Medical image recognition method, device, equipment and medium based on deep learning

Abstract

The present invention relates to the field of intelligent decision-making technology, and discloses a medical image recognition method based on deep learning, including: obtaining at least two sequence images of a target user; dividing the at least two sequence images into a main sequence image and an auxiliary sequence image; grouping the at least two sequence images to obtain at least two sets of sequence image sets; inputting the at least two sets of sequence image sets into a network model based on deep learning for recognition processing, and obtaining recognition results corresponding to each set of sequence image sets; summarizing and fusing the recognition results corresponding to each set of sequence image sets to obtain the medical image recognition results of the target user. The present invention also relates to blockchain technology, and the medical image set can be stored in a blockchain node. The present invention also proposes a medical image recognition device, equipment and computer-readable storage medium based on deep learning. The present invention can improve the accuracy of medical image recognition.

Description

Technical Field

[0001] The present invention relates to the field of intelligent decision-making, and in particular to a medical image recognition method, device, electronic device and computer-readable storage medium based on deep learning. Background Art

[0002] With the development of computer technology, medical artificial intelligence technology has gradually developed. Due to the multi-modal and multi-sequence characteristics of medical images, the existing technology usually fuses the features of different medical images, and then uses the model to identify and process the fused images. However, this method has high requirements for the model and needs to have the ability to recognize various specific sequence combinations. In actual applications, it is often impossible to exhaust each specific sequence combination in advance, which leads to a low accuracy rate of medical image recognition through this method. Summary of the invention

[0003] The present invention provides a medical image recognition method, device, electronic device and computer-readable storage medium based on deep learning, aiming to improve the accuracy of medical image recognition.

[0004] To achieve the above object, the present invention provides a medical image recognition method based on deep learning, comprising:

[0005] Acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user;

[0006] Dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images;

[0007] Grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image;

[0008] Inputting at least two groups of the sequence image sets into a deep learning-based network model for recognition processing, respectively, to obtain recognition results corresponding to each group of the sequence image sets;

[0009] The recognition results corresponding to each group of the sequence image sets are aggregated and fused to obtain the medical image recognition results of the target user.

[0010] Optionally, the sequence images are CT images or MRI images, and after dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images, the method further includes:

[0011] Get the preset grayscale transformation function;

[0012] The pixel values ​​of the main sequence images are input into the preset grayscale transformation function, and the main sequence images are reconstructed using the enhanced pixel values ​​output by the grayscale transformation function.

[0013] Optionally, dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the features of the sequence images includes:

[0014] Identify a main sequence image in at least two of the sequence images according to the names of the sequence images;

[0015] The sequence images other than the main sequence images are determined as auxiliary sequence images.

[0016] Optionally, the identifying a main sequence image in at least two of the sequence images according to the names of the sequence images includes:

[0017] Get the number of sequence images whose names match the preset main sequence name;

[0018] If the number of sequence images whose names match the preset main sequence name is greater than a first number, randomly selecting sequence images whose names match the preset main sequence name and are less than the first number as main sequence images;

[0019] If the number of sequence images whose names match the preset main sequence name is less than the first number, the sequence images corresponding to the names matching the preset main sequence name are determined as the main sequence images.

[0020] Optionally, the aggregating and fusing the recognition results corresponding to each group of the sequence image sets includes:

[0021] Identifying the degree of pseudo ghosting of the auxiliary sequence images;

[0022] When the number of the auxiliary sequence images is at least two, determining the fusion weight of the recognition results corresponding to each group of the sequence image sets according to the pseudo ghosting degree of each auxiliary sequence image;

[0023] The recognition results corresponding to each group of the sequence image sets are weightedly calculated according to the fusion weights.

[0024] Optionally, the deep learning based network model is a Densenet model.

[0025] Optionally, the deep learning-based network model is used to identify whether a lesion exists. After obtaining the medical image recognition result of the target user, the method further includes:

[0026] If the medical image recognition result of the user is that a lesion exists, a region of interest in the main sequence image and the auxiliary sequence image is extracted, and type information of the lesion is determined according to the region of interest.

[0027] In order to solve the above problems, the present invention also provides a medical image recognition device based on deep learning, the device comprising:

[0028] An image acquisition module, used to acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user;

[0029] An image type classification module, used for classifying at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images;

[0030] An image grouping module, used for grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image;

[0031] An image recognition module, used to input at least two groups of the sequence image sets into a network model based on deep learning for recognition processing, and obtain recognition results corresponding to each group of the sequence image sets;

[0032] The result fusion module is used to aggregate and fuse the recognition results corresponding to each group of the sequence image sets to obtain the medical image recognition results of the target user.

