Method and system for generating CTA image by using CTP reconstruction based on deep learning

Abstract

The invention relates to the field of image processing, and discloses a method and system for generating a CTA image through CTP reconstruction based on deep learning, and the method comprises the steps: carrying out the preprocessing of an original CTP image, and generating an MIP image; performing automatic segmentation on the MIP image by using a deep learning network, and segmenting arteriovenous regions in the CTP image; calculating a time density attenuation curve of all points in the arteriovenous region; according to the time density attenuation curve of all the points, the optimal arteriovenous point is screened out; obtaining a period number corresponding to a peak phase through the time density attenuation curve of the optimal arteriovenous point, and extracting an image corresponding to the period in the CTP original image; and based on the image corresponding to the period, reconstructing and generating a CTA image. According to the method, the acquisition process of the CTA image is accelerated by designing an automatic process based on deep learning, the time and the radiation dose required by patient examination are reduced, the quality and the diagnostic value of the CTA image are improved, and more reference bases are provided for prognosis improvement of patients.

Description

Technical Field

[0001] This invention relates to the field of image processing, and more particularly to a method and system for generating CTA images based on deep learning using CTP reconstruction. Background Technology

[0002] Acute ischemic stroke (AIS) is one of the leading causes of disability and death worldwide, affecting approximately 600 million people globally. Timely assessment of AIS and large vessel occlusion (LVO) can guide mechanical thrombectomy, thereby salvaging damaged brain tissue and reducing clinical functional impairment.

[0003] For patients suspected of having a stroke, non-contrast-enhanced CT is used first to rule out hemorrhagic stroke. Computed tomography perfusion (CTP) images are used to differentiate between the irreversible core infarct area and salvageable ischemic penumbra, while computed tomography angiography (CTA) images are used to assess the degree of vascular stenosis. The combined use of multiple modalities can significantly improve the diagnostic accuracy and specificity of AIS. However, traditional single-phase CTA lacks temporal resolution, leading to inaccurate assessment of collateral circulation. Furthermore, incorrect selection of contrast agent triggering timing also greatly affects image quality. The multi-phase nature of CTP offers advantages in assessing collateral circulation and thrombus hemodynamics, while eliminating the need for manual triggering timing, reducing the possibility of poor image quality.

[0004] Therefore, developing a technology that uses CTP reconstruction to generate CTA can not only effectively solve the above problems, but also avoid patients undergoing CTP and CTA examinations at the same time, reducing the radiation dose received by patients and the time required for the examination, and improving diagnostic efficiency.

[0005] Several studies have explored methods for generating CTA using CTP reconstruction. However, these methods focus on image quality and vascular stenosis assessment, neglecting other assessments such as collateral circulation scoring, thrombus burden, and aneurysm. Furthermore, these methods require manual evaluation of the CTP temporal density decay curve before manually selecting the first-phase image corresponding to the peak phase of the curve for CTA reconstruction, which introduces observer bias and fails to achieve a fully automated end-to-end processing flow. Summary of the Invention

[0006] The main objective of this invention is to address the technical problems of observer bias and the inability to achieve fully automated end-to-end processing in existing technologies. A method for generating CTA images using CTP reconstruction based on deep learning includes the following steps: The original CTP image is preprocessed to generate a MIP image; a deep learning network is used to automatically segment the MIP image to extract the arterial and venous regions in the CTP image; the temporal density decay curves of all points in the arterial and venous regions are calculated; the optimal arterial and venous points are selected based on the temporal density decay curves of all points; the period corresponding to the peak phase is obtained through the temporal density decay curve of the optimal arterial and venous points, and the image corresponding to that period in the original CTP image is extracted; a CTA image is reconstructed based on the image corresponding to that period.

[0007] Preprocessing of the raw CTP image includes: All phases of CTP except phase 1 were registered to phase 1 using image registration technology to ensure spatial consistency of all phases of CTP images; skull was removed using a thresholding method to eliminate interference from the skull on blood vessels; images of corresponding positions in each phase of CTP were extracted and subjected to maximum density projection processing, and the maximum pixel value of corresponding positions in multiple images was selected and projected onto a two-dimensional plane to enhance the display of blood vessels.

