Tumor recognition auxiliary system based on artificial intelligence

Abstract

The invention relates to a tumor identification auxiliary system based on artificial intelligence, and relates to the technical field of data processing, and the system comprises an information obtaining module which is used for obtaining medical record information and tumor identification position information of a patient; the image acquisition module is used for determining initial CT parameters of the patient according to the medical record information and the tumor recognition position information of the patient, acquiring an initial CT image of the patient based on the initial CT parameters of the patient, extracting image features of the initial CT image of the patient, determining multiple groups of target CT parameters of the patient according to the image features of the initial CT image of the patient, and sending the target CT parameters to the tumor recognition position information of the patient; acquiring a plurality of target CT images of the patient according to the plurality of groups of target CT parameters of the patient; and the identification auxiliary module is used for establishing and training a tumor identification model, performing tumor identification based on the multiple target CT images of the patient through the tumor identification model, and generating tumor identification auxiliary information, and has the advantage of realizing automatic analysis of the medical image.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to an artificial intelligence-based tumor recognition auxiliary system. Background Art

[0002] Currently, imaging examinations have become one of the most direct and effective ways to identify tumors. The clinical manifestations of tumors on CT images are the presence of several tissue areas of varying sizes and irregular morphological structures within the organ parenchyma, and the boundaries of these areas are relatively vague. Since tumor patients have no obvious symptoms in the early stages or their imaging manifestations are atypical, they are difficult to detect and diagnose. As a result, most of them are diagnosed in the middle and late stages, missing the best time for treatment. Therefore, early diagnosis and identification of tumors are crucial for cancer patients. In the early diagnosis of tumors, multi-slice spiral CT can clearly display the characteristics of lesions in the cross-section, sagittal and coronal planes through reconstruction technology. In the mid-term diagnosis, spiral CT diagnosis combined with surface masking and multi-planar reconstruction can clearly display the tumor location, internal structure, edge characteristics, blood supply, degree of invasion of surrounding tissues, and changes in surrounding tissues, with a high diagnostic accuracy. Therefore, CT images have become an important reference for tumor diagnosis and identification.

[0003] However, the current process of diagnosing tumors through medical images is mainly manual, and the diagnosis process mainly relies on experienced doctors. This process is subjective. In addition, the manual interpretation of medical images may also result in interpretation errors due to the doctor's cognitive limitations or fatigue, resulting in misdiagnosis. These defects all show the importance of using effective medical image analysis technology to improve the accuracy of disease diagnosis results.

[0004] Therefore, it is necessary to provide an artificial intelligence-based tumor recognition auxiliary system for automatic analysis of medical images. Summary of the Invention

[0005] The present invention provides an artificial intelligence-based tumor recognition auxiliary system, comprising: an information acquisition module for acquiring a patient's medical history information and tumor recognition location information; an image acquisition module for determining the patient's initial CT parameters based on the patient's medical history information and tumor recognition location information, acquiring the patient's initial CT image based on the patient's initial CT parameters, extracting image features of the patient's initial CT image, determining multiple sets of target CT parameters of the patient based on the image features of the patient's initial CT image, and acquiring multiple target CT images of the patient based on the multiple sets of target CT parameters of the patient; and an identification auxiliary module for establishing and training a tumor recognition model, performing tumor recognition based on the patient's multiple target CT images through the tumor recognition model, and generating tumor recognition auxiliary information.

[0006] Furthermore, the image acquisition module determines the initial CT parameters of the patient based on the patient's medical history information and tumor identification location information, including: for each tumor identification location, obtaining the medical history factors affecting CT imaging of the tumor identification location and historical imaging data, wherein the historical imaging data of the tumor identification location includes the medical history information, CT images and CT parameters of multiple historical patients; based on the medical history factors affecting CT imaging of the tumor identification location, the multiple historical patients are divided into multiple historical patient groups, and the initial CT parameters of each historical patient group are determined; based on the patient's tumor identification location information, multiple historical patient groups are queried; based on the patient's tumor identification location information, a target historical patient group is queried from multiple historical patient groups based on the medical history factors affecting CT imaging of the patient's tumor identification location and the patient's medical history information; and based on the initial CT parameters of the target historical patient group, the initial CT parameters of the patient are determined.