[0033] In order to solve the above problem, the present invention further provides an electronic device, the electronic device comprising:

[0034] at least one processor; and,

[0035] a memory communicatively connected to the at least one processor; wherein,

[0036] The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the deep learning-based medical image recognition method as described above.

[0037] In order to solve the above problems, the present invention also provides a computer-readable storage medium, including a data storage area and a program storage area, the data storage area stores created data, and the program storage area stores a computer program; wherein, when the computer program is executed by a processor, the medical image recognition method based on deep learning as described above is implemented.

[0038] In the embodiment of the present invention, the at least two acquired sequence images are divided into a main sequence image and an auxiliary sequence image, and the at least two sequence images are grouped, and the at least two groups of the sequence image sets are respectively input into a network model based on deep learning for recognition processing, without the need to fuse the sequence images in advance, the recognition processing can be performed based on different sequence image features, thereby improving the accuracy of model recognition, and at the same time, the fusion processing is performed according to the results of different groups of sequence image sets to obtain the final recognition result, thereby avoiding the problem of inaccurate overall results caused by inaccurate recognition of a single main sequence image and an auxiliary sequence image, thereby improving the recognition accuracy. Therefore, the embodiment of the present invention can achieve the purpose of improving the accuracy of medical image recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 A schematic diagram of a process flow of a medical image recognition method based on deep learning provided by an embodiment of the present invention;

[0040] Figure 2 A schematic diagram of a module of a medical image recognition device based on deep learning provided by an embodiment of the present invention;

[0041] Figure 3 A schematic diagram of the internal structure of an electronic device for implementing a medical image recognition method based on deep learning provided by an embodiment of the present invention;

[0042] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0043] It should be understood that the specific embodiments described herein are only used to explain the present invention, and are not used to limit the present invention.

[0044] The embodiment of the present application provides a medical image recognition method based on deep learning. The execution subject of the medical image recognition method based on deep learning includes but is not limited to at least one of the electronic devices such as a server, a terminal, etc. that can be configured to execute the method provided by the embodiment of the present application. In other words, the medical image recognition method based on deep learning can be executed by software or hardware installed on a terminal device or a server device, and the software can be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.

[0045] Reference Figure 1 FIG. 1 is a flow chart of a medical image recognition method based on deep learning provided by an embodiment of the present invention. In this embodiment, the medical image recognition method based on deep learning includes:

[0046] S110. Acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user.

[0047] In this embodiment, the medical image set is a plurality of medical images of the same user, and the plurality of medical images are all images related to a certain medical examination. For example, the medical image set is a plurality of CT images of a certain part of the target user.

[0048] S120 , dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the features of the sequence images.

[0049] In an optional embodiment, the feature of the sequence image may be the clarity of the sequence image. For example, the clarity of each sequence image is calculated, and the sequence image with the highest clarity is determined as the main sequence image, and the images other than the main sequence image are the auxiliary sequence images.

[0050] In another optional embodiment, the feature of the sequence image may also be the type of the sequence image. For example, if at least two sequence images include a CT image and an MRI image, the CT image is determined to be the main sequence image and the MRI image is determined to be the auxiliary sequence image. The CT image is a tomographic image established with X-rays as the radiation source. It is an analog image reconstructed through digital conversion and has a high density resolution, and can clearly display the appearance and state of each organ. The MRI image is reconstructed by a computer using the signal generated by the resonance of hydrogen nuclei in a magnetic field. There are often multiple imaging parameters in the MRI image, which can provide rich feature information.

[0051] In a preferred embodiment, dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the features of the sequence images comprises:

[0052] Identify a main sequence image in at least two of the sequence images according to the names of the sequence images;

[0053] The sequence images other than the main sequence images are determined as auxiliary sequence images.

[0054] For example, a medical image set includes six sequence images of a target user. It is determined whether the names of the six sequence images are preset main sequence names, and the sequence images with the main sequence names are determined as main sequence images, and the sequence images without the preset main sequence names are determined as auxiliary sequence images.

[0055] Specifically, if the sequence image is an MRI image, it is determined that the main sequence is an image named T2.

[0056] Usually, when a medical imaging device is used for testing, it will collect a series of sequential images. When saving these sequential images, the medical imaging device will name them according to the operation buttons selected by the doctor on the medical imaging device. For example, when taking a bone X-ray of a certain part of a user, multiple images of the same part from different angles, images of related parts, and images under different acquisition modes will be saved.

[0057] Therefore, the method of dividing at least two sequence images into a main sequence image and an auxiliary sequence image by the name of the sequence image can indirectly determine the main sequence image and the auxiliary sequence image according to the doctor's intention, making the division of the main sequence image and the auxiliary sequence image more reasonable and accurate, which is conducive to improving the accuracy of recognition based on the image.

