Automatic key layer identification method based on aortic dissection CTA image

Abstract

The invention discloses a key level automatic identification method based on aortic dissection CTA images, and the method comprises the steps: marking an aortic dissection CTA image sequence set based on a classification system comprising six types of positive samples and one type of negative samples, and dividing the marked aortic dissection CTA image sequence set into a training set, a verification set and a test set; performing preprocessing and data amplification processing on the aortic dissection CTA image sequence set to obtain an amplification training set; carrying out model training by utilizing a classification network architecture based on the amplification training set, carrying out performance evaluation through the verification set, and determining an optimal classification model; performing classification processing on the aortic dissection CTA image sequence of each patient in the test set according to the optimal classification model, and obtaining a classification prediction result through a softmax function; and extracting each category of probability sequence based on a classification prediction result, performing threshold filtering and isolated point removal to obtain a candidate image sequence, and determining an optimal key level from the candidate image sequence according to a preset selection rule. The key level identification accuracy and efficiency can be improved.

Description

Technical Field

[0001] This application relates to the field of deep learning-based medical image processing technology, and relates to, but is not limited to, a method for automatic identification of key layers in aortic dissection CTA images. Background Technology

[0002] Aortic dissection (AD) is a disease in which the intima of the aorta ruptures, allowing blood to enter and separate the aortic lumen into a true and false lumen. It is an exceptionally dangerous cardiovascular disease characterized by rapid onset, rapid progression, and high mortality. In acute AD patients, the mortality rate increases by 1%-2% per hour after symptom onset if left untreated. Computed tomography angiography (CTA) provides precise information about the aortic anatomy and is crucial for the diagnosis, surgical planning, and postoperative follow-up management of dissection. Currently, clinical selection of key CTA images and acquisition of relevant anatomical parameters rely entirely on manual measurement, resulting in low efficiency, significant heterogeneity, and difficulty in achieving large-scale data analysis. Automatic identification of key images and automatic measurement of relevant parameters can save doctors' time interpreting images, assist in faster diagnosis, and improve efficiency in surgical planning and long-term follow-up monitoring, thus making the establishment of large-scale data possible.

[0003] In existing technologies, research on the identification and localization of key layers includes deep learning-based methods for locating aortic landmarks and methods for predicting the cross-sectional orientation of aortic landmarks based on convolutional neural networks. Specifically, aortic landmarks can be located through aortic segmentation and heatmap regression tasks based on deep learning, but this method requires the aortic centerline to calculate a reconstructed plane perpendicular to the centerline, resulting in high computational complexity and long processing time. The cross-sectional orientation of aortic landmarks can be predicted based on convolutional neural networks and uncertainty quantization methods, but this method requires manual selection of landmarks.

[0004] Therefore, there is an urgent need for an automated and more accurate method for identifying key layers. This method would utilize image classification and post-processing to automatically identify key layers in aortic dissection CTA images, addressing issues such as low efficiency, high computational complexity, and low automation in existing technologies. This would enable efficient, direct, and accurate identification of key layers, saving doctors' image reading time, assisting them in making rapid diagnoses, and improving work efficiency in surgical planning and long-term follow-up monitoring. Summary of the Invention

[0005] This application provides a method for automatic identification of key layers based on CTA images of aortic dissection.

[0006] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide an automatic identification method for key layers based on aortic dissection CTA images. The method includes: establishing a classification system; labeling a set of aortic dissection CTA image sequences based on the classification system; dividing the labeled set of aortic dissection CTA image sequences into a training set, a validation set, and a test set, wherein the classification system includes six classes of positive samples and one class of negative samples; preprocessing the set of aortic dissection CTA image sequences to obtain a standardized image set; and performing data augmentation processing on the aortic dissection CTA images in the standardized image set that belong to the training set to obtain an augmented training set; and training a model using a classification network architecture based on the augmented training set. The system is trained, and the performance of the classification model is evaluated using a validation set to determine the optimal classification model. Based on the optimal classification model, the aortic dissection CTA image sequences of each patient in the test set are classified, and the classification prediction results are obtained using a softmax function. The classification prediction results include the probability that each aortic dissection CTA image belongs to each category. For each patient's aortic dissection CTA image sequence, probability sequences for each category are extracted based on the classification prediction results. Threshold filtering and outlier removal are performed on the probability sequences for each category to obtain candidate image sequences for each category. According to the preset selection rules for each category, the optimal key layer for each category is determined from the candidate image sequences for each category.

