Apparatus for ischemic stroke prognosis prediction and storage medium

Abstract

The application discloses an apparatus and a storage medium for ischemic stroke prognosis prediction. The apparatus comprises a processor and a memory, and the apparatus implements the following operations: performing image feature extraction and spatial feature extraction on a first cranial image respectively to obtain first image features and first spatial features; performing image feature extraction and spatial feature extraction on a second cranial image respectively to obtain second image features and second spatial features; performing fusion operations on the first image features and the first spatial features, the second image features and the second spatial features respectively to obtain corresponding first fusion features and second fusion features; and performing ischemic stroke prognosis prediction based on a splicing result of the first fusion features and the second fusion features to obtain a prediction result. The scheme of the application can efficiently fuse multi-modal information, improve spatial structure perception capability, and improve prediction accuracy.

Description

Technical Field

[0001] This application generally relates to the field of medical image analysis technology. More specifically, this application relates to a device and a computer-readable storage medium for predicting the prognosis of ischemic stroke. Background Technology

[0002] Ischemic stroke is caused by the interruption of blood flow to the brain, leading to hypoxia and ischemia in brain tissue, often resulting in neurological dysfunction and serious clinical consequences. Due to the complexity of patients' conditions and significant individual differences in pathological characteristics, accurate prognostic assessment is crucial for clinical treatment decisions. Current prognostic assessment mainly relies on physicians' comprehensive analysis of CT perfusion imaging (“CTP”), CT angiography (“CTA”), and clinical information. However, manual analysis is highly subjective and struggles to accurately model the complex relationships between multiple modalities.

[0003] While deep learning models perform well in image segmentation and classification tasks, their direct application to predicting the prognosis of ischemic stroke has significant limitations: their multimodal fusion methods are simplistic and fail to adequately model the interaction between images and structured data; they lack spatial structure perception capabilities, making it difficult to capture crucial information such as perfusion boundaries and brain region topology; and their prediction results are opaque, resulting in poor clinical interpretability. These issues limit the accuracy and practicality of existing methods in predicting the prognosis of ischemic stroke.

[0004] In view of this, this application provides a scheme for predicting the prognosis of ischemic stroke, which can efficiently integrate multimodal information, improve spatial structure perception, and enhance the interpretability of prediction results, thereby achieving accurate prediction of the prognosis of ischemic stroke patients. Summary of the Invention

[0005] In order to at least address one or more of the technical problems mentioned above, this application proposes a scheme for predicting the prognosis of ischemic stroke in several aspects.

[0006] In a first aspect, this application provides an apparatus for predicting the prognosis of ischemic stroke, including a processor and a memory, wherein the memory stores program instructions for predicting the prognosis of ischemic stroke. When the program instructions are executed by the processor, the apparatus performs the following operations: acquiring a first cranial image and a second cranial image; using a first feature module and a first geometry module to extract image features and spatial features from the first cranial image respectively, to obtain a first image feature and a first spatial feature; using a second feature module and a second geometry module to extract image features and spatial features from the second cranial image respectively, to obtain a second image feature and a second spatial feature; using a first fusion module and a second fusion module to perform fusion operations on the first image feature and the first spatial feature, and the second image feature and the second spatial feature respectively, to obtain corresponding first fusion features and second fusion features; and using a prediction module to predict the prognosis of ischemic stroke based on the splicing result of the first fusion features and the second fusion features, to obtain a prediction result.

[0007] In some embodiments, the first brain image includes CT perfusion imaging, and the second brain image includes CT angiography.

[0008] In some embodiments, the first spatial feature includes a gradient feature, and the second spatial feature includes a distance feature.

[0009] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations to obtain the first spatial feature: calculates gradient information of each pixel in the first cranial image using the first geometry module to obtain a gradient map; extracts a target region map under the target parameters from the first cranial image; and overlays the gradient map on the target region map to obtain the first spatial feature.

[0010] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations to obtain the second spatial feature: extracting the vascular centerline from the second cranial image using the second geometry module; calculating the distance from each pixel in the second cranial image to the vascular centerline to obtain the second spatial feature.