[0058] Further, in an optional embodiment of the present invention, the step of identifying a main sequence image in at least two of the sequence images according to the names of the sequence images includes:

[0059] Get the number of sequence images whose names match the preset main sequence name;

[0060] If the number of sequence images whose names match the preset main sequence name is greater than a first number, randomly selecting sequence images whose names match the preset main sequence name and are less than the first number as main sequence images;

[0061] If the number of sequence images whose names match the preset main sequence name is less than the first number, the sequence images corresponding to the names matching the preset main sequence name are determined as the main sequence images.

[0062] In this embodiment, the first number is one, that is, when there are at least two sequence images whose names match the preset main sequence name, one is randomly selected as the main sequence image; when there is only one sequence image whose name matches the preset main sequence name, this sequence image is selected as the main sequence image.

[0063] Through the operation of this embodiment, a unique main sequence image can be screened, the number of main sequence images can be simplified, the complexity of calculation can be reduced, the operation efficiency can be improved, and the recognition efficiency can be improved.

[0064] In a preferred embodiment, the sequence images are CT images or MRI images, and after dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images, the method further comprises:

[0065] Get the preset grayscale transformation function;

[0066] The pixel values ​​of the main sequence images are input into the preset grayscale transformation function, and the main sequence images are reconstructed using the enhanced pixel values ​​output by the grayscale transformation function.

[0067] In this embodiment, the preset grayscale transformation function is used to perform grayscale transformation on the main sequence image.

[0068] Specifically, the preset grayscale function is:

[0069]

[0070] Among them, b is a constant coefficient, x is the pixel value in the main sequence image, and y is the enhanced pixel value obtained after transformation.

[0071] The reconstructed main sequence image can be obtained by calculating all enhanced pixel values ​​at each point. Since the logarithmic transformation has the characteristic of compressing data, and most CT images and MRI images are grayscale images, which contain black, white, and gray from black to white, the use of logarithmic transformation for grayscale transformation can achieve the expansion and compression of grayscale images.

[0072] Specifically, through this embodiment, the low grayscale values ​​in the main sequence image can be expanded, and the high grayscale values ​​in the main sequence image can be compressed, so that the contrast of the low grayscale area in the main sequence image will be increased, and the details of the originally dark parts of the main sequence image will be enhanced, showing more delicate detail features, which is conducive to improving the accuracy of image recognition.

[0073] S130, grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image.

[0074] In this embodiment, after grouping, each set of sequence images contains at most two sequence images, and each set contains one main sequence image.

[0075] For example, a medical image set contains six sequential images (such as image A, image B, image C, image D, image E, and image F), one of which (such as image A) is the main sequential image, and the other five are auxiliary sequential images. The six sequential images are divided into six groups, each of which contains one main sequential image and a different auxiliary sequential image, and one group contains only one main sequential image, that is, the six groups of sequential image sets are: {(image A), (image A, image B), (image A, image C), (image A, image D), (image A, image E), (image A, image F)}.

[0076] S140, respectively inputting at least two groups of the sequence image sets into a network model based on deep learning for recognition processing, and obtaining recognition results corresponding to each group of the sequence image sets.

[0077] In this embodiment, the network model based on deep learning can be used for image classification and image detection. For example, the network model based on deep learning can determine whether there is a certain organ lesion in each set of input sequence images, and then obtain the recognition result of whether there is a certain organ lesion in each set of sequence images.

[0078] In the specific implementation, the network model based on deep learning is trained, and then at least two groups of sequential image sets are respectively input into the network model based on deep learning for recognition processing, and each group of sequential image sets is recognized and processed by the network model based on deep learning.

[0079] For example, if the six groups of sequential image sets are: {(image A), (image A, image B), (image A, image C), (image A, image D), (image A, image E), (image A, image F)}, then image A is input into the network model based on deep learning to obtain the recognition result of the model for image A, and image A and image B are input into the network model based on deep learning to obtain the recognition result of the model based on image A and image B. Similarly, the recognition result of the model based on each group of sequential image sets is obtained, for a total of 6 recognition results.

[0080] Preferably, the deep learning-based network model is a Densenet model.

[0081] The Densenet model is a convolutional neural network. In the Densenet model, the current layer has a connection with each subsequent layer. For each layer, the feature maps of all previous layers are used as input. Therefore, it improves the flow of information and gradients in the network and is easier to train. Therefore, the recognition processing in this scheme through the Densenet model is efficient and practical.

[0082] In other optional embodiments, the network model based on deep learning can also be a network model such as Resnet-50, Resnext-101, CenterNet, ATSS, 3DCE, etc. In this embodiment, the auxiliary sequence images are used to assist the main sequence images in analysis, and there is no need for the network model to learn the sequences of various image fusions. The model has high selectivity, and the network model based on deep learning has higher accuracy in identifying separate main sequence images and auxiliary sequence images.

