A method, system, device and storage medium for extracting medical image features

Abstract

The present invention provides a method, system, device, and storage medium for extracting medical image features. The method includes obtaining a medical original image; inputting the medical original image into a neural network; and splicing the medical original image to participate in convolution in a first feature vector. The first feature vector is the number of channels after convolution when the medical original image is input into the neural network. After splicing the medical original image to participate in convolution in the first feature vector, a corresponding number of slices of the medical original image are spliced to a second feature vector and then participate in convolution. The second feature vector is the number of channels output by each pooling layer after convolution. Wherein the width of the medical original image after slicing is equal to the width of the medical original image obtained after the previous convolution. Based on this method, a feature extraction system, device, and storage medium are also provided. The present invention adds the original image information to each layer of feature map obtained by convolution, maximally retains the original information of the image, and can extract image features more sufficiently.

Description

Technical Field

[0001] The present invention belongs to the technical field of medical image feature extraction, and particularly relates to a medical image feature extraction method, system, device, and storage medium. Background Art

[0002] Imaging features are graphical representations on medical image pictures that have clinical diagnostic significance. These image features are extracted and appropriately selected, and then through software processing, corresponding disease diagnosis references are obtained.

[0003] In the prior art, traditional feature extraction networks such as VGG or RESNET are used for medical image feature extraction. Since AlexNet won the championship with a far higher error rate than the second place in 2012, deep learning has become extremely popular, and deep learning networks have emerged in an endless stream. Compared with traditional medical images, medical image features are single. For deeper feature extraction networks such as VGG, due to the excessive number of convolutional layers, the phenomenon of gradient disappearance and feature disappearance is obvious, and the effect is not ideal. Summary of the Invention

[0004] In order to solve the above technical problems, the present invention proposes a medical image feature extraction method, system, device, and storage medium, which can extract image features more abundantly, thereby achieving better classification, detection, segmentation, and other effects.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A medical image feature extraction method includes the following steps:

[0007] Obtain a medical original image; input the medical original image into a neural network; and splice the medical original image in a first feature vector to participate in convolution; the first feature vector is the number of channels after the medical original image is input into the neural network and undergoes convolution;

[0008] After splicing the medical original image in the first feature vector to participate in convolution, splice the corresponding number of slices of the medical original image to a second feature vector and then participate in convolution; the second feature vector is the number of channels output by each pooling layer after convolution; wherein the width of the medical original image after slicing is equal to the width of the medical original image obtained after the previous convolution.

[0009] Further, the medical original image is a single-channel image.

[0010] Further, after splicing the corresponding number of slices of the medical original image to the second feature vector and then participating in convolution, if the number of output channels is greater than a first threshold, the slices around the medical original image are removed.

[0011] Further, the first threshold is such that the number of channels output after splicing is twice the number of channels output by the pooling layer before splicing.

[0012] Further, discard the slices around the medical original image, and perform windowing processing on the medical original image so that useful medical feature information is located at the center of the image.

[0013] The present invention also provides a medical image feature extraction system, including a first splicing module and a second splicing module;

[0014] The first splicing module is configured to obtain a medical original image; input the medical original image into a neural network; and splice the medical original image in a first feature vector to participate in convolution; the first feature vector is the number of channels after convolution when the medical original image is input into the neural network.

[0015] The second splicing module is configured to, after splicing the medical original image in the first feature vector to participate in convolution, splice a corresponding number of slices of the medical original image to a second feature vector and then participate in convolution; the second feature vector is the number of channels output by each pooling layer after convolution; wherein the width of the medical original image after slicing is equal to the width of the medical original image obtained after the previous convolution.

[0016] Further, in the second splicing module, after splicing a corresponding number of slices of the medical original image to the second feature vector and then participating in convolution, if the number of output channels is greater than the first threshold, discard the slices around the medical original image.

[0017] Further, in the second splicing module, after discarding the slices around the medical original image, perform windowing processing on the medical original image so that useful feature information is located at the center of the image.