[0011] In some embodiments, when the program instructions are executed by the processor, the apparatus further performs the following operations to obtain the first fusion feature: grouping the first image features according to the target parameters; using the first fusion module to perform attention fusion operations between the first image features of each group and the first spatial features respectively to obtain attention features corresponding to each group; and fusing the attention features corresponding to each group to obtain the first fusion feature.

[0012] In some embodiments, the target parameters include at least the time to peak contrast agent concentration, the amount of blood flowing through brain tissue per unit time, the blood volume within a unit brain tissue, and the average time for the contrast agent to pass through the vascular bed.

[0013] In some embodiments, when the program instructions are executed by the processor, the apparatus further performs the following operation to obtain the second fusion feature: using the second fusion module to perform an attention fusion operation on the second image feature and the second spatial feature to obtain the second fusion feature.

[0014] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operations: normalizing clinical information; extracting feature vectors of the normalized clinical information using a multilayer sensing module; and fusing the feature vectors with the concatenation result.

[0015] In some embodiments, the clinical information includes at least age, sex, blood pressure, blood glucose, blood oxygen saturation, atrial fibrillation, time from onset to treatment, disease history, and NIHSS score.

[0016] In some embodiments, when the program instructions are executed by the processor, the device further performs the following operation: performing a dimensionality reduction operation on the splicing result.

[0017] In a second aspect, this application provides a computer-readable storage medium having stored thereon computer program instructions for predicting the prognosis of ischemic stroke, which, when executed by one or more processors, cause one or more embodiments of the first aspect to be implemented.

[0018] The above-described scheme for predicting the prognosis of ischemic stroke, in this embodiment, acquires first and second cranial images and performs feature extraction and spatial feature extraction respectively. Combined with a fusion module and a prediction module, it achieves accurate prediction of the prognosis of ischemic stroke, solves the problems of insufficient multimodal fusion, lack of spatial structure perception and poor interpretability in the prior art, and improves the accuracy and clinical applicability of prognosis prediction. Attached Figure Description

[0019] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein:

[0020] Figure 1This is an exemplary structural block diagram illustrating an apparatus 100 for predicting the prognosis of ischemic stroke according to an embodiment of this application.

[0021] Figure 2 This is an exemplary flowchart illustrating the operation 200 performed by a device for predicting the prognosis of ischemic stroke according to an embodiment of this application;

[0022] Figure 3 This is an exemplary schematic diagram illustrating the overall operation performed by a device for predicting the prognosis of ischemic stroke according to an embodiment of this application.

[0023] Figure 4 This is an exemplary structural block diagram illustrating a device according to an embodiment of this application. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] It should be understood that the terms "comprising" and "including" used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or collections thereof.

[0026] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0027] As used in this specification and claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if [described condition or event] is detected" may be interpreted, depending on the context, as "once determined," "in response to determination," "once [described condition or event] is detected," or "in response to detection of [described condition or event]."

[0028] The specific embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0029] Figure 1 This is an exemplary structural block diagram illustrating an apparatus 100 for predicting the prognosis of ischemic stroke according to an embodiment of this application. Figure 1 As shown, the device 100 may include a processor 101 and a memory 102. The processor 101 may include, for example, a general-purpose processor (“CPU”) or a dedicated graphics processor (“GPU”), and the memory 102 stores program instructions executable on the processor. In some embodiments, the memory 102 may include, but is not limited to, resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), and enhanced dynamic random access memory (EDRAM).

[0030] Furthermore, the aforementioned memory 102 may store program instructions for predicting the prognosis of ischemic stroke. When the program instructions are executed by the processor, the device 100 performs the following operations: acquiring a first cranial image and a second cranial image; using a first feature module and a first geometry module to extract image features and spatial features from the first cranial image respectively, to obtain first image features and first spatial features; using a second feature module and a second geometry module to extract image features and spatial features from the second cranial image respectively, to obtain second image features and second spatial features; using a first fusion module and a second fusion module to perform fusion operations on the first image features and first spatial features, and the second image features and second spatial features respectively, to obtain corresponding first fusion features and second fusion features; and using a prediction module to predict the prognosis of ischemic stroke based on the splicing result of the first fusion features and the second fusion features, to obtain a prediction result.