[0083] S150, summarizing and fusing the recognition results corresponding to the sequence image sets of each group to obtain the medical image recognition result of the target user.

[0084] In this embodiment, the recognition results corresponding to each group of sequential image sets may be weighted averaged.

[0085] Furthermore, in a preferred embodiment, the step of aggregating and fusing the recognition results corresponding to each group of the sequence image sets comprises:

[0086] Identifying the degree of pseudo ghosting of the auxiliary sequence images;

[0087] When the number of the auxiliary sequence images is at least two, determining the fusion weight of the recognition results corresponding to each group of the sequence image sets according to the pseudo ghosting degree of each auxiliary sequence image;

[0088] The recognition results corresponding to each group of the sequence image sets are weightedly calculated according to the fusion weights.

[0089] In this embodiment, identifying the pseudo ghosting degree of the auxiliary sequence images specifically refers to identifying the degree of pseudo ghosting of the auxiliary sequence images.

[0090] Specifically, the blur degree or overlap degree of the auxiliary sequence image may be calculated. If the blur degree is severe, the pseudo ghosting degree is determined to be the first level. If the blur degree is mild, the pseudo ghosting degree is determined to be the second level. When the pseudo ghosting degree of the auxiliary sequence image is the first level, it is determined that the weight of the recognition result corresponding to the set of sequence images where the auxiliary sequence image is located is less than a preset threshold (such as the preset threshold is 0.5). When the pseudo ghosting degree of the auxiliary sequence image is the first level, it is determined that the weight of the recognition result corresponding to the set of sequence images where the auxiliary sequence image is located is less than the preset threshold.

[0091] Due to reasons related to medical imaging equipment parameters or user factors when acquiring images (such as wearing metal interference objects or natural physiological characteristics of the human body), pseudo ghosting is inevitable. Therefore, in this embodiment, the weights of different groups of sequence image sets are determined according to the degree of pseudo ghosting, which can improve the accuracy of the summary and fusion processing, thereby improving the accuracy of the medical image recognition results of the target user.

[0092] Furthermore, in an embodiment of the present invention, after obtaining the medical image recognition result of the target user, the method further includes:

[0093] The medical image recognition result is sent to the user end in encrypted form.

[0094] In this embodiment, the medical image recognition result generates a medical result report which is sent in encrypted form to the user or relevant personnel such as a health management personnel.

[0095] Furthermore, in an embodiment of the present invention, after obtaining the medical image recognition result of the target user, the method further includes:

[0096] If the medical image recognition result of the user is that a lesion exists, a region of interest in the main sequence image and the auxiliary sequence image is extracted, and type information of the lesion is determined according to the region of interest.

[0097] In this embodiment, the extracted region of interest is the region containing the lesion.

[0098] Specifically, the image segmentation technology can be used to extract the region of interest, and then the features of the region of interest are extracted, and the similarity between the features and the sample features in the preset feature library (different sample features in the preset feature library correspond to different lesion types) is calculated, and the lesion type corresponding to the sample feature with the highest similarity is selected as the type information of the lesion in the region of interest.

[0099] In the embodiment of the present invention, the at least two acquired sequence images are divided into a main sequence image and an auxiliary sequence image, and the at least two sequence images are grouped, and the at least two groups of the sequence image sets are respectively input into a network model based on deep learning for recognition processing, without the need to fuse the sequence images in advance, the recognition processing can be performed based on different sequence image features, thereby improving the accuracy of model recognition, and at the same time, the fusion processing is performed according to the results of different groups of sequence image sets to obtain the final recognition result, thereby avoiding the problem of inaccurate overall results caused by inaccurate recognition of a single main sequence image and an auxiliary sequence image, thereby improving the recognition accuracy. Therefore, the embodiment of the present invention can achieve the purpose of improving the accuracy of medical image recognition.

[0100] like Figure 2 The figure shows a module schematic diagram of a medical image recognition device based on deep learning according to the present invention.

[0101] The medical image recognition device 200 based on deep learning described in the present invention can be installed in an electronic device. According to the functions implemented, the medical image recognition device based on deep learning can include an image acquisition module 201, an image type classification module 202, an image grouping module 203, an image recognition module 204 and a result fusion module 205. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by an electronic device processor and can complete fixed functions, which are stored in the memory of the electronic device.

[0102] In this embodiment, the functions of each module / unit are as follows:

[0103] The image acquisition module 201 is used to acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user.

[0104] In this embodiment, the medical image set is a plurality of medical images of the same user, and the plurality of medical images are all images related to a certain medical examination. For example, the medical image set is a plurality of CT images of a certain part of the target user.

[0105] The image type classification module 202 is used to classify at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the features of the sequence images.

