ASPECTS scoring method for post-circulation area

Abstract

The invention discloses an ASPECTS scoring method for a post-circulation area, and relates to the technical field of computer image processing. The method specifically comprises the following steps: acquiring image information of a post-circulation acute ischemic stroke patient; constructing a preset number of standard areas in the graphic information to generate sample data; constructing a model according to the sample data, and generating an ASPECTS scoring model; and acquiring image information of an unprocessed patient suffering from the posterior circulation acute ischemic stroke, and scoring the image information of the unprocessed patient suffering from the posterior circulation acute ischemic stroke through the ASPECTS scoring model. The objective of the invention is to improve the accuracy, objectivity and consistency of scanning scoring.

Description

Technical Field

[0001] The invention relates to the technical field of computer image processing, and in particular to an ASPECTS scoring method for a posterior circulation area. Background Art

[0002] With the increasing aging of the global population, stroke has become one of the major diseases threatening human health. According to statistics, stroke is the leading cause of disability and death worldwide, seriously affecting the patient's quality of life and socioeconomic burden. The occurrence of stroke is mainly divided into two categories, namely ischemic stroke and hemorrhagic stroke, of which ischemic stroke accounts for about 80% of all stroke cases. The occurrence of ischemic stroke is usually due to cerebral vascular obstruction, resulting in reduced or interrupted blood supply to the corresponding area, thereby causing ischemic necrosis of brain tissue. Early identification and treatment of ischemic stroke is crucial to improving the patient's prognosis and quality of life. Computed tomography (CT) has become a key tool widely used in clinical practice.

[0003] Computed tomography (CT) technology is an important imaging examination method and is widely used in the diagnosis and evaluation of stroke. Through CT scanning, changes in brain tissue structure can be observed more clearly, providing doctors with a basis for diagnosis. In the early stages of ischemic stroke, subtle low-density image changes (early ischemic changes, EIC) may appear on CT images, which reflect the ischemic state of brain tissue. However, early EIC is usually more subtle and subtle, especially in the posterior circulation area, where its imaging manifestations are more complex. In order to improve the accuracy of diagnosis, a scanning score (PC-ASPECTS) tool has appeared on the market. This tool is one of the commonly used scoring tools for clinically evaluating acute ischemic stroke in the posterior circulation area. This scoring system scores the imaging manifestations of specific anatomical areas of the posterior circulation (such as the midbrain, pons, cerebellum, thalamus, etc.) to evaluate the scope and severity of ischemic stroke in a semi-quantitative manner, thereby providing decision support for clinicians.

[0004] The traditional PC-ASPECTS scoring relies on the subjective judgment of doctors, who need to manually score early ischemic changes (EIC) in the posterior circulation region based on CT imaging data. Because the human visual system is not sensitive enough to subtle changes in non-enhanced CT images, especially early ischemic changes in the posterior circulation region, the traditional PC-ASPECTS scoring method has large errors. Studies have shown that even experienced neurologists may have limited accuracy in identifying subtle EICs. Especially in the early stages of the disease, the imaging manifestations are often not significant enough, and the anatomical complexity of the posterior circulation region and the hidden nature of the imaging features make the accuracy of manual scoring have great uncertainty in actual operation. In addition, the scoring process is relatively cumbersome, and doctors need to evaluate the imaging features of multiple anatomical regions one by one, which is not only time-consuming and labor-intensive, but also easily interfered by multiple factors such as bone artifacts and imaging quality, further affecting the accuracy of diagnosis and the stability of scoring.

[0005] Therefore, how to improve the accuracy, objectivity and consistency of ASPECTS scoring has become a technical problem that needs to be solved urgently. Summary of the invention

[0006] The main purpose of the present invention is to provide an ASPECTS scoring method for the posterior circulation region, aiming to improve the accuracy, objectivity and consistency of scan scoring.

[0007] In order to achieve the above object, the present invention proposes an ASPECTS scoring method for the posterior circulation area, comprising the following steps: Obtain image information of patients with posterior circulation acute ischemic stroke; constructing a predetermined number of standard areas in the graphic information to generate sample data; Building a model based on the sample data to generate an ASPECTS scoring model; The image information of the untreated posterior circulation acute ischemic stroke patient is obtained, and the image information of the untreated posterior circulation acute ischemic stroke patient is scored by using the ASPECTS scoring model.

[0008] In one embodiment of the present application, acquiring image information of a patient with posterior circulation acute ischemic stroke also includes: The image information of patients with posterior circulation acute ischemic stroke was screened according to predetermined criteria.

[0009] In one embodiment of the present application, the predetermined standard is: The efficacy of the image information is greater than or equal to 0.8 and the effect value of the image information is less than or equal to 0.3 and the significance standard is 0.05.

