A deep learning-based tumor image analysis method

Abstract

The application discloses a tumor image analysis method based on deep learning, and relates to the technical field of image processing; a plurality of tumor images respectively corresponding to a plurality of tumor types are acquired, tumor images with tumor region areas greater than a preset tumor region area and tumor images with tumor region aggregation degrees greater than a preset aggregation degree are respectively screened out from the tumor images, important tumor images are determined in combination with tumor types respectively corresponding to the tumor images, and important tumor image analysis results are obtained based on the important tumor images. Therefore, the analysis method not only combines tumor images of multiple types, but also realizes analysis and screening of the tumor images in combination with a deep learning algorithm, so that effective analysis of tumor images of multiple tumor types can be realized, and accurate and effective screening of important tumor images can be realized.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a deep learning-based method for tumor image analysis. Background Technology

[0002] With the development of image processing technology, it has been increasingly applied in medical settings. In some medical contexts, image processing techniques are used to analyze various medical images to obtain corresponding medical image analysis results; these results can assist doctors in making better diagnoses.

[0003] Currently, common methods for analyzing tumor images include segmenting the tumor region using image processing techniques and further determining the tumor type. While this method can analyze tumor images, its approach is relatively limited. Summary of the Invention

[0004] The purpose of this application is to provide a deep learning-based tumor image analysis method for the effective analysis of tumor images of various tumor types, thereby enabling accurate and effective screening of important tumor images.

[0005] To achieve the above objectives, embodiments of this application provide a deep learning-based tumor image analysis method, comprising: acquiring multiple tumor images corresponding to various tumor types, each tumor image including a tumor region corresponding to the tumor type; selecting multiple first tumor images from the multiple tumor images corresponding to various tumor types according to a preset first tumor image screening model, wherein the tumor region area of ​​the multiple first tumor images is greater than a preset tumor region area; selecting multiple second tumor images from the multiple tumor images corresponding to various tumor types according to a preset second tumor image screening model, wherein the tumor region clustering degree of the multiple second tumor images is greater than a preset clustering degree; determining multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images; and generating tumor image analysis results based on the multiple important tumor images.

[0006] In one possible implementation, the deep learning-based tumor image analysis method further includes: acquiring a first training dataset, the first training dataset including multiple sample tumor images corresponding to various tumor types, each sample tumor image including a tumor region corresponding to the tumor type, the multiple sample tumor images including sample tumor images with a first area label and sample tumor images with a second area label, the first area label being used to indicate that the tumor region area is greater than the preset tumor region area, and the second area label being used to indicate that the tumor region area is less than or equal to the preset tumor region area; and training an initial first tumor image screening model based on the first training dataset to obtain the preset first tumor image screening model.

[0007] In one possible implementation, the deep learning-based tumor image analysis method further includes: acquiring a second training dataset, the second training dataset including multiple sample tumor images corresponding to various tumor types, each sample tumor image including a tumor region corresponding to the tumor type, the multiple sample tumor images including sample tumor images with a first clustering degree label and sample tumor images with a second clustering degree label, the first clustering degree label being used to characterize that the tumor region clustering degree is greater than a preset clustering degree, and the second area label being used to characterize that the tumor region clustering degree is less than or equal to the preset clustering degree; and training an initial second tumor image screening model based on the second training dataset to obtain the preset second tumor image screening model.

[0008] In one possible implementation, determining multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images includes: determining a first target tumor image and a second target tumor image based on the multiple first tumor images and the multiple second tumor images, wherein the first target tumor image belongs to both the multiple first tumor images and the multiple second tumor images, and the second target tumor image does not belong to both the multiple first tumor images and belongs to the multiple second tumor images; determining whether the tumor type corresponding to the first target tumor image includes all tumor types among the multiple tumor types; if the tumor type corresponding to the first target tumor image includes all tumor types among the multiple tumor types, then determining the multiple important tumor images based on at least two second target tumor images from the first target tumor image and the second target tumor image.

[0009] In one possible implementation, the deep learning-based tumor image analysis method further includes: if the tumor type corresponding to the first target tumor image does not include all of the multiple tumor types, then, based on the tumor type corresponding to the second target tumor image, determining tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image from the second target tumor image and other tumor images excluding the first target tumor image and the second target tumor image, wherein the selection priority of the second target tumor image is higher than that of the other tumor images excluding the first target tumor image and the second target tumor image; and determining the multiple important tumor images based on the first target tumor image and the tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image.