[0031] In some embodiments, the first cranial image may include CT perfusion imaging (“CTP”), and the second cranial image may include CT angiography (“CTA”). The CT perfusion imaging includes multiple plots under target parameters, which may be, for example, time to peak contrast agent (“Tmax”), blood flow through brain tissue per unit time (“CBF”), blood volume within a unit brain tissue (“CBV”), and mean time to cross the contrast agent vascular bed (“MTT”).

[0032] Based on the extracted CTP and CTA, image feature extraction and spatial feature extraction are performed on the first cranial image using a first feature module and a first geometric module, respectively, to obtain first image features and first spatial features. Similarly, image feature extraction and spatial feature extraction are performed on the second cranial image using a second feature module and a second geometric module, respectively, to obtain second image features and second spatial features. In some implementations, the aforementioned first and second feature modules can be, for example, CNN encoders. By performing multi-scale encoding operations on the CTP and CTA using a CNN encoder, multi-scale first and second image features can be obtained.

[0033] In some embodiments, the first spatial feature includes a gradient feature, and the second spatial feature includes a distance feature. For the first spatial feature, in some embodiments, a first geometric module can be used to calculate the gradient information of each pixel in the first cranial image to obtain a gradient map. Then, a target region map under the target parameters is extracted from the first cranial image, and the gradient map is superimposed on the target region map to obtain the first spatial feature. In some implementation scenarios, the aforementioned first geometric module can be, for example, a Sobel 2D filter to calculate the gradient intensity information of each pixel to obtain a gradient map. The extracted target region map under the target parameters includes, for example, target region maps corresponding to Tmax, CBF, CBV, and MTT. As an example, a target region map of tmax > 6s can be extracted, and the specific values ​​of each parameter (e.g., tmax > 6s) can be set according to actual needs; this application does not impose any limitations in this regard. Next, the first spatial feature (i.e., gradient feature or boundary gradient map) is obtained by superimposing (e.g., multiplying corresponding pixels) the gradient map on the target region map.

[0034] For the second spatial features, in some embodiments, a second geometry module can be used to extract the vessel centerline from the second cranial image, and then calculate the distance from each pixel in the second cranial image to the vessel centerline to obtain the second spatial features. In some implementation scenarios, the second geometry module can be, for example, a vessel segmentation model. The vessel centerline is extracted by combining the vessel segmentation model with, for example, the VMTK algorithm. The second spatial features (e.g., distance features or distance map) are obtained by calculating the Euclidean distance from each pixel in the second cranial image to the vessel centerline.

[0035] In some implementation scenarios, before inputting the first and second cranial images into the corresponding feature modules and geometry modules, preprocessing operations can be performed on the first and / or second cranial images. These preprocessing operations can include, for example, standardization and spatial registration, interpolation, or zeroing, to ensure uniform value range and improve the accuracy of subsequent processing.

[0036] Based on the obtained first image features and first spatial features, second image features and second spatial features, a first fusion module and a second fusion module can be used to perform fusion operations on the first image features and first spatial features, the second image features and second spatial features, respectively, to obtain corresponding first fusion features and second fusion features. Specifically, in some embodiments, the first image features are first grouped according to target parameters, and the first fusion module is used to perform attention fusion operations on the first image features of each group with the first spatial features to obtain attention features corresponding to each group. The attention features corresponding to each group are then fused to obtain the first fusion feature. In other embodiments, the second fusion module is used to perform attention fusion operations on the second image features and second spatial features to obtain the second fusion feature.

[0037] In some implementation scenarios, the first and second fusion modules can be, for example, a geometrically self-attention-based cross-modal fusion module (“GSA”). This GSA-based cross-modal fusion module can be a module that models cross-scale dependencies between multiple modalities using the self-attention mechanism proposed in the DFormerv2 model. In the implementation scenario, for CTP parameter maps, image features are first grouped according to different parameter maps (e.g., Tmax group, CBF group). The image features of each group are then subjected to an attention operation with the calculated gradient map, and then concatenated to obtain the first fused feature. Based on this, the unique information of different parameter maps can be preserved, while the gradient map enhances the focus on lesion boundaries. For CTA, the image features and distance map are subjected to an attention operation to obtain the second fused feature that integrates vascular morphology and spatial distance information. This simplifies the calculation process and uses the distance map to guide the model to focus on the spatial relationship between blood vessels and their surrounding areas.