[0106] In an optional embodiment, the feature of the sequence image may be the clarity of the sequence image. For example, the clarity of each sequence image is calculated, and the sequence image with the highest clarity is determined as the main sequence image, and the images other than the main sequence image are the auxiliary sequence images.

[0107] In another optional embodiment, the feature of the sequence image may also be the type of the sequence image. For example, if at least two sequence images include a CT image and an MRI image, the CT image is determined to be the main sequence image and the MRI image is determined to be the auxiliary sequence image. The CT image is a tomographic image established with X-rays as the radiation source. It is an analog image reconstructed through digital conversion and has a high density resolution, and can clearly display the appearance and state of each organ. The MRI image is reconstructed by a computer using the signal generated by the resonance of hydrogen nuclei in a magnetic field. There are often multiple imaging parameters in the MRI image, which can provide rich feature information.

[0108] In a preferred embodiment, the image type classification module 202 includes:

[0109] an identification unit, configured to identify a main sequence image in at least two of the sequence images according to the names of the sequence images;

[0110] The determining unit is used to determine the sequence images other than the main sequence images as auxiliary sequence images.

[0111] For example, a medical image set includes six sequence images of a target user. It is determined whether the names of the six sequence images are preset main sequence names, and the sequence images with the main sequence names are determined as main sequence images, and the sequence images without the preset main sequence names are determined as auxiliary sequence images.

[0112] Specifically, if the sequence image is an MRI image, it is determined that the main sequence is an image named T2.

[0113] Usually, when a medical imaging device is used for testing, it will collect a series of sequential images. When saving these sequential images, the medical imaging device will name them according to the operation buttons selected by the doctor on the medical imaging device. For example, when taking a bone X-ray of a certain part of a user, multiple images of the same part from different angles, images of related parts, and images under different acquisition modes will be saved.

[0114] Therefore, the method of dividing at least two sequence images into a main sequence image and an auxiliary sequence image by the name of the sequence image can indirectly determine the main sequence image and the auxiliary sequence image according to the doctor's intention, making the division of the main sequence image and the auxiliary sequence image more reasonable and accurate, which is conducive to improving the accuracy of recognition based on the image.

[0115] Furthermore, in an optional embodiment of the present invention, the identification unit is specifically used for:

[0116] Get the number of sequence images whose names match the preset main sequence name;

[0117] If the number of sequence images whose names match the preset main sequence name is greater than a first number, randomly selecting sequence images whose names match the preset main sequence name and are less than the first number as main sequence images;

[0118] If the number of sequence images whose names match the preset main sequence name is less than the first number, the sequence images corresponding to the names matching the preset main sequence name are determined as the main sequence images.

[0119] In this embodiment, the first number is one, that is, when there are at least two sequence images whose names match the preset main sequence name, one is randomly selected as the main sequence image; when there is only one sequence image whose name matches the preset main sequence name, this sequence image is selected as the main sequence image.

[0120] Through the operation of this embodiment, a unique main sequence image can be screened, the number of main sequence images can be simplified, the complexity of calculation can be reduced, the operation efficiency can be improved, and the recognition efficiency can be improved.

[0121] In a preferred embodiment, the sequence of images is a CT image or an MRI image, and the device further comprises an image processing module for:

[0122] After dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images, a preset grayscale transformation function is obtained;

[0123] The pixel values ​​of the main sequence images are input into the preset grayscale transformation function, and the main sequence images are reconstructed using the enhanced pixel values ​​output by the grayscale transformation function.

[0124] In this embodiment, the preset grayscale transformation function is used to perform grayscale transformation on the main sequence image.

[0125] Specifically, the preset grayscale function is:

[0126]

[0127] Among them, b is a constant coefficient, x is the pixel value in the main sequence image, and y is the enhanced pixel value obtained after transformation.

[0128] The reconstructed main sequence image can be obtained by calculating all enhanced pixel values ​​at each point. Since the logarithmic transformation has the characteristic of compressing data, and most CT images and MRI images are grayscale images, which contain black, white, and gray from black to white, the use of logarithmic transformation for grayscale transformation can achieve the expansion and compression of grayscale images.

[0129] Specifically, through this embodiment, the low grayscale values ​​in the main sequence image can be expanded, and the high grayscale values ​​in the main sequence image can be compressed, so that the contrast of the low grayscale area in the main sequence image will be increased, and the details of the originally dark parts of the main sequence image will be enhanced, showing more delicate detail features, which is conducive to improving the accuracy of image recognition.

[0130] The image grouping module 203 is used to group at least two of the sequence images to obtain at least two sequence image sets, each of which includes at least one main sequence image and at most one auxiliary sequence image.

