An infection detection method, device, equipment and medium based on a multi-stage model

Abstract

The application provides an infection detection method and device based on a multi-stage model, equipment and a medium, which are used for judging whether an infection exists in a detection site, comprising: acquiring a three-dimensional image to be detected, and inputting the three-dimensional image to be detected into a pre-trained infection detection model, wherein the infection detection model comprises a pre-trained site detection model and a pre-trained infection classification model; acquiring a first predicted position of the detection site in the three-dimensional image to be detected by using the site detection model, and acquiring a three-dimensional ROI image to be detected according to the first predicted position; identifying the three-dimensional ROI image to be detected by using the infection classification model, judging whether an infection exists in the detection site, and outputting a judgment result. After the above technical scheme is used, the multi-stage model can first extract a region of interest (ROI) image, and then perform infection classification on the region of interest image, so as to avoid mutual influence of model parameters in different steps, and realize high-accuracy infection detection.

Description

Technical Field

[0001] The present invention relates to the field of medical imaging, and in particular to an infection detection method, device, equipment and medium based on a multi-stage model. Background Art

[0002] Spinal infection is a rare but dangerous disease. Its pathogens can be transmitted to the spine through blood, lymphatic vessels, trauma or surgery, leading to infectious diseases such as osteomyelitis, spinal meningitis, and discitis. In recent years, the prevalence of nonspecific spinal infection has gradually increased and has developed into a global health problem. Clinically, early diagnosis and treatment are crucial to preventing further development of the disease and the recovery of patients. However, the symptoms of spinal infection are atypical and easily confused with other orthopedic diseases, such as tumors and spinal tuberculosis. In addition, conventional examination methods such as X-rays and CT scans are often unable to accurately identify early spinal infections, so an accurate and rapid method is needed to assist in diagnosis.

[0003] With the rapid development of deep learning technology, the use of computer vision methods for medical image recognition has become a hot research topic. In recent years, the application of deep learning algorithms in medical image recognition has made significant progress, particularly in tumor diagnosis and orthopedic disease diagnosis. Deep learning algorithms have also been widely used in the diagnosis of spinal infections, and have demonstrated excellent performance in some studies. Currently, some researchers have begun to investigate spinal infection recognition methods based on deep learning algorithms, such as using convolutional neural networks (CNNs) and recurrent neural networks (RNNs) for image analysis to achieve rapid and accurate spinal infection diagnosis. However, these methods typically require large amounts of medical image data for training and require appropriate data preprocessing and feature extraction methods. Furthermore, these methods require further optimization to improve their accuracy and robustness across diverse datasets. Furthermore, consideration is needed to determine how these deep learning algorithms can be applied in real-world clinical settings to help physicians quickly and accurately diagnose spinal infections. Summary of the Invention

[0004] In order to overcome the above technical defects, the purpose of the present invention is to provide an infection detection method, device, equipment and medium based on a multi-stage model for determining whether there is an infection in the part to be inspected.

[0005] The present invention discloses an infection detection method based on a multi-stage model, comprising the following steps:

[0006] Acquire a three-dimensional image to be inspected, and input the three-dimensional image to be inspected into a pre-trained infection detection model, wherein the infection detection model includes a site detection model and an infection classification model;

[0007] Using the part detection model to obtain a first predicted position of the part to be inspected within the three-dimensional image to be inspected, and obtaining a three-dimensional ROI image to be inspected according to the first predicted position;

[0008] The infection classification model is used to identify the three-dimensional ROI image to be inspected, determine whether there is infection in the part to be inspected, and output the judgment result.

[0009] Preferably, the infection detection method further comprises:

[0010] The three-dimensional image to be inspected is composed of a plurality of two-dimensional images to be inspected, the part detection model is used to obtain a second predicted position where the part to be inspected is located in any of the plurality of two-dimensional images to be inspected, a plurality of two-dimensional ROI images to be inspected are obtained based on the plurality of second predicted positions, and the plurality of two-dimensional ROI images to be inspected constitute the three-dimensional ROI image to be inspected;

[0011] Using the infection classification model to identify the multiple two-dimensional ROI images to be inspected, and obtaining a single-layer predicted infection probability corresponding to any one of the multiple two-dimensional ROI images to be inspected;

[0012] Based on the predicted infection probabilities of the multiple single layers, it is determined whether there is infection in the part to be inspected.

