Adrenal functional tumor classification and positioning equipment, device and storage medium

Abstract

The present invention relates to artificial intelligence and digital medical technology, and an embodiment thereof provides an adrenal functional tumor classification and positioning device, an apparatus and a storage medium, and the method executed by the device includes: inputting pre-processed adrenal CT data into a first nnUnet network for learning segmentation processing to obtain kidney mask data; inputting the kidney mask data and the adrenal CT data into a second nnUnet network for learning segmentation processing to obtain a first segmentation result of the adrenal tumor; inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map; constructing a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer; comparing the mean vector and the standard deviation vector with the mean and standard deviation values ​​of a normal distribution through the KL divergence to obtain a difference result; and training the network according to the difference result to obtain a kidney image analysis result. The recognition accuracy of adrenal functional tumor images can be improved.

Description

Technical Field

[0001] The present invention relates to the field of digital medicine, and in particular to an adrenal functional tumor typing and positioning device, apparatus and storage medium. Background Art

[0002] Adrenal tumors are one of the most common clinical tumors, which can lead to excessive hormone secretion and cause various endocrine disorders and related clinical symptoms. At present, the clinical diagnosis of adrenal tumors requires a combination of image analysis and comprehensive endocrine hormone examinations. However, the detection of hormones such as adrenocorticotropic hormone, aldosterone, and metanephrine is expensive and time-consuming, the procedures are complicated, and they are greatly affected by food and drug factors. Some hospitals do not even have the ability to test endocrine hormones, so it is difficult to ensure timely and accurate diagnosis of patients.

[0003] Doctors and researchers have explored the potential of CT scan radiological parameters in predicting functional tumors with hormone activity, using deep convolutional neural networks to distinguish between benign and malignant chest nodules and segment liver and pancreatic tumors. However, there is a lack of research and application on using CT images to detect and distinguish types of functional adrenal tumors. In addition, for large three-dimensional images such as adrenal CT, data collection and annotation are very difficult, which will seriously affect the performance and robustness of the model. Summary of the invention

[0004] The following is a summary of the subject matter described in detail herein. This summary is not intended to limit the scope of the claims.

[0005] The main purpose of the embodiment of the present invention is to provide a method for locating and typing adrenal functional tumors and improve the image recognition accuracy of adrenal functional tumors.

[0006] In a first aspect, an embodiment of the present invention provides an adrenal functional tumor typing and positioning device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the method implemented when the processor executes the computer program comprises:

[0007] The pre-processed adrenal CT data is input into the first nnUnet network for learning and segmentation processing to obtain kidney mask data;

[0008] Inputting the kidney mask data and the adrenal gland CT data into a second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor;

[0009] Inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map;

[0010] The latent variable feature map is constructed into a mean vector and a standard deviation vector through a fully connected layer;

[0011] Comparing the mean vector and the standard deviation vector with the mean and standard deviation values ​​of a normal distribution by using the KL divergence to obtain a difference result;

[0012] The network is trained according to the difference results to obtain kidney image analysis results.

[0013] In one embodiment, before the method of inputting the preprocessed adrenal CT data into the first nnUnet network for learning segmentation processing to obtain kidney mask data, the method executed by the processor includes:

[0014] Acquire CT data of the adrenal gland, wherein the CT data includes arterial phase data, venous phase data, and plain scan phase data;

[0015] The arterial phase data, the venous phase data and the plain scan phase data are preprocessed to obtain the preprocessed arterial phase data, the venous phase data and the plain scan phase data.

[0016] In one embodiment, the method performed by the processor of preprocessing the arterial phase data, the venous phase data, and the plain scan phase data to obtain the preprocessed arterial phase data, the venous phase data, and the plain scan phase data includes:

[0017] Aligning the origins of the arterial phase data, the venous phase data, and the plain scan phase data;

[0018] The voxel values ​​and voxel spacings of the arterial phase data, the venous phase data and the plain scan phase data after the source point alignment processing are normalized to obtain the pre-processed arterial phase data, the venous phase data and the plain scan phase data.

[0019] In one embodiment, the formula of the loss function set in the first nnUnet network or the second nnUnet network includes:

[0020] L nnUnet =L dc +L ce ;

[0021] Among them, L ce is the cross entropy loss function, L ce is the diceloss function for three types of functional adrenal tumors.