[0008] A deep learning network is used to automatically segment the preprocessed CTP image, identifying the arterial and venous regions within the CTP image, including: A deep learning network model was built using the PyTorch framework and Python language, with the 3D ResU-Net model selected as the network structure. Training data was collected and manually labeled to identify the arterial and venous regions before being fed into the network for training. After training, the network performance was validated using test data, resulting in the trained deep learning network model. The trained deep learning network model was then used to automatically segment the maximum density projection image of CTP, identifying the arterial and venous regions.

[0009] Calculate the time density decay curve for each point in the arteriovenous region, including: The time density curve is a curve with time as the horizontal axis and CT value as the vertical axis, representing the enhancement changes of a specific region, such as brain tissue or tumor, as the contrast agent passes through. The formula is: TDC(t) = CT value(t), where: t is the time from the start of contrast agent injection to a certain moment; CT value(t) is the average CT value of the region of interest at time point t.

[0010] In time-series images, a temporal density curve is constructed for each pixel. A score for each pixel is obtained by weighting it using selection criteria, ultimately yielding the optimal arteriovenous point. These selection criteria include: a) Using the known standard time density curve as a template, a correlation analysis was performed between the time density curve of each candidate region and the template. This weighting term is positively correlated with the score. b) The time difference between the peak time of each curve and the peak times of all curves; this weighting is positively correlated with the score. c) The pixel peak value of each curve. This weighting term is positively correlated with the score. The region with the highest peak enhancement in the entire time series is identified as a candidate blood vessel point. Morphological analysis, such as connected component detection, is then used to confirm whether it is a blood vessel structure. This weighting term is positively correlated with the score. d) The number of peaks for each curve; this weighting term is negatively correlated with the score. e) The half-width of each curve at the half-peak, this weighting term is positively correlated with the score.

[0011] The method also includes outputting results on a display interface for doctors' reference, the output of which includes: Generate vessel segmentation maps, VR maps, MIP maps, CPR maps, and MPR maps for CTA images; Mark the location and degree of stenosis in the image, as well as the location of the thrombus; Automatically score lateral circulation; It automatically detects and segments aneurysms on blood vessels and provides the size of the aneurysm.

[0012] A second aspect of the present invention provides a system for generating CTA images based on deep learning using CTP reconstruction, comprising: The preprocessing unit preprocesses the original CTP image to generate a MIP image; The automatic segmentation unit is used to automatically segment MIP images using a deep learning network to segment the arterial and venous regions in CTP images; The calculation unit is used to calculate the time density decay curves of all points in the arteriovenous region; The filtering unit is used to select the best arteriovenous points based on the time density decay curves of all points. The extraction unit is used to obtain the period corresponding to the peak phase through the time density decay curve of the optimal arteriovenous point, and extract the image corresponding to that period from the original CTP image; The reconstruction generation unit is used to reconstruct and generate a CTA image based on the image corresponding to the current period.

[0013] A third aspect of the present invention provides an electronic device comprising: a memory and at least one processor, the memory storing instructions, the memory and the at least one processor being interconnected via a circuit; the at least one processor calling the instructions in the memory to cause the electronic device to execute the system method described above for generating CTA images based on deep learning using CTP reconstruction.

[0014] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the system method described above for generating CTA images based on deep learning using CTP reconstruction.

[0015] The present invention has the following beneficial effects: This invention accelerates the acquisition of CTA images by designing an automated process based on deep learning, reduces the time required for patient examinations and the radiation dose, improves the quality and diagnostic value of CTA images, and provides more reference for improving patient prognosis. Attached Figure Description

[0016] Figure 1 Flowchart for generating CTA from CTP reconstruction.

[0017] Figure 2 This is a diagram of the 3D ResU-Net network structure.

[0018] Figure 3 To reconstruct and compare the diagnostic results of the original CTA and the generated CTA.

