Cognitive state detection method, model training method, apparatus, device, and medium

Abstract

Embodiments of the present application provide a cognitive state detection method, a model training method, an apparatus, a device and a medium. The apparatus can obtain a to-be-detected medical image containing a basal ganglion of a to-be-detected person. Then, feature extraction is performed on each voxel in the to-be-detected medical image to obtain first radiomics features. Next, image segmentation is performed on the to-be-detected medical image again according to the first radiomics features to obtain at least two sub-regions contained in the basal ganglion, and feature extraction is performed on the at least two sub-regions to obtain second radiomics features. Finally, the cognitive level of the to-be-detected person is determined according to the second radiomics features. That is, the use of the first radiomics features can accurately segment the at least two sub-regions. Further feature extraction on the at least two sub-regions can enable the second radiomics features to contain rich information describing the state of each sub-region, and therefore, the use of the second radiomics features can accurately determine the cognitive level of the to-be-detected person.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a cognitive state detection method, model training method, apparatus, device, and medium. Background Technology

[0002] Cognitive impairment refers to a state in which the brain experiences difficulty or damage to higher cognitive functions such as processing information, understanding concepts, learning new knowledge, reasoning, and making judgments. In reality, any factor that causes abnormalities in the basal ganglia of the brain can lead to cognitive impairment, such as Alzheimer's disease, chronic kidney disease or complications from other diseases, genetic factors, and so on.

[0003] Based on this, improving the accuracy of cognitive state detection has become an urgent problem to be solved. Summary of the Invention

[0004] In view of this, embodiments of the present invention provide a cognitive state detection method, a model training method, an apparatus, a device, and a medium to improve the accuracy of cognitive state detection.

[0005] In a first aspect, embodiments of the present invention provide a cognitive state detection device, comprising:

[0006] The first acquisition module is used to acquire medical images of the subject's basal ganglia;

[0007] The first feature extraction module is used to extract features from each voxel in the medical image to be tested in order to obtain the first radiomics features.

[0008] The first image segmentation module is used to segment the medical image to be tested according to the first radiomics features to obtain at least two sub-regions contained in the basal ganglia;

[0009] The second feature extraction module is used to extract features from the at least two sub-regions to obtain second radiomics features;

[0010] The first determining module is used to determine the cognitive level of the subject based on the second radiomics features.

[0011] Secondly, embodiments of the present invention provide a cognitive state detection method, including:

[0012] Acquire medical images containing the basal ganglia of the subject;

[0013] Feature extraction is performed on each voxel in the medical image to be tested to obtain the first radiomics feature;

[0014] Based on the first radiomics features, the medical image to be tested is segmented to obtain at least two sub-regions contained in the basal ganglia;

[0015] Feature extraction is performed on the at least two sub-regions to obtain second radiomics features;

[0016] The cognitive level of the subject is determined based on the second radiomics features.

[0017] Thirdly, embodiments of the present invention provide a model training method, including:

[0018] Obtain training images containing basal ganglia and reference classification results of the training images;

[0019] Feature extraction is performed on each voxel in the training image to obtain the first training feature;

[0020] Based on the first training feature, the training image is segmented to obtain at least two sub-regions contained in the basal segment;

[0021] Feature extraction is performed on the at least two sub-regions to obtain the second training features;

[0022] The second training feature is input into the prediction model so that the prediction model outputs the prediction score corresponding to the training image;

[0023] Based on the prediction scores corresponding to the training images, the prediction classification results of the training images are determined.

[0024] The prediction model is trained based on the loss calculation result between the predicted classification result and the reference classification result.

[0025] Fourthly, embodiments of the present invention provide a model training apparatus, comprising:

[0026] The second acquisition module is used to acquire training images containing basal ganglia and reference classification results of the training images;

[0027] The fourth feature extraction module is used to extract features from each voxel in the training image to obtain the first training feature;

[0028] The second image segmentation module is used to segment the training image according to the first training features to obtain at least two sub-regions contained in the basal segment;

[0029] The fifth feature extraction module is used to extract features from the at least two sub-regions to obtain the second training features;

[0030] An input module is used to input the second training features into a prediction model so that the prediction model outputs a prediction score corresponding to the training image.

[0031] The second determining module is used to determine the prediction classification result of the training image based on the prediction score corresponding to the training image.

[0032] The second training module is used to train the prediction model based on the loss calculation result between the predicted classification result and the reference classification result.

[0033] Fifthly, embodiments of the present invention provide an electronic device, including a processor and a memory, wherein the memory is used to store one or more computer instructions, wherein the one or more computer instructions, when executed by the processor, implement the cognitive state detection method in the second aspect above, or execute the model training method in the third aspect above. The electronic device may also include a communication interface for communicating with other devices or communication systems.

[0034] In a sixth aspect, embodiments of the present invention provide a non-transitory machine-readable storage medium storing executable code, wherein when the executable code is executed by a processor of an electronic device, the processor is able to implement at least the cognitive state detection method as described in the second aspect above, or the model training method as described in the third aspect above.

[0035] In a seventh aspect, embodiments of the present invention provide a program product, the computer program product including a computer program or instructions, which, when executed by a processor, cause the processor to implement the cognitive state detection method as described in the second aspect above, or the model training method as described in the third aspect above.

[0036] In the solution provided by this embodiment of the invention, a first acquisition module acquires a medical image containing the basal ganglia of the subject. Then, a first feature extraction module extracts features from each voxel in the medical image to obtain first radiomics features. Next, a first image segmentation module segments the medical image based on the first radiomics features to obtain at least two sub-regions contained in the basal ganglia. Then, a second feature extraction module extracts features from the at least two sub-regions to obtain second radiomics features. Finally, a first determination module determines the cognitive level of the subject based on the second radiomics features, thus achieving the detection of the subject's cognitive state.