[0007] Furthermore, based on the medical history factors affecting CT imaging of the tumor identification position, multiple historical patients are divided into multiple historical patient groups, and the initial CT parameters of each historical patient group are determined, including: determining the medical history feature vector of the historical patient based on the medical history factors affecting CT imaging of the tumor identification position and the medical history information of the historical patient; dividing the multiple historical patients into multiple historical patient groups based on the medical history feature vectors of the multiple historical patients through a clustering algorithm; establishing a fitness function; for each historical patient group, determining the fitness of the CT parameters of the historical patient based on the fitness function and the CT image of each historical patient included in the historical patient group, and determining the initial CT parameters of the historical patient group based on the CT parameters of each historical patient included in the historical patient group and the fitness of the CT parameters through a genetic algorithm.

[0008] Furthermore, establishing a fitness function includes: determining multiple CT image quality assessment factors; determining a radiation dose calculation function, wherein the independent variables of the radiation dose calculation function include CT parameters; and establishing a fitness function based on the multiple CT image quality assessment factors and the radiation dose calculation function.

[0009] Furthermore, image features of the patient's initial CT image are extracted, including: for each tumor identification position, determining multiple tumor identification image feature factors associated with the tumor identification position; and extracting image features of the patient's initial CT image based on the multiple tumor identification image feature factors associated with the patient's tumor identification position.

[0010] Furthermore, determining multiple tumor recognition image feature factors associated with the tumor recognition position includes: determining multiple tumor recognition image feature factors; obtaining initial CT images of multiple historical users of the historical patient group corresponding to the tumor recognition position; for each tumor recognition image feature factor, based on the initial CT images of multiple historical users, calculating the difference value of the tumor recognition image feature factor at the tumor recognition position; and determining multiple tumor recognition image feature factors associated with the tumor recognition position based on the difference value of each tumor recognition image feature factor at the tumor recognition position.

[0011] Furthermore, based on the image features of the patient's initial CT image, multiple sets of target CT parameters of the patient are determined, including: obtaining initial CT images and multiple sets of target CT parameters of multiple historical users of the historical patient group corresponding to the tumor identification position; determining the target historical user based on the image features of the patient's initial CT image and the image features of the initial CT images of the historical users of the patient's tumor identification position, and determining the multiple sets of target CT parameters of the patient based on the target historical user.

[0012] Furthermore, the tumor recognition model includes multiple feature extraction units, multimodal feature fusion units, clinical information embedding units and classification decision units, wherein one feature extraction unit corresponds to a set of target CT parameters, the feature extraction unit is used to extract the feature vector of the input target CT image, the multimodal feature fusion unit is used to fuse the feature vectors extracted by multiple feature extraction units and output the fused feature vector, the clinical information embedding unit is used to splice the patient's medical record information with the fused feature vector to generate a comprehensive feature vector, and the classification decision unit is used to output the tumor recognition result based on the comprehensive feature vector.

[0013] Furthermore, the feature extraction unit includes a first convolutional layer, a first activation function and multiple dense residual attention blocks, wherein the dense residual attention block includes an input layer, a dense connection layer, a point-by-point convolution layer, an attention mechanism layer, a residual connection layer and an output layer.

[0014] Furthermore, the loss function used to train the tumor recognition model includes classification loss and orthogonal constraint loss, wherein the orthogonal constraint loss is calculated based on the feature vectors of the input target CT image extracted by multiple feature extraction units.

[0015] Compared with the existing technology, the artificial intelligence-based tumor recognition auxiliary system provided in this specification has at least the following beneficial effects:

[0016] 1. Dynamically adjust subsequent CT scan parameters (such as contrast agent dose, scanning phase, and slice thickness) based on the image features of the initial CT image (such as tumor location, size, and density), obtain multiple sets of target CT images, reduce unnecessary radiation dose, avoid the use of uniform high-dose scans, improve the visualization of tumor boundaries and internal structures, and provide higher-quality data for subsequent models. The tumor recognition model uses a multi-branch network or fusion network architecture, which can simultaneously process multiple sets of target CT images, extract multimodal features, avoid the limitations of single-modality information, improve classification performance, and automatically complete CT parameter optimization, image feature extraction, and preliminary tumor identification. Doctors only need to review the results, reducing repetitive work. Through standardized processes and deep learning models, diagnostic differences between different doctors can be reduced.

[0017] 2. Based on the patient's tumor location and medical history (such as age, weight, and medical history), a target patient group is screened from historical data and their optimized CT parameters are inherited, avoiding a "one-size-fits-all" scanning plan. The fitness function is used to quantitatively evaluate the pros and cons of CT parameters, balancing image quality (such as signal-to-noise ratio and contrast) and radiation dose. A genetic algorithm is used to globally search for optimal CT parameters, avoiding local optimal solutions. Through personalized parameter optimization, high-quality images can be obtained on the first scan, reducing repeated scans caused by inappropriate parameters.