[0010] In one possible implementation, the deep learning-based tumor image analysis method further includes: selecting multiple third tumor images from multiple tumor images corresponding to the multiple tumor types according to a preset third tumor image screening model, wherein the tumor region diffusion of the multiple third tumor images is greater than a preset diffusion; and determining multiple important tumor images according to the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images, including: determining multiple important tumor images according to the multiple first tumor images, the multiple second tumor images, the multiple third tumor images, the tumor types corresponding to the multiple first tumor images, the tumor types corresponding to the multiple second tumor images, and the tumor types corresponding to the multiple third tumor images.

[0011] In one possible implementation, determining multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the multiple third tumor images, the tumor types corresponding to the multiple first tumor images, the tumor types corresponding to the multiple second tumor images, and the tumor types corresponding to the multiple third tumor images, includes: determining a third target tumor image and a fourth target tumor image based on the multiple first tumor images, the multiple second tumor images, and the multiple third tumor images, wherein the third target tumor image belongs to the multiple first tumor images, the multiple second tumor images, and the multiple third tumor images, and the fourth target tumor image belongs to any two of the multiple first tumor images, the multiple second tumor images, and the multiple third tumor images; determining whether the tumor type corresponding to the third target tumor image includes all tumor types among the multiple tumor types; if the tumor type corresponding to the third target tumor image includes all tumor types among the multiple tumor types, then determining the multiple important tumor images based on a preset number of fourth target tumor images among the third target tumor images and the fourth target tumor images, wherein the preset number is less than or equal to one-third of the number of third target tumor images.

[0012] In one possible implementation, the deep learning-based tumor image analysis method further includes: if the tumor type corresponding to the third target tumor image does not include all of the multiple tumor types, then, based on the tumor type corresponding to the fourth target tumor image, determining tumor images corresponding to tumor types not included in the tumor type corresponding to the third target tumor image from the fourth target tumor image and other tumor images excluding the third target tumor image and the fourth target tumor image, wherein the selection priority of the fourth target tumor image is higher than that of the other tumor images excluding the third target tumor image and the fourth target tumor image; and determining the multiple important tumor images based on the third target tumor image and the tumor images corresponding to tumor types not included in the tumor type corresponding to the third target tumor image.

[0013] In one possible implementation, the deep learning-based tumor image analysis method further includes: acquiring patient medical record information corresponding to multiple important tumor images, the patient medical record information including: patient identifier, number of cases, duration of illness, and treatment duration; determining multiple first important tumor images from the multiple important tumor images based on the patient identifier, the multiple first important tumor images corresponding to the same patient identifier; determining at least one second important tumor image from the multiple important tumor images based on the number of cases, the second important tumor image corresponding to a number of cases greater than a preset number; determining at least one third important tumor image from the multiple important tumor images based on the duration of illness and the treatment duration, the duration of illness and the treatment duration corresponding to the third important tumor image satisfying a preset duration relationship; correspondingly, generating tumor image analysis results based on the important tumor images includes: generating tumor image analysis results based on the multiple first important tumor images, the at least one second important tumor image, and the at least one third important tumor image.

[0014] In one possible implementation, the tumor image analysis result includes: multiple important tumor images and their corresponding importance and influence levels. The step of generating the tumor image analysis result based on the multiple first important tumor images, the at least one second important tumor image, and the at least one third important tumor image includes: determining the importance level of the multiple first important tumor images as a first importance level, determining the importance level of the at least one second important tumor image as a second importance level, and determining the importance level of the at least one third important tumor image as a third importance level, wherein the first importance level is higher than the second importance level, and the second importance level is higher than the third importance level; determining the influence level of the multiple first important tumor images as a first influence level, determining the importance level of the at least one second important tumor image as a second influence level, and determining the importance level of the at least one third important tumor image as a third influence level, wherein the second influence level is higher than the third influence level, and the third influence level is higher than the first influence level; and generating the tumor image analysis result based on the multiple first important tumor images and their corresponding importance and influence levels, the at least one second important tumor image and its corresponding importance and influence levels, and the at least one third important tumor image and its corresponding importance and influence levels.

[0015] Compared with existing technologies, the deep learning-based tumor image analysis method provided in this application has the following technical advantages:

[0016] By acquiring multiple tumor images corresponding to various tumor types, this method filters out tumor images with tumor areas larger than a preset area and tumor regions with clustering greater than a preset density. Then, by combining the tumor types corresponding to these images, important tumor images are identified, and analysis results are obtained based on these important tumor images. Therefore, this analysis method not only combines multiple types of tumor images but also incorporates deep learning algorithms to achieve effective analysis of tumor images across various tumor types, thus enabling accurate and effective selection of important tumor images. Attached Figure Description

[0017] Figure 1 This is a schematic diagram of a first tumor image analysis scenario according to an embodiment of this application;

[0018] Figure 2 This is a schematic diagram of a second tumor image analysis scenario according to an embodiment of this application;

[0019] Figure 3 This is a flowchart of a deep learning-based tumor image analysis method according to an embodiment of this application;

[0020] Figure 4 This is a schematic diagram of the structure of a deep learning-based tumor image analysis device according to an embodiment of this application;

[0021] Figure 5 This is a schematic diagram of the structure of a terminal device according to an embodiment of this application. Detailed Implementation

[0022] The specific embodiments of this application will be described in detail below with reference to the accompanying drawings, but it should be understood that the scope of protection of this application is not limited to the specific embodiments.