[0037] As can be seen, the above process can subdivide the basal ganglia into subregions based on voxel-level first radiomics features. Since these voxel-level first radiomics features can accurately describe the image information contained in each pixel of the medical image under test, their use can accurately segment at least two subregions. Subsequently, feature extraction can be performed on each of these at least two subregions, resulting in second radiomics features containing rich information describing the state of each subregion. Furthermore, since cognitive level is directly affected by the state of different subregions in the basal ganglia, the use of second radiomics features can accurately determine the cognitive level of the subject. Attached Figure Description

[0038] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0039] Figure 1 This is a schematic diagram of the structure of a cognitive state detection device provided in an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of another cognitive state detection device provided in an embodiment of the present invention;

[0041] Figure 3 A schematic diagram illustrating the working process of a cognitive state detection device provided in an embodiment of the present invention;

[0042] Figure 4 This is a schematic diagram of the structure of another cognitive state detection device provided in an embodiment of the present invention;

[0043] Figure 5 A flowchart of a cognitive state detection method provided in an embodiment of the present invention;

[0044] Figure 6 A flowchart of a model training method provided in an embodiment of the present invention;

[0045] Figure 7 A flowchart of an electronic device provided in an embodiment of the present invention;

[0046] Figure 8 This is a schematic diagram of the structure of a model training device provided in an embodiment of the present invention;

[0047] Figure 9 A flowchart of another electronic device provided in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0049] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” used in the embodiments of this invention and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. “Multiple” generally includes at least two, but does not exclude the inclusion of at least one.

[0050] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0051] Depending on the context, the words “if” or “suppose” as used here can be interpreted as “when” or “in response to determination” or “in response to identification.” Similarly, depending on the context, the phrases “if determination” or “if identification (of the condition or event of the statement)” can be interpreted as “when determination” or “in response to determination” or “when identification (of the condition or event of the statement)” or “in response to identification (of the condition or event of the statement).”

[0052] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this invention are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0053] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a product or system comprising a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a product or system. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the product or system that includes said element.

[0054] Before providing a detailed description of the various embodiments provided by the present invention, the concepts involved in the present invention can also be explained:

[0055] Quantitative susceptibility mapping (QSM) images are an emerging technique in the field of magnetic resonance imaging (MRI) that can reflect the magnetic susceptibility distribution of the basal ganglia.

[0056] Magnetic susceptibility is a physical quantity that characterizes the degree to which a material is magnetized in an external magnetic field. Changes in magnetic susceptibility can reflect changes in the tissue composition and microstructure of the basal ganglia. For example, an increase in magnetic susceptibility may lead to an increase in the iron content of the basal ganglia, thereby causing cognitive impairment in the subject.

[0057] The basal ganglia are a key region of the brain that can influence a subject's cognitive abilities. Specifically, they can include the globus pallidus, putamen, and caudate nucleus.

[0058] Before describing in detail the cognitive state detection device provided in the various embodiments of the present invention, the application scenarios of cognitive state detection can also be illustrated:

[0059] As described in the background section, in practice, any factor that causes abnormalities in the basal ganglia of the brain can lead to cognitive impairment, such as Alzheimer's disease, complications of chronic kidney disease or other diseases, genetic factors, etc. The basal ganglia in the brain can be considered as the target area to be detected. In this case, the cognitive state detection device can perform cognitive state detection on medical images containing the basal ganglia of the subject.

[0060] However, in practice, the structure of the basal ganglia is usually quite complex. In order to improve the accuracy of cognitive state detection, the cognitive state detection device provided in the following embodiments of the present invention can be used.

[0061] The following detailed description of some embodiments of the present invention, in conjunction with the accompanying drawings, is provided. Where there is no conflict between the embodiments, the following embodiments and the features and steps described therein can be combined with each other. Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0062] Figure 1 This is a schematic diagram of a cognitive state detection device provided in an embodiment of the present invention. Optionally, the cognitive state detection device may include a device on any of the following devices: terminal device, medical device, smart wearable device, etc. Specifically, as shown... Figure 1 As shown, the cognitive state detection device may include a first acquisition module 11, a first feature extraction module 12, a first image segmentation module 13, a second feature extraction module 14, and a first determination module 15. Furthermore, it should be noted that the multiple modules of this cognitive state detection device may be integrated into the same device or integrated into different devices; this embodiment of the invention does not impose specific limitations on this.

[0063] The first acquisition module 11 can acquire a medical image containing the basal ganglia of the subject. Optionally, the subject can be a patient with Alzheimer's disease or a patient with chronic kidney disease. Optionally, the medical image can be an MRI image or a QSM image. It is understood that MRI images and QSM images reflect different granularities of basal ganglia information. Furthermore, the medical image can be a three-dimensional image.

[0064] Optionally, the first acquisition module 11 can also acquire medical images corresponding to the brain of the test subject, and extract the medical image containing the basal ganglia from the medical images corresponding to the brain of the test subject.

[0065] For example, suppose the subject has chronic kidney disease, and the medical image to be tested is a QSM image. Optionally, for acquiring the medical image, the first acquisition module 11 can first acquire the medical image corresponding to the brain of the subject with chronic kidney disease. Then, a denoising model or algorithm can be used to denoise the medical image corresponding to the brain. Afterwards, a QSM image containing the basal ganglia of the subject can be extracted from the denoising result.

[0066] Optionally, the denoising model may include Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and so on. The denoising algorithm may include variable, complex harmonic artifact removal algorithms.