[0018] The present invention also provides a device, including:

[0019] A memory for storing a computer program;

[0020] A processor for implementing the method steps when executing the computer program.

[0021] The present invention also provides a readable storage medium, on which a computer program is stored, and the computer program implements the method steps when executed by a processor.

[0022] The effects provided in the summary of the invention are only the effects of the embodiments, rather than all the effects of the invention. One of the above technical solutions has the following advantages or beneficial effects:

[0023] The present invention provides a method, system, device and storage medium for extracting medical image features. The method includes obtaining a medical original image; inputting the medical original image into a neural network; and splicing the medical original image in a first feature vector to participate in convolution. The first feature vector is the number of channels after convolution when the medical original image is input into the neural network. After splicing the medical original image in the first feature vector to participate in convolution, corresponding slices of the medical original image are spliced into a second feature vector and then participate in convolution. The second feature vector is the number of channels output by each pooling layer after convolution. Wherein the width of the sliced medical original image is equal to the width of the medical original image obtained after the previous convolution. Based on a method for extracting medical image features, a system, device and storage medium for extracting medical image features are also provided. In the present invention, the original image information that is completely lossless (without convolution and pooling) is added to each layer of feature map obtained by convolution, and the original information of the image is maximally retained during the convolution process, effectively preventing the loss of the original information of the image during the convolution process, resulting in the loss of the original information of the image as the convolution deepens and finally the effect deteriorates. Compared with VGG and Resnet, the present invention can achieve better detection and segmentation effects. It can extract image features more sufficiently, thereby achieving better classification, detection, segmentation and other effects. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Such as Figure 1 is a flowchart of a method for extracting medical image features according to Embodiment 1 of the present invention;

[0025] Such as Figure 2 is a schematic diagram of the neural network after splicing in Example 1 of Embodiment 1 of the present invention;

[0026] Such as Figure 3 is a schematic diagram of the change of the convolution process channels after splicing with VGG16 in Example 2 of Embodiment 1 of the present invention;

[0027] Such as Figure 4 is a schematic diagram of a system for extracting medical image features according to Embodiment 2 of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0028] To clearly illustrate the technical features of the present solution, the present invention will be described in detail below through specific embodiments and in conjunction with their accompanying drawings. The following disclosure provides many different embodiments or examples for implementing different structures of the present invention. To simplify the disclosure of the present invention, the components and settings of specific examples are described below. In addition, the present invention may repeat reference numerals and / or letters in different examples. This repetition is for the purpose of simplification and clarity, and does not itself indicate the relationship between the various embodiments and / or settings discussed. It should be noted that the components illustrated in the drawings are not necessarily drawn to scale. The present invention omits the description of well-known components and processing techniques and processes to avoid unnecessarily limiting the present invention.

[0029] Example 1

[0030] Example 1 of the present invention proposes a method for extracting medical image features. The original image information that is completely lossless (without convolution and pooling) is added to each layer of the feature map obtained by convolution. The original information of the image is maximally retained during the convolution process, effectively preventing the loss of the original image information during convolution, which may lead to the loss of the original image information as the convolution deepens and the final effect deteriorates. Compared with VGG and Resnet, the present invention can achieve better detection and segmentation effects.

[0031] As Figure 1 is a flowchart of a method for extracting medical image features according to Example 1 of the present invention;

[0032] In step S101, obtain the original medical image; input the original medical image into the neural network; and splice the original medical image into the first feature vector to participate in convolution; the first feature vector is the number of channels after the original medical image is input into the neural network and undergoes convolution;

[0033] In step S102, after splicing the original medical image into the first feature vector to participate in convolution, splice the corresponding number of slices of the original medical image into the second feature vector and then participate in convolution; the second feature vector is the number of channels output by each pooling layer after convolution; wherein the width of the original medical image after slicing is equal to the width of the original medical image obtained after the previous convolution.

