Classification method and classification device for abdominal cavity infectious effusion

Abstract

The invention discloses a classification method and a classification device for abdominal cavity infectious effusion. The classification method comprises the following steps: segmenting a plurality of non-enhanced CT images by using a pre-trained segmentation model; for each liquid accumulation region, applying a morphological dilation operation of a first size to produce a first lesion surrounding region, and performing a morphological dilation operation of a second size to produce a second lesion surrounding region; extracting a quantitative morphological feature, a 3D radiomics feature, a plurality of focus surrounding features, a plurality of local pathological features and a plurality of global context features; selecting a plurality of target features, and calculating a radiomics score of each non-enhanced CT image; inputting the radiomics score and the clinical parameters into a logistic regression model for training to obtain a trained logistic regression model; and using the trained logistic regression model to classify non-enhanced CT images to be classified. According to the method, the classification of the infectious hydrops in the abdominal cavity can be quickly and accurately realized at low cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence, and in particular to a classification method and device for intra-abdominal infectious fluid, a medium, an electronic device, and a computer program product. BACKGROUND

[0002] Accurate diagnosis of intra-abdominal infectious fluid (IAFCs) is challenging. The traditional “pneumobullae sign” on computed tomography (CT) as a diagnostic marker for infection has low sensitivity (18-56%), leading to delayed diagnosis of infectious cases, which can progress to sepsis and multiple organ failure. Moreover, the specificity of the pneumobullae sign is limited because of the introduction of iatrogenic gas or residual gas during drainage surgery, which can cause false positives. This diagnostic uncertainty leads to the widespread empirical use of antibiotics in clinical practice, despite evidence that prophylactic broad-spectrum antibiotics do not significantly reduce infection rates or improve mortality. This exacerbates antibiotic resistance, increases healthcare costs, and can produce adverse reactions.

[0003] Existing artificial intelligence research has limitations in this field, mainly manifested as single-center design, time-consuming manual region of interest (ROI) delineation, and lack of rigorous external validation, which severely limits its clinical applicability and universality. Unlike previous radiomic studies that required manual ROI delineation, existing technologies have not yet achieved fully automated, real-time classification and diagnosis of intra-abdominal infectious fluid, which is a barrier to clinical application. SUMMARY

[0004] The embodiments of the present application provide a classification method and device for intra-abdominal infectious fluid, a medium, an electronic device, and a computer program product.

[0005] In a first aspect, the embodiments of the present application provide a classification method for intra-abdominal infectious fluid, for an electronic device, the classification method comprising:

[0006] a segmentation step of segmenting, using a pre-trained segmentation model, a plurality of non-contrast CT images from a plurality of patients to segment a plurality of regions of interest in each non-contrast CT image, the plurality of regions of interest including a liquid accumulation region, a gas bubble region within the fluid collection, a pancreatic parenchyma region, and a gas bubble bed region, the gas bubble bed region defining a morphological dilation region around the accumulated gas bubbles;

[0007] a generation step of applying a morphological dilation operation of a first size to generate a first lesion surrounding region and a morphological dilation operation of a second size to generate a second lesion surrounding region for each liquid accumulation region, wherein the first size is smaller than the second size;

[0008] extracting quantitative morphological features, 3D radiomic features from the multiple regions of interest, extracting multiple per-lesion features from the first per-lesion region and the second per-lesion region, and extracting multiple local pathological features and multiple global contextual features from the liquid accumulation region for each non-contrast CT image;

[0009] selecting multiple target features from the quantitative morphological features, the 3D radiomic features, the multiple per-lesion features, the multiple local pathological features, the multiple global contextual features, and calculating a radiomic score of each non-contrast CT image;

[0010] training a logistic regression model by inputting the radiomic scores of the multiple non-contrast CT images and the clinical parameters of the multiple patients, and obtaining a trained logistic regression model;

[0011] classifying the non-contrast CT image to be classified using the trained logistic regression model.

[0012] In a second aspect, the present application provides a classification device for abdominal cavity infectious effusion, the classification device comprising:

[0013] a segmentation unit configured to segment multiple non-contrast CT images from multiple patients using a pre-trained segmentation model to segment multiple regions of interest in each non-contrast CT image, the multiple regions of interest comprising a liquid accumulation region, a gas bubble region in the effusion, a pancreatic parenchyma region, and a gas bubble bed region, the gas bubble bed region defining a morphological expansion region around the gathered gas bubbles;

[0014] a generating unit configured to, for each liquid accumulation region, apply a morphological dilation operation of a first size to generate a first per-lesion region and a morphological dilation operation of a second size to generate a second per-lesion region, wherein the first size is smaller than the second size;

[0015] an extracting unit configured to extract quantitative morphological features, 3D radiomic features from the multiple regions of interest, extract multiple per-lesion features from the first per-lesion region and the second per-lesion region, and extract multiple local pathological features and multiple global contextual features from the liquid accumulation region for each non-contrast CT image;

[0016] a selecting unit configured to select multiple target features from the quantitative morphological features, the 3D radiomic features, the multiple per-lesion features, the multiple local pathological features, the multiple global contextual features, and calculate a radiomic score of each non-contrast CT image.