[0010] In one embodiment of the present application, after the image information of patients with posterior circulation acute ischemic stroke is screened according to predetermined standards, it also includes: determining whether the image information is in JPEG format, and if not, converting the image information format into JPEG format.

[0011] In one embodiment of the present application, the ASPECTS scoring model is constructed using the YOLO algorithm.

[0012] In one embodiment of the present application, after building a model based on the sample data and generating an ASPECTS scoring model, the method further includes: The ASPECTS scoring model is evaluated using a predetermined indicator, and an evaluation result is obtained, and it is determined whether the evaluation result meets a preset condition, and if so, the model is retained.

[0013] In one embodiment of the present application, the predetermined indicator includes at least one of AUC, specificity, and sensitivity.

[0014] In one embodiment of the present application, the predetermined number is 8.

[0015] The above technical solution can accurately identify early ischemic changes through deep learning models, reduce human errors in the scoring process, and ensure the objectivity and consistency of the scoring results. The automated scoring system quickly completes image analysis and scoring through the YOLO V8 model, and can generate PC-ASPECTS scores within seconds, significantly improving scoring efficiency. Through the automated scoring system, doctors do not need to manually evaluate the EIC of the posterior circulation area one by one, reducing the workload. Doctors can focus more on treatment decisions and patient management, improving the utilization of medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The present invention is described in detail below with reference to specific embodiments and accompanying drawings, wherein: Figure 1 It is a schematic structural diagram of the first embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the present invention clearer, the present invention is described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the following specific embodiments are only used to explain the present invention and do not constitute a limitation to the present invention.

[0018] like Figure 1 As shown, in order to achieve the above purpose, the present invention proposes an ASPECTS scoring method for the posterior circulation area, comprising the following steps: Obtain image information of patients with posterior circulation acute ischemic stroke; constructing a predetermined number of standard areas in the graphic information to generate sample data; Building a model based on the sample data to generate an ASPECTS scoring model; The image information of the untreated posterior circulation acute ischemic stroke patient is obtained, and the image information of the untreated posterior circulation acute ischemic stroke patient is scored by using the ASPECTS scoring model.

[0019] Specifically, ASPECTS refers to the Alberta Stroke Program Early Computed Scan Score. Relevant imaging information is obtained from patients with acute ischemic stroke in the posterior circulation. Specifically, the patient's NCCT (non-enhanced CT) images are used for analysis. These image data are acquired through a standardized CT scanning process, including the setting of positioning, range, slice thickness and scanning parameters. A 256-slice or similar high-resolution CT scanner is used to ensure that the image quality meets the model analysis requirements. The acquired NCCT images need to be clear enough so that subsequent automatic analysis can effectively identify early ischemic changes (EIC).

[0020] By adopting consistent scanning specifications, high-quality, comparable image data can be obtained, providing a solid foundation for subsequent analysis.

[0021] In the acquired imaging information, a predetermined number of standard regions are constructed to generate the sample data required for the model. According to the PC-ASPECTS scoring system, the scoring of the posterior circulation region is based on the evaluation of eight anatomical regions, including the midbrain, pons, cerebellum, thalamus, etc. These regions are the key locations that clinicians need to pay attention to when manually scoring.

[0022] In order to automate the scoring, an experienced team of doctors manually marked the eight scoring areas on each NCCT image using standard templates for delineation and segmentation. Each area was clearly marked, and sample data was generated for model training. Doctors recorded the coordinate information and related features of each marked area using professional labeling software (such as labelImg) to ensure that these areas can be accurately identified and evaluated during model training.

[0023] Based on the sample data generated above, an automated scoring model is constructed. Specifically, the YOLO V8 deep learning model is used to train the sample data. The YOLO (You Only Look Once) model is an advanced target detection algorithm that can quickly and accurately detect specific areas in an image. By training on a large number of annotated posterior circulation NCCT images, the model learns how to identify and analyze EIC in the posterior circulation area.

[0024] In the process of model training, in order to further optimize the model performance, optimization algorithms such as Distribution Focal Loss and Clou_Loss were used to improve the recognition accuracy of EIC. Through continuous iterative training and model parameter optimization, a model that can automatically score unprocessed NCCT images is gradually constructed.

[0025] After the model was built, new, unprocessed NCCT images of patients with acute ischemic stroke in the posterior circulation were obtained. These unprocessed images were scored using the above-mentioned automatic scoring model. The model first identified eight standard regions in the image, and then the ischemic probability score of each region was calculated by evaluating the EIC (early ischemic change) of each region.

[0026] Finally, the model aggregates the scores of each region and automatically generates the PC-ASPECTS score. This score can quickly and accurately reflect the patient's ischemic condition and provide decision support for clinicians. The entire scoring process is fully automated, and it only takes a few seconds from image input to score generation, greatly improving scoring efficiency.