[0131] In this embodiment, after grouping, each set of sequence images contains at most two sequence images, and each set contains one main sequence image.

[0132] For example, a medical image set contains six sequential images (such as image A, image B, image C, image D, image E, and image F), one of which (such as image A) is the main sequential image, and the other five are auxiliary sequential images. The six sequential images are divided into six groups, each of which contains one main sequential image and a different auxiliary sequential image, and one group contains only one main sequential image, that is, the six groups of sequential image sets are: {(image A), (image A, image B), (image A, image C), (image A, image D), (image A, image E), (image A, image F)}.

[0133] The image recognition module 204 is used to input at least two groups of the sequence image sets into the network model based on deep learning for recognition processing, so as to obtain the recognition results corresponding to each group of the sequence image sets.

[0134] In this embodiment, the network model based on deep learning can be used for image classification and image detection. For example, the network model based on deep learning can determine whether there is a certain organ lesion in each set of input sequence images, and then obtain the recognition result of whether there is a certain organ lesion in each set of sequence images.

[0135] In the specific implementation, the network model based on deep learning is trained, and then at least two groups of sequential image sets are respectively input into the network model based on deep learning for recognition processing, and each group of sequential image sets is recognized and processed by the network model based on deep learning.

[0136] For example, if the six groups of sequential image sets are: {(image A), (image A, image B), (image A, image C), (image A, image D), (image A, image E), (image A, image F)}, then image A is input into the network model based on deep learning to obtain the recognition result of the model for image A, and image A and image B are input into the network model based on deep learning to obtain the recognition result of the model based on image A and image B. Similarly, the recognition result of the model based on each group of sequential image sets is obtained, for a total of 6 recognition results.

[0137] Preferably, the deep learning-based network model is a Densenet model.

[0138] The Densenet model is a convolutional neural network. In the Densenet model, the current layer has a connection with each subsequent layer. For each layer, the feature maps of all previous layers are used as input. Therefore, it improves the flow of information and gradients in the network and is easier to train. Therefore, the recognition processing in this scheme through the Densenet model is efficient and practical.

[0139] In other optional embodiments, the network model based on deep learning can also be a network model such as Resnet-50, Resnext-101, CenterNet, ATSS, 3DCE, etc. In this embodiment, the auxiliary sequence images are used to assist the main sequence images in analysis, and there is no need for the network model to learn the sequences of various image fusions. The model has high selectivity, and the network model based on deep learning has higher accuracy in identifying separate main sequence images and auxiliary sequence images.

[0140] The result fusion module 205 is used to aggregate and fuse the recognition results corresponding to each group of the sequence image sets to obtain the medical image recognition result of the target user.

[0141] In this embodiment, the recognition results corresponding to each group of sequential image sets may be weighted averaged.

[0142] Furthermore, in a preferred embodiment, the result fusion module 205 is specifically used for:

[0143] Identifying the degree of pseudo ghosting of the auxiliary sequence images;

[0144] When the number of the auxiliary sequence images is at least two, determining the fusion weight of the recognition results corresponding to each group of the sequence image sets according to the pseudo ghosting degree of each auxiliary sequence image;

[0145] The recognition results corresponding to each group of the sequence image sets are weightedly calculated according to the fusion weights to obtain the medical image recognition results of the target user.

[0146] In this embodiment, identifying the pseudo ghosting degree of the auxiliary sequence images specifically refers to identifying the degree of pseudo ghosting of the auxiliary sequence images.

[0147] Specifically, the blur degree or overlap degree of the auxiliary sequence image may be calculated. If the blur degree is severe, the pseudo ghosting degree is determined to be the first level. If the blur degree is mild, the pseudo ghosting degree is determined to be the second level. When the pseudo ghosting degree of the auxiliary sequence image is the first level, it is determined that the weight of the recognition result corresponding to the set of sequence images where the auxiliary sequence image is located is less than a preset threshold (such as the preset threshold is 0.5). When the pseudo ghosting degree of the auxiliary sequence image is the first level, it is determined that the weight of the recognition result corresponding to the set of sequence images where the auxiliary sequence image is located is less than the preset threshold.

[0148] Due to reasons related to medical imaging equipment parameters or user factors when acquiring images (such as wearing metal interference objects or natural physiological characteristics of the human body), pseudo ghosting is inevitable. Therefore, in this embodiment, the weights of different groups of sequence image sets are determined according to the degree of pseudo ghosting, which can improve the accuracy of the summary and fusion processing, thereby improving the accuracy of the medical image recognition results of the target user.

[0149] Furthermore, in an embodiment of the present invention, after obtaining the medical image recognition result of the target user, the medical image recognition result may be sent to the user end in an encrypted form.

[0150] In this embodiment, the medical image recognition result generates a medical result report which is sent in encrypted form to the user or relevant personnel such as a health management personnel.