[0018] 3. Assigning a separate feature extraction unit to each target CT parameter set extracts tumor features under different target CT parameters to avoid information loss. The clinical information embedding unit combines patient medical records (such as age and medical history) with image features to improve the model's ability to identify tumors and reduce false positives (such as benign nodules in elderly patients being misdiagnosed as tumors). Orthogonal constraint loss is used to enforce orthogonality in the output feature vectors of multiple feature extraction units, reducing redundant information and enhancing the complementarity of multimodal features. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] This specification will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, like numbers represent like structures, wherein:

[0020] Figure 1 This is a module diagram of an artificial intelligence-based tumor recognition assistance system shown in one embodiment of the present application;

[0021] Figure 2 It is a structural diagram of a tumor recognition model shown in one embodiment of the present application. DETAILED DESCRIPTION

[0022] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly introduces the drawings required for describing the embodiments.

[0023] Figure 1 This is a module diagram of an artificial intelligence-based tumor recognition auxiliary system shown in one embodiment of the present application. Figure 1 As shown, an artificial intelligence-based tumor recognition assistance system may include an information acquisition module, an image acquisition module, and an identification assistance module.

[0024] The information acquisition module can be used to obtain the patient's medical history information and tumor identification location information.

[0025] Specifically, the patient's medical record information may include the patient's basic information (eg, gender, age, BMI, blood test results, biochemical indicators, etc.), medical history, etc. The information acquisition module may acquire the patient's medical record information from the hospital's data system.

[0026] The tumor identification location information refers to the location that needs to be scanned by CT. The information acquisition module can obtain the patient's tumor identification location information from the doctor's terminal.

[0027] In some embodiments, the image acquisition module determines the patient's initial CT parameters based on the patient's medical history information and tumor identification location information, including:

[0028] For each tumor identification location, obtaining medical history factors affecting CT imaging and historical imaging data of the tumor identification location, wherein the historical imaging data of the tumor identification location includes medical history information, CT images, and CT parameters of multiple historical patients; dividing the multiple historical patients into multiple historical patient groups based on the medical history factors affecting CT imaging of the tumor identification location, and determining initial CT parameters for each historical patient group; wherein the CT imaging locations of the multiple historical patients are all the tumor identification location;

[0029] Based on the patient's tumor identification location information, multiple historical patient groups are queried. For example, historical patient groups 1-3 correspond to tumor identification location A, and historical patient groups 4-7 correspond to tumor identification location B. Based on the patient's tumor identification location information, if the patient's tumor identification location is A, historical patient groups 1-3 are used as the multiple historical patient groups obtained by the query.

[0030] Based on the medical record factors affecting the patient's tumor identification position on CT imaging and the patient's medical record information, a target historical patient group is queried from multiple historical patient groups. For example, based on the medical record factors affecting the patient's tumor identification position on CT imaging and the patient's medical record information, a medical record feature vector of the patient is determined, and the Euclidean distance between the patient's medical record feature vector and the medical record feature vectors of historical patients sampled from the historical patient group is calculated. The historical patient group to which the historical patients whose Euclidean distance is less than a Euclidean distance threshold belong is taken as the target historical patient group.

[0031] Based on the initial CT parameters of the target historical patient group, the initial CT parameters of the patient are determined, wherein the CT parameters may include parameters such as tube voltage, tube current, and layer thickness. For example, the initial CT parameters of the target historical patient group are used as the initial CT parameters of the patient.

[0032] Specifically, medical history factors affecting CT imaging of the tumor identification location may be factors that affect CT imaging of the tumor identification location, such as gender, age, BMI, and disease. For example, sex hormones (such as estrogen and testosterone) may affect the growth and metabolism of certain tumors, thereby affecting tumor characteristics (such as density and enhancement pattern) in CT imaging. With aging, tissue density may change (such as decreased bone density and increased adipose tissue), affecting the measurement of CT values ​​(Hounsfield Unit, HU). Patients with high BMI have increased fat and soft tissue thickness, resulting in increased X-ray attenuation, which may reduce the image signal-to-noise ratio (SNR) and contrast-to-noise ratio (CNR). Surgery, radiotherapy, chemotherapy, etc. may change the anatomical structure or biological characteristics of the tumor. Comorbidities (such as diabetes and cardiovascular disease) may affect the patient's hemodynamics and metabolic status, indirectly affecting CT imaging.