[0022] In one embodiment, the formula of the loss function set in the nnUnet-VAE network includes:

[0023] L nnUnet-vae =L nnUnet +L cls +L vae ;

[0024] L vae =L KL +L MSE ;

[0025] Among them, L cls is the cross entropy loss function for classification, L KL represents the KL divergence, L MSE It is the mean square error loss function between the model input image and the output of the nnUnet-VAE network.

[0026] In one embodiment, the first nnUnet network or the second nnUnet network includes an encoder and a decoder that is jump-connected to the encoder, the encoder is used to perform multiple downsampling operations and construct multiple feature map operations, and the decoder is used to perform multiple consecutive upsampling operations starting from the feature map of the deepest network.

[0027] In one embodiment, the method executed by the processor for constructing a mean vector and a standard deviation vector from a latent variable feature map through a fully connected layer includes:

[0028] The latent variable feature map is constructed into a 1024-dimensional mean vector and a 1024-dimensional standard deviation vector through a fully connected layer.

[0029] In one embodiment, the method executed by the processor further includes: visualizing the kidney image indication result and displaying it on a screen.

[0030] In a second aspect, an embodiment of the present invention provides a device for locating and typing adrenal gland functional tumors, comprising:

[0031] The first segmentation module is used to input the preprocessed adrenal CT data into the first nnUnet network for learning and segmentation processing to obtain kidney mask data;

[0032] A second segmentation module is used to input the kidney mask data and the adrenal gland CT data into a second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor;

[0033] A generation module, used for inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map;

[0034] A construction module is used to construct a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer;

[0035] A comparison module, used for comparing the mean vector and the standard deviation vector with the mean and the standard deviation of the normal distribution through the KL divergence to obtain a difference result;

[0036] The training module is used to train the network according to the difference result to obtain the judgment result.

[0037] In one embodiment, it further includes an acquisition module, the acquisition module is used to acquire CT data of the adrenal gland, the CT data including arterial phase data, venous phase data and plain scan phase data;

[0038] The arterial phase data, the venous phase data and the plain scan phase data are preprocessed to obtain the preprocessed arterial phase data, the venous phase data and the plain scan phase data.

[0039] In one embodiment, the acquisition module is also used to align the origins of the arterial phase data, the venous phase data, and the plain scan phase data; and normalize the voxel values ​​and voxel spacing of the arterial phase data, the venous phase data, and the plain scan phase data after the source point alignment processing to obtain the pre-processed arterial phase data, the venous phase data, and the plain scan phase data.

[0040] In one embodiment, the formula of the loss function set in the first nnUnet network or the second nnUnet network includes:

[0041] L nnUnet =L dc +L ce ;

[0042] Among them, L ce is the cross entropy loss function, L ce is the diceloss function for three types of functional adrenal tumors.

[0043] In one embodiment, the formula of the loss function set in the nnUnet-VAE network includes:

[0044] L nnUnet-vae =L nnUnet +L cls +L vae ;

[0045] L vae =L KL +L MSE ;

[0046] Among them, L cls is the cross entropy loss function for classification, L KL represents the KL divergence, L MSEIt is the mean square error loss function between the model input image and the output of the nnUnet-VAE network.

[0047] In one embodiment, the first nnUnet network or the second nnUnet network includes an encoder and a decoder that is jump-connected to the encoder, the encoder is used to perform multiple downsampling operations and construct multiple feature map operations, and the decoder is used to perform multiple consecutive upsampling operations starting from the feature map of the deepest network.

[0048] In one embodiment, the construction module is further used to construct a 1024-dimensional mean vector and a 1024-dimensional standard deviation vector from the latent variable feature map through a fully connected layer.

[0049] In a fourth aspect, a computer-readable storage medium stores computer-executable instructions, wherein the computer-executable instructions are used to execute the method performed by the adrenal functional tumor classification device described in the first aspect.

[0050] The embodiment of the present invention includes: the method performed by the adrenal functional tumor classification and positioning device includes the following steps: inputting the pre-processed adrenal CT data into the first nnUnet network for learning segmentation processing to obtain kidney mask data; inputting the kidney mask data and the adrenal CT data into the second nnUnet network for learning segmentation processing to obtain the first segmentation result of the adrenal tumor; inputting the first segmentation result into the nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map; constructing a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer; comparing the mean vector and the standard deviation vector with the mean and standard deviation values ​​of the normal distribution through the KL divergence to obtain a difference result; training the network according to the difference result to obtain a kidney image analysis result. In the technical solution of this embodiment, the proposed new deep learning convolutional network can complete the classification and positioning of adrenal functional tumors using only CT scan images, avoiding the interference of external factors in traditional endocrine detection and simplifying the inspection process. The nnUnet network can automatically adapt to any dataset without human intervention, while the variational autoencoder can ensure the robustness of the model trained on a small dataset and the accuracy of adrenal functional tumor image recognition.