[0067] The first feature extraction module 12 can use a feature extraction algorithm to extract features from each voxel in the acquired medical image to obtain a first radiomics feature. This first radiomics feature is a voxel-level feature, which can be represented as an array.

[0068] Optionally, the feature extraction algorithm may include at least one of the following: entropy algorithm, mean absolute deviation algorithm, median algorithm, gray-level difference average algorithm, gray-level difference entropy algorithm, gray-level difference variance algorithm, information metric correlation algorithm 1, information metric correlation algorithm 2, inverse variance algorithm, joint energy algorithm, joint entropy algorithm, and total entropy algorithm.

[0069] Among them, the entropy algorithm is used to reflect the randomness of the gray-level distribution of the medical image under test. The mean absolute deviation algorithm is used to reflect the average deviation between the gray-level values ​​of the medical image under test and the average gray-level values. The median algorithm is used to calculate the median of the gray-level values ​​of the medical image under test. The gray-level difference average algorithm is used to describe the average gray-level difference of pixel pairs in the medical image under test. The gray-level difference entropy algorithm is used to measure the gray-level difference of pixel pairs in the medical image under test. The gray-level difference variance algorithm is used to reflect the variability of the gray-level difference of pixel pairs in the medical image under test. Information measurement correlation algorithm 1 and information measurement correlation algorithm 2 are two different algorithms used to measure the texture irregularity of the medical image under test. The inverse variance algorithm is used to measure the texture uniformity of the medical image under test. The joint energy algorithm is used to reflect the uniformity and repeatability of the texture of the medical image under test. The joint entropy algorithm is used to measure the irregularity and complexity of the texture of the medical image under test. The total entropy algorithm is used to combine the entropy sum of all elements in the gray-level co-occurrence matrix of the medical image under test.

[0070] Continuing with the example above, assuming the medical image to be tested is a QSM image, the feature extraction algorithms include entropy algorithms and inverse variance algorithms. Each voxel in a QSM image includes a magnetic susceptibility value. Therefore, the first radiomics feature of the array can specifically include the magnetic susceptibility value, entropy value, and inverse variance value corresponding to that voxel.

[0071] Furthermore, the first image segmentation module 13 can perform image segmentation on the medical image under test based on the first radiomics features extracted by the first feature extraction module 12 to obtain at least two sub-regions contained in the basal ganglia. The sub-regions may include the globus pallidus, putamen, and caudate nucleus.

[0072] The second feature extraction module 14 can extract features from at least two sub-regions contained in the basal segment to obtain second radiomics features. Optionally, the feature extraction algorithm used in determining the second radiomics features may include at least one of the aforementioned entropy algorithm, gray-level difference entropy algorithm, and inverse variance algorithm.

[0073] Finally, the first determining module 15 can analyze the second radiomics feature using a preset algorithm or its built-in prediction model to determine the cognitive level of the subject. The specific determining process can be found in the description of the following embodiments.

[0074] Optionally, the aforementioned cognitive state detection device may further include a display module. This display module can be used to display the detection results of the cognitive state detection device, for example, displaying the subject's cognitive level on the screen of a terminal device.

[0075] In this embodiment, the first acquisition module 11 acquires a medical image containing the basal ganglia of the test subject. Then, the first feature extraction module 12 extracts features from each voxel in the medical image to obtain first radiomics features. Next, the first image segmentation module 13 segments the medical image based on the first radiomics features to obtain at least two sub-regions contained in the basal ganglia. Then, the second feature extraction module 14 extracts features from the at least two sub-regions to obtain second radiomics features. Finally, the first determination module 15 determines the cognitive level of the test subject based on the second radiomics features, thus achieving cognitive state detection of the test subject.

[0076] As can be seen, the above process can subdivide the basal ganglia into subregions based on voxel-level first radiomics features. Since these voxel-level first radiomics features can accurately describe the image information contained in each pixel of the medical image under test, their use can accurately segment at least two subregions. Subsequently, feature extraction can be performed on each of these at least two subregions, resulting in second radiomics features containing rich information describing the state of each subregion. Furthermore, since cognitive level is directly affected by the state of different subregions in the basal ganglia, the use of second radiomics features can accurately determine the cognitive level of the subject.

[0077] Figure 1 As mentioned in the illustrated embodiment, at least two sub-regions contained in the basal ganglia of the medical image under test can be determined based on the first radiomics features. Specifically, the first image segmentation module 13 can use the k-means clustering algorithm to cluster the first radiomics features to obtain at least two sub-regions contained in the basal ganglia.

[0078] The clustering process is as follows: First, three initial cluster centers are preset. Then, three voxels are randomly selected from the first radiomics features as initial cluster centers. Next, the distance between each voxel in the first radiomics features and the three initial cluster centers is calculated, and each voxel is assigned to the cluster containing the nearest initial cluster center. After that, the cluster center point is recalculated based on the mean of all voxels in each cluster. This process is iterated until the distance between the newly calculated center point and the original initial cluster center is less than a preset threshold, thus obtaining the three sub-regions contained in the basal segment.

[0079] In this embodiment, clustering the first radiomics features using a clustering algorithm can better divide the first radiomics features into at least two sub-regions with similar features, thereby improving the accuracy of image segmentation.

[0080] Figure 1 The illustrated embodiment already mentions that the second radiomics feature can be analyzed using a preset algorithm or its built-in prediction model to determine the cognitive level of the test subject. Specifically, the first determining module 15 can input the second radiomics feature into the prediction model, so that the prediction model outputs a predicted score corresponding to the medical image to be tested. Then, the predicted score corresponding to the medical image to be tested can be compared with a preset threshold, and the cognitive level of the test subject can be determined based on the comparison result.