[0013] Preferably, the infection detection method further comprises:

[0014] Acquire an initial prediction map for representing the corresponding relationship between the number of layers of the plurality of to-be-inspected two-dimensional ROI images and the plurality of single-layer predicted infection probabilities;

[0015] Using a median filtering method to eliminate abnormal points in the initial prediction graph to obtain a corrected prediction graph;

[0016] According to the correction prediction graph, a KNN clustering algorithm is used to calculate and obtain an adaptive threshold;

[0017] The adaptive threshold and the plurality of single-layer predicted infection probabilities are respectively compared to determine whether infection exists in the part to be inspected, and a determination result is output.

[0018] Preferably, the infection detection method further comprises:

[0019] When the judgment result is that there is infection in the part to be inspected, the number of layers corresponding to the two-dimensional ROI image to be inspected whose corresponding single-layer predicted infection probability is greater than the adaptive threshold is further output.

[0020] Preferably, the part detection model is pre-trained through the following steps:

[0021] Acquire a plurality of training two-dimensional images and a part detection model to be trained, mark a training position where the part to be inspected is located in any one of the plurality of training two-dimensional images, and input the plurality of training two-dimensional images into the part detection model to be trained;

[0022] Using the part detection model to be trained, obtaining a third predicted position of the part to be detected within any one of the plurality of training two-dimensional images;

[0023] Comparing the training position corresponding to any one of the plurality of training two-dimensional images with the third predicted position, respectively, obtaining a first loss function, and performing back propagation to optimize parameters of the part detection model to be trained;

[0024] Repeat the above steps until the first loss function converges to obtain the part detection model.

[0025] Preferably, the training step further comprises:

[0026] Performing data enhancement processing on the plurality of training two-dimensional images, wherein the data enhancement processing includes horizontal mirroring, random rotation, random brightness, random contrast, and random scale transformation;

[0027] The plurality of training two-dimensional images after data enhancement processing are normalized and input into the part detection model to be trained.

[0028] Preferably, the training step further comprises:

[0029] Acquire a plurality of test two-dimensional images and input them into the part detection model, and acquire a test position where the part to be inspected is located in any one of the plurality of test two-dimensional images;

[0030] Using the part detection model to obtain a fourth predicted position where the part to be inspected is located in any one of the plurality of training two-dimensional images;

[0031] Compare the test position and the fourth predicted position corresponding to any one of the multiple test two-dimensional images respectively, obtain some test two-dimensional images as difficult sample images, add the multiple difficult sample images to the training two-dimensional image, and use the FocalLoss function to update the first loss function.

[0032] The present invention also discloses an infection detection device based on a multi-stage model, comprising:

[0033] A preprocessing module, which obtains a three-dimensional image to be inspected and inputs the three-dimensional image to be inspected into a pre-trained infection detection model, wherein the infection detection model includes a site detection model and an infection classification model;

[0034] a position prediction module, which uses the part detection model to obtain a first predicted position of the part to be inspected within the three-dimensional image to be inspected, and obtains a three-dimensional ROI image to be inspected based on the first predicted position;

[0035] The infection classification module uses the infection classification model to identify the three-dimensional ROI image to be inspected, determines whether there is infection in the part to be inspected, and outputs a determination result.

[0036] The present invention also discloses an electronic device, characterized in that the electronic device includes a memory storing computer-executable instructions and a processor, and when the instructions are executed by the processor, the electronic device implements the infection detection method.

[0037] The present invention also discloses a computer-readable storage medium on which a computer program is stored, characterized in that when the instruction is run on a computer, the computer is enabled to execute the infection detection method.

[0038] Compared with the existing technology, the above technical solution has the following beneficial effects:

[0039] 1. Build a multi-stage infection detection model. Using different networks, we first extract region of interest (ROI) images and then classify them for infection. This avoids the mutual influence of model parameters at different steps, resulting in higher accuracy than single-stage infection detection models.

[0040] 2. Use the median filter method and KNN clustering algorithm to post-process the prediction results of the infection classification model to eliminate abnormal values, which can effectively prevent false positives in the infection detection model;

[0041] 3. By performing data augmentation on the training two-dimensional images and using difficult sample images to update the part detection model, the image size required for training can be reduced, solving the problems of insufficient training data and high labeling costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 A schematic flow chart of an infection detection method based on a multi-stage model disclosed in the present invention;

[0043] Figure 2 A schematic diagram of an initial prediction graph disclosed in the present invention;

[0044] Figure 3 This is a schematic diagram of a correction prediction graph disclosed in the present invention. DETAILED DESCRIPTION

[0045] The advantages of the present invention are further described below with reference to the accompanying drawings and specific embodiments.