[0007] The technical solution provided in this application establishes a classification system and annotates a set of aortic dissection CTA image sequences based on this system. The annotated aortic dissection CTA image sequence set is then divided into training, validation, and test sets, providing a high-quality and reliable data foundation for subsequent classification models. This effectively prevents model overfitting, ensures the generalization ability of the recognition results, and allows the model to learn more robust anatomical features, guaranteeing the accuracy and clinical reliability of subsequent automatic recognition results. Most importantly, the original three-dimensional aortic dissection CTA images are divided into two-dimensional aortic dissection CTA image sequences, thereby addressing the challenge of identifying key layers in three-dimensional images. The problem is transformed into a classification and optimal selection problem for two-dimensional images, which greatly simplifies the process of key layer identification in aortic dissection CTA images, thereby improving the accuracy and real-time performance of the final identification results to a certain extent. The aortic dissection CTA image sequence set is preprocessed to obtain a standardized image set, eliminating differences in spatial resolution, grayscale values, and sizes among aortic dissection CTA images from different sources. This reduces the model's dependence on specific scanning equipment and improves the model's generalization performance. Furthermore, the aortic dissection CTA images in the standardized image set that belong to the training set are subjected to data augmentation to obtain an augmented training set, effectively increasing the diversity of training samples and solving the problem of class imbalance in the dataset. To further improve the robustness and recognition accuracy of the model, the model was trained using a classification network architecture based on an expanded training set, and its performance was evaluated using a validation set to determine the optimal classification model. Based on the optimal classification model, the aortic dissection CTA image sequences of each patient in the test set were classified, and the classification prediction results were obtained using a softmax function. The classification prediction results included the probability of each aortic dissection CTA image belonging to each category. For each patient's aortic dissection CTA image sequence, probability sequences for each category were extracted based on the classification prediction results. Threshold filtering and outlier removal were performed on the probability sequences for each category to optimize the model's initial prediction results and obtain candidate image sequences for each category. This method effectively eliminates unreliable predictions, significantly improving the accuracy and robustness of the final key layer localization. Based on preset selection rules for each category, it determines the optimal key layer for each category from candidate image sequences, thereby achieving precise identification of the unique optimal key layer from a series of similar images. The identification results fully comply with the anatomical standards for clinical diagnosis, ultimately achieving a unity of high automation, high real-time performance, and high accuracy in the identification process. This saves doctors' image reading time, assists them in making diagnoses more quickly, and improves work efficiency in surgical planning and long-term follow-up monitoring. It also helps to realize intelligent information feature extraction of CTA images and accelerates the development and application of artificial intelligence technology in the field of aortic disease.

[0008] Optionally, the six positive sample categories include: the aortic sinus structure at the level of the origin of the ascending aorta, the aortic arch structure at the level of the end of the ascending aorta and the origin of the descending aorta, the pulmonary artery bifurcation level, the celiac trunk level and the superior mesenteric artery level, the renal artery level, and the abdominal aortic bifurcation level; the aortic dissection CTA image sequence set includes aortic dissection CTA image sequences from multiple patients; the classification system is established based on predefined anatomical features; the dataset division unit of the aortic dissection CTA image sequence set is the aortic dissection CTA image sequence of each patient.

[0009] Optionally, the preprocessing of the aortic dissection CTA image sequence set to obtain a standardized image set includes: resampling the XY plane of the aortic dissection CTA image sequence to unify the resolution to a preset size; normalizing the gray values ​​of the resampled aortic dissection CTA image sequence to a preset range; and cropping the normalized aortic dissection CTA image sequence to a fixed pixel size based on the image center to obtain the standardized image set.

[0010] Optionally, the fully connected layer of the classification model has an output dimension of 7, corresponding to the six positive samples and one negative sample, respectively. The classification model does not load pre-trained weights based on natural images during training. The classification network architecture includes, but is not limited to, Xception, Inception_V3, and ResNet18, as well as optimized classification network architectures that add attention modules or other modules to the traditional classification network architecture.

[0011] Optionally, the optimal key layer for each category is obtained through the following process: extracting the probability sequence of the current category based on the classification prediction results; selecting aortic dissection CTA images with probability values ​​greater than the preset probability threshold from the probability sequence of the current category based on a preset probability threshold, and recording the location index of the selected aortic dissection CTA images; removing isolated points from the image location indexes to obtain at least one continuous location index sequence, and determining the candidate image sequence for the current category based on the location index sequence; selecting one or more aortic dissection CTA images as the optimal key layer for the current category from the candidate image sequence of the current category according to the preset selection rules for the current category.

[0012] Optionally, the preset selection rules for each category include: for the ascending aorta origin level and the abdominal aorta bifurcation level, selecting the uppermost layer in the candidate image sequence of the current category as the optimal key layer for the ascending aorta origin level and the abdominal aorta bifurcation level; for the ascending aorta termination level and the descending aorta origin level, selecting the lowermost layer in the candidate image sequence of the current category as the optimal key layer for the ascending aorta termination level and the descending aorta origin level; for the pulmonary artery bifurcation level, selecting the aortic dissection CTA image layer corresponding to the maximum probability in the probability sequence of the current category as the optimal key layer for the pulmonary artery bifurcation level; if the maximum probability is not unique, then selecting the middle layer in the candidate image sequence of the current category as the optimal key layer for the pulmonary artery bifurcation level; for the celiac trunk level and the superior mesenteric artery level, dividing the candidate image sequence of the current category into two groups, and selecting the middle layer of each group of candidate image sequences as the optimal key layer for the celiac trunk level and the superior mesenteric artery level; for the renal artery level, finding the probability peak in the probability sequence of the current category, and for each probability peak, selecting the middle layer of the candidate image sequence corresponding to each probability peak as the optimal key layer for the renal artery level.