[0033] CT imaging-influencing medical history factors can be determined manually (e.g., experts, doctors, etc.) to identify the location of the tumor.

[0034] In some embodiments, based on the medical history factors affecting the CT imaging of the tumor identification location, multiple historical patients are divided into multiple historical patient groups, and initial CT parameters of each historical patient group are determined, including:

[0035] Determine the medical record feature vector of the historical patient based on the medical record factors affecting the tumor identification location and the medical record information of the historical patient;

[0036] By using a clustering algorithm, multiple historical patients are divided into multiple historical patient groups according to their medical record feature vectors;

[0037] Establish a fitness function;

[0038] For each historical patient group, the fitness of the historical patient's CT parameters is determined based on the fitness function and the CT images of each historical patient included in the historical patient group, and the initial CT parameters of the historical patient group are determined through a genetic algorithm based on the CT parameters of each historical patient included in the historical patient group and the fitness of the CT parameters.

[0039] Specifically, determining the medical record feature vector of the historical patient based on the medical record factors affecting the tumor identification position and the medical record information of the historical patient may include the following steps:

[0040] S11. Medical Record Information Extraction: Extract information from the patient's electronic medical records that corresponds to factors affecting CT imaging of tumor location, such as demographic characteristics (gender, age, and BMI). Laboratory tests (hemoglobin, platelets), and biochemical parameters (liver and renal function). Disease characteristics (tumor type, stage, location, and size). Previous treatments (surgery, radiotherapy, and chemotherapy).

[0041] S12. Feature quantification: Convert non-numeric features (such as gender and tumor type) into numerical codes. For example: Gender: Male = 1, Female = 0. Tumor type: Lung cancer = 1, Liver cancer = 2, Breast cancer = 3.

[0042] S13. Feature vector construction: All quantitative features are combined into a multidimensional vector. For example, feature vector = [sex, age, BMI, hemoglobin, tumor type, tumor size]. Example: male, 55 years old, BMI = 25, hemoglobin = 130g / L, lung cancer, tumor size = 3cm, the corresponding medical record feature vector is: [1, 55, 25, 130, 1, 3].

[0043] Use the K-means clustering algorithm to cluster multiple historical patients and divide them into multiple historical patient groups. The specific steps include:

[0044] Initialization: Randomly select K historical patients as center points (K is the preset number of groups).

[0045] Allocation: The Euclidean distance between the feature vectors of patients who are not selected as center points and the feature vectors of the center points can be calculated, and the feature vector of each patient is allocated to the group where the center point with the smallest Euclidean distance is located.

[0046] Update Center Points: Recalculate the center point (mean eigenvector) of each group.

[0047] Iteration: Repeat the assignment and update steps until the center point no longer changes or the maximum number of iterations is reached.

[0048] In some embodiments, establishing a fitness function includes:

[0049] Determine multiple CT image quality assessment factors, such as spatial resolution, noise level, contrast, artifact level, etc.;

[0050] determining a radiation dose calculation function, wherein the independent variables of the radiation dose calculation function include CT parameters;

[0051] A fitness function is established based on multiple CT image quality assessment factors and radiation dose calculation functions.

[0052] For example, the radiation dose calculation function can be:

[0053]

[0054] in, is the dose-length product, reflecting the total radiation dose of a single scan. is the tube current, is the tube voltage, is the pitch, a, b, c, d are empirical coefficients related to the CT equipment model and scanning protocol, is the scan length, a ranges from 0.001 to 0.1, characterizing the contribution of tube current to dose, b ranges from 0.1 to 10, and is used to adjust the basic dose level. Its value varies depending on the device model and scan protocol, c ranges from 1.5 to 3.0, reflecting the degree of influence of tube voltage on dose, and d ranges from 0.5 to 2.0, reflecting the degree of influence of pitch on dose.

[0055] For example, the fitness function can be:

[0056]

[0057] in, is the fitness function, 、 、 、 and is the weight, 、 、 、 and greater than 0, , example, =0.2, =0.1, =0.2, =0.2, =0.3, is the normalized score of spatial resolution (0-1). The higher the CT image quality, the higher the normalized score of spatial resolution. is the normalized score of contrast (0-1). The higher the CT image quality, the higher the normalized score of contrast. is the normalized score of the noise level (0-1). The higher the CT image quality, the lower the normalized score of the noise level. is the normalized score of the artifact degree (0-1). The higher the CT image quality, the lower the normalized score of the artifact degree. is the normalized score of radiation dose (0~1), After normalization, we get , The larger the , the higher the normalized score of radiation dose.