[0019] Figure 4 To reconstruct and generate CTA and original CTA images showing and measuring thrombi.

[0020] Figure 5 This is a graph evaluating the robustness and accuracy of the optimal point selection scheme. Detailed Implementation

[0021] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0022] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1-4 The first embodiment of the present invention includes: A method for generating CTA images based on CTP using deep learning includes the following steps: The original CTP image is preprocessed to generate a MIP image; a deep learning network is used to automatically segment the MIP image to extract the arterial and venous regions in the CTP image; the temporal density decay curves of all points in the arterial and venous regions are calculated; the optimal arterial and venous points are selected based on the temporal density decay curves of all points; the period corresponding to the peak phase is obtained through the temporal density decay curve of the optimal arterial and venous points, and the image corresponding to that period in the original CTP image is extracted; a CTA image is reconstructed based on the image corresponding to that period.

[0023] Because CTP involves multiple phases of scanning, slight patient movement during the scan inevitably causes spatial position shifts in each phase. Therefore, all phases of CTP except the first phase are first registered to the first phase using image registration technology to ensure spatial consistency across all phases and meet the requirements of subsequent processing. Since the pixel values ​​of the skull are much higher than those of blood vessels, a thresholding method is used to remove skull fragments, eliminating interference from the skull and preventing misidentification of the skull as blood vessels. Maximum density projection is then performed on the corresponding locations of each CTP phase, selecting the image with the highest pixel value at each location from multiple images and projecting it onto a two-dimensional plane to enhance the visualization of blood vessels.

[0024] A deep learning network is used to automatically segment the maximum density projection image of CTP to identify the arterial and venous regions. The method for obtaining this network is as follows: a) Use the PyTorch framework and Python language to build a deep learning network model, and select the 3DResU-Net model as the network structure.

[0025] b) Collect training data, manually label the training data, mark the arterial and venous regions, and send them into the network for training.

[0026] c) After training, use test data to verify the network performance to ensure that it can accurately and stably segment the arterial and venous regions.

[0027] Preprocessing of the raw CTP image includes: All phases of CTP except phase 1 were registered to phase 1 using image registration technology to ensure spatial consistency of all phases of CTP images; skull was removed using a thresholding method to eliminate interference from the skull on blood vessels; images of corresponding positions in each phase of CTP were extracted and subjected to maximum density projection processing, and the maximum pixel value of corresponding positions in multiple images was selected and projected onto a two-dimensional plane to enhance the display of blood vessels.

[0028] A deep learning network is used to automatically segment the preprocessed CTP image, identifying the arterial and venous regions within the CTP image, including: A deep learning network model was built using the PyTorch framework and Python language, with the 3D ResU-Net model selected as the network structure. Training data was collected and manually labeled to identify the arterial and venous regions before being fed into the network for training. After training, the network performance was validated using test data, resulting in the trained deep learning network model. The trained deep learning network model was then used to automatically segment the maximum density projection image of CTP, identifying the arterial and venous regions.

[0029] Calculate the time density decay curve for each point in the arteriovenous region, including: The time density curve is a curve with time as the horizontal axis and CT value as the vertical axis, representing the enhancement changes of a specific region, such as brain tissue or tumor, as the contrast agent passes through. The formula is: TDC(t) = CT value(t), where: t is the time from the start of contrast agent injection to a certain moment; CT value(t) is the average CT value of the region of interest at time point t.

[0030] This invention utilizes optimal point TDC (Trans-Cellular Directional Control) to intelligently determine the peak vascular phase: like Figure 3 and 4 The optimal time density curve (TDC) of the selected arteriovenous points is used to determine the phase in the entire CTP sequence at which vascular enhancement reaches its peak. This is an intelligent selection based on actual intravascular contrast agent dynamics, rather than a fixed phase or empirical estimation. Selecting the original CTP image at this peak phase as the source data for reconstructed CTA ensures that the reconstructed CTA image displays the vascular structure most clearly.