[0017] a training unit configured to input the radiomic scores of the plurality of non-contrast CT images and the clinical parameters of the plurality of patients into a logistic regression model to train the logistic regression model, to obtain a trained logistic regression model;

[0018] a classification unit configured to use the trained logistic regression model to classify the non-contrast CT image to be classified.

[0019] In a third aspect, an embodiment of the present application provides a computer readable storage medium, the storage medium storing instructions, the instructions causing a computer to execute the method for classifying intraperitoneal infectious effusion according to any one of the first aspect when executed on the computer.

[0020] In a fourth aspect, an embodiment of the present application provides an electronic device, comprising: one or more processors; one or more memories; the one or more memories storing one or more programs, the one or more programs causing the electronic device to execute the method for classifying intraperitoneal infectious effusion according to the first aspect when executed by the one or more processors.

[0021] In a fifth aspect, an embodiment of the present application provides a computer program product comprising computer executable instructions for implementing the method for classifying intraperitoneal infectious effusion according to the first aspect when executed by a processor.

[0022] In the present application, quantitative morphological features, 3D radiomic features, a plurality of lesion surrounding features, a plurality of local pathological features, and a plurality of global context features are extracted from each non-contrast CT image, realizing extremely rich multi-dimensional feature extraction. Then, a plurality of target features are selected from the extracted multi-dimensional features, radiomic scores of each non-contrast CT image are calculated according to the target features, and finally, a logistic regression model is trained (constructed) based on the radiomic scores of each non-contrast CT image and the clinical parameters of the patient corresponding to each non-contrast CT image, to obtain a classification model. In this way, according to the output of the classification model, the classification of intraperitoneal infectious effusion can be realized quickly, accurately and at low cost.

[0023] In the present application, the features used to train the logistic regression model include features related to the bubble bed region. The bubble bed region can capture the inflammatory or necrotic microenvironment around the bubble, and these microenvironments can change the surrounding tissue density and texture pattern, thereby reflecting changes in tissue density, increased vascular permeability, inflammatory cell infiltration and protein denaturation. Therefore, the classification model trained by the present application can more accurately classify intraperitoneal infectious effusion. BRIEF DESCRIPTION OF DRAWINGS

[0024] Figure 1According to embodiments of the present application, a flow chart of a method for classifying abdominal infectious fluid is shown.

[0025] Figure 2 According to embodiments of the present application, a structural diagram of a device for classifying abdominal infectious fluid is shown.

[0026] Figure 3 According to embodiments of the present application, a block diagram of an electronic device is shown. DETAILED DESCRIPTION

[0027] Illustrative embodiments of the present application include, but are not limited to, a method for classifying abdominal infectious fluid, a device for classifying abdominal infectious fluid, a medium, an electronic device, and a computer program product.

[0028] Embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

[0029] Figure 1 A flow chart of a method for classifying abdominal infectious fluid of embodiments of the present application is shown, for an electronic device. Specifically, in a segmentation step S11, a pre-trained segmentation model is used to segment a plurality of non-contrast CT images from a plurality of patients respectively, to segment a plurality of regions of interest (ROIs) in each non-contrast CT image, including a fluid accumulation region, a gas bubble region within the fluid collection, a pancreatic parenchyma region, and a gas bubble bed region. The regions of interest can also include a drain region (if any).

[0030] The gas bubble bed region defines a morphological expansion region around the gas bubbles, to capture the inflammatory or necrotic microenvironment around the bubbles. The gas bubble bed is conceptually similar to the tumor bed concept in oncology research. It is hypothesized that gas-producing bacteria will not only produce discrete gas bubbles, but will also establish a local inflammatory microenvironment, extending beyond the immediate gas-containing region. This microenvironment is characterized by changes in tissue density, increased vascular permeability, inflammatory cell infiltration, and protein denaturation, which should be detected by advanced image analysis techniques. Operationally, the gas bubble bed region is defined as a morphological expansion region around the aggregated bubbles, capturing the region of infection-related tissue changes. Analyzing this region enables the capture of infection-specific information, which goes beyond the simple binary assessment of the presence or absence of gas.