[0058] As you can understand, the fitness function comprehensively evaluates CT image quality (spatial resolution, contrast, noise level, and artifact severity) and radiation dose through a weighted summation. Higher normalized scores for spatial resolution and contrast correspond to larger fitness function values; higher normalized scores for noise, artifact, and radiation dose correspond to smaller fitness function values. This fitness function ensures that CT image quality meets diagnostic requirements while minimizing radiation dose and improving patient safety and comfort.

[0059] In some embodiments, determining the initial CT parameters of the historical patient group using a genetic algorithm based on the CT parameters of each historical patient included in the historical patient group and the fitness of the CT parameters may include the following steps:

[0060] (1) Initialize the population

[0061] The historical imaging data of the tumor identification position includes CT parameters of a plurality of historical patients as a plurality of initialized individuals.

[0062] (2) Calculating fitness

[0063] For each individual (CT parameter), calculate its fitness value:

[0064] The radiation dose was calculated based on the CT parameters.

[0065] Image quality factors (such as spatial resolution, noise level, etc.) were evaluated based on historical patient CT images.

[0066] Substitute into the fitness function to calculate the fitness value.

[0067] (3) Select

[0068] Select excellent individuals based on their fitness values ​​(individuals with higher fitness values ​​are more likely to be selected).

[0069] (4) Cross

[0070] Randomly select two parent individuals and generate offspring individuals through crossover operation.

[0071] Crossover mode: single-point crossover, multi-point crossover, etc.

[0072] (5) Variation

[0073] Some CT parameters of offspring individuals are randomly perturbed to increase population diversity.

[0074] (6) Termination conditions

[0075] Repeat steps (2) to (5) until the termination condition is met: the maximum number of iterations is reached or the fitness value converges.

[0076] (7) Output the optimal solution

[0077] The individual with the highest fitness value is returned as the initial CT parameter of the historical patient group.

[0078] The image acquisition module can be used to determine the patient's initial CT parameters based on the patient's medical history information and tumor identification location information, obtain the patient's initial CT image based on the patient's initial CT parameters, extract image features of the patient's initial CT image, determine the patient's multiple sets of target CT parameters based on the patient's image features, and obtain multiple target CT images of the patient based on the patient's multiple sets of target CT parameters.

[0079] In some embodiments, the image acquisition module extracts image features of the initial CT image of the patient, including:

[0080] For each tumor identification position, determining multiple tumor identification image feature factors associated with the tumor identification position, wherein the multiple tumor identification image feature factors are used to quantify the morphology, texture, metabolism and other characteristics of the tumor, such as morphological feature factors (e.g., the number of pixels of the tumor in the two-dimensional image, the pixel length of the tumor boundary, etc.), texture feature factors (e.g., gray-level co-occurrence matrix, gray-level run-length matrix), intensity feature factors (e.g., average gray value, maximum / minimum gray value, skewness, kurtosis, etc.), etc.;

[0081] Image features of the patient's initial CT image are extracted based on a plurality of tumor recognition image feature factors associated with the patient's tumor recognition position.

[0082] Specifically, image segmentation or object detection algorithms (such as U-Net and Mask R-CNN) are used to locate suspected tumor areas in the patient's initial CT image. The grayscale values ​​of the suspected tumor area are normalized to a fixed range (such as [0, 1] or [-1, 1]). Gaussian filtering or median filtering is then used to remove noise, unify the image resolution, and ensure feature consistency. Based on the tumor mask, morphological features are extracted from the patient's initial CT image. Based on the pixel values ​​of the patient's initial CT image, texture features and intensity features are extracted from the patient's initial CT image. In other words, the image features of the patient's initial CT image include morphological features, texture features, and intensity features.

[0083] In some embodiments, the image acquisition module determines multiple tumor identification image feature factors associated with the tumor identification location, including:

[0084] Determining multiple tumor recognition image feature factors. The multiple tumor recognition image feature factors can be determined manually (e.g., by a doctor or expert);

[0085] Obtaining initial CT images of multiple historical users of a historical patient group corresponding to a tumor identification location, wherein the initial CT images of the historical users may be CT images obtained with a CT scanning location corresponding to the tumor identification location and under CT scanning parameters corresponding to the historical patient group;

[0086] For each tumor recognition image feature factor, based on the initial CT images of multiple historical users, the difference value of the tumor recognition image feature factor at the tumor recognition position is calculated. For example, the difference value of the tumor recognition image feature factor at each sample can be calculated based on the tumor recognition image feature factor;

[0087] Based on the difference value of each tumor recognition image feature factor at the tumor recognition position, multiple tumor recognition image feature factors associated with the tumor recognition position are determined. For example, the tumor recognition image feature factor with a difference value greater than a difference value threshold can be used as the multiple tumor recognition image feature factors associated with the tumor recognition position.