[0023] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0024] With the development of image processing technology, it has been increasingly applied in medical settings. In some medical scenarios, image processing techniques are used to analyze various medical images to obtain corresponding medical image analysis results; these results can assist doctors in making better diagnoses. In other medical scenarios, these medical image analysis results can be used for tumor research, possessing corresponding research and analytical significance.

[0025] Currently, common methods for analyzing tumor images include segmenting the tumor region using image processing techniques and further determining the tumor type. While this method can analyze tumor images, its approach is relatively limited.

[0026] As can be seen from the above application scenarios, the research and analysis of tumor images are also of great significance. Therefore, if we can select some important tumor images from a large number of tumor images and analyze them, we can improve the effectiveness of tumor image research.

[0027] Based on this, embodiments of this application provide a deep learning-based tumor image analysis scheme. This scheme not only combines multiple types of tumor images but also incorporates deep learning algorithms to analyze and filter tumor images. This enables effective analysis of tumor images of various tumor types, thereby achieving accurate and effective screening of important tumor images. Since the ultimately identified important tumor images can cover multiple tumor types, it can analyze not only important images of a single tumor type but also integrate images of multiple tumor types for analysis, thus improving the effectiveness of tumor image analysis.

[0028] Please refer to Figure 1 This is a schematic diagram of a first tumor image analysis scenario provided in the embodiments of this application. In this tumor image analysis scenario, there are a terminal device and multiple image acquisition devices, and the terminal device is communicatively connected to the multiple image acquisition devices respectively.

[0029] In some embodiments, multiple image acquisition devices are used to acquire different types of images. For example, the image acquisition devices may include: brain image acquisition devices, chest image acquisition devices, lung image acquisition devices, etc., corresponding to image acquisition devices for different body parts. These image acquisition devices may be CT (Computed Tomography) image acquisition devices, MRI image acquisition devices, etc.

[0030] In some embodiments, these multiple image acquisition devices can not only acquire images but also perform image processing, such as determining whether the acquired image is a tumor image.

[0031] Furthermore, images acquired by multiple image acquisition devices can be transmitted to a terminal device, which can then analyze these images to determine the corresponding image analysis results. In addition, these images can also be used by doctors for diagnosis, etc., without further limitation.

[0032] Please refer to Figure 2This is a schematic diagram of a second tumor image analysis scenario provided in the embodiments of this application. In this tumor image analysis scenario, there are a terminal device and a tumor image database, and the terminal device is communicatively connected to the tumor image database.

[0033] In this application scenario, the tumor image database contains multiple tumor images corresponding to various tumor types. When the terminal device has image analysis requirements, it can directly obtain multiple tumor images corresponding to various tumor types from this database.

[0034] Please refer to Figure 3 This application provides a deep learning-based tumor image analysis method, which can be applied to... Figure 1 or Figure 2 The terminal device shown. The method includes:

[0035] Step 301: Obtain multiple tumor images corresponding to different tumor types.

[0036] Step 302: Based on the preset first tumor image screening model, select multiple first tumor images from multiple tumor images corresponding to various tumor types.

[0037] Step 303: Based on the preset second tumor image screening model, select multiple second tumor images from multiple tumor images corresponding to various tumor types.

[0038] Step 304: Based on the multiple first tumor images, multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images, determine multiple important tumor images.

[0039] Step 305: Generate tumor image analysis results based on multiple important tumor images.

[0040] In step 301, each tumor image includes a tumor region corresponding to the tumor type. These multiple tumor images can be tumor images directly sent by the corresponding image acquisition device; or they can be images obtained by the terminal device after first acquiring the images sent by the image acquisition device, then performing tumor region segmentation and tumor type determination. Alternatively, they can be tumor images directly obtained by the terminal device from a tumor image database.

[0041] In some embodiments, tumor region segmentation and tumor type determination can be referred to mature technologies in the field, and will not be described in detail here.

[0042] In the embodiments of this application, various tumor types can cover all or some of the tumor types in medical scenarios, and they will not be listed one by one here.

[0043] In step 302, the tumor region area of ​​multiple first tumor images is larger than the preset tumor region area. Therefore, the first tumor image screening model is used to filter out tumor images with larger tumor region areas.

[0044] In some embodiments, the preset tumor area can be set according to different application scenarios. It can be the area of ​​a tumor region with a high degree of impact, and it can be set by experts.