[0051] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the description, claims and drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Figure 1It is a schematic diagram of an adrenal gland functional tumor classification and positioning device for executing an adrenal gland functional tumor classification and positioning method provided by an embodiment of the present invention;

[0053] Figure 2 It is a flow chart of a method for typing and locating adrenal functional tumors provided by one embodiment of the present invention;

[0054] Figure 3 It is a flow chart of acquiring and preprocessing CT images in a method for locating adrenal functional tumor typing provided by an embodiment of the present invention;

[0055] Figure 4 is a flow chart of preprocessing CT images in a method for locating and typing adrenal functional tumors provided by another embodiment of the present invention;

[0056] Figure 5 is a schematic diagram of a pedestrian re-identification device based on posture guidance provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. 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.

[0058] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims or the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0059] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0060] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0061] At present, adrenal tumors are one of the most common tumors in clinical practice, which can lead to excessive hormone secretion and cause various endocrine disorders and related clinical symptoms. At present, the clinical diagnosis of adrenal tumors requires a combination of image analysis and comprehensive endocrine hormone examinations. However, the detection of hormones such as adrenocorticotropic hormone, aldosterone, and metanephrine is expensive and time-consuming, the procedures are complicated, and they are greatly affected by food and drug factors. Some hospitals do not even have the ability to test endocrine hormones, so it is difficult to ensure timely and accurate diagnosis of patients.

[0062] Doctors and researchers have explored the potential of CT scan radiological parameters in predicting functional tumors with hormone activity, using deep convolutional neural networks to distinguish between benign and malignant chest nodules and segment liver and pancreatic tumors. However, there is a lack of research and application on using CT images to detect and distinguish types of functional adrenal tumors. In addition, for large three-dimensional images such as adrenal CT, data collection and annotation are very difficult, which will seriously affect the performance and robustness of the model.

[0063] To solve the above problems, an embodiment of the present invention provides a method for categorizing and locating adrenal functional tumors. The pedestrian re-identification method includes the following steps: inputting the pre-processed adrenal CT data into a first nnUnet network for learning segmentation processing to obtain kidney mask data; inputting the kidney mask data and adrenal CT data into a second nnUnet network for learning segmentation processing to obtain a first segmentation result of the adrenal tumor; inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map; constructing a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer; comparing the mean vector and the standard deviation vector with the mean and standard deviation values ​​of the normal distribution through KL divergence to obtain a difference result; training the network according to the difference result to obtain a kidney image analysis result. In the technical solution of this embodiment, the proposed new deep learning convolutional network can complete the classification and positioning of adrenal functional tumors using only CT scan images, avoiding the interference of external factors in traditional endocrine testing and simplifying the inspection process. The nnUnet network can automatically adapt to any dataset without human intervention, while the variational autoencoder can ensure the robustness of the model trained on a small dataset and the accuracy of adrenal functional tumor image recognition.

[0064] First, some terms used in this application are explained:

[0065] nnU-Net is applied to 3D biomedical image segmentation. nnU-Net can automatically adapt to any dataset and achieve out-of-the-box segmentation due to the following two key factors: the pipeline optimization problem is formulated according to the data fingerprint (data fingerprint, representing the key attributes of the dataset) and the pipeline fingerprint (pipeline fingerprint, representing the key design choices of the segmentation algorithm); the relationship between domain knowledge is made explicit by condensing them into a set of heuristic rules, which will robustly generate high-quality pipeline fingerprints from the corresponding data fingerprints while considering the associated hardware constraints. nnUnet can fully utilize the characteristics of the dataset to train the basic U-Net model without manual intervention, and has the ability to surpass various U-Net architectures.

[0066] The variational autoencoder is an unsupervised generative model, and the application conditions of VAE are also very loose. As long as there is unlabeled training data and a prior probability distribution is set for the latent variable, training can be performed. After the training is completed, there are two results: (1) Simulation of the training data. As a generative model, VAE can generate results with the same distribution as the training data. (2) The latent variable itself. VAE can generate a latent variable z for each data in the data set. By analyzing the distribution of the latent variable, some conclusions about the data distribution can be obtained.