[0088] Specifically, for each historical patient group corresponding to the tumor identification location, the variance of the tumor identification image feature factor values ​​for each historical user in the historical patient group can be calculated to serve as the difference value of the tumor identification image feature factor for the historical patient group corresponding to the tumor identification location. The difference values ​​of the tumor identification image feature factor for each historical patient group corresponding to the tumor identification location are averaged to serve as the difference value of the tumor identification image feature factor for the tumor identification location.

[0089] In some embodiments, the image acquisition module determines multiple sets of target CT parameters of the patient based on image features of the initial CT image of the patient, including:

[0090] Obtaining initial CT images and multiple sets of target CT parameters of multiple historical users of a historical patient group corresponding to a tumor identification location, wherein the multiple sets of target CT parameters of the historical patient group corresponding to the tumor identification location can be manually determined (e.g., by a doctor or expert);

[0091] A target historical user is determined based on image features of the patient's initial CT image and image features of initial CT images of historical users of the patient's tumor identification position, and multiple sets of target CT parameters of the patient are determined based on the target historical users.

[0092] Specifically, the cosine similarity between the image features of the patient's initial CT image and the image features of the initial CT images of historical users in the patient's target historical patient group can be calculated, and the historical user with the largest cosine similarity is used as the target historical user. The multiple sets of target CT parameters of the target historical user are used as the multiple sets of target CT parameters of the patient.

[0093] The recognition auxiliary module can be used to establish and train a tumor recognition model, perform tumor recognition based on multiple target CT images of the patient through the tumor recognition model, and generate tumor recognition auxiliary information.

[0094] Figure 2 This is a structural diagram of a tumor recognition model shown in an embodiment of the present application. Figure 2 As shown, in some embodiments, the tumor recognition model includes multiple feature extraction units, a multimodal feature fusion unit, a clinical information embedding unit and a classification decision unit, wherein one feature extraction unit corresponds to a set of target CT parameters, the feature extraction unit is used to extract tumor-related feature vectors of the input target CT image, which may include morphological, texture, intensity and other features, the multimodal feature fusion unit is used to fuse the feature vectors extracted by multiple feature extraction units, output the fused feature vector, generate a fused feature vector, directly splice multiple feature vectors into a longer vector to integrate tumor information under different parameters, the clinical information embedding unit is used to splice the patient's medical record information with the fused feature vector to generate a comprehensive feature vector, the classification decision unit is used to output the tumor recognition result according to the comprehensive feature vector, and the classification decision unit may include a support vector machine.

[0095] In some embodiments, the feature extraction unit includes a first convolutional layer, a first activation function and multiple dense residual attention blocks, wherein the dense residual attention block includes an input layer, a dense connection layer, a point-by-point convolution layer, an attention mechanism layer, a residual connection layer and an output layer.

[0096] Specifically, the first convolutional layer performs preliminary feature extraction on the input CT image (such as using a 3×3 convolution kernel) to generate a basic feature map. The first activation function is ReLU to enhance the expressiveness of the model. Multiple stacked dense residual attention blocks are used to further extract high-level features, while alleviating the gradient vanishing problem and enhancing feature expression capabilities.

[0097] The dense residual attention block is a module that combines dense connections, residual connections, and attention mechanisms. Its structure is as follows:

[0098] 1. Input layer:

[0099] Receives the feature map from the previous layer as input.

[0100] 2. Dense Connection Layer:

[0101] Function: Dense connection is used to concatenate the input of the current layer with the output of all previous layers to achieve feature reuse and gradient flow.

[0102] accomplish:

[0103] Assume that the current layer is the lth layer, and its input is the concatenation of the outputs of all previous layers: xl=[x0,x1,…,xl−1].

[0104] Dense connections can reduce the number of parameters, alleviate the vanishing gradient problem, and enhance feature propagation.

[0105] Output: concatenated feature map.

[0106] 3. Pointwise Convolution Layer:

[0107] Function: Performs channel dimensionality reduction or dimensionality increase on densely connected feature maps through 1×1 convolution, reducing the amount of computation and adjusting the number of channels.