[0034] As Figure 2 is a schematic diagram of the neural network after splicing in Example 1 of the present invention;

[0035] The input medical image is a single-channel image, i.e., 224×224×1; the process of the image size and channel change in the entire network is as follows:

[0036] (1) 224×224×1① --> 224×224×(64 + 1)② --> 112×112×(128 + 4)③ --> 56×56×(256 + 16)④ --> 28×28×(512 + 64)⑤ --> 14×14×(512 + 156)⑥ --> The subsequent operations are as Figure 2 in the network.

[0037] In the VGG network, the number of channels is 64. Here, the original image is directly spliced behind the feature layer to participate in convolution.

[0038] In the VGG network, the number of channels is 128. Here, 4 is the four pictures obtained by directly slicing the original image and splicing them behind the first pooling layer to participate in convolution

[0039] In the VGG network, the number of channels is 256. Here, 16 means directly slicing the original image into sixteen pictures and splicing them behind the second pooling layer to participate in convolution.

[0040] In VGG, the number of channels is 512. Here, 64 means directly slicing the original image into sixty - four pictures and splicing them behind the third pooling layer to participate in convolution. According to the above rules, it should originally be 14×14×(512 + 256). The operation here is 156 because medical images have been window - leveled, and the target attributes are all in the center position of the image. When the original image is split into 256 pictures, 25 pictures are removed from each of the top, bottom, left, and right. This is used in medical segmentation to effectively reduce the tendency of segmentation to be more inclined to the large - sample direction due to the imbalance between positive and negative samples.

[0041] In the present invention, after splicing the corresponding number of slices of the medical original image to the second feature vector and participating in convolution, if the number of output channels is greater than the first threshold, the slices around the medical original image are removed. The first threshold is that the number of output channels after splicing is twice the number of output channels of the pooling layer before splicing.

[0042] Remove the slices around the medical original image, and perform window - level processing on the medical original image so that useful feature information is all located in the center position of the image.

[0043] Example 1 of the present invention also gives another example to illustrate the implementation process of the present invention. The protection scope of the present invention is not limited to the content listed in the embodiments or examples.

[0044] Referring to VGG16, for example, there are a total of 16 layers, 13 convolutional layers and 3 fully - connected layers. After two convolutions with 64 convolutional kernels for the first time, one pooling is adopted. After two convolutions with 128 convolutional kernels for the second time, pooling is adopted again. After repeating two times of three convolutions with 512 convolutional kernels, pooling is adopted again. Finally, after three fully - connected layers, the rest of the structure is the same as that of VGG16. Figure 3 It is a schematic diagram of the channel change during the convolution process after splicing using VGG16 in Example 2 of Embodiment 1 of the present invention

[0045] The medical original image is 512×512×1; the input channel = 1, the output channel = 64, the convolution kernel size=(3, 3), the convolution stride=(1, 1), the feature map padding width=(1, 1); the present invention uses the ReLU function.

[0046] The image after the first convolution is 512×512×65; input channels = 65, output channels = 65, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1); The present invention uses the ReLU function. Among them, the extra channel here is the original image, that is, (64 (feature maps) + 1 (medical original image)).

[0047] The image after the second convolution is 256×256×132; MaxPool2d; Conv2d(65, 128, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1)); The present invention uses the ReLU function; Among them, the extra 4 channels here are 4 slices of the original image.

[0048] The image after the third convolution is 256×256×132; Conv2d(132, 132, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1)); The present invention uses the ReLU function, and among them, the extra 4 channels here are 4 slices of the original image.

[0049] The image after the third convolution is 128×128×272; MaxPool2d; Conv2d(132, 256, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1)); The present invention uses the ReLU function, and among them, the extra 16 channels here are 16 slices of the original image.

[0050] The image after the fourth convolution is 128×128×272; Conv2d(272, 272, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1)); The present invention uses the ReLU function, and among them, the extra 16 channels here are 16 slices of the original image.

[0051] The image after the fifth convolution is 64×64×576; MaxPool2d; Conv2d(272, 512, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1))); The present invention uses the ReLU function; Among them, the extra 64 channels here are 64 slices of the original image.