[0046] Exemplary embodiments will be described in detail herein, with examples illustrated in the accompanying drawings. In the following description, when referring to the drawings, identical numerals in different figures represent identical or similar elements, unless otherwise indicated. The embodiments described in the following exemplary embodiments are not intended to represent all possible embodiments consistent with the present disclosure. Rather, they are merely examples of apparatus and methods consistent with certain aspects of the present disclosure, as detailed in the appended claims.

[0047] The terms used in this disclosure are for the purpose of describing specific embodiments only and are not intended to limit the disclosure. As used in this disclosure and the appended claims, the singular forms "a," "an," "the," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and encompasses any and all possible combinations of one or more of the associated listed items.

[0048] It should be understood that although the terms first, second, third, etc. may be used in this disclosure to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of this disclosure, first information may also be referred to as second information, and similarly, second information may also be referred to as first information. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining."

[0049] In the description of the present invention, it should be understood that the terms "longitudinal", "transverse", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention.

[0050] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal communication between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.

[0051] In the following description, the suffixes such as "module", "component" or "unit" used to represent elements are only used to facilitate the description of the present invention and have no specific meaning. Therefore, "module" and "component" can be used interchangeably.

[0052] See also Figure 1 As shown, the present invention discloses an infection detection method based on a multi-stage model, comprising the following steps:

[0053] S100, obtaining a three-dimensional image to be inspected, and inputting the three-dimensional image to be inspected into a pre-trained infection detection model, wherein the infection detection model includes a site detection model and an infection classification model;

[0054] S200, using the part detection model to obtain a first predicted position of the part to be inspected within the three-dimensional image to be inspected, and obtaining a three-dimensional ROI image to be inspected based on the first predicted position;

[0055] S300: Using the infection classification model to identify the three-dimensional ROI image to be inspected, determine whether there is infection in the part to be inspected, and output a determination result.

[0056] Preferably, the infection detection method further comprises:

[0057] The three-dimensional image to be inspected is composed of a plurality of two-dimensional images to be inspected, the part detection model is used to obtain a second predicted position where the part to be inspected is located in any of the plurality of two-dimensional images to be inspected, a plurality of two-dimensional ROI images to be inspected are obtained based on the plurality of second predicted positions, and the plurality of two-dimensional ROI images to be inspected constitute the three-dimensional ROI image to be inspected;

[0058] Using the infection classification model to identify the multiple two-dimensional ROI images to be inspected, and obtaining a single-layer predicted infection probability corresponding to any one of the multiple two-dimensional ROI images to be inspected;

[0059] Based on the predicted infection probabilities of the multiple single layers, it is determined whether there is infection in the part to be inspected.

[0060] Specifically, a part detection model is used to predict multiple two-dimensional images to be inspected, and multiple second predicted positions are obtained. Based on the multiple second predicted positions, multiple two-dimensional ROI images to be inspected are obtained. For example, for any two-dimensional image to be inspected, a number of pixels are expanded in each dimension of the length and width corresponding to the second predicted position to obtain a region of interest. The portion outside the region of interest in the two-dimensional image to be inspected is regarded as noise and cropped to obtain a two-dimensional ROI image to be inspected that only contains the region of interest. Repeat the above steps to obtain multiple two-dimensional ROI images to be inspected, and splice the multiple two-dimensional ROI images to be inspected along the Z-axis direction and circumscribe the rectangular boundary to form a three-dimensional ROI image to be inspected. The infection classification model is used to identify the three-dimensional ROI image to be inspected, that is, to identify the multiple two-dimensional ROI images to be inspected therein, and obtain a single-layer predicted infection probability of any two-dimensional ROI image to be inspected. The infection classification model can adopt a deep neural network model with an EfficientNet structure.