[0045] In some embodiments, the first tumor image screening model is a deep learning-based model, which may employ a corresponding deep learning algorithm, and the deep learning algorithm may refer to mature technologies in the field.

[0046] As an optional implementation, the training process of the first tumor image screening model may include: acquiring a first training dataset, which includes multiple sample tumor images corresponding to various tumor types, each sample tumor image including a tumor region corresponding to the tumor type, the multiple sample tumor images including sample tumor images with a first area label and sample tumor images with a second area label, the first area label being used to indicate that the tumor region area is greater than a preset tumor region area, and the second area label being used to indicate that the tumor region area is less than or equal to the preset tumor region area; and training the initial first tumor image screening model based on the first training dataset to obtain a preset first tumor image screening model.

[0047] It is understandable that by training the model using the first training dataset, the trained model can be used to filter out tumor images with a tumor area larger than a preset tumor area.

[0048] In some embodiments, the training of the model can also employ certain implementation methods to improve the model's performance. For example, preset training iterations, preset test datasets, etc.

[0049] In step 303, the tumor region clustering degree of multiple second tumor images is greater than a preset clustering degree. It can be understood that tumor region clustering degree can characterize the number of neighboring tumor regions or the distance between multiple tumor regions. Generally speaking, the higher the clustering degree, the more neighboring tumor regions there are; or, the smaller the distance between multiple tumor regions.

[0050] In some embodiments, the preset aggregation degree can be set according to different application scenarios. It can be the aggregation degree of tumor areas with high impact, which can be set by experts.

[0051] In some embodiments, the second tumor image screening model may also be a deep learning-based model, and the deep learning algorithm used may be the same as or different from that of the first tumor image screening model.

[0052] As an optional implementation, the training process of the second tumor image screening model may include: acquiring a second training dataset, which includes multiple sample tumor images corresponding to various tumor types. Each sample tumor image includes a tumor region corresponding to the tumor type. The multiple sample tumor images include sample tumor images with a first clustering degree label and sample tumor images with a second clustering degree label. The first clustering degree label is used to indicate that the clustering degree of the tumor region is greater than a preset clustering degree, and the second area label is used to indicate that the clustering degree of the tumor region is less than or equal to the preset clustering degree. Based on the second training dataset, the initial second tumor image screening model is trained to obtain the preset second tumor image screening model.

[0053] It is understandable that by training the model using the second training dataset, the trained model can be used to filter out tumor images with a tumor area larger than a preset tumor area.

[0054] In some embodiments, the training of the model can also employ certain implementation methods to improve the model's performance. For example, preset training iterations, preset test datasets, etc.

[0055] In step 304, multiple important tumor images are determined based on the multiple first tumor images, multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images.

[0056] As an optional implementation, step 304 includes: determining a first target tumor image and a second target tumor image based on multiple first tumor images and multiple second tumor images, wherein the first target tumor image belongs to multiple first tumor images and also belongs to multiple second tumor images, and the second target tumor image does not belong to multiple first tumor images but belongs to multiple second tumor images; determining whether the tumor type corresponding to the first target tumor image includes all tumor types among multiple tumor types; if the tumor type corresponding to the first target tumor image includes all tumor types among multiple tumor types, then determining multiple important tumor images based on at least two second target tumor images among the first target tumor image and the second target tumor image.

[0057] In some embodiments, if a tumor image belongs to multiple first tumor images and multiple second tumor images, it indicates that the tumor region of the tumor image is large and the tumor region is highly clustered. In this case, it can be identified as the first target tumor image.

[0058] Furthermore, if a tumor image does not belong to multiple first tumor images but to multiple second tumor images, it indicates that the tumor region of the image has a high degree of aggregation. In this case, it can be identified as the second target tumor image.

[0059] Furthermore, it can be determined whether the tumor type corresponding to the first target tumor image includes all tumor types among multiple tumor types. If so, it means that the first target tumor image has covered all tumor types. Multiple important tumor images can be determined based on at least two second target tumor images from the first target tumor image and the second target tumor image.

[0060] In some embodiments, at least two second target tumor images may be images randomly selected from a plurality of second target tumor images.

[0061] If not, it indicates that the first target tumor image does not cover all tumor types. In this case, the method may further include: if the tumor type corresponding to the first target tumor image does not include all tumor types among multiple tumor types, then, based on the tumor type corresponding to the second target tumor image, determining tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image from the second target tumor image and other tumor images excluding the first and second target tumor images, wherein the selection priority of the second target tumor image is higher than that of the other tumor images excluding the first and second target tumor images; and determining multiple important tumor images based on the first target tumor image and the tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image.