[0052] The image after the sixth convolution is 64×64×576 (MaxPool2d after repeating 2 times); Conv2d(576, 576, convolution kernel size = (3, 3), convolution stride = (1, 1), feature map padding width = (1, 1)); ReLU; The present invention uses the ReLU function.

[0053] After the previous step, it is the same as VGG16; when the number of slices of the original image added after the channel is too large, the sliced parts of the original image on the top, bottom, left, and right can be removed, because medical images have been windowed, and the useful feature information is in the central position of the image. This can not only reduce the amount of calculation, but also effectively reduce the classification result tending to the large sample direction due to the imbalance between positive and negative samples when used for medical image segmentation; for example, when the original image is split into 256 slices, 25 slices are removed from the top and bottom, and 23 slices are removed from the left and right.

[0054] A medical image feature extraction method proposed in Embodiment 1 of the present invention adds the original image information that is completely lossless (without convolution and pooling) to each layer of feature map obtained by convolution. During the convolution process, the original information of the image is maximally retained, effectively preventing the loss of the original information of the image during the convolution process, resulting in the loss of the original information of the image as the convolution deepens and the final effect deteriorates. Compared with VGG and Resnet, the present invention can achieve better detection and segmentation effects. It can extract image features more abundantly, thereby achieving better classification, detection, segmentation and other effects.

[0055] Embodiment 2

[0056] Based on the medical image feature extraction method proposed in Embodiment 1 of the present invention, Embodiment 2 of the present invention also proposes a medical image feature extraction system, as Figure 4 is a schematic diagram of a medical image feature extraction system according to Embodiment 2 of the present invention. The system includes a first splicing module and a second splicing module;

[0057] The first splicing module is used to obtain the medical original image; input the medical original image into the neural network; and splice the medical original image in the first feature vector to participate in convolution; the first feature vector is the number of channels after the medical original image is input into the neural network and undergoes convolution;

[0058] The second splicing module is used to splice the corresponding number of slices of the medical original image to the second feature vector and then participate in convolution after splicing the medical original image in the first feature vector to participate in convolution; the second feature vector is the number of channels output by each pooling layer after convolution; where the width of the sliced medical original image is equal to the width of the medical original image obtained after the previous convolution.

[0059] In the second splicing module, after splicing the corresponding number of slices of the medical original image to the second feature vector and then participating in convolution, if the output number of channels is greater than the first threshold, the slices around the medical original image are removed. The first threshold is that the output number of channels after splicing is twice the number of channels output by the pooling layer before splicing.

[0060] In the second splicing module, the slices around the medical original image are removed, and the medical original image is windowed so that the useful feature information is located at the center of the image.

[0061] Such as Figure 2 is the schematic diagram of the neural network after splicing in Example 1 of the present invention;

[0062] The input medical image is a single-channel image, i.e., 224×224×1; the process of image size and channel change in the whole network is as follows:

[0063] (1) 224×224×1① --> 224×224×(64 + 1)② --> 112×112×(128 + 4)③ --> 56×56×(256 + 16)④ --> 28×28×(512 + 64)⑤ --> 14×14×(512 + 156)⑥ --> The subsequent operations are as Figure 2 in the network

[0064] In the VGG network, the number of channels is 64. Here, the original image is directly spliced behind the feature layer to participate in convolution.

[0065] In the VGG network, the number of channels is 128. Here, 4 means that the original image slices are directly divided into four pictures and spliced behind the first pooling layer to participate in convolution.

[0066] In the VGG network, the number of channels is 256. Here, 16 means that the original image slices are directly divided into sixteen pictures and spliced behind the second pooling layer to participate in convolution.