[0031] In time-series images, a temporal density curve is constructed for each pixel. A score for each pixel is obtained by weighting it using selection criteria, ultimately yielding the optimal arteriovenous point. These selection criteria include: a) Using the known standard time density curve as a template, a correlation analysis was performed between the time density curve of each candidate region and the template. This weighting term is positively correlated with the score. b) The time difference between the peak time of each curve and the peak times of all curves; this weighting is positively correlated with the score. c) The pixel peak value of each curve. This weighting term is positively correlated with the score. The region with the highest peak enhancement in the entire time series is identified as a candidate blood vessel point. Morphological analysis, such as connected component detection, is then used to confirm whether it is a blood vessel structure. This weighting term is positively correlated with the score. d) The number of peaks for each curve; this weighting term is negatively correlated with the score. e) The half-width of each curve at the half-peak, this weighting term is positively correlated with the score.

[0032] This invention calculates the time density curve (TDC) for each point within the segmented arteriovenous region and innovatively proposes a set of comprehensive, quantitative, and weighted scoring screening criteria to identify the "optimal" arteriovenous point (representing typical arterial / venous inflow / outflow).

[0033] Multi-dimensional feature fusion: The screening criteria incorporate multiple dimensions of TDC features. Morphological similarity (a): Correlation with the standard template.

[0034] Temporal consistency (b): The difference between peak time and group average (the smaller the better).

[0035] Signal strength (c): Peak CT value (the higher the better) and morphological verification (connected regions).

[0036] Curve complexity (d): Number of peaks (the fewer the better, representing a single peak).

[0037] Curve width (e): Half-peak width. This comprehensive, weighted, multi-parameter screening mechanism is more reliable in identifying representative vascular points than a single standard (such as the highest peak value). It is a key innovation to ensure the accuracy of subsequent steps and is a quantitative and robust solution designed to reliably identify typical vascular dynamic characteristics in complex tissue backgrounds and under noise interference.

[0038] The method also includes outputting results on a display interface for doctors' reference, the output of which includes: Generate vessel segmentation maps, VR maps, MIP maps, CPR maps, and MPR maps for CTA images; Mark the location and degree of stenosis in the image, as well as the location of the thrombus; Automatically score lateral circulation; It automatically detects and segments aneurysms on blood vessels and provides the size of the aneurysm.

[0039] The methods in the embodiments of the present invention have been described above. The apparatus in the embodiments of the present invention is described below: The preprocessing unit preprocesses the original CTP image to generate a MIP image; The automatic segmentation unit is used to automatically segment MIP images using a deep learning network to segment the arterial and venous regions in CTP images; The calculation unit is used to calculate the time density decay curves of all points in the arteriovenous region; The filtering unit is used to select the best arteriovenous points based on the time density decay curves of all points. The extraction unit is used to obtain the period corresponding to the peak phase through the time density decay curve of the optimal arteriovenous point, and extract the image corresponding to that period from the original CTP image; The reconstruction generation unit is used to reconstruct and generate a CTA image based on the image corresponding to the current period.

[0040] The evaluation results of the solution of the present invention are as follows: Subjective assessment: The results of physician image quality assessments were compared. The assessed images included two sets of corresponding original axial images and post-processed images used for diagnosis—including volume rendering (VR), maximum intensity projection (MIP), curved planar reconstruction (CPR), and curved multiple planar reformations (MPR). Original axial images were subjectively assessed using a 5-point Likert scale, while post-processed images were scored using a 3-point scale. Higher scores indicated better image quality.

[0041] Objective evaluation metrics: The quality of the two sets of images was compared using Noise, SNR, and CNR metrics.

[0042] Finally, the scan images of a total of 288 patients were evaluated and compared, and the comparison results are shown in Table 1.

[0043] Table 1 The data in the table are presented as mean ± standard deviation. P < 0.05 indicates a statistically significant difference between the two groups. The results show that the generated CTA images outperform traditional CTA images in terms of post-processed images and signal-to-noise ratio (CNR). Examples of comparison are shown below. Figure 3 As shown.