[0061] Preferably, the infection detection method further comprises:

[0062] Acquire an initial prediction map for representing the corresponding relationship between the number of layers of the plurality of to-be-inspected two-dimensional ROI images and the plurality of single-layer predicted infection probabilities;

[0063] Using a median filtering method to eliminate abnormal points in the initial prediction graph to obtain a corrected prediction graph;

[0064] According to the correction prediction graph, a KNN clustering algorithm is used to calculate and obtain an adaptive threshold;

[0065] The adaptive threshold and the plurality of single-layer predicted infection probabilities are respectively compared to determine whether infection exists in the part to be inspected, and a determination result is output.

[0066] Specifically, if Figure 2 As shown in the figure, the initial prediction map is constructed with the layer number corresponding to any two-dimensional ROI image to be inspected as the horizontal axis and the corresponding single-layer predicted infection probability as the vertical axis. The layer number represents the order in which the two-dimensional ROI image to be inspected is spliced ​​along the Z axis in the corresponding three-dimensional ROI image to be inspected. The median filter method is used to eliminate outliers in the initial prediction map, and the following is obtained: Figure 3 Corrected prediction plot shown.

[0067] For example Figure 3The corrected prediction graph shown uses a KNN clustering algorithm to calculate an adaptive threshold based on the corrected prediction graph. The adaptive threshold is then compared with the multiple single-layer predicted infection probabilities to determine whether the area to be inspected is infected. For example, a first threshold representing the minimum infection determination probability and a second threshold representing the minimum number of infection determination images can be pre-set. When the adaptive threshold is greater than the first threshold, and the number of the corresponding two-dimensional ROI images to be inspected with a single-layer predicted infection probability greater than the adaptive threshold is greater than the second threshold, the area to be inspected is determined to be infected.

[0068] Preferably, the infection detection method further comprises:

[0069] When the judgment result is that there is infection in the part to be inspected, the number of layers corresponding to the two-dimensional ROI image to be inspected whose corresponding single-layer predicted infection probability is greater than the adaptive threshold is further output.

[0070] Preferably, the part detection model is pre-trained through the following steps:

[0071] Acquire a plurality of training two-dimensional images and a part detection model to be trained, mark a training position where the part to be inspected is located in any one of the plurality of training two-dimensional images, and input the plurality of training two-dimensional images into the part detection model to be trained;

[0072] Using the part detection model to be trained, obtaining a third predicted position of the part to be detected within any one of the plurality of training two-dimensional images;

[0073] Comparing the training position corresponding to any one of the plurality of training two-dimensional images with the third predicted position, respectively, obtaining a first loss function, and performing back propagation to optimize parameters of the part detection model to be trained;

[0074] Repeat the above steps until the first loss function converges to obtain the part detection model.

[0075] Specifically, the multiple training two-dimensional images are medical images that have been diagnosed with infection and contain the part to be inspected. For example, for spinal infection detection, multiple transverse spinal CT images with infection are used as training two-dimensional images, and the position of the spine in any training two-dimensional image is marked as the training position. The part detection model can select a YOLO network model, randomly initialize the parameters of the YOLO network model to be trained, use the YOLO network model to predict any training two-dimensional image, and obtain the third predicted position of the spine in any training two-dimensional image. Compare the training position and the third predicted position corresponding to any one of the multiple training two-dimensional images respectively, obtain the first loss function, and perform back propagation to optimize the parameters of the YOLO network model to be trained.

[0076] Preferably, the training step further comprises:

[0077] Performing data enhancement processing on the plurality of training two-dimensional images, wherein the data enhancement processing includes horizontal mirroring, random rotation, random brightness, random contrast, and random scale transformation;

[0078] The plurality of training two-dimensional images after data enhancement processing are normalized and input into the part detection model to be trained.

[0079] Preferably, the training step further comprises:

[0080] Acquire a plurality of test two-dimensional images and input them into the part detection model, and acquire a test position where the part to be inspected is located in any one of the plurality of test two-dimensional images;

[0081] Using the part detection model to obtain a fourth predicted position where the part to be inspected is located in any one of the plurality of training two-dimensional images;

[0082] Compare the test position and the fourth predicted position corresponding to any one of the multiple test two-dimensional images respectively, obtain some test two-dimensional images as difficult sample images, add the multiple difficult sample images to the training two-dimensional image, and use the FocalLoss function to update the first loss function.