[0067] In the VGG network, the number of channels is 512. Here, 64 means that the original image slices are directly divided into sixty-four pictures and spliced behind the third pooling layer to participate in convolution. According to the above rules, it should originally be 14×14×(512 + 256). Here, the operation with 156 is because the medical images have all been windowed, and the target attributes are all at the center of the image. When the original image is split into 256 pictures, 25 pictures are removed from each of the top, bottom, left, and right, which can effectively reduce the tendency of segmentation to be more inclined to the large sample direction due to the imbalance between positive and negative samples during medical segmentation.

[0068] A medical image feature extraction system proposed in Example 2 of the present invention adds the original image information that is completely lossless (without convolution and pooling) to each layer of the feature map obtained by convolution. During the convolution process, the original information of the image is maximally retained, effectively preventing the loss of the original information of the image during the convolution process, resulting in the loss of the original information of the image as the convolution deepens and the final effect deteriorates. Compared with VGG and Resnet, the present invention can achieve better detection and segmentation effects. It can extract image features more abundantly, thereby achieving better classification, detection, segmentation and other effects.

[0069] Example 3

[0070] The present invention also provides a device, comprising:

[0071] a memory for storing a computer program;

[0072] a processor for implementing the following method steps when executing the computer program:

[0073] As Figure 1 is a flowchart of a method for extracting medical image features according to Embodiment 1 of the present invention;

[0074] In step S101, obtain a medical original image; input the medical original image into a neural network; and splice the medical original image in a first feature vector to participate in convolution; the first feature vector is the number of channels after convolution of the medical original image input into the neural network;

[0075] In step S102, after splicing the medical original image in the first feature vector to participate in convolution, splice a corresponding number of slices of the medical original image to a second feature vector and then participate in convolution; the second feature vector is the number of channels output by each pooling layer after convolution; wherein the width of the medical original image after slicing is equal to the width of the medical original image obtained after the previous convolution.

[0076] As Figure 2 is a schematic diagram of the neural network after splicing in Example 1 of the present invention;

[0077] The input medical image is a single-channel image, i.e., 224×224×1; the process of the image size and channels changing in the whole network is:

[0078] (1) 224×224×1① --> 224×224×(64 + 1)② --> 112×112×(128 + 4)③ --> 56×56×(256 + 16)④ --> 28×28×(512 + 64)⑤ --> 14×14×(512 + 156)⑥ --> The subsequent operations are as Figure 2 in the network

[0079] In the VGG network, the number of channels is 64, and here the original image is directly spliced behind the feature layer to participate in convolution.

[0080] In the VGG network, the number of channels is 128, and here 4 is the four pictures obtained by directly slicing the original image and splicing them behind the first pooling layer to participate in convolution

[0081] In the VGG network, the number of channels is 256, and here 16 is the sixteen pictures obtained by directly slicing the original image and splicing them behind the second pooling layer to participate in convolution

[0082] In VGG, the number of channels in the network is 512. Here, 64 means that the original image is directly sliced into sixty-four pictures and spliced behind the third pooling layer to participate in convolution. According to the above rules, it should originally be 14×14×(512 + 256). The operation here is 156 because medical images have all undergone windowing processing, and the target attributes are all at the center position of the image. When the original image is split into 256 pictures, 25 pictures are removed from the top, bottom, left, and right respectively, which can effectively reduce the tendency of segmentation to be more inclined to the large sample direction due to the imbalance between positive and negative samples during medical segmentation.

[0083] In the present invention, after splicing the corresponding number of slices of the medical original image to the second feature vector and participating in convolution, if the number of output channels is greater than the first threshold, the slices around the medical original image are removed. The first threshold is that the number of output channels after splicing is twice the number of output channels of the pooling layer before splicing.

[0084] Remove the slices around the medical original image, and perform windowing processing on the medical original image so that useful feature information is all located at the center position of the image.

[0085] A medical image feature extraction device proposed in Embodiment 3 of the present invention adds the original image information that is completely lossless (without convolution and pooling) to each layer of the feature map obtained by convolution. During the convolution process, the original information of the image is retained to the greatest extent, effectively preventing the loss of the original information of the image during the convolution process, resulting in the loss of the original information of the image as the convolution deepens and the final effect deteriorates. Compared with VGG and Resnet, the present invention can achieve better detection and segmentation effects. It can extract image features more abundantly, thereby achieving better classification, detection, segmentation and other effects.