[0062] In this implementation, tumor images can be selected from a second target tumor image and other tumor images excluding the first and second target tumor images. Prioritize selecting tumor images from the second target tumor image. If tumor images representing all remaining un-included tumor types can be identified from the second target tumor image, then at least two tumor images (which may correspond to different tumor types than the first target tumor image) can be arbitrarily selected from the remaining tumor images excluding the first and second target tumor images. If tumor images representing all remaining unincluded tumor types cannot be identified from the second target tumor image, then tumor images representing tumor types not yet included are first identified from the remaining tumor images excluding the first and second target tumor images.

[0063] Furthermore, based on the first target tumor image and tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image, multiple important tumor images are identified.

[0064] In this embodiment, in addition to considering the area and aggregation of the tumor region, the degree of tumor spread can also be considered. The degree of tumor spread requires consideration of two conditions: first, the neighboring tumor regions surrounding the tumor region; and second, the location of the tumor region. Generally, the tumor region is located at the edge of the organ, and there are a large number of neighboring tumor regions.

[0065] Therefore, as an optional implementation, the method further includes: selecting multiple third tumor images from multiple tumor images corresponding to multiple tumor types according to a preset third tumor image screening model, wherein the tumor region diffusion of the multiple third tumor images is greater than a preset diffusion; step 304 may include: determining multiple important tumor images according to multiple first tumor images, multiple second tumor images, multiple third tumor images, tumor types corresponding to multiple first tumor images, tumor types corresponding to multiple second tumor images, and tumor types corresponding to multiple third tumor images.

[0066] In some embodiments, the preset diffusion degree can be set according to different application scenarios. It can be the diffusion degree of tumor areas with high impact, which can be set by experts.

[0067] In some embodiments, the third tumor image screening model is a deep learning-based model, which may employ a corresponding deep learning algorithm, and the deep learning algorithm may refer to mature technologies in the field.

[0068] As an optional implementation, the training process of the third tumor image screening model may include: acquiring a third training dataset, which includes multiple sample tumor images corresponding to various tumor types. Each sample tumor image includes a tumor region corresponding to the tumor type. The multiple sample tumor images include sample tumor images with a first diffusion label and sample tumor images with a second diffusion label. The first diffusion label is used to indicate that the diffusion of the tumor region is greater than a preset tumor region diffusion, and the second diffusion label is used to indicate that the diffusion of the tumor region is less than or equal to the preset tumor region diffusion. Based on the third training dataset, the initial third tumor image screening model is trained to obtain the preset third tumor image screening model.

[0069] It is understandable that by training the model using a third training dataset, the trained model can be used to filter out tumor images where the tumor region diffusion exceeds a preset tumor region diffusion.

[0070] In some embodiments, the training of the model can also employ certain implementation methods to improve the model's performance. For example, preset training iterations, preset test datasets, etc.

[0071] Further, based on multiple first tumor images, multiple second tumor images, multiple third tumor images, and the tumor types corresponding to multiple first tumor images, multiple second tumor images, and multiple third tumor images, multiple important tumor images are determined, including: determining a third target tumor image and a fourth target tumor image based on multiple first tumor images, multiple second tumor images, and multiple third tumor images, wherein the third target tumor image belongs to multiple first tumor images, multiple second tumor images, and multiple third tumor images, and the fourth target tumor image belongs to any two of multiple first tumor images, multiple second tumor images, and multiple third tumor images; determining whether the tumor type corresponding to the third target tumor image includes all tumor types among multiple tumor types; if the tumor type corresponding to the third target tumor image includes all tumor types among multiple tumor types, then multiple important tumor images are determined based on a preset number of fourth target tumor images among the third target tumor images and the fourth target tumor images, wherein the preset number is less than or equal to one-third of the number of third target tumor images.

[0072] In some embodiments, if the tumor type corresponding to the third target tumor image does not include all tumor types among multiple tumor types, then based on the tumor type corresponding to the fourth target tumor image, tumor images corresponding to tumor types not included in the tumor type corresponding to the third target tumor image are determined from the fourth target tumor image and the remaining tumor images other than the third target tumor image and the fourth target tumor image, wherein the selection priority of the fourth target tumor image is higher than that of the remaining tumor images other than the third target tumor image and the fourth target tumor image; multiple important tumor images are determined based on the third target tumor image and the tumor images corresponding to tumor types not included in the tumor type corresponding to the third target tumor image.

[0073] It is understood that the logic for determining multiple important tumor images here can refer to the aforementioned embodiments. That is, based on ensuring that the tumor types can cover all tumor types, important tumor images are determined sequentially from the third target tumor image, the fourth target tumor image, and tumor images other than the third target tumor image and the fourth target tumor image.

[0074] Step 305: Generate tumor image analysis results based on multiple important tumor images.