[0081] In one optional scenario, if the predicted score corresponding to the medical image to be tested is less than a preset threshold, the determining module can determine that the cognitive level of the test subject is normal. In another optional scenario, if the predicted score corresponding to the medical image to be tested is greater than a preset threshold, the determining module can determine that the cognitive level of the test subject is abnormal.

[0082] Optionally, the prediction model may include Extreme Gradient Boosting (XGBoost) models, CNN models, RNN models, etc.

[0083] In this embodiment, by using a prediction model to analyze the second radiomics features, the prediction score corresponding to the medical image to be tested can be accurately predicted. Then, based on the comparison between the prediction score and a preset threshold, it is possible to accurately determine whether there is an abnormality in the cognitive level of the subject.

[0084] As mentioned in the above embodiments, the cognitive level of the test subject can be determined based on a second radiomics feature extracted from at least two sub-regions of the basal ganglia. Based on this, to further improve the accuracy of determining the cognitive level of the test subject, the following... Figure 2The device shown is used to determine the cognitive level of the test subject.

[0085] like Figure 2 As shown, the cognitive state detection device may further include a third feature extraction module 16. This third feature extraction module 16 is used to extract features from medical images corresponding to the basal ganglia of the test subject to obtain third radiomics features. That is, the third radiomics features target the entire region of the basal ganglia. The second radiomics features, on the other hand, target at least two sub-regions of the basal ganglia. Clearly, the granularity of the second and third radiomics features is different.

[0086] Similar to the process described above for determining the cognitive level of a test subject based on second radiomics features, the first determining module 15 can analyze second and third radiomics features of different granularities to determine the cognitive level of the test subject. Specifically, the first determining module 15 can input the second and third radiomics features into a prediction model, so that the prediction model outputs a predicted score corresponding to the medical image to be tested. Then, the first determining module 15 can determine the cognitive level of the test subject based on the comparison result between the predicted score corresponding to the medical image to be tested and a preset threshold. For the specific process of determining the cognitive level of the test subject, please refer to the description in the above embodiments, which will not be repeated here.

[0087] In this embodiment, since the finer-grained second radiomics features can reflect the local details of at least two sub-regions in the medical image under test, and the coarser-grained third radiomics features can reflect the overall information of the basal ganglia region in the medical image under test, inputting the second and third radiomics features of different granularities into the prediction model makes the predicted score of the medical image under test output by the prediction model more accurate. Furthermore, based on the comparison between the predicted score of the medical image under test and a preset threshold, the accuracy of determining the cognitive level of the test subject can be improved.

[0088] In practice, chronic kidney disease (CKD) is associated with cognitive impairment, making it crucial to accurately determine the cognitive level of CKD patients. Optionally, when the test subject is specifically a CKD patient, for ease of understanding, it can be combined with... Figure 3 The operation of the cognitive state detection device is illustrated by example. Figure 3 This is a schematic diagram illustrating the working process of a cognitive state detection device provided in an embodiment of the present invention.

[0089] Optionally, the medical image to be tested can be a QSM image containing the basal ganglia of the patient with chronic kidney disease. Each voxel in the QSM image contains a magnetic susceptibility value. Then, an entropy algorithm can be used to extract features from each voxel in the QSM image to obtain a first feature value. Simultaneously, a mean absolute deviation algorithm can also be used to extract features from each voxel in the QSM image to obtain a second feature value. That is, the first radiomics feature, represented as an array, can include the magnetic susceptibility value, the first feature value, and the second feature value.

[0090] Furthermore, the first radiomics feature can be clustered to obtain the globus pallidus, putamen, and caudate nucleus contained in the basal ganglia of the QSM image. Then, at least one feature extraction algorithm can be used to extract features from the globus pallidus, putamen, and caudate nucleus respectively to obtain the second radiomics feature. The feature extraction algorithm may include the entropy algorithm and the mean absolute deviation algorithm mentioned above, and may also include other feature extraction algorithms. Specific types of feature extraction algorithms can be found in the descriptions of the above embodiments.

[0091] Finally, the second radiomics feature can be input into the prediction model, which outputs a predicted score for the QSM image. Based on the comparison between the predicted score and a preset threshold, the cognitive level of the chronic kidney disease patient is determined. Specifically, if the predicted score is less than the preset threshold, the patient's cognitive level is considered normal. If the predicted score is greater than the preset threshold, the patient's cognitive level is considered abnormal, potentially indicating cognitive impairment.

[0092] Furthermore, to further improve the accuracy of cognitive level determination, feature extraction can also be performed on the QSM image corresponding to the basal ganglia of the chronic kidney disease patient to obtain third radiomics features. These third radiomics features and second radiomics features are then input into the prediction model, which outputs the prediction score corresponding to the QSM image to further determine the cognitive level of the chronic kidney disease patient.

[0093] As described in the above embodiments, the predicted score corresponding to the medical image to be tested can be output by the prediction model. Optionally, such as... Figure 4 As shown, the cognitive state detection device may also include a first training module 17. The working process of the first training module 17 will be described below.

[0094] Optionally, the first training module 17 can acquire training images containing the basal ganglia and reference classification results of the training images from the training image set. Optionally, the training images may include MRI images containing the basal ganglia or QSM images containing the basal ganglia. The reference classification results of the training images may include normal or abnormal cognitive levels.

[0095] Then, the first training module 17 can extract features from each voxel in the training image to obtain first training features. Optionally, the feature extraction algorithm may include at least one. Specific types of feature extraction algorithms can be found in the description of the above embodiments. Next, the first training module 17 can perform clustering processing on the first training features to obtain at least two sub-regions contained in the basal segment. Optionally, the clustering algorithm used in the clustering process may include the k-means clustering algorithm mentioned above. Afterwards, the first training module 17 can use at least one feature extraction algorithm to further extract features from the at least two sub-regions to obtain second training features.