[0031] In this embodiment, for example, the pre-trained segmentation model is based on the nnMamba model. The nnMamba model is a hybrid framework combining convolutional neural networks (CNN) and state-space models (SSM), specifically designed for 3D medical image analysis. This architecture addresses the limitations of traditional CNNs in capturing long-range dependencies while maintaining computational efficiency, which is particularly important for analyzing complex anatomical relationships in abdominal imagery. Using the nnMamba model, better understanding of the global contextual information of the image can be achieved, significantly improving the segmentation accuracy.

[0032] In the generating step S12, for each liquid accumulation region, a morphological dilation operation of a first size is applied to generate a first lesion surrounding region, and a morphological dilation operation of a second size is performed to generate a second lesion surrounding region. The first size is smaller than the second size, in this embodiment, for example, the first size is 2mm, and the second size is 5mm.

[0033] The liquid accumulation region is also referred to as a lesion. The infection-related inflammatory process extends beyond the primary lesion boundary, resulting in detectable changes in the surrounding tissue structure. Two expanded edges (2mm and 5mm) use morphological dilation operations to capture immediate and extended lesion surrounding changes.

[0034] In the extracting step S13, for each non-enhanced CT image, quantitative morphological features, 3D radiomic features are extracted from the plurality of regions of interest, a plurality of lesion surrounding features are extracted from the first lesion surrounding region and the second lesion surrounding region, and a plurality of local pathological features and a plurality of global contextual features are extracted from the liquid accumulation region (lesion). The details of these feature extraction are described below.

[0035] Regarding the extraction of quantitative morphological features:

[0036] Quantitative feature extraction is performed from the above-mentioned five ROIs segmented to extract a plurality of quantitative morphological features, which represent the morphological and density characteristics of the lesion. Specifically, the volume, three-dimensional maximum diameter, average CT value, maximum CT value, and minimum CT value of the lesion are extracted from the gas pocket region, which provide morphological and density distribution information of the lesion region. The total volume, number of bubbles, average volume, maximum volume, minimum volume, minimum distance between bubbles, and minimum enclosing sphere diameter of the bubbles in the effusion region are extracted, which provide the distribution pattern of the bubbles and the individual bubbles. In addition, the area and diameter of the contact boundary between the drainage tube region and the liquid accumulation region are calculated, which describe the connectivity and distribution pattern between the bubbles and the drainage tube, and help to determine whether the bubbles are derived from infection.

[0037] Regarding the extraction of 3D radiomic features:

[0038] In this embodiment, 3D (three-dimensional) radiomic features are extracted from the above-mentioned 5 ROIs, for example using the PyRadiomics library, to capture subtle textural and morphological features. These 3D radiomic features include, for example, first-order statistics describing intensity distribution, shape features describing geometric properties, and various textural features derived from the gray level co-occurrence matrix (GLCM), gray level run length matrix (GLRLM), gray level size zone matrix (GLSZM), neighborhood gray tone difference matrix (NGTDM), and gray level dependence matrix (GLDM). Features are extracted from the original images and filtered images using wavelet, Gaussian-Laplacian, square-root, square-root and log transformation to capture multi-scale textural information. This multi-dimensional approach generates, for example, 1,409 3D radiomic features that provide reliable quantitative features for patterns of lesion morphology, intensity distribution, and textural complexity.

[0039] Regarding the extraction of the plurality of lesion surrounding features:

[0040] The plurality of lesion surrounding features includes radiomic features (3D radiomic features) extracted from the first lesion surrounding region and the second lesion surrounding region. The plurality of lesion surrounding features represents microenvironmental features of inflammation associated with infection to capture microenvironmental changes of inflammation. That is, these lesion surrounding features extracted from the two main lesion surrounding regions can capture microenvironmental features of inflammation associated with infection, providing supplementary information of the microenvironment of inflammation associated with infection.

[0041] Regarding the extraction of the plurality of local pathological features and the plurality of global contextual features:

[0042] For each non-contrast CT image, a plurality of local pathological features (also referred to as “ResNet18 features”) are extracted from a plurality of consecutive axial slices centered on the maximum cross-section of the fluid accumulation region (lesion) using a lightweight ResNet18 network, and wherein a plurality of global contextual features (also referred to as “Big Model features”) are extracted from the plurality of consecutive axial slices using a DINOv2 network. The plurality of local pathological features and the plurality of global contextual features are referred to as deep learning features.

[0043] The present invention employs two complementary deep learning architectures for feature extraction to capture local and global imaging patterns. Specifically, the present invention employs a 2.5D deep learning feature extraction approach to extract information from five key cross-sectional slices of the lesion (fluid accumulation region).

[0044] The two complementary deep learning architectures include two complementary models: a ResNet18 network and a DINOv2 network.