[0067] The loss function or cost function is a function that maps the value of a random event or its related random variables to a non-negative real number to represent the "risk" or "loss" of the random event. In applications, the loss function is usually associated with the optimization problem as a learning criterion, that is, the model is solved and evaluated by minimizing the loss function.

[0068] The cross entropy loss function is a smooth function, which is essentially the application of cross entropy in information theory to classification problems. From the definition of cross entropy, minimizing cross entropy is equivalent to minimizing the relative entropy of the observed value and the estimated value, that is, the Kullback-Leibler divergence of the probability distribution of the two: , so it is a proxy loss that provides unbiased estimation. The cross entropy loss function is the most widely used proxy loss in the table, and the corresponding classifier examples include logistic regression, artificial neural network, and support vector machine with probability output.

[0069] The embodiments of the present invention are further described below in conjunction with the accompanying drawings.

[0070] like Figure 1 As shown, Figure 1FIG. 1 is a schematic diagram of a device 100 for performing adrenal functional tumor typing and positioning according to an embodiment of the present invention.

[0071] exist Figure 1 In the example, the adrenal gland function tumor typing and positioning device 100 is provided with a processor 110 and a memory 120, wherein the processor 110 and the memory 120 can be connected via a bus or other means. Figure 1 The example of connecting through bus is taken in the following.

[0072] The memory 120, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory 120 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory 120 may optionally include a memory remotely arranged relative to the processor 110, and these remote memories may be connected to the adrenal function tumor typing and positioning device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0073] Those skilled in the art will appreciate that the adrenal functional tumor typing and positioning device can be applied to 5G communication network systems and subsequently evolved mobile communication network systems, etc., and this embodiment does not specifically limit this.

[0074] It can be understood by those skilled in the art that Figure 1 The adrenal functional tumor typing and positioning device shown in the figure does not constitute a limitation on the embodiments of the present invention, and may include more or fewer components than shown in the figure, or a combination of certain components, or a different arrangement of components.

[0075] The adrenal function tumor typing and positioning device 100 can be an independent adrenal function tumor typing and positioning device, or it can be a cloud adrenal function tumor typing and positioning device 100 that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (Content Delivery Network, CDN), and big data and artificial intelligence platforms.

[0076] The device may also include: a radio frequency (RF) circuit, an input unit, a display unit, a sensor, an audio circuit, a wireless fidelity (WiFi) module, a processor, a power supply, and other components. Those skilled in the art will appreciate that this embodiment does not limit the structure of the device, and may include more or fewer components than this embodiment, or combine certain components, or arrange components differently.

[0077] RF circuits can be used to receive and send signals during the process of sending and receiving information or making calls. In particular, after receiving the downlink information of the base station, it is processed by the processor; in addition, the designed uplink data is sent to the base station. Usually, the RF circuit includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (Low Noise Amplifier, LNA for short), a duplexer, etc. In addition, the RF circuit can also communicate with the network and other devices through wireless communication. The above-mentioned wireless communication can use any communication standard or protocol, including but not limited to the Global System of Mobile communication (GSM for short), General Packet Radio Service (GPRS for short), Code Division Multiple Access (CDMA for short), Wideband Code Division Multiple Access (WCDMA for short), Long Term Evolution (LTE for short), email, Short Messaging Service (SMS for short), etc.

[0078] The memory 120 can also be used to store software programs and modules. The processor executes various functional applications and data processing of the device by running the software programs and modules stored in the memory. The memory can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the device (such as audio data, a phone book, etc.), etc. In addition, the memory can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0079] The input unit can be used to receive input digital or character information, and to generate key signal input related to the setting and function control of the device. Specifically, the input unit may include a touch panel and other input devices. The touch panel, also known as a touch screen, can collect touch operations on or near it (such as using any suitable object or accessory such as a finger, stylus, etc. on the touch panel or near the touch panel), and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into the contact coordinates, and then sends it to the processor, and can receive the command sent by the processor and execute it. In addition, the touch panel can be implemented in various categories such as resistive, capacitive, infrared and surface acoustic waves. In addition to the touch panel, the input unit may also include other input devices. Specifically, other input devices may include but are not limited to one or more of a physical keyboard, a function key (such as a volume control key, a switch key, etc.), a trackball, a mouse, a joystick, etc.