[0038] In some implementations, during the aforementioned attention operation, a bias matrix can be constructed by calculating the Manhattan distance (i.e., the sum of the absolute values ​​of the coordinate differences) between any two locations in the image features. By incorporating spatial distance information into the attention calculation, more attention is paid to spatially adjacent regions or regions with specific geometric relationships.

[0039] Furthermore, a prediction module is used to predict the prognosis of ischemic stroke based on the concatenation result of the first and second fusion features to obtain a prediction result. In some implementation scenarios, the prediction module can be, for example, a fully connected layer, through which the final prediction result is output. This prediction result is a probability value of the prognosis for ischemic stroke patients.

[0040] In some embodiments, when the program instructions are executed by the processor, the device 100 further performs the following operations: normalizing the clinical information, extracting the feature vector of the normalized clinical information using a multilayer sensing module, and fusing the feature vector with the splicing result. That is, clinical indicators are added before outputting the prediction result. Clinical indicators reflect the patient's overall condition and disease progression, complementing imaging features (such as blood perfusion and vascular morphology) to improve prediction accuracy. In some embodiments, the clinical information includes at least age, sex, blood pressure, blood glucose, blood oxygen saturation, atrial fibrillation, time from onset to treatment, disease history, and NIHSS score.

[0041] In some embodiments, when the program instructions are executed by the processor, the device 100 further performs the following operation: dimensionality reduction of the splicing result. For example, dimensionality reduction is performed by using a certain number of 1*1 convolutional blocks (1*1 convolution + batch normalization layer + ReLU activation layer), and the dimensionality-reduced splicing result is input into the prediction module to predict the prognosis of ischemic stroke patients, thereby obtaining the prediction result. Based on this, by mapping the high-dimensional features of CTP, CTA, and clinical data spliced ​​to a low-dimensional space (e.g., 64-dimensional), key diagnostic information such as ischemic boundary gradient, vascular distance features, and NIHSS scores can be preserved while reducing model parameters, improving computational efficiency, and effectively avoiding overfitting problems under small sample medical data. In addition, the dimensionality-reduced features, after nonlinear transformation, can more accurately capture the interaction relationship of multiple factors such as "degree of vascular occlusion × ischemic penumbra range × patient age," thereby improving prediction accuracy.

[0042] As described above, this application's embodiments acquire first cranial images (such as CTP) and second cranial images (such as CTA), extracting image features and spatial features (such as gradient features and distance features) using feature modules and geometric modules respectively. After processing by a fusion module, these features are combined with a prediction module to achieve prognostic prediction. This effectively solves the problems of simple multimodal fusion and insufficient spatial structure perception in existing technologies, improving the modeling ability for key information such as ischemic boundaries and vascular distribution. Furthermore, the generation of gradient and distance maps enhances the spatial representation of perfusion boundaries and vascular topology. Grouped attention fusion and normalization processing of clinical information (including age, NIHSS score, etc.) not only refine the cross-modal interaction of multi-scale features, but also improves the personalized adaptation capability of the prediction model by introducing structured clinical data. Dimensionality reduction and fully connected layer prediction optimize computational efficiency while ensuring feature integrity, ultimately achieving accurate prediction of the prognosis (e.g., 90-day mRS score) for ischemic stroke patients. Moreover, the prediction process can be associated with specific brain region structures, enhancing clinical interpretability.

[0043] Figure 2This is an exemplary flowchart illustrating operations 200 performed by a device for predicting the prognosis of ischemic stroke according to an embodiment of this application. It should be understood that... Figure 2 The above Figure 1 One specific embodiment, therefore the above regarding Figure 1 The description also applies to Figure 2 .

[0044] like Figure 2 As shown, in steps S201 and S202, a first cranial image and a second cranial image are acquired, respectively. In some embodiments, the first cranial image may be, for example, a CTP (Cognitive Processing) image, and the second cranial image may be, for example, a CTA (Cognitive Aspect-Oriented Image). For the first cranial image, in steps S203 and S204, its first image features and first spatial features are extracted, respectively. In some implementation scenarios, image feature extraction can be performed using, for example, a first feature module (e.g., a CNN encoder) to obtain the first image features. Spatial feature extraction can be performed using a first geometry module (e.g., a Sobel 2D filter) to obtain the first spatial features. Specifically, a Sobel 2D filter can be used to calculate the gradient intensity information of each pixel to obtain a gradient map. A target region map under the target parameters is extracted from the first cranial image, and the gradient map is superimposed on the target region map to obtain the first spatial features (i.e., gradient features or boundary gradient map).