[0086] It should be noted that the technical solution of the present invention also provides an electronic device, including: a communication interface capable of interacting with other devices such as network devices; a processor connected to the communication interface to realize information interaction with other devices, and when running a computer program, it executes a medical image feature extraction method provided by one or more of the above technical solutions, and the computer program is stored on a memory. Of course, in actual application, each component in the electronic device is coupled together through a bus system. It can be understood that the bus system is used to realize the connection and communication between these components. In addition to the data bus, the bus system also includes a power bus, a control bus, and a status signal bus. The memory in the embodiments of the present application is used to store various types of data to support the operation of the electronic device. Examples of these data include: any computer program for operating on the electronic device. It can be understood that the memory can be a volatile memory or a non-volatile memory, or can include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM, Read-Only Memory), a programmable read-only memory (PROM, Programmable Read-Only Memory), an erasable programmable read-only memory (EPROM, Erasable Programmable Read-Only Memory), an electrically erasable programmable read-only memory (EEPROM, Electrically Erasable Programmable Read-Only Memory), a ferromagnetic random access memory (FRAM, ferromagnetic random access memory), a flash memory (Flash Memory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM, Compact Disc Read-Only Memory); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM, Random Access Memory), which is used as an external cache.By way of example and not limitation, many forms of RAM are available, such as static random access memory (SRAM), synchronous static random access memory (SSRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), sync link dynamic random access memory (SLDRAM), direct rambus random access memory (DRRAM). The memories described in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories. The methods disclosed in the embodiments of the present application above can be applied to a processor or implemented by a processor. The processor may be an integrated circuit chip with the ability to process signals. In the implementation process, the steps of the above methods can be completed by the integrated logic circuit of the hardware in the processor or the instructions in the form of software. The above processor may be a general-purpose processor, a DSP (Digital Signal Processing, that is, a chip capable of implementing digital signal processing technology), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The processor can implement or execute the various methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. Combining the steps of the methods disclosed in the embodiments of the present application, it can be directly embodied as being executed and completed by a hardware decoding processor, or by a combination of the hardware and software modules in the decoding processor. The software module may be located in a storage medium, and this storage medium is located in the memory. The processor reads the program in the memory and combines its hardware to complete the steps of the foregoing methods. When the processor executes the program, it implements the corresponding processes in the various methods of the embodiments of the present application. For the sake of brevity, it will not be elaborated here.

[0087] Embodiment 4

[0088] The present invention also provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the method steps are as follows:

[0089] Such as Figure 1 is a flowchart of a method for extracting medical image features in Embodiment 1 of the present invention;

[0090] In step S101, obtain a medical original image; input the medical original image into a neural network; and splice the medical original image in the first feature vector to participate in convolution; the first feature vector is the number of channels after convolution of the medical original image input into the neural network;

[0091] In step S102, after splicing the medical original image in the first feature vector to participate in convolution, splice the corresponding number of slices of the medical original image to the second feature vector and then participate in convolution; the second feature vector is the number of channels output by each pooling layer after convolution; wherein the width of the medical original image after slicing is equal to the width of the medical original image obtained after the previous convolution.

[0092] Such as Figure 2 is a schematic diagram of the neural network after splicing in Example 1 of the present invention;

[0093] The input medical image is a single-channel image, i.e., 224×224×1; the process of the image size and channels changing in the whole network is as follows:

[0094] (1) 224×224×1① --> 224×224×(64 + 1)② --> 112×112×(128 + 4)③ --> 56×56×(256 + 16)④ --> 28×28×(512 + 64)⑤ --> 14×14×(512 + 156)⑥ --> The subsequent operations are as Figure 2 in the network

[0095] In the VGG network, the number of channels is 64. Here, the original image is directly spliced behind the feature layer to participate in convolution.