[0080] The display unit can be used to display the input information or the provided information and various menus of the device. The display unit may include a display panel, and optionally, the display panel may be configured in the form of a liquid crystal display (Liquid Crystal Display, LCD), an organic light-emitting diode (Organic Light-Emitting Diode, OLED), etc. Further, the touch panel may cover the display panel, and when the touch panel detects a touch operation on or near it, it is transmitted to the processor to determine the category of the touch event, and then the processor provides a corresponding visual output on the display panel according to the category of the touch event. Although the touch panel and the display panel are implemented as two independent components to implement the input and output functions of the device, in some embodiments, the touch panel and the display panel can be integrated to implement the input and output functions of the device.

[0081] The device may also include at least one sensor, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel according to the brightness of the ambient light, and the proximity sensor may turn off the display panel and / or backlight when the device is moved to the ear. As a type of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that identify the posture of the device (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors that can be configured on the device, such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., they will not be repeated here.

[0082] The audio circuit, speaker, and microphone can provide an audio interface. The audio circuit can convert the received audio data into an electrical signal and transmit it to the speaker, which converts it into a sound signal for output; on the other hand, the microphone converts the collected sound signal into an electrical signal, which is received by the audio circuit and converted into audio data. The audio data is then processed by the output processor and sent to another device through the RF circuit, or the audio data is output to a memory for further processing.

[0083] WiFi is a short-range wireless transmission technology. Devices can send and receive emails, browse web pages, and access streaming media through WiFi modules. It provides wireless broadband Internet access. WiFi modules are not a necessary component of the device and can be omitted as needed without changing the essence of the invention.

[0084] The processor 110 is the control center of the device. It uses various interfaces and lines to connect various parts of the entire device. It executes various functions of the device and processes data by running or executing software programs and / or modules stored in the memory, and calling data stored in the memory, so as to monitor the device as a whole. Optionally, the processor may include one or more processing units; preferably, the processor may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, operating interface, and application programs, and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor.

[0085] The device also includes a power source (such as a battery) for supplying power to each component. Preferably, the power source can be logically connected to the processor through a power management system, so that the power management system can manage charging, discharging, power consumption and other functions.

[0086] Although not shown, the device may also include a camera, a Bluetooth module, etc., which will not be described in detail here.

[0087] Based on the above-mentioned adrenal functional tumor typing and positioning device, various embodiments of the adrenal functional tumor typing and positioning method of the present invention are proposed below.

[0088] like Figure 2 As shown, Figure 2 It is a flow chart of a method for categorizing and localizing adrenal functional tumors provided by an embodiment of the present invention. The method for categorizing and localizing adrenal functional tumors is applied to the above-mentioned architecture platform, and the method for categorizing and localizing adrenal functional tumors includes but is not limited to step S100, step S200, step S300, step S400, step S500, step S600, step S700 and step S800.

[0089] Step S100, inputting the pre-processed adrenal CT data into the first nnUnet network for learning and segmentation processing to obtain kidney mask data;

[0090] Step S200, inputting the kidney mask data and the adrenal gland CT data into the second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor;

[0091] Step S300, inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map;

[0092] Step S400, constructing a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer;

[0093] Step S500, comparing the mean vector and the standard deviation vector with the mean and standard deviation of the normal distribution through KL divergence to obtain a difference result;

[0094] Step S600, training the network according to the difference result to obtain the kidney image analysis result.

[0095] In one embodiment, the pre-processed adrenal CT data is first input into the first nnUnet network for learning segmentation processing to obtain kidney mask data, and then the kidney mask data and adrenal CT data are input into the second nnUnet network for learning segmentation processing to obtain the first segmentation result of the adrenal tumor, and then the first segmentation result is input into the nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map, and then the latent variable feature map is constructed into a mean vector and a standard deviation vector through a fully connected layer, and then the mean vector and the standard deviation vector are compared with the mean and standard deviation values ​​of the normal distribution through KL divergence to obtain a difference result, and finally the network is trained according to the difference result to obtain a kidney image analysis result. In the technical solution of this embodiment, the proposed new deep learning convolutional network can complete the classification and positioning of adrenal functional tumors using only CT scan images, avoiding the interference of external factors in traditional endocrine detection and simplifying the inspection process. The nnUnet network can automatically adapt to any dataset without human intervention, while the variational autoencoder can ensure the robustness of the model trained on a small dataset and the accuracy of adrenal functional tumor image recognition.