[0075] As an optional implementation, the method further includes: acquiring patient medical record information corresponding to multiple important tumor images, the patient medical record information including: patient identifier, number of lesions, duration of illness, and treatment duration; determining multiple first important tumor images from the multiple important tumor images based on the patient identifier, wherein the multiple first important tumor images correspond to the same patient identifier; determining at least one second important tumor image from the multiple important tumor images based on the number of lesions, wherein the number of lesions corresponding to the second important tumor image is greater than a preset number; and determining at least one third important tumor image from the multiple important tumor images based on the duration of illness and treatment duration, wherein the duration of illness and treatment duration corresponding to the third important tumor image satisfy a preset duration relationship.

[0076] It is understandable that each important tumor image is associated with corresponding patient medical record information. Therefore, the patient's medical record information can be obtained to obtain more analytical results.

[0077] Patient identification can be understood as a patient's identity identifier, such as the patient's name or medical card number.

[0078] The number of illnesses can be understood as the number of different types of illnesses a patient has contracted. This information can be directly found in the medical records or determined through appropriate statistical methods.

[0079] Duration of illness can be understood as the length of time a patient has been suffering from an illness. It can be information directly found in the medical records or information determined through appropriate statistical methods.

[0080] Treatment duration can be understood as the length of time a patient is treated for a disease. It can be calculated from the first dose of medication. It can be information directly available in the medical record or information determined through appropriate statistical methods.

[0081] In some embodiments, if multiple important tumor images correspond to the same patient identifier, then that image is identified as the first important tumor image.

[0082] In some embodiments, the preset quantity can be 2 or more, and is not limited here.

[0083] In some embodiments, the preset duration relationship may be: the ratio of illness duration to treatment duration is greater than 1; or, the illness duration is equal to the treatment duration, and both the illness duration and the treatment duration are greater than the preset duration, such as one month.

[0084] Furthermore, the tumor image analysis results include: multiple important tumor images and the importance and impact levels corresponding to the multiple important tumor images, respectively.

[0085] Then, based on multiple images of the most important tumor, at least one image of the second most important tumor, and at least one image of the third most important tumor, a tumor image analysis result is generated, including: determining the importance level of the multiple images of the most important tumor as the first importance level, determining the importance level of at least one image of the second most important tumor as the second importance level, and determining the importance level of at least one image of the third most important tumor as the third importance level, wherein the first importance level is higher than the second importance level, and the second importance level is higher than the third importance level; determining the influence level of the multiple images of the most important tumor as the first influence level, determining the importance level of at least one image of the second most important tumor as the second influence level, and determining the importance level of at least one image of the third most important tumor as the third influence level, wherein the second influence level is higher than the third influence level, and the third influence level is higher than the first influence level; and generating a tumor image analysis result based on the multiple images of the most important tumor and their corresponding importance and influence levels, at least one image of the second most important tumor and its corresponding importance and influence levels, and at least one image of the third most important tumor and its corresponding importance and influence levels.

[0086] As described in the embodiments of this application, by acquiring multiple tumor images corresponding to various tumor types, tumor images with tumor regions larger than a preset tumor region area and tumor images with tumor region clustering greater than a preset clustering degree are selected. Then, by combining the tumor types corresponding to these images, important tumor images are identified, and based on these important tumor images, important tumor image analysis results are obtained. Therefore, this analysis method not only combines multiple types of tumor images but also incorporates deep learning algorithms to achieve tumor image analysis and selection, thereby enabling effective analysis of tumor images of various tumor types and achieving accurate and effective selection of important tumor images.

[0087] Please refer to Figure 4 Embodiments of this application also provide a deep learning-based tumor image analysis device, comprising:

[0088] The acquisition module 401 is used to acquire multiple tumor images corresponding to various tumor types, and each tumor image includes the tumor region corresponding to the tumor type.

[0089] Analysis module 402 is used to: select multiple first tumor images from multiple tumor images corresponding to various tumor types according to a preset first tumor image screening model, wherein the tumor area of ​​the multiple first tumor images is greater than a preset tumor area; select multiple second tumor images from multiple tumor images corresponding to various tumor types according to a preset second tumor image screening model, wherein the tumor region clustering of the multiple second tumor images is greater than a preset clustering degree; determine multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images; and generate tumor image analysis results based on the multiple important tumor images.

[0090] The implementation of this deep learning-based tumor image analysis device can be referred to the aforementioned method embodiments, and will not be repeated here.

[0091] Please refer to Figure 5 The embodiments of this application also provide a terminal device, which can be a computer, mobile phone or other electronic devices with image processing functions.

[0092] The terminal device includes a processor 501 and a memory 502, which are communicatively connected.