[0108] accomplish:

[0109] Using a 1×1 convolution kernel, each position of the feature map is independently convolved without changing the spatial dimension and only adjusting the number of channels.

[0110] Output: Feature map after channel number adjustment.

[0111] 4. Attention Mechanism Layer:

[0112] Function: Dynamically adjust the importance of different positions or channels in the feature map through attention mechanisms (such as channel attention, spatial attention, or mixed attention).

[0113] accomplish:

[0114] Channel Attention: Use global average pooling (GAP) and fully connected layers (FC) to generate channel weights and weight each channel of the feature map.

[0115] Spatial Attention: Use convolutional layers to generate a spatial weight map that weights each spatial position of the feature map.

[0116] Hybrid Attention: Combining channel attention and spatial attention to generate more refined weight maps.

[0117] Output: Attention-weighted feature map.

[0118] 5. Residual Connection Layer:

[0119] Function: Add the input feature map and the attention-weighted feature map through the residual connection to achieve feature reuse and gradient flow.

[0120] accomplish:

[0121] Residual connection formula: y=F(x)+x, where F(x) is the feature map after attention weighting and x is the input feature map.

[0122] Residual connections can alleviate the gradient vanishing problem of deep networks and accelerate model convergence.

[0123] 6. Output: Feature map after residual connection.

[0124] Output layer:

[0125] Output the final feature map of the current dense residual attention block as the input of the next layer.

[0126] In some embodiments, the loss function used to train the tumor recognition model includes classification loss and orthogonal constraint loss, wherein the orthogonal constraint loss is calculated based on the feature vectors of the input target CT image extracted by multiple feature extraction units.

[0127] For example, the loss function used to train a tumor recognition model is:

[0128]

[0129]

[0130]

[0131]

[0132] in, is the total loss, is the classification loss, is the orthogonality constraint loss, is a hyperparameter used to balance the weight of classification loss and orthogonality constraint loss, is the number of categories, is the true label, is the predicted probability of the cth category of the kth training sample, is the orthogonal constraint loss of the k-th training sample, is the batch size, is the total number of feature extraction units, is the feature vector of the input target CT image extracted by the i-th feature extraction unit, is the feature vector of the input target CT image extracted by the j-th feature extraction unit, for and The inner product of is the eigenvector The L2 norm of is the eigenvector The L2 norm of is the square of the L2 norm, which is used to measure the cosine similarity between two feature vectors.

[0133] As can be understood, in the above formula, the classification loss is used to optimize the classification performance of the model, and the orthogonality constraint loss is used to ensure diversity between the feature vectors extracted by different feature extraction units. The classification loss is calculated using the cross-entropy loss and is used to measure the difference between the model's predicted probability and the true label. The orthogonality constraint loss is calculated using the squared cosine similarity of the feature vectors and is used to constrain the orthogonality between the feature vectors extracted by different feature extraction units. The squared cosine similarity measures the directional consistency of two feature vectors (0 for orthogonal and 1 for parallel). By minimizing the orthogonality constraint loss, the feature vectors extracted by different feature extraction units are forced to be orthogonal, avoiding feature redundancy. By combining the classification loss and the orthogonality constraint loss, the model can extract more discriminative and diverse features while optimizing classification performance, thereby improving the robustness and accuracy of tumor recognition tasks.

[0134] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. An artificial intelligence-based tumor recognition auxiliary system, characterized in that: include: Information acquisition module, used to obtain the patient's medical history information and tumor identification location information; an image acquisition module, configured to determine initial CT parameters of the patient based on the patient's medical history information and tumor identification location information, acquire an initial CT image of the patient based on the initial CT parameters of the patient, extract image features of the initial CT image of the patient, determine multiple sets of target CT parameters of the patient based on the image features of the initial CT image of the patient, and acquire multiple target CT images of the patient based on the multiple sets of target CT parameters of the patient; The recognition auxiliary module is used to establish and train a tumor recognition model, perform tumor recognition based on multiple target CT images of the patient through the tumor recognition model, and generate tumor recognition auxiliary information.

2. The artificial intelligence-based tumor recognition auxiliary system according to claim 1, characterized in that: The image acquisition module determines the patient's initial CT parameters based on the patient's medical history information and tumor identification location information, including: For each tumor identification location, obtaining medical history factors affecting CT imaging and historical imaging data of the tumor identification location, wherein the historical imaging data of the tumor identification location includes medical history information, CT images, and CT parameters of multiple historical patients; dividing the multiple historical patients into multiple historical patient groups based on the medical history factors affecting CT imaging of the tumor identification location, and determining initial CT parameters for each historical patient group; Query multiple historical patient groups based on the patient's tumor identification location information; Querying a target historical patient group from multiple historical patient groups based on medical history factors affecting the patient's tumor identification location on CT imaging and the patient's medical history information; Based on the initial CT parameters of the target historical patient group, the initial CT parameters of the patient are determined.