[0083] Specifically, the multiple test two-dimensional images are medical images that have been diagnosed with infection and contain the part to be tested. For example, for spinal infection detection, multiple transverse spinal CT images with infection that are completely different from the above-mentioned training two-dimensional images are used as test two-dimensional images, and the position of the spine in the test two-dimensional image is marked as the test position. The YOLO network model trained and converged through the above steps is used to predict any test two-dimensional image to obtain the fourth predicted position of the spine in any test two-dimensional image. The deviations between the test position and the fourth predicted position corresponding to any of the multiple test two-dimensional images are compared respectively, and some test two-dimensional images with larger deviations are obtained as difficult sample images. The training two-dimensional images are expanded using the difficult sample images, and the YOLO network model is continued to be trained. The FocalLoss function is used instead of the first loss function to enhance the YOLO network model's ability to recognize difficult samples.

[0084] The present invention also discloses an infection detection device based on a multi-stage model, comprising:

[0085] A preprocessing module, which obtains a three-dimensional image to be inspected and inputs the three-dimensional image to be inspected into a pre-trained infection detection model, wherein the infection detection model includes a site detection model and an infection classification model;

[0086] a position prediction module, which uses the part detection model to obtain a first predicted position of the part to be inspected within the three-dimensional image to be inspected, and obtains a three-dimensional ROI image to be inspected based on the first predicted position;

[0087] The infection classification module uses the infection classification model to identify the three-dimensional ROI image to be inspected, determines whether there is infection in the part to be inspected, and outputs a determination result.

[0088] The present invention also discloses an electronic device, characterized in that the electronic device includes a memory storing computer-executable instructions and a processor, and when the instructions are executed by the processor, the electronic device implements the infection detection method.

[0089] The present invention also discloses a computer-readable storage medium on which a computer program is stored, characterized in that when the instruction is run on a computer, the computer is enabled to execute the infection detection method.

[0090] It should be noted that the embodiments of the present invention have better practicability and do not impose any form of limitation on the present invention. Any technician familiar with the field may use the technical content disclosed above to change or modify it into an equivalent effective embodiment. However, any modification or equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solution of the present invention are still within the scope of the technical solution of the present invention.

Claims

1. A multi-stage model-based infection detection method for determining whether an area to be inspected has an infection, characterized in that: The following steps are involved: Acquire a three-dimensional image to be inspected, and input the three-dimensional image to be inspected into a pre-trained infection detection model, wherein the infection detection model includes a pre-trained site detection model and a pre-trained infection classification model; Using the part detection model to obtain a first predicted position of the part to be inspected within the three-dimensional image to be inspected, and obtaining a three-dimensional ROI image to be inspected according to the first predicted position; Using the infection classification model to identify the three-dimensional ROI image to be inspected, determine whether there is infection in the part to be inspected, and output the determination result; The infection detection method further includes: the three-dimensional image to be inspected is composed of a plurality of two-dimensional images to be inspected, using the part detection model to obtain a second predicted position where the part to be inspected is located within any one of the plurality of two-dimensional images to be inspected, obtaining a plurality of two-dimensional ROI images to be inspected based on the plurality of second predicted positions, and the plurality of two-dimensional ROI images to be inspected constitute the three-dimensional ROI image to be inspected; Using the infection classification model to identify the multiple two-dimensional ROI images to be inspected, and obtaining a single-layer predicted infection probability corresponding to any one of the multiple two-dimensional ROI images to be inspected; Determining whether there is infection in the part to be inspected based on the multiple single-layer predicted infection probabilities; Infection detection methods also include: Obtaining an initial prediction map for representing the correspondence between the number of layers of the multiple two-dimensional ROI images to be inspected and the multiple single-layer predicted infection probabilities, including constructing the initial prediction map with the layer number corresponding to any two-dimensional ROI image to be inspected as the abscissa and the corresponding single-layer predicted infection probability as the ordinate, where the layer number represents the order in which the two-dimensional ROI image to be inspected is spliced ​​along the Z-axis direction in its corresponding three-dimensional ROI image to be inspected; Using a median filtering method to eliminate abnormal points in the initial prediction graph to obtain a corrected prediction graph; According to the correction prediction graph, a KNN clustering algorithm is used to calculate and obtain an adaptive threshold; The adaptive threshold and the plurality of single-layer predicted infection probabilities are respectively compared to determine whether infection exists in the part to be inspected, and a determination result is output.