[0045] For the second cranial image, in steps S205 and S206, its second image features and second spatial features are extracted, respectively. In some implementation scenarios, image feature extraction can be performed using, for example, a second feature module (e.g., a CNN encoder) to obtain the second image features. The vessel centerline is extracted using a second geometry module (e.g., a vessel segmentation model) combined with the VMTK algorithm, and the Euclidean distance from each pixel in the second cranial image to the vessel centerline is calculated to obtain the second spatial features (e.g., distance features or a distance map).

[0046] Next, in steps S207 and S208, a fusion operation is performed on the first image features and the first spatial features to obtain the corresponding first fused features; a fusion operation is performed on the second image features and the second spatial features to obtain the corresponding second fused features. In some implementation scenarios, the first fusion module and the second fusion module can be, for example, a cross-modal fusion module based on geometric self-attention (“GSA”). Specifically, for the CTP parameter map, the first fusion module is used to group the first image features according to the target parameters, and the first fusion module performs attention fusion operations on the first image features of each group with the first spatial features to obtain the attention features corresponding to each group. The attention features corresponding to each group are then fused to obtain the first fused features. For CTA, the second fusion module performs attention fusion operations on the second image features and the second spatial features to obtain the second fused features.

[0047] Further, in step S209, the first fusion feature and the second fusion feature are concatenated to obtain a concatenation result. Finally, in step S210, based on this concatenation result, the prediction module outputs a prediction result for the prognosis of ischemic stroke patients. In some implementations, the prediction module can be, for example, a fully connected layer. In some embodiments, the concatenation result can also be dimensionality reduced and / or clinical indicator features can be concatenated to obtain the final prediction result. For more details, please refer to the above. Figure 1 The content described herein will not be repeated here.

[0048] Figure 3 This is an exemplary schematic diagram illustrating the overall operation performed by a device for predicting the prognosis of ischemic stroke according to an embodiment of this application. It should be understood that... Figure 3 The above Figure 1 Another specific embodiment, therefore the above regarding Figure 1 The description also applies to Figure 3 .

[0049] like Figure 3 As shown, the first image features 303 are extracted by inputting CTP 301 into CNN encoder 302, and then gradient map 305 is obtained by inputting it into Sobel 2D filter 304. Gradient map 305 is then superimposed on target region map 306 to obtain boundary gradient map 307. Next, attention fusion operation is performed through GSA 308 to obtain first fused features 309.

[0050] The figure further illustrates how the second image features 312 are extracted by inputting CTA 310 into a CNN encoder 311, inputting them into a blood vessel segmentation model 313, and combining them with the VMTK algorithm to obtain the blood vessel centerline 314. A distance map 315 is then obtained by calculating the Euclidean distance from each pixel in the second cranial image to the blood vessel centerline. Next, an attention fusion operation is performed using GSA 316 to obtain the second fused feature 317. Furthermore, the first fused feature 309 and the second fused feature 317 are concatenated to obtain the concatenated result 318.

[0051] In some implementation scenarios, feature vectors extracted based on clinical information 319 can also be concatenated to obtain the final concatenated result. The final concatenated result is then dimensionality-reduced 320 and input into the prediction module 321 for ischemic stroke prognosis prediction to obtain the prediction result 322. For more details, please refer to the above. Figure 1 The content described herein will not be repeated here.

[0052] By utilizing the embodiments of this application, first and second cranial images are acquired, and image features and spatial features (such as gradient features and distance features) are extracted using feature modules and geometric modules, respectively. After processing by the fusion module, these features are combined with the prediction module to achieve prognostic prediction of ischemic stroke. This effectively solves the problems of simple multimodal fusion and insufficient spatial structure perception in existing technologies, and improves the modeling ability for key information such as ischemic boundaries and vascular distribution. Furthermore, by combining the complementarity of CTP and CTA multimodal data, the spatial representation of perfusion boundaries and vascular topology by gradient maps and distance maps, group attention fusion based on target parameters, normalization processing of clinical information (including age, NIHSS score, etc.), and feature vector fusion, the cross-modal interaction of multi-scale features is refined and the personalized adaptation ability of the prediction model is improved. At the same time, the computational efficiency is optimized through dimensionality reduction operations, ultimately achieving accurate prediction of the 90-day mRS score of ischemic stroke patients. Moreover, the prediction process can be associated with specific brain region structures, enhancing clinical interpretability.