[0045] The 2.5D deep learning feature extraction method in the application is specifically as follows: first, five continuous axial slices (specifically-4, -2, 0, +2, +4 layers in this embodiment) centered on the largest cross-section of the lesion are selected, and a lightweight ResNet18 network is used to extract multiple local pathological features from the five continuous axial slices, aiming to capture local three-dimensional information (local pathological features) of the lesion. The ResNet18 network is pre-trained on ImageNet and is good at identifying micro lesions and classifying tissues, and can accurately capture local detail features of the lesion. For example, 512 features are extracted from each slice, and a total of 2560 feature vectors are generated for each lesion.

[0046] Secondly, a visual base model (DINOv2) trained on a large-scale natural image dataset is used to capture global patterns and complex context information. Feature extraction is performed on each of the five slices using DINOv2, for example, 5,120 features (1,024 features x 5 slices) are extracted for each lesion. DINOv2 shows particularly effective in a wide range of lesion representation, thereby being able to capture subtle spatial relationships and situational patterns that may be related to the infection state.

[0047] The application uses a 2.5D deep learning feature extraction method, thereby capturing 3D spatial context information through a multi-layer slice information fusion mechanism while maintaining the efficiency of 2D convolution calculation.

[0048] In the selection step S14, for each non-enhanced CT image, a plurality of target features are selected from the extracted quantitative morphological features, 3D radiomics features, a plurality of lesion surrounding features, a plurality of local pathological features, and a plurality of global context features, and a radiomics score of each non-enhanced CT image is calculated.

[0049] Specifically, for each non-enhanced CT image, a plurality of target features are selected from the extracted features through the following three stages.

[0050] 1) Univariate screening: The extracted quantitative morphological features, 3D radiomics features, a plurality of lesion surrounding features, a plurality of local pathological features, and a plurality of global context features are subjected to univariate screening using a non-parametric test, to obtain screened partial features.

[0051] For example, the Mann-Whitney U test is used to preliminarily screen the above-mentioned features, and partial features having statistical differences (P<0.05) between the infection group and the non-infection group are screened out. This step significantly reduces the feature space dimension and removes noise features irrelevant to the target features.

[0052] 2) Correlation analysis: Calculate the Spearman correlation coefficient between the above-mentioned part of the selected features, identify the highly correlated feature pairs according to the Spearman correlation coefficient, and retain the features with larger information content (correlation coefficient > 0.9) in the feature pairs according to the principle of maximum variance, so as to eliminate redundant features.

[0053] 3) LASSO regularization selection: Use LASSO logistic regression for feature selection to select multiple target features from the retained features, and calculate the radiomics score of each non-enhanced CT image.

[0054] LASSO logistic regression is used to further select the features that can best predict infection, and finally multiple target features (λ = 0.0155) are selected.

[0055] LASSO (Least Absolute Shrinkage and Selection Operator) logistic regression is a linear model with L1 regularization, which completes the final feature selection task in this phase. Its unique feature is that it can compress some coefficients to zero, enabling simultaneous feature selection and model construction, especially suitable for high-dimensional data.

[0056] Here, 10-fold cross-validation is used to determine the optimal regularization parameter λ. In each cross-validation fold, the LASSO logistic regression model is fitted using the training data, and the performance is evaluated on the validation data. The λ value that maximizes the validation AUC is selected as the optimal parameter. The final LASSO model usually selects the most important features. The LASSO method can select the simplest feature subset while ensuring prediction performance.

[0057] In this embodiment, for example, the selected multiple target features include (here only the top ten features are given as an example): wavelet.LLL_glcm_Imc1 (2 cm around the bubble), resnet18_192 (bubble), square_glszm_SmallAreaLowGrayLevelEmphasis (2 mm around the liquid accumulation), squareroot_glrlm_ShortRunHighGrayLevelEmphasis (liquid accumulation), big_model_feature607 (liquid accumulation), big_model_feature459 (bubble), resnet18_171 (bubble bed), big_model_feature44 (bubble bed), resnet18_481 (bubble bed), and resnet18_490 (bubble).