[0044] The accuracy of identifying vascular stenosis and cerebral aneurysm on reconstructed CTA images (CTPA) is as follows: All images were independently evaluated by two experienced radiologists (both with over 5 years of experience) without prior knowledge of clinical indications, imaging reports, or other examination results. In case of disagreement, a third senior radiologist (with over 15 years of experience) was consulted to reach a consensus. The final diagnoses of vascular stenosis and cerebral aneurysms were compared with the gold standard, digital angiography (DSA). The evaluation focused on vascular regions within the CTPA scan field, covering key stroke-related vessels, including the distal bilateral internal carotid arteries (ICAs), middle cerebral arteries (MCAs), and basilar artery (BA). Stenosis severity was graded as: mild (<50%), moderate (50%–69%), severe (70%–99%), and complete occlusion (100%). For cerebral aneurysms, their location and size were recorded. A total of 390 vessels from 82 patients were evaluated, all with DSA results. The CTPA-based diagnoses of vascular stenosis showed a high degree of concordance with CTA results. At the 50% threshold, CTPA demonstrated diagnostic efficacy comparable to CTA in patient-level analyses (all p>0.05); however, at the 70% threshold, CTPA outperformed CTA in the analysis (p<0.05). CTPA also showed excellent performance in the visualization and diagnosis of cerebral aneurysms. Specific diagnostic efficacy parameters are detailed in Table 2. Table 2 The data in the table above are presented as percentages and 95% confidence intervals. P < 0.05 indicates a statistically significant difference between the two groups.

[0045] Robustness and accuracy evaluation of the optimal point selection scheme: In a dataset of 288 CTP scan images, both algorithms were able to generate CTPA in all cases with a 100% reconstruction success rate, demonstrating their strong robustness. Based on the clinical diagnostic usability assessment results evaluated by physicians, the original images were scored on a 5-point scale, with a score of 3 or higher considered satisfactory; the post-processed images were scored on a 3-point scale, with a score of 2 or higher considered satisfactory. The comparison results of the two methods are as follows: Figure 5 As shown, in both the original image and the post-processing reconstruction, the optimal selection point scheme (Group 1) has better overall image quality than the single peak selection scheme (Group 2), with a higher proportion of qualified images, especially in the original image and VR image.

[0046] This invention also provides an electronic device, which can vary significantly due to differences in configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media can be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0047] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that the electronic device structure in this embodiment does not constitute a limitation on the electronic device itself, and may include more or fewer components, or combinations of certain components, or different component arrangements.

[0048] This invention provides an electronic device structure that can vary significantly depending on configuration or performance. It may include one or more central processing units (CPUs) (e.g., one or more processors) and memory, and one or more storage media (e.g., one or more mass storage devices) for storing applications or data. The memory and storage media can be temporary or persistent storage. The program stored in the storage media may include one or more modules, each module including a series of instruction operations on the electronic device. Furthermore, the processor may be configured to communicate with the storage media and execute the series of instruction operations stored in the storage media on the electronic device.

[0049] The electronic device may also include one or more power supplies, one or more wired or wireless network interfaces, one or more input / output interfaces, and / or one or more operating systems, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that the structure of the electronic device does not constitute a limitation on the electronic device itself, and may include more or fewer components than described above, or combine certain components, or have different component arrangements.

[0050] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when executed on a computer, cause the computer to perform the steps of the aforementioned method.

[0051] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0052] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0053] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. 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. Such 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 method for generating CTA images based on deep learning using CTP reconstruction, characterized in that, Includes the following steps: Preprocess the original CTP image to generate a MIP image; A deep learning network was used to automatically segment MIP images and extract the arteriovenous region from CTP images. Calculate the time density decay curves for all points in the arteriovenous region; The optimal arteriovenous points were selected based on the time density decay curves of all points. By using the time density decay curve of the optimal arteriovenous point, the period corresponding to the peak phase is obtained, and the image corresponding to that period in the original CTP image is extracted. Based on the corresponding image for that period, a CTA image is reconstructed.