[0151] Furthermore, in an embodiment of the present invention, the device further includes a type determination module, which is used to:

[0152] If the medical image recognition result of the user is that a lesion exists, a region of interest in the main sequence image and the auxiliary sequence image is extracted, and type information of the lesion is determined according to the region of interest.

[0153] In this embodiment, the extracted region of interest is the region containing the lesion.

[0154] Specifically, the image segmentation technology can be used to extract the region of interest, and then the features of the region of interest are extracted, and the similarity between the features and the sample features in the preset feature library (different sample features in the preset feature library correspond to different lesion types) is calculated, and the lesion type corresponding to the sample feature with the highest similarity is selected as the type information of the lesion in the region of interest.

[0155] In the embodiment of the present invention, the at least two acquired sequence images are divided into a main sequence image and an auxiliary sequence image, and the at least two sequence images are grouped, and the at least two groups of the sequence image sets are respectively input into a network model based on deep learning for recognition processing, without the need to fuse the sequence images in advance, the recognition processing can be performed based on different sequence image features, thereby improving the accuracy of model recognition, and at the same time, the fusion processing is performed according to the results of different groups of sequence image sets to obtain the final recognition result, thereby avoiding the problem of inaccurate overall results caused by inaccurate recognition of a single main sequence image and an auxiliary sequence image, thereby improving the recognition accuracy. Therefore, the embodiment of the present invention can achieve the purpose of improving the accuracy of medical image recognition.

[0156] like Figure 3 The figure is a schematic diagram of the structure of an electronic device for implementing a medical image recognition method based on deep learning according to the present invention.

[0157] The electronic device 1 may include a processor 10, a memory 11 and a bus, and may also include a computer program stored in the memory 11 and executable on the processor 10, such as a medical image recognition program 12 based on deep learning.

[0158] Among them, the memory 11 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, disk, optical disk, etc. The memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1. The memory 11 may also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 1. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can not only be used to store application software and various types of data installed in the electronic device 1, such as a code of a medical image recognition program 12 based on deep learning, but also be used to temporarily store data that has been output or is to be output.

[0159] The processor 10 may be composed of an integrated circuit in some embodiments, for example, a single packaged integrated circuit, or a plurality of packaged integrated circuits with the same or different functions, including one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and a combination of various control chips, etc. The processor 10 is the control core (Control Unit) of the electronic device, and uses various interfaces and lines to connect various components of the entire electronic device, and executes or executes programs or modules stored in the memory 11 (for example, executing a medical image recognition program based on deep learning, etc.), and calls data stored in the memory 11 to execute various functions of the electronic device 1 and process data.

[0160] The bus may be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. The bus is configured to realize connection and communication between the memory 11 and at least one processor 10, etc.

[0161] Figure 3 Only an electronic device with components is shown, and those skilled in the art will understand that Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure, or combine certain components, or arrange the components differently.

[0162] For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to each component. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, so that the power management device can realize functions such as charging management, discharging management, and power consumption management. The power source may also include any components such as one or more DC or AC power sources, recharging devices, power failure detection circuits, power converters or inverters, power status indicators, etc. The electronic device 1 may also include a variety of sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.

[0163] Furthermore, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a WI-FI interface, a Bluetooth interface, etc.), which is generally used to establish a communication connection between the electronic device 1 and other electronic devices.

[0164] Optionally, the electronic device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), or a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, and an OLED (Organic Light-Emitting Diode) touch device. The display may also be appropriately referred to as a display screen or a display unit, which is used to display information processed in the electronic device 1 and to display a visual user interface.

[0165] It should be understood that the embodiment is for illustration only and the scope of the patent application is not limited to this structure.

[0166] The medical image recognition program 12 based on deep learning stored in the memory 11 of the electronic device 1 is a combination of multiple computer programs. When running in the processor 10, it can achieve:

[0167] Acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user;

[0168] Dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images;

[0169] Grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image;

[0170] Inputting at least two groups of the sequence image sets into a deep learning-based network model for recognition processing, respectively, to obtain recognition results corresponding to each group of the sequence image sets;

[0171] The recognition results corresponding to each group of the sequence image sets are aggregated and fused to obtain the medical image recognition results of the target user.

[0172] Furthermore, if the module / unit integrated in the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be volatile or non-volatile. For example, the computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a computer memory, and a read-only memory (ROM).

[0173] Furthermore, the computer-usable storage medium may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function, etc.; the data storage area may store data created according to the use of the blockchain node, etc.