[0058] wavelet.LLL_glcm_Imcl (2 cm around the air bubbles) is a feature of the gray level co-occurrence matrix (GLCM) representing 3D radiomic features extracted from the air bubble region within the effusion. resnet18_192 (air bubbles) represents the 192th layer ResNet18 feature representing air bubble related local pathological features extracted using ResNet18. square_root_glrlm_ShortRunHighGrayLevelEmphasis (2 mm around the liquid accumulation) is a feature of the gray level run length matrix (GLRLM) which is a texture feature computed on a three-dimensional volume representing 3D radiomic features extracted from the region around the first lesion. square_root_glrlm_ShortRunHighGrayLevelEmphasis (liquid accumulation) is a feature of the gray level run length matrix (GLRLM) which is a texture feature computed on a three-dimensional volume representing 3D radiomic features extracted from the liquid accumulation region. big_model_feature607 (liquid accumulation) represents the 607th layer big model (DINOv2) feature representing global contextual features related to the liquid accumulation (i.e. the lesion) extracted using DINOv2. Similarly, big_model_feature459 (air bubbles) represents global contextual features related to the air bubbles, resnet18_171 (air bubble bed) represents local pathological features related to the air bubble bed, big_model_feature44 (air bubble bed) represents global contextual features related to the air bubble bed, resnet18_481 (air bubble bed) represents local pathological features related to the air bubble bed, resnet18_490 (air bubbles) represents local pathological features related to the air bubbles.

[0059] It can be understood that the ResNet18 related features represent local pathological features, and the big_model related features represent global contextual features.

[0060] It can be understood that the target features listed above are only examples, and other features can also be included without limitation.

[0061] It can be understood that the finally selected target features include local pathological features, global contextual features, 3D radiomic features, lesion surrounding features, quantitative morphological features. It can be understood that the finally selected target features include features related to the air bubble bed region, such as global contextual features related to the air bubble bed, local pathological features.

[0062] Next, based on the selected multiple target features, the radiomic score DLR-Score of each non-contrast CT image is calculated by a LASSO regression equation. These target features are features with significant correlation.

[0063] It can be understood that for each non-enhanced CT image, a corresponding radiomics score DLR-Score can be calculated.

[0064] In the training step S15, the radiomics score DLR-Score of each non-enhanced CT image and the clinical parameters of each patient are input into the logistic regression model for training to obtain the trained logistic regression model (i.e., the classification model).

[0065] The clinical parameters include the clinical information of each patient, for example, including demographic characteristics (age, gender, body mass index), vital signs, hematological parameters, biochemical markers, inflammatory markers, and coagulation parameters, etc.

[0066] In the classification step S16, the trained logistic regression model is used to classify the non-enhanced CT image to be classified.

[0067] For example, the non-enhanced CT image to be classified of a certain patient is input into the trained logistic regression model, and based on the output radiomics score, the patient is classified as a high-risk or low-risk abdominal infectious effusion for the medical staff to refer to and perform corresponding medical treatment.

[0068] The radiomics scoring method of the present application constructs a more perfect and practical classification model through multi-dimensional feature fusion, clinical parameter integration, and dynamic optimization, which is used for effective classification of abdominal infectious effusion.

[0069] In addition, in this embodiment, the performance of the classification model is comprehensively verified by various methods such as ROC curve, Kaplan-Meier survival analysis, and decision curve analysis, and its clinical practicability is evaluated.

[0070] In the present application, based on each non-enhanced CT image, quantitative morphological features, 3D radiomics features, multiple lesion surrounding features, multiple local pathological features, and multiple global context features are extracted to realize extremely rich multi-dimensional feature extraction. Then, multiple target features are selected from the extracted multi-dimensional features, the radiomics score of each non-enhanced CT image is calculated according to the target features, and finally the logistic regression model is trained (constructed) based on the radiomics score of each non-enhanced CT image and the clinical parameters of the patient corresponding to each non-enhanced CT image to obtain the classification model. In this way, according to the output of the classification model, the classification of abdominal infectious effusion can be quickly, accurately, and low-costly realized.

[0071] In the present application, the features used to train the logistic regression model include features related to the bubble bed region. The bubble bed region can capture the inflammatory or necrotic microenvironment around the gas bubbles, which can change the surrounding tissue density and texture pattern, thus reflecting changes in tissue density, increased vascular permeability, inflammatory cell infiltration, and protein denaturation. Therefore, the classification model trained by the present application can more accurately classify abdominal infectious effusion.

[0072] The present application also provides a classification device 20 for abdominal infectious effusion, as shown in Figure 2 The classification device 20 comprises:

[0073] The segmentation unit 201 uses a pre-trained segmentation model to segment a plurality of non-enhanced CT images from a plurality of patients respectively to segment a plurality of regions of interest in each non-enhanced CT image, the plurality of regions of interest including a liquid accumulation region, a bubble region in the effusion, a pancreatic parenchyma region, a drainage tube region, and a bubble bed region, the bubble bed region defining a morphological dilation region around the accumulated gas bubbles;

[0074] The generation unit 202 applies a morphological dilation operation of a first size to generate a first lesion surrounding region and a morphological dilation operation of a second size to generate a second lesion surrounding region for each liquid accumulation region, wherein the first size is smaller than the second size;