[0096] It should be noted that the adrenal CT data includes arterial phase data, venous phase data and plain scan phase data, and may also include data from other phases, which are not limited in this embodiment.

[0097] It should be noted that the pre-processed adrenal CT data is sent to the nnUnet network to obtain the location of the kidney (kidney mask data), and the kidney mask data is used as a guide for the subsequent positioning of the adrenal glands and tumors. The ultimate goal of using the algorithm model is to segment and classify functional adrenal tumors. The first step is to obtain the location of the kidney area to guide the model to reduce the input data size for subsequent identification of adrenal tumors. By first obtaining the kidney area, there is no need to analyze the entire CT image. When identifying adrenal tumors later, only the kidney area obtained in this step needs to be analyzed, thereby improving the accuracy and efficiency of AI analysis. It can be understood that nnUnet itself is a segmentation model. The result of inputting adrenal CT data into the nnUnet model is the kidney position (kidney mask data).

[0098] It should be noted that nnUnet is composed of an encoder and a decoder. The encoder includes four downsampling operations and constructs four feature maps. The decoder performs four consecutive upsamplings starting from the feature map of the deepest layer of the network and will be jump-connected with the encoder of the same layer depth. The encoder and decoder in nnUnet may also include other functional modules, which are not limited to them in this embodiment.

[0099] It should be noted that constructing a mean vector and a standard deviation vector from a latent variable feature map through a fully connected layer can be to construct a 1024-dimensional mean vector and a 1024-dimensional standard deviation vector from the latent variable feature map through a fully connected layer. This embodiment does not make a specific and unique limitation on the number of dimensions of the constructed mean vector and standard deviation vector.

[0100] In one embodiment, the pre-processed adrenal CT data is sent to the nnUnet network to obtain the position of the kidney (kidney mask data) as a guide for the subsequent positioning of the adrenal glands and tumors, and then the kidney mask data obtained by the previous segmentation step and the original input adrenal CT data are input into another nnUnet network together to obtain a rough segmentation result of the adrenal tumor. The model will extract the cubic area containing the complete adrenal tumor obtained in the previous stage and send it to the nnUnet network with a variational autoencoder VAE, referred to as nnUnet-VAE. In the VAE branch, the feature map generated by the deepest network is called the latent variable, and constraints are added to the encoder so that the latent variables generated by the network obey the normal distribution, and then the KL divergence is used to measure the difference between the two vectors and the mean and standard deviation of the normal distribution and train the nnUnet-VAE network. Thus, the kidney image analysis results can be obtained through training. In the technical solution of this embodiment, the proposed new deep learning convolutional network can complete the classification and positioning of adrenal functional tumors using only CT scan images, avoiding the interference of external factors in traditional endocrine testing and simplifying the inspection process. The nnUnet network can automatically adapt to any data set without manual intervention, and the variational autoencoder can ensure the robustness of the model training under small data sets and the accuracy of adrenal functional tumor image recognition.

[0101] It should be noted that the formula of the loss function set in the first nnUnet network or the second nnUnet network includes:

[0102] L nnUnet =L dc +L ce ;

[0103] Among them, L ce is the cross entropy loss function, L ce is the diceloss function for three types of functional adrenal tumors.

[0104] It should be noted that the formula of the loss function set in the nnUnet-VAE network includes:

[0105] L nnUnet-vae =L nnUnet +L cls +L vae ;

[0106] L vae =L KL +L MSE ;

[0107] Among them, L cls is the cross entropy loss function for classification, L KL represents the KL divergence, L MSE It is the mean square error loss function between the model input image and the output of the nnUnet-VAE network.

[0108] In one embodiment, the device can also visualize the kidney image indication result to obtain an indication image, and display the indication image on the screen. The indication image includes the image information generated by the preprocessed adrenal CT data and indication information corresponding to the image information, and the indication information represents the indication result for providing the user with a decision on adrenal functional tumor.

[0109] Reference Figure 3 In one embodiment, the process before step S100 includes but is not limited to step S310 and step S320.

[0110] Step S310, acquiring adrenal CT data, the CT data including arterial phase data, venous phase data and plain scan phase data;

[0111] Step S320 , preprocessing the arterial phase data, the venous phase data, and the plain scan phase data to obtain preprocessed arterial phase data, venous phase data, and plain scan phase data.

[0112] Specifically, the CT data of each adrenal patient is mainly divided into three phases: arterial phase, venous phase and plain phase. Before training the data, the arterial phase data, venous phase data and plain phase data can be preprocessed to obtain preprocessed arterial phase data, venous phase data and plain phase data that meet the training requirements, which can improve the accuracy of model training.