[0096] In the VGG network, the number of channels is 128. Here, 4 is the four pictures obtained by directly slicing the original image and splicing them behind the first pooling layer to participate in convolution

[0097] In the VGG network, the number of channels is 256. Here, 16 is the sixteen pictures obtained by directly slicing the original image and splicing them behind the second pooling layer to participate in convolution

[0098] In VGG, the number of channels in the network is 512. Here, 64 means that the original image is directly sliced into sixty-four pictures and spliced behind the third pooling layer to participate in convolution. According to the above rules, it should originally be 14×14×(512 + 256). The operation here with a value of 156 is because medical images have been windowed, and the target attributes are all at the center position of the image. When the original image is split into 256 pictures, 25 pictures are removed from each of the top, bottom, left, and right. This is used in medical segmentation to effectively reduce the segmentation bias towards the large sample direction due to the imbalance between positive and negative samples.

[0099] In the present invention, after splicing the corresponding number of slices of the medical original image to the second feature vector and participating in convolution, if the number of output channels is greater than the first threshold, the slices around the medical original image are removed. The first threshold is that the number of output channels after splicing is twice the number of output channels of the pooling layer before splicing.

[0100] Remove the slices around the medical original image, and perform windowing processing on the medical original image so that useful feature information is all located at the center position of the image.

[0101] A storage medium for extracting medical image features proposed in Embodiment 4 of the present invention adds the original image information that is completely lossless (without going through convolution and pooling) to each layer of feature map obtained by convolution. During the convolution process, the original information of the image is maximally retained, effectively preventing the loss of the original information of the image during the convolution process, which may lead to the loss of the original information of the image as the convolution deepens and the final effect deteriorates. Compared with VGG and Resnet, the present invention can achieve better detection and segmentation effects. It can extract image features more abundantly, thereby achieving better classification, detection, segmentation and other effects.

[0102] The embodiment of the present application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, such as a memory storing a computer program. The above computer program can be executed by a processor to complete the steps of the foregoing method. The computer-readable storage medium can be a FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface memory, optical disc, or CD-ROM, etc.

[0103] Those of ordinary skill in the art can understand that all or part of the steps of implementing the above method embodiments can be completed by hardware related to program instructions. The foregoing program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps including those of the above method embodiments. The foregoing storage medium includes various media that can store program codes, such as removable storage devices, ROM, RAM, magnetic disks, or optical discs. Alternatively, if the above integrated units of the present application are implemented in the form of software function modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the embodiments of the present application, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing an electronic device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the various embodiments of the present application. The foregoing storage medium includes various media that can store program codes, such as removable storage devices, ROM, RAM, magnetic disks, or optical discs.

[0104] For the description of the relevant parts in a medical image feature extraction device and a storage medium provided in the embodiments of the present application, reference can be made to the detailed description of the corresponding parts in a medical image feature extraction method provided in Embodiment 1 of the present application, which will not be elaborated here.

[0105] It should be noted that in this text, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device that includes a series of elements includes the inherent elements thereof. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device that includes the element. In addition, the parts of the above technical solutions provided in the embodiments of the present application that are the same as the corresponding technical solutions in the prior art in terms of implementation principles are not described in detail to avoid excessive elaboration.

[0106] Although the specific implementation manners of the present invention are described above in conjunction with the accompanying drawings, it is not a limitation on the protection scope of the present invention. For those skilled in the art, other different forms of modifications or deformations can be made based on the above description. It is not necessary and impossible to list all the implementation manners here. Based on the technical solution of the present invention, various modifications or deformations that can be made by those skilled in the art without creative efforts are still within the protection scope of the present invention.