2. The infection detection method according to claim 1, characterized in that include: When the judgment result is that there is infection in the part to be inspected, the number of layers corresponding to the two-dimensional ROI image to be inspected whose corresponding single-layer predicted infection probability is greater than the adaptive threshold is further output.

3. The infection detection method according to claim 1, characterized in that The part detection model is pre-trained by the following steps: Acquire a plurality of training two-dimensional images and a part detection model to be trained, mark a training position where the part to be inspected is located in any one of the plurality of training two-dimensional images, and input the plurality of training two-dimensional images into the part detection model to be trained; Using the part detection model to be trained, obtaining a third predicted position of the part to be detected within any one of the plurality of training two-dimensional images; Comparing the training position corresponding to any one of the plurality of training two-dimensional images with the third predicted position, respectively, obtaining a first loss function, and performing back propagation to optimize parameters of the part detection model to be trained; Repeat the above steps until the first loss function converges to obtain the pre-trained part detection model.

4. The infection detection method according to claim 3, characterized in that: The training step further comprises: Performing data enhancement processing on the plurality of training two-dimensional images, wherein the data enhancement processing includes horizontal mirroring, random rotation, random brightness, random contrast, and random scale transformation; The plurality of training two-dimensional images after data enhancement processing are normalized and input into the part detection model to be trained.

5. The infection detection method according to claim 3, characterized in that: The training step further comprises: Acquire a plurality of test two-dimensional images and input them into the part detection model, and acquire a test position where the part to be inspected is located in any one of the plurality of test two-dimensional images; Using the part detection model to obtain a fourth predicted position where the part to be inspected is located in any one of the plurality of training two-dimensional images; Compare the test position and the fourth predicted position corresponding to any one of the multiple test two-dimensional images respectively, obtain some test two-dimensional images as difficult sample images, add the multiple difficult sample images to the training two-dimensional image, and use the FocalLoss function to update the first loss function.

6. An infection detection device based on a multi-stage model, used to determine whether an area to be inspected has an infection, characterized in that: include: A preprocessing module, which obtains a three-dimensional image to be inspected and inputs the three-dimensional image to be inspected into a pre-trained infection detection model, wherein the infection detection model includes a site detection model and an infection classification model; a position prediction module, which uses the part detection model to obtain a first predicted position of the part to be inspected within the three-dimensional image to be inspected, and obtains a three-dimensional ROI image to be inspected based on the first predicted position; An infection classification module, which uses the infection classification model to identify the three-dimensional ROI image to be inspected, determines whether there is infection in the part to be inspected, and outputs a judgment result; The infection detection device further includes: the three-dimensional image to be inspected is composed of a plurality of two-dimensional images to be inspected, the part detection model is used to obtain a second predicted position where the part to be inspected is located within any one of the plurality of two-dimensional images to be inspected, and a plurality of two-dimensional ROI images to be inspected are obtained based on the plurality of second predicted positions, wherein the plurality of two-dimensional ROI images to be inspected constitute the three-dimensional ROI image to be inspected; Using the infection classification model to identify the multiple two-dimensional ROI images to be inspected, and obtaining a single-layer predicted infection probability corresponding to any one of the multiple two-dimensional ROI images to be inspected; Determining whether there is infection in the part to be inspected based on the multiple single-layer predicted infection probabilities; The infection detection device also includes: Obtaining an initial prediction map for representing the correspondence between the number of layers of the multiple two-dimensional ROI images to be inspected and the multiple single-layer predicted infection probabilities, including constructing the initial prediction map with the layer number corresponding to any two-dimensional ROI image to be inspected as the abscissa and the corresponding single-layer predicted infection probability as the ordinate, where the layer number represents the order in which the two-dimensional ROI image to be inspected is spliced ​​along the Z-axis direction in its corresponding three-dimensional ROI image to be inspected; Using a median filtering method to eliminate abnormal points in the initial prediction graph to obtain a corrected prediction graph; According to the correction prediction graph, a KNN clustering algorithm is used to calculate and obtain an adaptive threshold; The adaptive threshold and the plurality of single-layer predicted infection probabilities are respectively compared to determine whether infection exists in the part to be inspected, and a determination result is output.

7. An electronic device, characterized in that: The electronic device includes a memory storing computer-executable instructions and a processor. When the instructions are executed by the processor, the electronic device implements the infection detection method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed on a computer, the computer is enabled to execute the infection detection method according to any one of claims 1 to 5.

Images