[0096] After obtaining the second training feature, the first training module 17 can input the second training feature as training data into the prediction model, so that the prediction model outputs a prediction score corresponding to the training image. Then, the first training module 17 can determine the predicted classification result of the training image based on the comparison result between the predicted score corresponding to the training image and a preset threshold. Optionally, if the predicted score corresponding to the training image is less than the preset threshold, the predicted classification result of the training image can be determined as normal cognitive level. If the predicted score of the training image is greater than the preset threshold, the predicted classification result of the training image can be determined as abnormal cognitive level.

[0097] Finally, the first training module 17 can calculate the loss between the predicted classification result and the reference classification result, and train the prediction model based on the loss calculation result. Optionally, the prediction model can include any of the XGBoost model, CNN model, and RNN model mentioned above. Optionally, the loss can be calculated using a relative entropy loss function or a cross-entropy loss function, etc.

[0098] The following uses the XGBoost model as an example to illustrate the specific training process of the above prediction model:

[0099] Step 1: The first training module 17 can first use constant values ​​to initialize the model:

[0100]

[0101] in, This represents the initial model. Optionally, the initial model can be a randomly chosen constant value. θ represents the model parameters, which may specifically include the tree structure and the weights of the leaf nodes. L(y iy(θ) represents a differentiable loss function, which can be either the relative entropy loss function or the cross-entropy loss function mentioned above. The loss function is mainly used to calculate the difference between the model's predicted values ​​and the true labels. This difference is used to guide the model training process, i.e., to update the model parameters. N represents the number of training images. i This represents the reference classification result for the i-th training image. Indicates to make The value of θ or the set of values ​​that achieves the minimum value.

[0102] Step 2: Based on the initial model, the first training module 17 can undergo iterative training, as follows:

[0103] In each iteration, the gradient of the loss function, i.e. the first derivative, can be calculated using formula (2).

[0104]

[0105] In the above formula (2), This represents the first derivative (gradient) of the loss function with respect to the model output, which reflects the current state of the model in x. i The direction and magnitude of the error at x. i Let f(x) represent the second training feature corresponding to the i-th training image. i ) represents the output of the current model relative to x. i The corresponding predicted value. m represents the number of iterations. This refers to the predicted value obtained after the (m-1)th iteration. That is, each iteration trains a new predicted value based on the predicted value obtained in the previous iteration.

[0106] Meanwhile, the Hessian matrix, i.e. the second derivative matrix, can be calculated using formula (3).

[0107]

[0108] In the above formula (3), This represents the second derivative of the loss function with respect to the model output (i.e., the diagonal elements of the Hessian matrix), which reflects the current model performance on x. i Error curvature information at the location.

[0109] Furthermore, the first training module 17 can optimize the XGBoost model by fitting a base learner (i.e., a decision tree) based on the calculated gradient and Hessian matrix. Specifically, as shown in formula (4):

[0110]

[0111] The optimization objective of the above formula (4) is to find the base learner from the set Φ that minimizes the loss function. Here, Φ represents the set of all possible base learners. Specifically, we can first calculate φ(x) i )and The mean squared error between the values ​​is used to obtain the mean squared error result. Wherein, φ(x) i ) indicates that the base learner is in x i The predicted value at that location. This is the ratio between the gradient and the diagonal elements of the Hessian matrix, used to provide the model with information on the direction and step size for parameter updates. Then, the diagonal elements of the Hessian matrix can be... With mean squared error results Perform a multiplication operation to obtain the first product result. Alternatively, to simplify the calculation process, the first product result can be... The product is multiplied by the standard coefficients to obtain the second product. Optionally, the standard coefficients can be 1 / 2. Finally, the second product results corresponding to all training images are summed to determine the parameter φ that minimizes the summation from the set Φ, and this parameter φ is defined as... That is, to complete the base learner Fitting.

[0112] After obtaining the base learner Then, as shown in formula (5), the predicted values ​​of the base learner can be... Multiply by the learning rate to obtain the third product. The result of the third product This represents the prediction result output by the model in the m-th iteration. The learning rate can be represented as a.

[0113]

[0114] Finally, the prediction results output by the m-1 round model will be... The prediction results output by the m-round model The results are added together to obtain the predicted result f of the model output in the m-th iteration. m (x).

[0115]

[0116] Step 3: Based on the prediction result f output by the model in the m-th iteration... m (x), determine the XGBoost model.

[0117] For details, please refer to the following formula (7).

[0118]

[0119] To simplify the calculation process, optionally, the prediction result f output by the model in the m-th iteration can be used. m (x) is multiplied by the standard coefficient to obtain the fourth product result. Optionally, the standard coefficient can also be 1 / 2. Then, the fourth product results from all M rounds (i.e., all rounds) are summed to obtain the final sum. The summation result That is, a pre-trained XGBoost model, which can also be used express.

[0120] In this embodiment, a supervised training method is used, that is, the prediction model is trained with reference classification results, which makes the training effect of the prediction model better. Furthermore, during the model use phase, the prediction score corresponding to the medical image to be tested output by the prediction model is also more accurate, thereby improving the accuracy of determining the cognitive level of the test subject.

[0121] exist Figure 4 Based on the illustrated embodiment, to further improve the training effect of the prediction model, the first training module 17 can also extract features from the training image containing the basal ganglia to obtain a third training feature. Then, the first training module 17 can input the second and third training features into the prediction model, so that the prediction model can output the prediction score corresponding to the training image. The specific extraction and training processes can be found in the descriptions of the above embodiments, and will not be repeated here.