[0174] The present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor of an electronic device, the computer program can implement:

[0175] Acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user;

[0176] Dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the characteristics of the sequence images;

[0177] Grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image;

[0178] Inputting at least two groups of the sequence image sets into a deep learning-based network model for recognition processing, respectively, to obtain recognition results corresponding to each group of the sequence image sets;

[0179] The recognition results corresponding to each group of the sequence image sets are aggregated and fused to obtain the medical image recognition results of the target user.

[0180] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0181] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0182] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0183] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0184] Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any associated diagram mark in the claims should not be regarded as limiting the claims involved.

[0185] The blockchain referred to in this invention is a new application mode of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm, etc. Blockchain is essentially a decentralized database, a string of data blocks generated by cryptographic methods. Each data block contains a batch of network transaction information, which is used to verify the validity of its information (anti-counterfeiting) and generate the next block. Blockchain can include the underlying blockchain platform, platform product service layer, and application service layer.

[0186] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0187] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A medical image recognition method based on deep learning, characterized in that: The method comprises: Acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user; Dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the clarity or type of the sequence images, inputting the pixel values ​​of the main sequence images into a preset grayscale transformation function, and reconstructing the main sequence images using the enhanced pixel values ​​output by the grayscale transformation function; Grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image; Inputting at least two groups of the sequence image sets into a deep learning-based network model for recognition processing, respectively, to obtain recognition results corresponding to each group of the sequence image sets; Identify the pseudo-ghosting degree of the auxiliary sequence images. When the number of the auxiliary sequence images is at least two, determine the fusion weight of the recognition results corresponding to each group of the sequence image sets according to the pseudo-ghosting degree of each auxiliary sequence image, and perform weighted calculation on the recognition results corresponding to each group of the sequence image sets according to the fusion weight to obtain the medical image recognition result of the target user.

2. The medical image recognition method based on deep learning according to claim 1, characterized in that: The step of dividing at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the clarity or type of the sequence images comprises: Identify a main sequence image in at least two of the sequence images according to the names of the sequence images; The sequence images other than the main sequence images are determined as auxiliary sequence images.

3. The medical image recognition method based on deep learning as claimed in claim 2, characterized in that: The step of identifying a main sequence image in at least two of the sequence images according to the names of the sequence images comprises: Get the number of sequence images whose names match the preset main sequence name; If the number of sequence images whose names match the preset main sequence name is greater than a first number, randomly selecting sequence images whose names match the preset main sequence name and are less than the first number as main sequence images; If the number of sequence images whose names match the preset main sequence name is less than the first number, the sequence images corresponding to the names matching the preset main sequence name are determined as the main sequence images.

4. The medical image recognition method based on deep learning according to any one of claims 1 to 3, characterized in that: The network model based on deep learning is a Densenet model.

5. The medical image recognition method based on deep learning according to any one of claims 1 to 3, characterized in that: The deep learning-based network model is used to identify whether a lesion exists. After obtaining the medical image recognition result of the target user, the method further includes: If the medical image recognition result of the user is that a lesion exists, a region of interest in the main sequence image and the auxiliary sequence image is extracted, and type information of the lesion is determined according to the region of interest.

6. A medical image recognition device based on deep learning, characterized in that: The device comprises: An image acquisition module, used to acquire a medical image set of a target user collected by a medical imaging device, wherein the medical image set includes at least two sequential images of the target user; An image type classification module, used for classifying at least two of the sequence images into a main sequence image and an auxiliary sequence image according to the clarity or type of the sequence images, inputting the pixel values ​​of the main sequence images into a preset grayscale transformation function, and reconstructing the main sequence images using the enhanced pixel values ​​output by the grayscale transformation function; An image grouping module, used for grouping at least two of the sequence images to obtain at least two sequence image sets, each sequence image set including at least one main sequence image and at most one auxiliary sequence image; An image recognition module, used to input at least two groups of the sequence image sets into a network model based on deep learning for recognition processing, and obtain recognition results corresponding to each group of the sequence image sets; The result fusion module is used to identify the pseudo ghosting degree of the auxiliary sequence images. When the number of the auxiliary sequence images is at least two, the fusion weight of the recognition result corresponding to each group of the sequence image sets is determined according to the pseudo ghosting degree of each auxiliary sequence image, and the recognition result corresponding to each group of the sequence image sets is weightedly calculated according to the fusion weight to obtain the medical image recognition result of the target user.

7. An electronic device, characterized in that: The electronic device comprises: at least one processor; and, a memory communicatively connected to the at least one processor; wherein, The memory stores computer program instructions that can be executed by the at least one processor, and the computer program instructions are executed by the at least one processor so that the at least one processor can execute the deep learning-based medical image recognition method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, comprising a data storage area and a program storage area, wherein the data storage area stores created data and the program storage area stores a computer program; characterized in that: When the computer program is executed by a processor, the deep learning-based medical image recognition method as described in any one of claims 1 to 5 is implemented.

Images