[0093] The processor 501 and the memory 502 are electrically connected directly or indirectly to enable data transmission or interaction. For example, these components can be electrically connected via one or more communication buses or signal buses. The method steps executed by each of the aforementioned modules or interactive terminals each include at least one software functional module that can be stored in the memory 502 in the form of software or firmware.

[0094] Processor 501 can be an integrated circuit chip with signal processing capabilities. Processor 501 can be a general-purpose processor, including a CPU (Central Processing Unit), NP (Network Processor), etc.; it can also be a digital signal processor, application-specific integrated circuit, off-the-shelf programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component. It can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of this invention. The general-purpose processor can be a microprocessor or any conventional processor.

[0095] The memory 502 can store various software programs and modules. The processor 501 executes various functional applications and data processing by running the software programs and modules stored in the memory 502, thereby implementing the various steps in the embodiments of this application.

[0096] The memory 502 may include, but is not limited to, RAM (Random Access Memory), ROM (Read Only Memory), PROM (Programmable Read-Only Memory), EPROM (Erasable Programmable Read-Only Memory), EEPROM (Electric Erasable Programmable Read-Only Memory), etc.

[0097] Understandable. Figure 5 The structure shown is for illustrative purposes only; the terminal device may also include components that are more advanced than those shown. Figure 5 The more or fewer components shown, or having the same Figure 5 The different configurations shown.

[0098] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0099] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0100] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0101] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0102] The foregoing description of specific exemplary embodiments of this application is for illustrative and explanatory purposes. These descriptions are not intended to limit this application to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of this application and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of this application, as well as various different choices and variations. The scope of this application is intended to be defined by the claims and their equivalents.

Claims

1. A tumor image analysis method based on deep learning, characterized in that, include: Acquire multiple tumor images corresponding to various tumor types, with each tumor image including the tumor region corresponding to the tumor type; According to the preset first tumor image screening model, multiple first tumor images are selected from multiple tumor images corresponding to the various tumor types, and the tumor area of ​​the multiple first tumor images is greater than the preset tumor area. According to the preset second tumor image screening model, multiple second tumor images are selected from multiple tumor images corresponding to the various tumor types, and the tumor region clustering degree of the multiple second tumor images is greater than the preset clustering degree. Based on the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images, multiple important tumor images are determined; Based on the multiple important tumor images, tumor image analysis results are generated.

2. The tumor image analysis method based on deep learning according to claim 1, characterized in that, The deep learning-based tumor image analysis method further includes: Obtain a first training dataset, which includes multiple sample tumor images corresponding to various tumor types. Each sample tumor image includes a tumor region corresponding to the tumor type. The multiple sample tumor images include sample tumor images with a first area label and sample tumor images with a second area label. The first area label is used to indicate that the area of ​​the tumor region is greater than the preset tumor region area, and the second area label is used to indicate that the area of ​​the tumor region is less than or equal to the preset tumor region area. Based on the first training dataset, the initial first tumor image screening model is trained to obtain the preset first tumor image screening model.

3. The tumor image analysis method based on deep learning according to claim 1, characterized in that, The deep learning-based tumor image analysis method further includes: Obtain a second training dataset, which includes multiple sample tumor images corresponding to various tumor types. Each sample tumor image includes a tumor region corresponding to the tumor type. The multiple sample tumor images include sample tumor images with a first clustering degree label and sample tumor images with a second clustering degree label. The first clustering degree label is used to indicate that the clustering degree of the tumor region is greater than a preset clustering degree, and the second area label is used to indicate that the clustering degree of the tumor region is less than or equal to the preset clustering degree. Based on the second training dataset, the initial second tumor image screening model is trained to obtain the preset second tumor image screening model.

4. The tumor image analysis method based on deep learning according to claim 1, characterized in that, The step of determining multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images includes: Based on the plurality of first tumor images and the plurality of second tumor images, a first target tumor image and a second target tumor image are determined, wherein the first target tumor image belongs to the plurality of first tumor images and also belongs to the plurality of second tumor images, and the second target tumor image does not belong to the plurality of first tumor images but belongs to the plurality of second tumor images; Determine whether the tumor type corresponding to the first target tumor image includes all of the multiple tumor types; If the tumor type corresponding to the first target tumor image includes all of the multiple tumor types, then the multiple important tumor images are determined based on at least two second target tumor images from the first target tumor image and the second target tumor image.

5. The tumor image analysis method based on deep learning according to claim 4, characterized in that, The deep learning-based tumor image analysis method also includes: If the tumor type corresponding to the first target tumor image does not include all of the multiple tumor types, then based on the tumor type corresponding to the second target tumor image, tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image are determined from the second target tumor image and the remaining tumor images other than the first target tumor image and the second target tumor image, wherein the selection priority of the second target tumor image is higher than that of the remaining tumor images other than the first target tumor image and the second target tumor image. The multiple important tumor images are determined based on the first target tumor image and tumor images corresponding to tumor types not included in the tumor type corresponding to the first target tumor image.