[0122] Figure 5 This is a flowchart illustrating a cognitive state detection method provided in an embodiment of the present invention. The cognitive state detection method provided in this embodiment can be executed by the aforementioned cognitive state detection device. It is understood that the cognitive state detection device can be implemented as software, or a combination of software and hardware.

[0123] like Figure 5 As shown, the method includes the following steps:

[0124] S101, Acquire the medical image of the subject, including the basal ganglia of the subject.

[0125] Optionally, the device can acquire a medical image containing the basal ganglia of the subject. Optionally, the subject can be a patient with Alzheimer's disease or a patient with chronic kidney disease. Optionally, the medical image to be acquired can be an MRI image or a QSM image. Furthermore, the medical image to be acquired can be a three-dimensional image.

[0126] Optionally, the device can also acquire medical images corresponding to the brain of the test subject, and extract the medical image containing the basal ganglia from the medical images corresponding to the brain of the test subject. For details, please refer to the description in the above embodiments, which will not be repeated here.

[0127] S102, extract features from each voxel in the medical image to be tested to obtain the first radiomics features.

[0128] S103, based on the first radiomics features, perform image segmentation on the medical image to be tested to obtain at least two sub-regions contained in the basal ganglia.

[0129] S104, extract features from at least two sub-regions to obtain second radiomics features.

[0130] S105, Determine the cognitive level of the subject based on the second radiomics characteristics.

[0131] The device can then extract features from each voxel in the medical image under test to obtain a first radiomics feature. This first radiomics feature is a voxel-level feature, which can be represented as an array. Optionally, the feature extraction algorithm can include at least one, and specific types can be found in [reference needed]. Figure 1 The description in the illustrated embodiment.

[0132] Furthermore, the device can perform image segmentation on the medical image under test based on this first radiomics feature to obtain at least two sub-regions contained in the basal ganglia. These sub-regions may include the globus pallidus, putamen, and caudate nucleus.

[0133] Subsequently, the device can use at least one of the aforementioned feature extraction algorithms to extract features from at least two sub-regions contained in the basal segment to obtain second radiomics features.

[0134] Ultimately, the device can use a preset algorithm or its built-in predictive model to analyze the second radiomics feature to determine the cognitive level of the subject.

[0135] In this embodiment, a medical image containing the basal ganglia of the subject is acquired. Then, features are extracted from each voxel in the medical image to obtain first radiomics features. Next, the medical image is segmented based on the first radiomics features to obtain at least two sub-regions contained in the basal ganglia. Finally, the cognitive level of the subject is determined based on the second radiomics features, thus achieving the detection of the subject's cognitive state.

[0136] As can be seen, the above process can subdivide the basal ganglia into subregions based on voxel-level first radiomics features. Since these voxel-level first radiomics features can accurately describe the image information contained in each pixel of the medical image under test, their use can accurately segment at least two subregions. Subsequently, feature extraction can be performed on each of these at least two subregions, resulting in second radiomics features containing rich information describing the state of each subregion. Furthermore, since cognitive level is directly affected by the state of different subregions in the basal ganglia, the use of second radiomics features can accurately determine the cognitive level of the subject.

[0137] In addition, the contents not described in detail in this embodiment and the technical effects that can be achieved can also be found in [reference needed]. Figure 1 The embodiments shown will not be described in detail here.

[0138] Figure 6 This is a flowchart illustrating a model training method provided in an embodiment of the present invention. The executing entity of this method can also be the aforementioned cognitive state detection device, specifically the first training module 17 within that cognitive state detection device, or a separate model training device, used to improve the training effect of the prediction model. Figure 6 As shown, the method may include the following steps:

[0139] S201, Obtain training images containing the basal ganglia and reference classification results for the training images.

[0140] This device can acquire training images containing the basal ganglia and reference classification results for the training images from a training image set. Optionally, the training images may include MRI images containing the basal ganglia or QSM images containing the basal ganglia. The reference classification results for the training images may include normal or abnormal cognitive levels.

[0141] S202, extract features from each voxel in the training image to obtain the first training features.

[0142] S203, based on the first training feature, perform image segmentation on the training image to obtain at least two sub-regions contained in the basal segment.

[0143] S204, extract features from at least two sub-regions to obtain second training features.

[0144] Then, the device can use at least one feature extraction algorithm to extract features from each voxel in the training image to obtain a first training feature. Optionally, the feature extraction algorithm may include at least one, as described in the above embodiments. Next, the first training feature can be clustered to obtain at least two sub-regions contained in the basal ganglia. The sub-regions may include the globus pallidus, putamen, and caudate nucleus. Optionally, the clustering process may also be described in the above embodiments. Afterward, at least one feature extraction algorithm can be used to extract features from the at least two sub-regions again to obtain a second training feature.

[0145] S205, input the second training feature into the prediction model so that the prediction model outputs the prediction score corresponding to the training image.

[0146] S206. Based on the prediction scores corresponding to the training images, determine the prediction classification results of the training images.

[0147] S207. Train the prediction model based on the loss calculation results between the predicted classification results and the reference classification results.

[0148] After obtaining the second training feature, this second training feature can be input into the prediction model so that the prediction model outputs the prediction score corresponding to the training image. Based on the comparison between the prediction score corresponding to the training image and the preset threshold, the prediction classification result of the training image is determined.

[0149] Finally, the training module can calculate the loss between the predicted classification result and the reference classification result, and train the prediction model based on the loss calculation result. Optionally, the prediction model can include any one of the XGBoost model, CNN model, and RNN model. Optionally, the loss can be calculated using KL divergence or cross-entropy loss function, etc.