[0053] Figure 4 This is an exemplary structural block diagram illustrating a device 400 according to an embodiment of this application. It is understood that the device 400 may include the apparatus of the embodiments of this application, and the device implementing the solution of this application may be a single device (e.g., a computing device) or a multifunctional device including various peripheral devices.

[0054] like Figure 4As shown, the device of this application may further include a central processing unit (“CPU”) 411, which may be a general-purpose CPU, a special-purpose CPU, or other information processing and program execution unit. Further, the device 400 may also include a mass storage 412 and a read-only memory (“ROM”) 413, wherein the mass storage 412 may be configured to store various types of data, including various data related to the first and second cranial images, image features, spatial features, fusion features, prediction results, algorithm data, intermediate results, and various programs required to run the device 400. The ROM 413 may be configured to store data and instructions required for the device 400's power-on self-test, the initialization of various functional modules in the system, the system's basic input / output drivers, and the booting of the operating system.

[0055] Optionally, device 400 may also include other hardware platforms or components, such as the tensor processing unit (“TPU”) 414, graphics processing unit (“GPU”) 415, field-programmable gate array (“FPGA”) 416, and machine learning unit (“MLU”) 417 shown. It is understood that although various hardware platforms or components are shown in device 400, they are merely exemplary and not limiting, and those skilled in the art can add or remove appropriate hardware as needed. For example, device 400 may implement the method for predicting the prognosis of ischemic stroke of this application by including only a CPU, associated storage devices, and interface devices.

[0056] In some embodiments, to facilitate data transmission and interaction with external networks, the device 400 of this application further includes a communication interface 418, through which it can connect to a local area network / wireless local area network (“LAN / WLAN”) 405, and further through the LAN / WLAN to connect to a local server 406 or to the Internet (“Internet”) 407. Alternatively or additionally, the device 400 of this application can also directly connect to the Internet or cellular network based on wireless communication technology, such as 3G (“3G”), 4G (“4G”), or 5G (“5G”) wireless communication technology, via the communication interface 418. In some application scenarios, the device 400 of this application can also access the server 408 and database 409 of an external network as needed to obtain various known algorithms, data, and modules, and can remotely store various data, such as various data or instructions used to present, for example, first and second cranial images, image features, spatial features, fusion features, prediction results, etc.

[0057] Peripherals of device 400 may include a display device 402, an input device 403, and a data transmission interface 404. In one embodiment, the display device 402 may include, for example, one or more speakers and / or one or more visual displays, configured to provide voice prompts and / or display images and videos for the purpose of predicting the prognosis of ischemic stroke according to this application. The input device 403 may include, for example, a keyboard, mouse, microphone, posture capture camera, and other input buttons or controls, configured to receive audio data input and / or user commands. The data transmission interface 404 may include, for example, a serial interface, parallel interface, or Universal Serial Bus interface (“USB”), Small Computer System Interface (“SCSI”), Serial ATA, FireWire (“FireWire”), PCI Express, and High Definition Multimedia Interface (“HDMI”), configured for data transmission and interaction with other devices or systems. According to the scheme of this application, the data transmission interface 404 can receive aneurysm images acquired from CT or MRI equipment and transmit to device 400 data or results including cranial images or various other types of data.

[0058] The CPU 411, mass storage 412, ROM 413, TPU 414, GPU 415, FPGA 416, MLU 417, and communication interface 418 of the device 400 of this application can be interconnected via bus 419, and can interact with peripheral devices through this bus. In one embodiment, the CPU 411 can control other hardware components in the device 400 and its peripheral devices through this bus 419.

[0059] The above combination Figure 4 A device that can be used to execute this application is described. It should be understood that the device structure or architecture described herein is merely exemplary, and the implementation method and implementation entity of this application are not limited thereto, but can be changed without departing from the spirit of this application.