[0075] The extraction unit 203 extracts quantitative morphological features, 3D radiomics features from the plurality of regions of interest, a plurality of lesion surrounding features from the first lesion surrounding region and the second lesion surrounding region, and a plurality of local pathological features and a plurality of global context features from the liquid accumulation region for each non-enhanced CT image;

[0076] The selection unit 204 selects a plurality of target features from the quantitative morphological features, the 3D radiomics features, the plurality of lesion surrounding features, the plurality of local pathological features, and the plurality of global context features for each non-enhanced CT image, and calculates a radiomics score for each non-enhanced CT image;

[0077] The training unit 205 inputs the radiomics score of each non-enhanced CT image and the clinical parameters of each patient into a logistic regression model for training to obtain a trained logistic regression model;

[0078] The classification unit 206 uses the trained logistic regression model to classify the non-enhanced CT image to be classified.

[0079] It can be understood that the segmentation unit 201, the generation unit 202, the extraction unit 203, the selection unit 204, the training unit 205, and the classification unit 206 can be implemented by the processor 102 in the electronic device 100 having the functions of these modules or units. Figure 3

[0080] The present application also provides a computer readable storage medium, the storage medium storing instructions, the instructions being executed on a computer to make the computer execute the method shown in Figure 1

[0081] The present application also provides a computer program product, including computer executable instructions, the instructions being executed by the processor 102 to implement the method shown in Figure 1

[0082] Reference is now made to Figure 3 , Figure 3 An example electronic device 1400 according to embodiments of the present application is shown schematically. In one embodiment, the electronic device 1400 can include one or more processors 1404, system control logic 1408 connected to at least one of the processors 1404, system memory 1412 connected to the system control logic 1408, non-volatile memory (NVM) 1416 connected to the system control logic 1408, and a network interface 1420 connected to the system control logic 1408.

[0083] In some embodiments, the processor 1404 can include one or more single core or multi core processors. In some embodiments, the processor 1404 can include any combination of general purpose processors and special purpose processors (e.g., graphics processors, application processors, baseband processors, etc.). In embodiments where the electronic device 1400 employs an eNB (Evolved Node B) or RAN (Radio Access Network) controller, the processor 1404 can be configured to perform various compliant embodiments, for example, as shown in Figure 1

[0084] In some embodiments, the system control logic 1408 can include any suitable interface controllers appropriate for communicating with at least one of the processors 1404 and / or any suitable devices or components in communication with the system control logic 1408.

[0085] ​​​​In some embodiments, system control logic 1408 can include one or more memory controllers to provide an interface to system memory 1412. System memory 1412 can be used to load and store data and / or instructions. In some embodiments, system memory 1412 of electronic device 1400 can include any suitable volatile memory, such as suitable dynamic random access memory (DRAM).

[0086] Non-volatile memory 1416 can include one or more tangible, non-transitory computer-readable media used to store data and / or instructions. In some embodiments, non-volatile memory 1416 can include any suitable non-volatile memory, such as flash memory, and / or any suitable non-volatile storage device, such as at least one of a Hard Disk Drive (HDD), a Compact Disc (CD) drive, a Digital Versatile Disc (DVD) drive.

[0087] Non-volatile memory 1416 can include storage resources installed on the device of electronic device 1400, or it can be accessible by electronic device but not necessarily part of electronic device. For example, non-volatile memory 1416 can be accessed over a network via network interface 1420.

[0088] In particular, system memory 1412 and non-volatile memory 1416 can include, respectively, a temporary copy and a permanent copy of instructions 1424. Instructions 1424 can include instructions that, when executed by at least one of processors 1404, cause electronic device 1400 to implement a method as shown in Figure 1 In some embodiments, instructions 1424, hardware, firmware, and / or software components thereof can additionally / alternatively be placed in system control logic 1408, network interface 1420, and / or processors 1404.

[0089] Network interface 1420 can include a transceiver to provide a radio interface for electronic device 1400 to communicate with any other suitable device (e.g., a front-end module, an antenna, etc.) over one or more networks. In some embodiments, network interface 1420 can be integrated with other components of electronic device 1400. For example, network interface 1420 can be integrated with at least one of processors 1404, system memory 1412, non-volatile memory 1416, and a firmware device (not shown) having instructions that, when executed by at least one of processors 1404, cause electronic device 1400 to implement a method as shown in Figure 1

[0090] ​The network interface 1420 can further include any suitable hardware and / or firmware to provide a multiple-input multiple-output radio interface. For example, the network interface 1420 can be a network adapter, a wireless network adapter, a telephone modem, and / or a wireless modem.