[0150] In this embodiment, a supervised training method is used, that is, the prediction model is trained with reference classification results, which makes the training effect of the prediction model better. Furthermore, during the model use phase, the prediction score corresponding to the medical image to be tested output by the prediction model is also more accurate, thereby improving the accuracy of determining the cognitive level of the test subject.

[0151] In addition, the contents not described in detail in this embodiment and the technical effects that can be achieved can also be found in [reference needed]. Figure 4 The embodiments shown will not be described in detail here.

[0152] In one possible design, Figure 5 The cognitive state detection method in the illustrated embodiment can be applied to an electronic device, such as... Figure 7As shown, the electronic device may include a first processor 21 and a first memory 22. The first memory 22 is used to store data supporting the electronic device in performing the above-described actions. Figure 5 In the cognitive state detection method provided in the illustrated embodiment, the first processor 21 is configured to execute the program stored in the first memory 22.

[0153] The program includes one or more computer instructions, wherein when executed by the first processor 21, the one or more computer instructions can perform the following steps:

[0154] Acquire medical images containing the basal ganglia of the subject;

[0155] Feature extraction is performed on each voxel in the medical image to be tested to obtain the first radiomics feature;

[0156] Based on the first radiomics features, the medical image to be tested is segmented to obtain at least two sub-regions contained in the basal ganglia;

[0157] Feature extraction is performed on the at least two sub-regions to obtain second radiomics features;

[0158] The cognitive level of the subject is determined based on the second radiomics features.

[0159] Optionally, the first processor 21 is also used to perform the aforementioned Figure 5 All or part of the steps in the illustrated embodiments.

[0160] The structure of the electronic device may also include a first communication interface 23 for the electronic device to communicate with other devices or communication systems.

[0161] In addition, embodiments of the present invention provide a computer storage medium for storing computer software instructions used by the aforementioned electronic device, which includes instructions for executing the above-mentioned... Figure 5 The procedure involved in the cognitive state detection method shown.

[0162] In addition, embodiments of the present invention provide a computer program product. This computer program product includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is able to perform the above-described functions. Figure 5 The steps or functions of the method shown.

[0163] Figure 8 This is a schematic diagram of the structure of a model training device provided in an embodiment of the present invention, as shown below. Figure 8 As shown, the device includes:

[0164] The second acquisition module 31 is used to acquire training images containing basal ganglia and reference classification results of the training images.

[0165] The fourth feature extraction module 32 is used to extract features from each voxel in the training image to obtain the first training features.

[0166] The second image segmentation module 33 is used to segment the training image according to the first training features to obtain at least two sub-regions contained in the basal segment.

[0167] The fifth feature extraction module 34 is used to extract features from the at least two sub-regions to obtain the second training features.

[0168] The input module 35 is used to input the second training feature into the prediction model so that the prediction model outputs the prediction score corresponding to the training image.

[0169] The second determining module 36 is used to determine the prediction classification result of the training image based on the prediction score corresponding to the training image.

[0170] The second training module 37 is used to train the prediction model based on the loss calculation result between the predicted classification result and the reference classification result.

[0171] Figure 8 The device shown can perform Figure 6 For the methods shown in the embodiments, the parts not described in detail in this embodiment can be referred to the following: Figure 6 The relevant descriptions of the illustrated embodiments are provided below. For the execution process and technical effects of this technical solution, please refer to [link / reference]. Figure 6 The descriptions in the illustrated embodiments will not be repeated here.

[0172] In one possible design, the model training methods provided in the above embodiments can be applied to another electronic device, such as... Figure 9 As shown, the electronic device may include a second processor 41 and a second memory 42. The second memory 42 is used to store data supporting the electronic device in performing the above-described actions. Figure 6 In the model training method program provided in the illustrated embodiment, the second processor 41 is configured to execute the program stored in the second memory 42.

[0173] The program includes one or more computer instructions, wherein the one or more computer instructions, when executed by the second processor 41, can perform the following steps:

[0174] Obtain training images containing basal ganglia and reference classification results of the training images;

[0175] Feature extraction is performed on each voxel in the training image to obtain the first training feature;

[0176] Based on the first training feature, the training image is segmented to obtain at least two sub-regions contained in the basal segment;

[0177] Feature extraction is performed on the at least two sub-regions to obtain the second training features;

[0178] The second training feature is input into the prediction model so that the prediction model outputs the prediction score corresponding to the training image;

[0179] Based on the prediction scores corresponding to the training images, the prediction classification results of the training images are determined.

[0180] The prediction model is trained based on the loss calculation result between the predicted classification result and the reference classification result.

[0181] Optionally, the second processor 41 is also used to perform the aforementioned Figure 6 All or part of the steps in the illustrated embodiments.

[0182] The structure of the electronic device may also include a second communication interface 43 for the electronic device to communicate with other devices or communication systems.

[0183] In addition, embodiments of the present invention provide a computer storage medium for storing computer software instructions used by the aforementioned electronic device, which includes instructions for executing the above-mentioned... Figure 6 The procedure involved in the model training method shown.

[0184] In addition, embodiments of the present invention provide a computer program product. This computer program product includes a computer program or instructions. When the computer program or instructions are executed by a processor, the processor is able to perform the above-described functions. Figure 6 The steps or functions of the method shown.

[0185] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cognitive state detection device, characterized in that, include: The first acquisition module is used to acquire a medical image containing the basal ganglia of the subject, the medical image including a quantitative magnetic susceptibility image; The first feature extraction module is used to extract features from each voxel in the medical image to be tested in order to obtain the first radiomics features. The first image segmentation module is used to segment the medical image to be tested according to the first radiomics features to obtain at least two sub-regions contained in the basal ganglia; The second feature extraction module is used to extract features from the at least two sub-regions to obtain second radiomics features; The first determining module is used to input the second radiomics feature and the third radiomics feature into the prediction model, so that the prediction model outputs the prediction score corresponding to the medical image to be tested; the third radiomics feature is obtained by the third feature extraction module through feature extraction of the medical image corresponding to the basal ganglia of the subject; the cognitive level of the subject is determined according to the comparison result between the prediction score corresponding to the medical image to be tested and the preset threshold.