2. The method for generating CTA images based on deep learning using CTP reconstruction according to claim 1, characterized in that, The process involves preprocessing the original CTP image to generate a MIP image. All phases of CTP except the first phase are registered to the first phase using image registration technology to ensure the spatial consistency of all phases of CTP images; The skull was removed using a threshold method to eliminate its interference with blood vessels. Images of corresponding locations in each phase of CTP are extracted and subjected to maximum density projection processing. The maximum pixel value of the corresponding location in multiple images is selected and projected onto a two-dimensional plane to enhance the display of blood vessels.

3. The method for generating CTA images based on deep learning using CTP reconstruction according to claim 1, characterized in that, The automatic segmentation of the preprocessed CTP image using a deep learning network to segment the arterial and venous regions in the CTP image includes: A deep learning network model was built using the PyTorch framework and Python language, with the 3D ResU-Net model selected as the network structure. Collect training data, manually label the training data, mark the arterial and venous regions, and send it into the network for training; After training, the network performance is validated using test data to obtain the trained deep learning network model; A trained deep learning network model is used to automatically segment the maximum density projection image of CTP, and the arterial and venous regions are segmented.

4. The method for generating CTA images based on deep learning using CTP reconstruction according to claim 1, characterized in that, The calculation of the time density decay curve for each point in the arteriovenous region includes: The time density curve is a curve with time as the horizontal axis and CT value as the vertical axis, representing the enhancement changes of a specific region, such as brain tissue or tumor, as the contrast agent passes through. The formula is: TDC(t) = CT value(t), where: t is the time from the start of contrast agent injection to a certain moment; CT value(t) is the average CT value of the region of interest at time point t.

5. The method for generating CTA images based on deep learning using CTP reconstruction according to claim 1, characterized in that, In the time-series image, a temporal density curve is constructed for each pixel, and a score for each pixel is obtained by weighting through filtering criteria, ultimately yielding the optimal arteriovenous point. The filtering criteria include: a) Using the known standard time density curve as a template, a correlation analysis was performed between the time density curve of each candidate region and the template. This weighting term is positively correlated with the score. b) The time difference between the peak time of each curve and the peak times of all curves; this weighting is positively correlated with the score. c) The pixel peak value of each curve. This weighting term is positively correlated with the score. The region with the highest peak enhancement in the entire time series is identified as a candidate blood vessel point. Morphological analysis, such as connected component detection, is then used to confirm whether it is a blood vessel structure. This weighting term is positively correlated with the score. d) The number of peaks for each curve; this weighting term is negatively correlated with the score. e) The half-width of each curve at the half-peak, this weighting term is positively correlated with the score.

6. The method for generating CTA images based on deep learning using CTP reconstruction according to claim 1, characterized in that, The method also includes outputting results on a display interface for doctors' reference, the output of which includes: Generate vessel segmentation maps, VR maps, MIP maps, CPR maps, and MPR maps for CTA images; Mark the location and degree of stenosis in the image, as well as the location of the thrombus; Automatically score lateral circulation; It automatically detects and segments aneurysms on blood vessels and provides the size of the aneurysm.

7. A system for generating CTA images based on deep learning using CTP reconstruction, characterized in that, The system includes: The preprocessing unit preprocesses the original CTP image to generate a MIP image; The automatic segmentation unit is used to automatically segment MIP images using a deep learning network to segment the arterial and venous regions in CTP images; The calculation unit is used to calculate the time density decay curves of all points in the arteriovenous region; The filtering unit is used to select the best arteriovenous points based on the time density decay curves of all points. The extraction unit is used to obtain the period corresponding to the peak phase through the time density decay curve of the optimal arteriovenous point, and extract the image corresponding to that period from the original CTP image; The reconstruction generation unit is used to reconstruct and generate a CTA image based on the image corresponding to the current period.

8. An electronic device comprising a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the electronic device to perform the steps of the method for generating CTA images based on deep learning using CTP reconstruction as described in any one of claims 1-6.

9. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the method for generating CTA images based on deep learning using CTP as described in any one of claims 1-6.

Images