[0060] Based on the foregoing description in conjunction with the accompanying drawings, those skilled in the art will understand that the embodiments of this application can also be implemented by software programs. Therefore, this application also provides a computer-readable storage medium storing computer-readable instructions for predicting the prognosis of ischemic stroke. When executed by one or more processors, these computer-readable instructions can be used to implement the embodiments of this application in conjunction with the accompanying drawings. Figure 1 The method described is for predicting the prognosis of ischemic stroke.

[0061] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0062] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of this application, they are used only to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0063] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0064] Although the embodiments of this application are described above, the content is merely an example adopted for the purpose of facilitating understanding of this application and is not intended to limit the scope and application scenarios of this application. Any person skilled in the art described in this application may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

[0065] Furthermore, the collection and acquisition of various data in this application comply with relevant laws and regulations and are authorized by the data providers. Any organization or individual that needs to obtain external data shall obtain authorization in accordance with the law and ensure data security, and shall not illegally collect, use, process, or transmit unauthorized or unprotected data, nor shall it illegally buy, sell, provide, or disclose unauthorized or unprotected data.

Claims

1. A device for predicting the prognosis of ischemic stroke, comprising: processor; as well as A memory storing program instructions for predicting the prognosis of ischemic stroke, which, when executed by the processor, cause the device to perform the following operations: Acquire a first brain image and a second brain image, wherein the first brain image includes CT perfusion imaging and the second brain image includes CT angiography. The first feature module and the first geometry module are used to extract image features and spatial features from the first cranial image to obtain first image features and first spatial features, wherein the first spatial features include gradient features. The second feature module and the second geometry module are used to extract image features and spatial features from the second cranial image to obtain second image features and second spatial features, wherein the second spatial features include distance features. The first fusion module is used to fuse the first image features and the first spatial features to obtain the corresponding first fused features; The second fusion module is used to fuse the second image feature and the second spatial feature to obtain the corresponding second fused feature. The prediction module uses the splicing result of the first fusion feature and the second fusion feature to predict the prognosis of ischemic stroke, and obtains the prediction result. When the program instructions are executed by the processor, the device further performs the following operations to obtain the first spatial feature: The first geometry module is used to calculate the gradient information of each pixel in the first cranial image to obtain a gradient map. Extract the target region map under the target parameters from the first cranial image; The gradient map is superimposed on the target region map to obtain the first spatial feature; When the program instructions are executed by the processor, the device further performs the following operations to obtain the second spatial feature: The second geometry module is used to extract the vascular centerline from the second cranial image; The distance from each pixel in the second cranial image to the center line of the blood vessel is calculated to obtain the second spatial feature.

2. The apparatus of claim 1, wherein when the program instructions are executed by the processor, the apparatus further performs the following operation to obtain the first fusion feature: The first image features are grouped according to the target parameters; The first fusion module is used to perform attention fusion operation between the first image features of each group and the first spatial features to obtain the attention features corresponding to each group. The attention features corresponding to each group are fused to obtain the first fused feature.

3. The apparatus according to claim 2, wherein the target parameters include at least the time to peak contrast agent concentration, the amount of blood flowing through brain tissue per unit time, the blood volume in a unit brain tissue, and the average time for the contrast agent to pass through the vascular bed.

4. The apparatus of claim 1, wherein when the program instructions are executed by the processor, the apparatus further performs the following operation to obtain the second fusion feature: The second fusion module is used to perform an attention fusion operation on the second image features and the second spatial features to obtain the second fused features.

5. The apparatus of claim 1, wherein when the program instructions are executed by the processor, the apparatus further performs the following operations: Normalize clinical information; The feature vector of the normalized clinical information was extracted using a multi-layer sensing module. The feature vector is then fused with the concatenation result.

6. The device according to claim 5, wherein the clinical information includes at least age, sex, blood pressure, blood glucose, blood oxygen saturation, atrial fibrillation, time from onset to treatment, disease history, and NIHSS score.

7. The apparatus of claim 1 or 5, wherein when the program instructions are executed by the processor, the apparatus further performs the following operations: The splicing result is then subjected to dimensionality reduction.

8. A computer-readable storage medium having stored thereon computer program instructions for predicting the prognosis of ischemic stroke, the computer program instructions being executed by one or more processors to cause the operation performed by the apparatus according to any one of claims 1-7 to be implemented.

Images