[0113] Reference Figure 4 In one embodiment, step S320 includes but is not limited to step S410 and step S420.

[0114] Step S410, aligning the origins of the arterial phase data, the venous phase data, and the plain scan phase data;

[0115] Step S420 , normalizing the voxel values ​​and voxel spacings of the arterial phase data, venous phase data, and plain scan phase data after the source point alignment processing to obtain pre-processed arterial phase data, venous phase data, and plain scan phase data.

[0116] Specifically, the pulse phase data, venous phase data and plain scan phase data are firstly processed for origin alignment, and then the voxel values ​​and voxel spacing of the three phase data are normalized to make the voxel spacing equal.

[0117] It should be noted that voxels are used to describe three-dimensional images, and their concept is similar to that of pixels in two-dimensional images. Normalization is performed on three-phase CT images of the same patient. The voxel spacing between phases may not be equal, and the voxel spacing needs to be equalized.

[0118] Based on the above adrenal gland function tumor classification and positioning method, various embodiments of the adrenal gland function tumor classification and positioning method device, controller and computer-readable storage medium of the present invention are respectively proposed below.

[0119] Reference Figure 5 An embodiment of the present invention further provides a pedestrian re-identification device based on posture guidance, comprising:

[0120] A first segmentation module 510 is used to input the pre-processed adrenal CT data into a first nnUnet network for learning and segmentation processing to obtain kidney mask data;

[0121] The second segmentation module 520 is used to input the kidney mask data and the adrenal gland CT data into the second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor;

[0122] A generation module 530 is used to input the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map;

[0123] A construction module 540 is used to construct a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer;

[0124] A comparison module 550 is used to compare the mean vector and the standard deviation vector with the mean and standard deviation values ​​of the normal distribution through KL divergence to obtain a difference result;

[0125] The training module 560 is used to train the network according to the difference result to obtain the judgment result.

[0126] In one embodiment, it further includes an acquisition module, the acquisition module is used to acquire CT data of the adrenal gland, the CT data including arterial phase data, venous phase data and plain scan phase data;

[0127] The arterial phase data, the venous phase data and the plain scan phase data are preprocessed to obtain the preprocessed arterial phase data, the venous phase data and the plain scan phase data.

[0128] In one embodiment, the acquisition module is also used to align the origins of the arterial phase data, venous phase data, and plain phase data; the voxel values ​​and voxel spacing of the arterial phase data, venous phase data, and plain phase data after the source point alignment are normalized to obtain the pre-processed arterial phase data, venous phase data, and plain phase data.

[0129] In one embodiment, the construction module is further used to construct a 1024-dimensional mean vector and a 1024-dimensional standard deviation vector from the latent variable feature map through a fully connected layer.

[0130] It should be noted that the various embodiments of the above-mentioned adrenal functional tumor classification and positioning device are consistent with the technical means used, the technical problems solved, and the technical effects achieved in the embodiments of the adrenal functional tumor classification and positioning method, and will not be described in detail here. For details, please refer to the embodiments of the adrenal functional tumor classification and positioning method.

[0131] In this embodiment, the processor included in the terminal device can execute the adrenal functional tumor typing and positioning method of the previous embodiment.

[0132] In addition, one embodiment of the present invention further provides a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are used to execute the above-mentioned adrenal gland function tumor classification and positioning method on the terminal side, for example, the above-described Figure 2 The method steps S100 to S600, Figure 3 Method steps S310 to S320 and Figure 4 Method steps S410 to S420 in.

[0133] It will be appreciated by those skilled in the art that all or some of the steps and systems in the disclosed method above may be implemented as software, firmware, hardware and appropriate combinations thereof. Some physical components or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor or a microprocessor, or may be implemented as hardware, or may be implemented as an integrated circuit, such as an application specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or a non-transitory medium) and a communication medium (or a temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, disk storage or other magnetic storage devices, or any other medium that may be used to store desired information and may be accessed by a computer. Furthermore, it is well known to those skilled in the art that communication media typically include computer readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0134] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above-mentioned implementation mode. Technical personnel familiar with the field can also make various equivalent deformations or substitutions under the shared conditions without violating the spirit of the present invention. These equivalent deformations or substitutions are all included in the scope defined by the claims of the present invention.

