is segmented based on the segmentation model, wherein the segmentation model includes a window-based transformer, a feature integration mechanism and a linear aggregation layer.

2. The medical image processing method according to claim 1, characterized in that: The active learning strategy includes: Randomly select some samples as the initial labeled sample set, and use the initial labeled sample set to train the active learning model to obtain a classifier; Using the classifier to perform prediction evaluation on each sample in the unlabeled sample set, and evaluating the uncertainty or information gain of each sample in the unlabeled sample set; Determining a sampling strategy, and using a mathematical model of the sampling strategy to calculate the potential value of each sample in the unlabeled sample set based on the uncertainty or the information gain; According to the potential value, a plurality of most valuable samples are selected to form a sample set, the formed sample set is manually labeled, and the manually labeled sample set is added to the initial labeled sample set to obtain a labeled sample set; Repeating the training of the active learning model according to the labeled sample set to obtain a classifier again; Evaluate the performance of the recovered classifier on the validation set, adjust the sampling strategy according to the performance of the recovered classifier, and repeat the steps of using the classifier to perform predictive evaluation on each sample in the unlabeled sample set and evaluate the uncertainty or information gain of each sample in the unlabeled sample set.

3. The medical image processing method according to claim 1, characterized in that: The selection formula includes: Among them, a is the number of labeled image samples with added noise, The confidence of the optimal class label in the jth labeled image sample with added noise for the i-th labeled image sample, D i is the judgment value.

4. The medical image processing method according to claim 1, characterized in that: The step of optimizing the labeled image samples based on the judgment value to form a stable sample set includes: If the judgment value is greater than the threshold, the labeled image sample before the preset parameter is selected to form a stable sample set; If the judgment value is less than the threshold, the labeled image sample after the preset parameters is selected to form a stable sample set.

5. The medical image processing method according to claim 1, characterized in that: In the step of optimizing the labeled image samples based on the judgment value to form a stable sample set, a stable formula is introduced to better select the labeled image samples; The stability formula includes: Represents the detection box of the labeled image sample The stability score is the average stability score of all the annotated image samples with added noise. a represents the number of annotated image samples with added noise. Represents the detection box of the labeled image sample The detection box of the nth annotated image sample with added noise The intersection-over-union ratio is an indicator to measure the degree of overlap between two detection boxes. ′ ) represents the prediction box y of the labeled image sample with added noise and the prediction box y of the labeled image sample ′ The Euclidean distance between the center points of the image, x represents the diagonal distance of the minimum enclosing area (i.e., the minimum circumscribed rectangle) containing the noise sample prediction box and the original sample prediction box, and y represents the center point of the prediction box of the annotated image sample with added noise. ′ Represents the center point of the prediction box of the labeled image sample, Representing and labeling image sample detection boxes The nth annotated image sample detection box with the highest DIOU.

6. The medical image processing method according to claim 1, characterized in that: The step of extracting features from the stable sample set using a multi-scale mixed dilated convolution module to obtain image features includes: Inputting the stable sample set into a convolutional network for preliminary processing to obtain a feature map; Inputting the feature maps into multiple convolution paths respectively to obtain multiple features, and concatenating and integrating the multiple features to obtain feature information; The feature information and the stable sample set are then subjected to feature extraction through a convolutional network to obtain image features of contextual information of different scales.

7. The medical image processing method according to claim 1, characterized in that: The segmentation model also includes an encoder, a bottleneck, a decoder, and a redesigned multi-size skip connections.

8. A medical image processing device, characterized in that: The device comprises: A labeling module is used to label image samples using an active learning strategy to obtain labeled image samples; an optimization module, configured to add noise to the annotated image samples, calculate the judgment value of the annotated image samples with added noise using a selection formula, and optimize the annotated image samples based on the judgment value to form a stable sample set; An extraction module, used for extracting features from the stable sample set using a multi-scale mixed dilated convolution module to obtain image features; A segmentation module is used to obtain a segmentation model based on the image feature training, acquire a medical image, and segment the segmentation model based on the segmentation model, wherein the segmentation model includes a window-based transformer, a feature integration mechanism and a linear aggregation layer.

9. A computer comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the medical image processing method according to any one of claims 1 to 7 is implemented.

10. A storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the medical image processing method according to any one of claims 1 to 7 is implemented.

Images