[0091] In one embodiment, at least one of the processors 1404 can be packaged together with logic for one or more controllers of the system control logic 1408 in a system in package (SiP). In one embodiment, at least one of the processors 1404 can be integrated on the same die with logic for one or more controllers of the system control logic 1408 to form a system on a chip (SoC).

[0092] The electronic device 1400 can further include an input / output (I / O) device 1432. The I / O device 1432 can include a user interface that enables a user to interact with the electronic device 1400; a peripheral component interface that enables peripheral components to interact with the electronic device 1400. In some embodiments, the electronic device 1400 also includes a sensor to determine at least one of environmental conditions and location information related to the electronic device 1400.

[0093] In some embodiments, the user interface can include, but is not limited to, a display (e.g., a liquid crystal display, a touch screen display, etc.), a speaker, a microphone, one or more cameras (e.g., a still image camera and / or a video camera), a flashlight (e.g., a light emitting diode flash), and a keypad.

[0094] Embodiments of the mechanisms disclosed herein can be implemented in hardware, software, firmware, or any combination thereof. Embodiments of the application can be implemented as computer programs or program code executing on programmable systems comprising at least one processor, a storage system (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device.

[0095] Program code can be applied to input instructions to perform the functions described herein and generate output information. The output information can be applied to one or more output devices, in known fashion. For purposes of this application, a processing system includes any system that has a processor, such as, for example; a digital signal processor (DSP), a microcontroller, an application specific integrated circuit (ASIC), or a microprocessor.

[0096] The program code can be implemented in a high level of programming language or a object-oriented programming language to communicate with a processing system. When necessary, the program code can be implemented in an assembly language or a machine language. In fact, the mechanisms described in this application are not limited to any specific programming language. In any case, the language can be a compiled or interpreted language.

[0097] In some cases, the disclosed embodiments can be implemented in hardware, firmware, software, or any combination thereof. The disclosed embodiments can also be implemented as instructions carried by or stored on a transitory or non-transitory machine-readable (e.g., computer-readable) medium, which can be read and executed by one or more processors. For example, a machine-readable medium can be a storage device or memory storage that is external to a processing system, such as a floppy disk, Zip® disk, compact disk, Blu-ray disk, magnetic tape, or other computer- readable medium. In another example, a machine-readable medium can be a storage device or memory storage that is internal to a processing system, such as read-only memory (ROM), random-access memory (RAM), electrically erasable programmable read-only memory (EEPROM), or other computer-readable memory. In another example, a machine-readable medium can be a transitory signal indicative of computer instructions, i.e., a propagation signal in the form of an electromagnetic wave, such as an electrical, optical, or acoustical wave. Thus, a machine-readable medium includes any mechanism that provides (i.e., stores and / or transmits) information in a form accessible by a machine (e.g., a computer), such as the Internet or other network.

[0098] In the drawings, some of the structural or methodological features can be shown in particular arrangements and / or orders. However, it should be understood that such particular arrangements and / or orders can not be required. Instead, in some embodiments, the features can be arranged in a different manner and / or order than shown in the illustrative drawings. Additionally, inclusion of structural or methodological features in a particular figure is not meant to imply that such features are required in all embodiments, and in some embodiments, these features can not be included or can be combined with other features.

[0099] It should be noted that each unit / module mentioned in each device embodiment of the present application is a logical unit / module, and in physical, one logical unit / module can be a physical unit / module, or a part of a physical unit / module, or be realized in a combination of multiple physical unit / modules, and the physical realization of these logical units / modules is not the most important, and the combination of functions implemented by these logical units / modules is the key to solving the technical problems proposed in the present application. In addition, in order to highlight the innovative part of the present application, the above-mentioned device embodiments of the present application do not introduce the units / modules which are not closely related to solving the technical problems proposed in the present application, which does not mean that the above-mentioned device embodiments do not have other units / modules.

[0100] It should be noted that in the examples and descriptions of the present patent, the relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the terms "include", "contain" or any other variants thereof are intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device. Without more limitations, the element defined by the statement "including one" does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0101] Although the present application has been illustrated and described with reference to certain preferred embodiments thereof, it should be understood by those skilled in the art that various changes in form and details can be made therein without departing from the spirit and scope of the present application.