Claims

1. Adrenal gland functional tumor typing and positioning equipment, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the method implemented when the processor executes the computer program includes: The pre-processed adrenal CT data is input into the first nnUnet network for learning and segmentation processing to obtain kidney mask data; Inputting the kidney mask data and the adrenal gland CT data into a second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor; Inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map; The latent variable feature map is constructed into a mean vector and a standard deviation vector through a fully connected layer; Compare the mean vector and the standard deviation vector with the mean and standard deviation values ​​of a normal distribution by KL divergence to obtain a difference result; The network is trained according to the difference results to obtain kidney image analysis results; The first nnUnet network and the second nnUnet network include an encoder and a decoder jump-connected to the encoder, the encoder is used to perform multiple downsampling operations and construct multiple feature graphs, and the decoder is used to perform multiple consecutive upsampling operations starting from the feature graph of the deepest network.

2. The adrenal gland functional tumor typing and positioning device according to claim 1, characterized in that: Before the processor executes the method of inputting the pre-processed adrenal CT data into the first nnUnet network for learning and segmentation processing to obtain kidney mask data, the method executed by the processor includes: Acquire adrenal CT data, wherein the CT data includes arterial phase data, venous phase data, and plain scan phase data; The arterial phase data, the venous phase data and the plain scan phase data are preprocessed to obtain the preprocessed arterial phase data, the venous phase data and the plain scan phase data.

3. The adrenal gland functional tumor typing and positioning device according to claim 2, characterized in that: The method of preprocessing the arterial phase data, the venous phase data and the plain scan phase data to obtain the preprocessed arterial phase data, the venous phase data and the plain scan phase data, performed by the processor, comprises: Aligning the origins of the arterial phase data, the venous phase data, and the plain scan phase data; The voxel values ​​and voxel spacings of the arterial phase data, the venous phase data and the plain scan phase data after origin alignment are normalized to obtain the pre-processed arterial phase data, the venous phase data and the plain scan phase data.

4. The adrenal gland functional tumor typing and positioning device according to claim 1, characterized in that: The method executed by the processor for constructing a mean vector and a standard deviation vector from a latent variable feature map through a fully connected layer includes: The latent variable feature map is constructed into a 1024-dimensional mean vector and a 1024-dimensional standard deviation vector through a fully connected layer.

5. The adrenal gland functional tumor typing and positioning device according to claim 1, characterized in that: The method executed by the processor also includes: visualizing the kidney image analysis result and displaying it on a screen.

6. A device for locating and typing adrenal gland functional tumors, characterized in that: include: The first segmentation module is used to input the preprocessed adrenal CT data into the first nnUnet network for learning and segmentation processing to obtain kidney mask data; A second segmentation module is used to input the kidney mask data and the adrenal gland CT data into a second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor; A generation module, used for inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map; A construction module is used to construct a mean vector and a standard deviation vector from the latent variable feature map through a fully connected layer; A comparison module is used to compare the mean vector and the standard deviation vector with the mean and standard deviation values ​​of the normal distribution through KL divergence to obtain the difference results; A training module, used for training the network according to the difference results to obtain kidney image analysis results; The first nnUnet network and the second nnUnet network include an encoder and a decoder jump-connected to the encoder, the encoder is used to perform multiple downsampling operations and construct multiple feature graphs, and the decoder is used to perform multiple consecutive upsampling operations starting from the feature graph of the deepest network.

7. A computer-readable storage medium storing a computer program, wherein the computer program is used to execute a method comprising: The pre-processed adrenal CT data is input into the first nnUnet network for learning and segmentation processing to obtain kidney mask data; Inputting the kidney mask data and the adrenal gland CT data into a second nnUnet network for learning and segmentation processing to obtain a first segmentation result of the adrenal gland tumor; Inputting the first segmentation result into a nnUnet-VAE network provided with a variational autoencoder to generate a latent variable feature map; The latent variable feature map is constructed into a mean vector and a standard deviation vector through a fully connected layer; Compare the mean vector and the standard deviation vector with the mean and standard deviation values ​​of a normal distribution by KL divergence to obtain a difference result; The network is trained according to the difference results to obtain kidney image analysis results; The first nnUnet network and the second nnUnet network include an encoder and a decoder jump-connected to the encoder, the encoder is used to perform multiple downsampling operations and construct multiple feature graphs, and the decoder is used to perform multiple consecutive upsampling operations starting from the feature graph of the deepest network.

Images