[0027] The above technical solution can accurately identify early ischemic changes through deep learning models, reduce human errors in the scoring process, and ensure the objectivity and consistency of the scoring results. The automated scoring system quickly completes image analysis and scoring through the YOLO V8 model, and can generate PC-ASPECTS scores within seconds, significantly improving scoring efficiency. Through the automated scoring system, doctors do not need to manually evaluate the EIC of the posterior circulation area one by one, reducing the workload. Doctors can focus more on treatment decisions and patient management, improving the utilization of medical resources.

[0028] In one embodiment of the present application, acquiring image information of a patient with posterior circulation acute ischemic stroke also includes: The image information of patients with posterior circulation acute ischemic stroke was screened according to predetermined criteria.

[0029] Specifically, after obtaining the original NCCT image, the image screening step is entered. According to the predetermined standards, the image information is screened to ensure the accuracy and consistency of the input data. This standard screening process usually includes the following aspects: By analyzing the image resolution, contrast, noise and other indicators, the image data that does not meet the standards are eliminated. For example, cases with low resolution, excessive noise, motion artifacts or blurred CT images will be excluded. This can avoid artifacts and noise in the image interfering with the analysis and interpretation of the model.

[0030] Ensure that the posterior circulation areas (such as the midbrain, pons, cerebellum, thalamus, etc.) are fully presented in the image. If these areas are not fully captured due to imaging range problems or the imaging quality of some areas is poor, such images will be eliminated.

[0031] Image data of acute stroke patients were screened for time limits, and images taken within a few hours after onset were screened. The identification of early ischemic changes (EIC) is very important for PC-ASPECTS scoring, and the time factor directly affects the appearance of EIC on CT. Screening out images that fit within the time window will help improve the accuracy of the score.

[0032] By adopting the above technical solution, by adding a screening step with predetermined standards after image acquisition, those image data that do not meet the standards are eliminated, which effectively avoids the impact of image quality differences on the scoring results. Screening out low-quality images or incomplete posterior circulation area data can provide the model with more representative and high-quality data, thereby improving the accuracy of the model in identifying EIC. The image screening step can eliminate poor image data caused by motion artifacts and noise interference, reduce the interference of these noise factors on model training and scoring, and thus improve the stability and robustness of the automatic scoring system. The high-quality image data after screening can ensure the consistency of the model training data and reduce the deviation caused by inconsistent data quality.

[0033] In one embodiment of the present application, the predetermined standard is: The efficacy of the image information is greater than or equal to 0.8 and the effect value of the image information is less than or equal to 0.3 and the significance standard is 0.05.

[0034] Specifically, power is an indicator of the ability of a test in statistics, which indicates the probability that the test can reject the null hypothesis (i.e., discover the actual effect) when the hypothesis is true. The higher the power, the stronger the ability of the test, that is, the more effective it can be in detecting real differences. In this scheme, the power value is set to 0.8, indicating that the model has an 80% probability of correctly identifying early ischemic changes (EIC) in the image.

[0035] By requiring the efficacy of image information to be greater than or equal to 0.8, the model can efficiently and accurately identify early ischemic changes during image analysis. This means that in the acquired images, the model has sufficient ability to determine whether ischemia occurs in these areas, reducing the risk of missed diagnosis.

[0036] The effect size refers to the actual effect size or degree of relationship between two variables. In this protocol, the effect size is set to be less than or equal to 0.3, indicating that the effect size of the model analysis should be relatively small, which means that the model is sensitive enough to detect subtle changes when identifying early ischemic changes.

[0037] By constraining the effect size to no more than 0.3, we ensured that the processing of the image data in the model would be sensitive enough to capture slight changes within the posterior circulation region.

[0038] The significance criterion, also known as the P-value threshold, is a criterion used in statistical tests to determine whether a result is statistically significant. The significance level is set at 0.05. In this scenario, the significance criterion is 0.05, indicating that the model's judgment error rate will be controlled below 5% when processing image information. The significance criterion controls the model's error rate during image analysis and ensures that the ischemic changes identified by the model are statistically significant, that is, when early ischemic changes are judged, the result is not accidental. This ensures that the model does not easily produce false positive results when processing images.

[0039] In one embodiment of the present application, after the image information of patients with posterior circulation acute ischemic stroke is screened according to predetermined standards, it also includes: determining whether the image information is in JPEG format, and if not, converting the image information format into JPEG format.

[0040] Using the above technical solution, the JPEG format is a widely used image format with good compatibility. By uniformly converting image information into the JPEG format, it can ensure that the image data meets the input requirements of the deep learning model and avoid analysis failures caused by format incompatibility.

[0041] In one embodiment of the present application, the ASPECTS scoring model is constructed using the YOLO algorithm.

[0042] Specifically, the present invention uses the YOLO V8 model as the core framework of the deep learning algorithm. YOLO V8 is the latest version of the YOLO series, has higher detection speed and accuracy, can process target detection tasks in images in real time, and is suitable for rapid diagnosis.