Claims

1. A method for classifying infectious peritoneal effusions, used in electronic devices, characterized in that, The classification method includes: The segmentation step uses a pre-trained segmentation model to segment multiple non-contrast CT images from multiple patients to segment multiple regions of interest in each non-contrast CT image. The multiple regions of interest include a fluid accumulation region, a bubble region within the fluid accumulation, a pancreatic parenchyma region, and a bubble bed region, wherein the bubble bed region defines a morphological expansion region surrounding the accumulated gas bubbles. The generation step involves applying a morphological expansion operation of a first size to generate a region surrounding a first lesion for each liquid accumulation region, and performing a morphological expansion operation of a second size to generate a region surrounding a second lesion, wherein the first size is smaller than the second size. The extraction step involves extracting quantitative morphological features and 3D radiomics features from the multiple regions of interest for each non-enhanced CT image, extracting multiple lesion periphery features from the region surrounding the first lesion and the region surrounding the second lesion, and extracting multiple local pathological features and multiple global context features from the liquid accumulation region. In the selection step, for each non-contrast CT image, multiple target features are selected from the quantitative morphological features, the 3D radiomics features, the multiple lesion perilesional features, the multiple local pathological features, and the multiple global contextual features, and a radiomics score is calculated for each non-contrast CT image. The training step involves inputting the radiomics scores of multiple non-enhanced CT images and the clinical parameters of the multiple patients into the logistic regression model for training, resulting in the trained logistic regression model. The classification step involves using the trained logistic regression model to classify the non-enhanced CT images to be classified.

2. The classification method according to claim 1, characterized in that, The bubble bed area is used to capture the inflammatory or necrotic microenvironment around the bubbles. The multiple target features include features associated with the bubble bed region, used to capture the inflammatory or necrotic microenvironment around the bubbles.

3. The classification method according to claim 1, characterized in that, The features surrounding the multiple lesions represent the microenvironmental characteristics of infection-related inflammatory changes.

4. The classification method according to claim 1, characterized in that, In the extraction step, the multiple lesion periphery features include radiomics features extracted from the periphery regions of the first and second lesions.

5. The classification method according to claim 1, characterized in that, In the extraction step, for each liquid accumulation region, a ResNet18 network is used to extract the multiple local pathological features from multiple consecutive axial slices centered on the largest cross-section of the liquid accumulation region, and a DINOv2 network is used to extract the multiple global contextual features from the multiple consecutive axial slices.

6. The classification method according to claim 1, characterized in that, The selection steps include: For each non-contrast CT image, the quantitative morphological features, the 3D radiomics features, the multiple lesion perilesional features, the multiple local pathological features, and the multiple global context features are screened using nonparametric tests to obtain the screened partial features. Calculate the Spearman correlation coefficient between the aforementioned features, identify highly correlated feature pairs based on the Spearman correlation coefficient, and retain the feature with greater information content in the feature pair by adopting the variance maximization principle; LASSO logistic regression was used for feature selection to select the multiple target features from the retained features, and a radiomics score was calculated for each non-contrast CT image.

7. The classification method according to claim 1, characterized in that, The clinical parameters include each patient's clinical information.

8. A sorting device for infectious peritoneal effusion, characterized in that, The sorting device includes: The segmentation unit uses a pre-trained segmentation model to segment multiple non-contrast CT images from multiple patients to segment multiple regions of interest in each non-contrast CT image. The multiple regions of interest include a fluid accumulation region, a bubble region within the fluid accumulation, a pancreatic parenchyma region, and a bubble bed region, wherein the bubble bed region defines a morphological expansion region surrounding the accumulated gas bubbles. The generating unit applies a morphological expansion operation of a first size to generate a region surrounding a first lesion for each liquid accumulation region, and performs a morphological expansion operation of a second size to generate a region surrounding a second lesion, wherein the first size is smaller than the second size. The extraction unit extracts quantitative morphological features and 3D radiomics features from the multiple regions of interest for each non-enhanced CT image, extracts multiple lesion periphery features from the region surrounding the first lesion and the region surrounding the second lesion, and extracts multiple local pathological features and multiple global context features from the liquid accumulation region. The selection unit selects multiple target features from the quantitative morphological features, the 3D radiomics features, the multiple lesion perilesional features, the multiple local pathological features, and the multiple global context features for each non-contrast CT image, and calculates the radiomics score for each non-contrast CT image. The training unit inputs the radiomics scores of multiple non-enhanced CT images and the clinical parameters of the multiple patients into the logistic regression model for training, and obtains the trained logistic regression model. The classification unit uses the trained logistic regression model to classify the unenhanced CT images to be classified.

9. A computer-readable storage medium, characterized in that, The storage medium stores instructions that, when executed on a computer, cause the computer to perform the classification method for infectious peritoneal effusion as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, include: One or more processors; One or more memories; the one or more memories storing one or more programs, which, when executed by the one or more processors, cause the electronic device to perform the classification method for infectious peritoneal effusion as described in any one of claims 1 to 7.

11. A computer program product comprising computer-executable instructions, characterized in that, The instructions are executed by the processor to implement the classification method for infectious peritoneal effusion as described in any one of claims 1 to 7.

Images