3. The artificial intelligence-based tumor recognition auxiliary system according to claim 2, characterized in that: Based on the medical history factors affecting the CT imaging of the tumor identification location, multiple historical patients were divided into multiple historical patient groups, and the initial CT parameters of each historical patient group were determined, including: Determine the medical record feature vector of the historical patient based on the medical record factors affecting the tumor identification location and the medical record information of the historical patient; By using a clustering algorithm, multiple historical patients are divided into multiple historical patient groups according to their medical record feature vectors; Establish a fitness function; For each historical patient group, the fitness of the historical patient's CT parameters is determined based on the fitness function and the CT images of each historical patient included in the historical patient group, and the initial CT parameters of the historical patient group are determined through a genetic algorithm based on the CT parameters of each historical patient included in the historical patient group and the fitness of the CT parameters.

4. The artificial intelligence-based tumor recognition auxiliary system according to claim 3, characterized in that: Establish a fitness function, including: Determine multiple CT image quality assessment factors; determining a radiation dose calculation function, wherein the independent variables of the radiation dose calculation function include CT parameters; A fitness function is established based on multiple CT image quality assessment factors and radiation dose calculation functions.

5. The artificial intelligence-based tumor recognition auxiliary system according to any one of claims 1 to 4, characterized in that: The image acquisition module extracts image features of the patient's initial CT image, including: For each tumor identification position, determining a plurality of tumor identification image feature factors associated with the tumor identification position; Image features of the patient's initial CT image are extracted based on a plurality of tumor recognition image feature factors associated with the patient's tumor recognition position.

6. The artificial intelligence-based tumor recognition auxiliary system according to claim 5, characterized in that: The image acquisition module determines multiple tumor recognition image feature factors associated with the tumor recognition location, including: Determine multiple tumor recognition image feature factors; Obtaining initial CT images of a plurality of historical users of a historical patient group corresponding to tumor identification locations; For each tumor recognition image feature factor, based on the initial CT images of multiple historical users, the difference value of the tumor recognition image feature factor at the tumor recognition position is calculated; According to the difference value of each tumor recognition image characteristic factor at the tumor recognition position, multiple tumor recognition image characteristic factors associated with the tumor recognition position are determined.

7. The artificial intelligence-based tumor recognition auxiliary system according to claim 6, characterized in that: The image acquisition module determines multiple sets of target CT parameters of the patient based on the image features of the patient's initial CT image, including: Obtaining initial CT images and multiple sets of target CT parameters of multiple historical users of a historical patient group corresponding to a tumor identification position; A target historical user is determined based on image features of the patient's initial CT image and image features of initial CT images of historical users of the patient's tumor identification position, and multiple sets of target CT parameters of the patient are determined based on the target historical users.

8. The artificial intelligence-based tumor recognition auxiliary system according to any one of claims 1 to 4, characterized in that: The tumor recognition model includes multiple feature extraction units, a multimodal feature fusion unit, a clinical information embedding unit and a classification decision unit, wherein one feature extraction unit corresponds to a set of target CT parameters, the feature extraction unit is used to extract the feature vector of the input target CT image, the multimodal feature fusion unit is used to fuse the feature vectors extracted by multiple feature extraction units and output the fused feature vector, the clinical information embedding unit is used to splice the patient's medical record information with the fused feature vector to generate a comprehensive feature vector, and the classification decision unit is used to output the tumor recognition result based on the comprehensive feature vector.

9. The artificial intelligence-based tumor recognition auxiliary system according to claim 8, characterized in that: The feature extraction unit includes a first convolutional layer, a first activation function and multiple dense residual attention blocks, wherein the dense residual attention block includes an input layer, a dense connection layer, a point-by-point convolution layer, an attention mechanism layer, a residual connection layer and an output layer.

10. The artificial intelligence-based tumor recognition auxiliary system according to claim 9, characterized in that: The loss function used to train the tumor recognition model includes classification loss and orthogonal constraint loss, wherein the orthogonal constraint loss is calculated based on the feature vectors of the input target CT image extracted by multiple feature extraction units.

Images