[0043] The preprocessed and annotated image data is fed into the YOLO model for training. During the training process, the model will learn how to identify the posterior circulation areas in NCCT images through the annotated information and determine whether there are early ischemic changes in each area.

[0044] During the model training process, Distribution Focal Loss and Clou_Loss are used as loss functions. These loss functions can balance the problem of class imbalance, and are especially suitable for the detection of EIC in NCCT images, because EIC is usually rare and difficult to distinguish in images. Through these optimization methods, the model can converge better during the training process and improve the detection accuracy.

[0045] The model parameters are updated using the gradient descent method. After each iteration, the model adjusts the weights and biases based on the feedback of the loss function, gradually improving the detection accuracy of EIC.

[0046] By adopting the above technical solution, the automatic scoring system using the YOLO algorithm can significantly shorten the PC-ASPECTS scoring time. Compared with traditional manual scoring, the automatic scoring process only takes a few seconds to complete. The scoring efficiency is greatly improved. The YOLO algorithm has a high degree of accuracy when processing NCCT images, can effectively capture subtle changes in the posterior circulation area, and quantify the presence of EIC through probability scoring. Compared with the doctor's subjective judgment, the automated scoring system can reduce manual errors and improve the stability and consistency of scoring.

[0047] In one embodiment of the present application, after building a model based on the sample data and generating an ASPECTS scoring model, the method further includes: The ASPECTS scoring model is evaluated using a predetermined indicator, and an evaluation result is obtained, and it is determined whether the evaluation result meets a preset condition, and if so, the model is retained.

[0048] Specifically, the classification effect of the model is evaluated through indicators such as AUC (area under the curve), specificity, and sensitivity. AUC is used to evaluate the overall classification ability of the model, while specificity and sensitivity measure the model's ability to identify positive and negative samples respectively.

[0049] The kappa coefficient was used to evaluate the consistency between the model scores and the manual scores of clinicians. A higher kappa value indicated that the model and clinicians had better consistency in their scores, further verifying the clinical practicality of the model.

[0050] The above technical solution was used to verify the reliability of the model in practical application by analyzing the consistency with the scores of clinicians. The kappa coefficient showed that the model had a high consistency with the scores of doctors, making it more reliable for use in clinical practice.

[0051] In one embodiment of the present application, the predetermined indicator includes at least one of AUC, specificity, and sensitivity.

[0052] In one embodiment of the present application, the predetermined number is 8.

[0053] The above description is only a preferred embodiment of the present invention, and does not limit the patent scope of the present invention. All equivalent structural changes made by using the contents of the present invention specification and drawings under the inventive concept of the present invention, or directly / indirectly applied in other related technical fields are included in the patent protection scope of the present invention.

Claims

1. An ASPECTS scoring method for the posterior circulation region, characterized in that: The following steps are involved: Obtain image information of patients with posterior circulation acute ischemic stroke; constructing a predetermined number of standard areas in the graphic information to generate sample data; Building a model based on the sample data to generate an ASPECTS scoring model; The image information of the untreated posterior circulation acute ischemic stroke patient is obtained, and the image information of the untreated posterior circulation acute ischemic stroke patient is scored by using the ASPECTS scoring model.

2. The ASPECTS scoring method for the posterior circulation area according to claim 1, characterized in that: Obtain image information of patients with posterior circulation acute ischemic stroke, including: The image information of patients with posterior circulation acute ischemic stroke was screened according to predetermined criteria.

3. The ASPECTS scoring method for the posterior circulation area according to claim 2, characterized in that: The predetermined criteria are: The efficacy of the image information is greater than or equal to 0.8 and the effect value of the image information is less than or equal to 0.3 and the significance standard is 0.

05.

4. The ASPECTS scoring method for the posterior circulation area according to claim 2, characterized in that: After the image information of the posterior circulation acute ischemic stroke patient is screened according to a predetermined standard, the method further includes: determining whether the image information is in JPEG format, and if not, converting the image information format into JPEG format.

5. The ASPECTS scoring method for the posterior circulation area according to claim 1, characterized in that: The ASPECTS scoring model is constructed using the YOLO algorithm.

6. The ASPECTS scoring method for the posterior circulation area according to claim 1, characterized in that: After building a model based on the sample data and generating the ASPECTS scoring model, the following steps are also included: The ASPECTS scoring model is evaluated using a predetermined indicator, and an evaluation result is obtained, and it is determined whether the evaluation result meets a preset condition, and if so, the model is retained.

7. The ASPECTS scoring method for the posterior circulation area according to claim 6, characterized in that: The predetermined index includes at least one of AUC, specificity, and sensitivity.

8. The ASPECTS scoring method for the posterior circulation area according to claim 1, characterized in that: The predetermined number is 8.

Images