2. The apparatus according to claim 1, characterized in that, The first acquisition module is used to acquire medical images of the brain of the subject, wherein the subject has chronic kidney disease; The medical images corresponding to the brain of the subject were subjected to noise reduction processing; From the noise reduction results, extract the medical image containing the basal ganglia of the subject.

3. The apparatus according to claim 1, characterized in that, The first image segmentation module is used to perform clustering processing on the first radiomics features to obtain at least two sub-regions contained in the basal ganglia of the medical image to be tested.

4. The apparatus according to claim 1, characterized in that, The first determining module is used to input the second radiomics features into the prediction model so that the prediction model outputs the prediction score corresponding to the medical image to be tested; The cognitive level of the test subject is determined by comparing the predicted score corresponding to the medical image under test with a preset threshold.

5. The apparatus according to claim 1, characterized in that, The device also includes a third feature extraction module; The third feature extraction module is used to extract features from the medical images corresponding to the basal ganglia of the subject to obtain third radiomics features. The first determining module is used to determine the cognitive level of the subject based on the second radiomics features and the third radiomics features.

6. The apparatus according to claim 1 or 4, characterized in that, The first determining module is used to determine that the cognitive level of the test subject is normal if the predicted score corresponding to the medical image to be tested is less than the preset threshold. If the predicted score corresponding to the medical image to be tested is greater than the preset threshold, then the cognitive level of the subject to be tested is determined to be abnormal.

7. The apparatus according to claim 4, characterized in that, The device further includes a first training module, the first training module being used for: Obtain training images containing basal ganglia and reference classification results of the training images; Feature extraction is performed on each voxel in the training image to obtain the first training feature; Based on the first training feature, the training image is segmented to obtain at least two sub-regions contained in the basal segment; Feature extraction is performed on the at least two sub-regions to obtain the second training features; The second training feature is input into the prediction model so that the prediction model outputs the prediction score corresponding to the training image; Based on the prediction scores corresponding to the training images, the prediction classification results of the training images are determined. The prediction model is trained based on the loss calculation result between the predicted classification result and the reference classification result.

8. The apparatus according to claim 7, characterized in that, The first training module is used for: Feature extraction is performed on the training images to obtain a third training feature; The second training feature and the third training feature are input into the prediction model so that the prediction model outputs the prediction score corresponding to the training image.

9. A method for detecting cognitive states, characterized in that, include: Acquire a medical image containing the basal ganglia of the subject, the medical image including a quantitative magnetic susceptibility image; Feature extraction is performed on each voxel in the medical image to be tested to obtain the first radiomics feature; Based on the first radiomics features, the medical image to be tested is segmented to obtain at least two sub-regions contained in the basal ganglia; Feature extraction is performed on the at least two sub-regions to obtain second radiomics features; The second and third radiomics features are input into the prediction model, and the prediction model outputs the predicted score corresponding to the medical image to be tested; the third radiomics feature is obtained by feature extraction of the medical image corresponding to the basal ganglia of the subject; the cognitive level of the subject is determined according to the comparison between the predicted score corresponding to the medical image to be tested and the preset threshold.

10. A model training method, characterized in that, include: Acquire training images containing the basal ganglia and reference classification results of the training images, wherein the training images include quantitative magnetic susceptibility images; Feature extraction is performed on each voxel in the training image to obtain the first training feature; Based on the first training feature, the training image is segmented to obtain at least two sub-regions contained in the basal segment; Feature extraction is performed on the at least two sub-regions to obtain the second training features; Feature extraction is performed on the training images to obtain a third training feature; The second training feature and the third training feature are input into the prediction model so that the prediction model outputs the prediction score corresponding to the training image. Based on the prediction scores corresponding to the training images, the prediction classification results of the training images are determined. The prediction model is trained based on the loss calculation result between the predicted classification result and the reference classification result.

11. A model training device, characterized in that, include: The second acquisition module is used to acquire training images containing the basal ganglia and reference classification results of the training images, wherein the training images include quantitative magnetic susceptibility images. The fourth feature extraction module is used to extract features from each voxel in the training image to obtain the first training feature; The second image segmentation module is used to segment the training image according to the first training features to obtain at least two sub-regions contained in the basal segment; The fifth feature extraction module is used to extract features from the at least two sub-regions to obtain the second training features; The fifth feature extraction module is further used to extract features from the training image to obtain the third training features. The input module is used to input the second training feature and the third training feature into the prediction model so that the prediction model outputs the prediction score corresponding to the training image; The second determining module is used to determine the prediction classification result of the training image based on the prediction score corresponding to the training image. The second training module is used to train the prediction model based on the loss calculation result between the predicted classification result and the reference classification result.

12. An electronic device, characterized in that, include: The memory and the processor; wherein the memory stores executable code, and when the executable code is executed by the processor, the processor performs the cognitive state detection method as described in claim 9 or the model training method as described in claim 10.

13. A non-transitory machine-readable storage medium, characterized in that, The non-transitory machine-readable storage medium stores executable code, which, when executed by a processor of an electronic device, causes the processor to perform the cognitive state detection method as described in claim 9 or the model training method as described in claim 10.

14. A computer program product, characterized in that, The computer program product includes a computer program or instructions that enable the computer program or instructions to perform the steps in the cognitive state detection method of claim 9 or the model training method of claim 10.

Images