6. The tumor image analysis method based on deep learning according to claim 1, characterized in that, The deep learning-based tumor image analysis method also includes: According to the preset third tumor image screening model, multiple third tumor images are selected from multiple tumor images corresponding to the various tumor types, and the tumor region diffusion of the multiple third tumor images is greater than the preset diffusion. The step of determining multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the tumor types corresponding to the multiple first tumor images, and the tumor types corresponding to the multiple second tumor images includes: Based on the multiple first tumor images, the multiple second tumor images, the multiple third tumor images, the tumor types corresponding to the multiple first tumor images, the tumor types corresponding to the multiple second tumor images, and the tumor types corresponding to the multiple third tumor images, multiple important tumor images are determined.

7. The tumor image analysis method based on deep learning according to claim 6, characterized in that, The step of determining multiple important tumor images based on the multiple first tumor images, the multiple second tumor images, the multiple third tumor images, the tumor types corresponding to the multiple first tumor images, the tumor types corresponding to the multiple second tumor images, and the tumor types corresponding to the multiple third tumor images includes: Based on the plurality of first tumor images, the plurality of second tumor images, and the plurality of third tumor images, a third target tumor image and a fourth target tumor image are determined. The third target tumor image belongs to the plurality of first tumor images, the plurality of second tumor images, and the plurality of third tumor images. The fourth target tumor image belongs to any two of the plurality of first tumor images, the plurality of second tumor images, and the plurality of third tumor images. Determine whether the tumor type corresponding to the third target tumor image includes all of the multiple tumor types; If the tumor type corresponding to the third target tumor image includes all the tumor types among the multiple tumor types, then the multiple important tumor images are determined based on a preset number of fourth target tumor images among the third target tumor image and the fourth target tumor image, wherein the preset number is less than or equal to one-third of the number of the third target tumor images.

8. The tumor image analysis method based on deep learning according to claim 7, characterized in that, The deep learning-based tumor image analysis method also includes: If the tumor type corresponding to the third target tumor image does not include all of the multiple tumor types, then based on the tumor type corresponding to the fourth target tumor image, tumor images corresponding to tumor types not included in the tumor type corresponding to the third target tumor image are determined from the fourth target tumor image and the remaining tumor images other than the third target tumor image and the fourth target tumor image, wherein the selection priority of the fourth target tumor image is higher than that of the remaining tumor images other than the third target tumor image and the fourth target tumor image; The multiple important tumor images are determined based on the third target tumor image and tumor images corresponding to tumor types not included in the tumor type corresponding to the third target tumor image.

9. The tumor image analysis method based on deep learning according to claim 1, characterized in that, The deep learning-based tumor image analysis method also includes: Obtain patient medical record information corresponding to multiple important tumor images, including: patient identifier, number of cases, duration of illness, and treatment duration; Based on the patient identifier, multiple first important tumor images are identified from multiple important tumor images, and the multiple first important tumor images correspond to the same patient identifier; Based on the number of lesions, at least one second important tumor image is determined from multiple important tumor images, wherein the number of lesions corresponding to the second important tumor image is greater than a preset number; Based on the duration of illness and the duration of treatment, at least one third important tumor image is determined from multiple important tumor images, wherein the duration of illness and the duration of treatment corresponding to the third important tumor image satisfy a preset duration relationship; Correspondingly, generating tumor image analysis results based on the important tumor images includes: Based on the multiple first-important tumor images, the at least one second-important tumor image, and the at least one third-important tumor image, tumor image analysis results are generated.

10. The tumor image analysis method based on deep learning according to claim 9, characterized in that, The tumor image analysis results include: multiple important tumor images and their respective importance and impact levels. The generation of tumor image analysis results based on the multiple first-important tumor images, at least one second-important tumor image, and at least one third-important tumor image includes: The importance level of the plurality of first important tumor images is determined to be first important level, the importance level of the at least one second important tumor image is determined to be second important level, and the importance level of the at least one third important tumor image is determined to be third important level, wherein the first important level is higher than the second important level, and the second important level is higher than the third important level; The influence level of the plurality of first important tumor images is determined as the first influence level, the importance level of the at least one second important tumor image is determined as the second influence level, and the importance level of the at least one third important tumor image is determined as the third influence level, wherein the second influence level is higher than the third influence level, and the third influence level is higher than the first influence level; The tumor image analysis results are generated based on the multiple first-important tumor images and their corresponding importance and impact levels, the at least one second-important tumor image and its corresponding importance and impact level, and the at least one third-important tumor image and its corresponding importance and impact level.

Images