Claims

1. A method for extracting medical image features, characterized in that, Completely lossless original image information is added to each layer of feature map obtained by convolution, including the following steps: Acquire a medical original image; input the medical original image into a neural network; and splice the medical original image into a first eigenvector to participate in convolution; the first eigenvector is the number of channels of the medical original image after convolution when input into the neural network; the medical original image is a single-channel image; After the original medical image is spliced in the first eigenvector for convolution, a corresponding number of slices of the original medical image are spliced to the second eigenvector for convolution; the second eigenvector is the number of channels output by each pooling layer after convolution; wherein the width of the original medical image after slicing is equal to the width of the original medical image obtained after the last convolution; The process of determining the corresponding number of slices of the original medical image includes: The input medical image is a single-channel image, i.e., 224×224×1; the image size and channel change process in the entire network is: (1)224×224×1①-->224×224×(64+1)②-->112×112×(128+4)③-->56×56×(256+16)④-->28×28×(512+64)⑤-->14×14×(512+156)； The number of channels in the VGG network is 64, so the original image is directly spliced to the feature layer to participate in the convolution. The number of channels in the VGG network is 128, where 4 means directly slicing the original image into four images and splicing them to the first pooling layer for convolution; The number of channels in the VGG network is 256, and the 16 here means directly dividing the original image slices into sixteen pictures and splicing them to the second pooling layer to participate in convolution.

2. The medical image feature extraction method according to claim 1, characterized in that, After a corresponding number of slices of the original medical image are spliced to the second eigenvector and then participate in the convolution, if the number of output channels is greater than a first threshold, the slices around the original medical image are removed.

3. The medical image feature extraction method according to claim 2, wherein, The method for determining the first threshold is: the number of channels output after splicing is twice the number of channels output by the pooling layer when not spliced.

4. A method for extracting medical image features according to claim 3, characterized in that, After the slices around the original medical image are removed, the method further includes: performing window adjustment processing on the original medical image so that all useful medical feature information is located at the center of the image.

5. A medical image feature extraction system for performing the medical image feature extraction method according to any one of claims 1 to 4, characterized in that, It includes a first splicing module and a second splicing module; The first splicing module is used to obtain a medical original image; input the medical original image into a neural network; and splice the medical original image in a first eigenvector to participate in convolution; the first eigenvector is the number of channels of the medical original image after convolution when input into the neural network; the medical original image is a single-channel image; The second splicing module is used to splice the original medical image in the first feature vector to participate in the convolution, and then splice the corresponding number of slices of the original medical image to the second feature vector to participate in the convolution; the second feature vector is the number of channels output by each pooling layer after the convolution; The width of the original medical image after slicing is equal to the width of the original medical image obtained after the last convolution; The process of determining the corresponding number of slices of the original medical image includes: The input medical image is a single-channel image, i.e., 224×224×1; The image size and channel change process in the entire network is: (1)224×224×1①-->224×224×(64+1)②-->112×112×(128+4)③-->56×56×(256+16)④-->28×28×(512+64)⑤-->14×14×(512+156)； The number of channels in the VGG network is 64, so the original image is directly spliced to the feature layer to participate in the convolution. The number of channels in the VGG network is 128, where 4 means directly slicing the original image into four images and splicing them to the first pooling layer for convolution; In the VGG network, the number of channels is 256, and here 16 means directly slicing the original image into sixteen pictures and splicing them behind the second pooling layer to participate in convolution.

6. The medical image feature extraction system according to claim 5, wherein In the second splicing module, after splicing the corresponding number of slices of the medical original image to the second feature vector and participating in convolution, if the number of output channels is greater than the first threshold, the slices around the medical original image are removed.

7. A medical image feature extraction system according to claim 6, characterized in that, In the second splicing module, after removing the slices around the medical original image, it further includes: performing windowing processing on the medical original image so that useful feature information is all located at the center position of the image.

8. A device, characterized in that, It includes: A memory for storing a computer program; A processor for implementing the method steps described in any one of claims 1 to 4 when executing the computer program.

9. A readable storage medium, characterized in that, A computer program is stored on the readable storage medium, and when the computer program is executed by the processor, the method steps described in any one of claims 1 to 4 are implemented.

Images