[0013] Secondly, embodiments of this application provide an electronic device, including a memory and a processor. The memory stores a computer program that can run on the processor. When the processor executes the program, it implements the steps in the above-described method for automatic identification of key layers based on aortic dissection CTA images.

[0014] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the steps in the above-described method for automatic identification of key layers based on aortic dissection CTA images.

[0015] The beneficial effects of the technical solutions provided in this application include at least the following: This application provides an automatic identification method for key layers based on CTA images of aortic dissection. A classification system is established, and a set of aortic dissection CTA image sequences is labeled based on this system. The labeled set of aortic dissection CTA image sequences is then divided into training, validation, and test sets, providing a high-quality and reliable data foundation for subsequent classification models. This effectively prevents model overfitting and ensures the generalization ability of the identification results. Furthermore, the labeled data based on the classification system enables the model to learn more robust anatomical features, ensuring the accuracy and clinical reliability of subsequent automatic identification results. Most importantly, the original three-dimensional aortic dissection CTA images are divided into two-dimensional aortic dissection CTA image sequences, thereby enabling the identification of key layers. The problem of key-level identification in 3D images is transformed into a classification and optimal selection problem for 2D images. This greatly simplifies the process of key-level identification in aortic dissection CTA images, thereby improving the accuracy and real-time performance of the final identification results. Preprocessing the aortic dissection CTA image sequence set yields a standardized image set, eliminating differences in spatial resolution, grayscale values, and size among images from different sources. This reduces the model's dependence on specific scanning equipment and improves its generalization performance. Furthermore, data augmentation is performed on the aortic dissection CTA images in the standardized image set that belong to the training set, resulting in an augmented training set. This effectively increases the diversity of training samples and addresses the issue of data fragmentation. To address the class imbalance issue, the robustness and recognition accuracy of the model are further improved. The model is trained using a classification network architecture on an expanded training set, and its performance is evaluated using a validation set to determine the optimal classification model. Based on the optimal model, aortic dissection CTA image sequences from each patient in the test set are classified, and the classification prediction results are obtained using a softmax function. These predictions include the probability of each aortic dissection CTA image belonging to a specific class. For each patient's aortic dissection CTA image sequence, probability sequences for each class are extracted based on the classification prediction results. Threshold filtering and outlier removal are applied to these probability sequences to optimize the model's initial predictions, resulting in candidate sequences for each class. By selecting image sequences to effectively eliminate unreliable predictions, the accuracy and robustness of the final key layer localization are significantly improved. Based on preset selection rules for each category, the optimal key layer for each category is determined from candidate image sequences, thereby achieving precise identification of the unique optimal key layer from a series of similar images. The identification results fully conform to the anatomical standards for clinical diagnosis, ultimately achieving a balance between high automation, high real-time performance, and high accuracy in the identification process. This saves doctors' image reading time, assists them in making faster diagnoses, and improves work efficiency in surgical planning and long-term follow-up monitoring. It also contributes to the intelligent information feature extraction of CTA images, accelerating the development and application of artificial intelligence technology in the field of aortic disease. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort, wherein: Figure 1 A flowchart illustrating an automatic identification method for key layers based on aortic dissection CTA images provided in this application embodiment; Figure 2 A schematic diagram of key layers in a CTA image of aortic dissection provided in this application embodiment; Figure 3 A flowchart illustrating the optimal critical level selection process for the ascending aorta origin level and the abdominal aorta bifurcation level, provided for embodiments of this application; Figure 4 A flowchart illustrating the process of selecting the optimal critical layer between the end of the ascending aorta and the beginning of the descending aorta, provided for an embodiment of this application; Figure 5 A flowchart illustrating the optimal critical layer selection process for the pulmonary artery bifurcation level, provided in an embodiment of this application; Figure 6 A flowchart illustrating the optimal key level selection process for the celiac trunk level and the superior mesenteric artery level, provided for embodiments of this application; Figure 7 A flowchart illustrating the optimal key level selection process at the renal artery level is provided in this application embodiment. Figure 8 This is a schematic diagram of the hardware entity of an electronic device provided in an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, 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 embodiments of this application, not all embodiments. The following embodiments are used to illustrate this application, but are not intended to limit the scope of this application. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0019] It should be noted that the terms "first, second, and third" used in the embodiments of this application are merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, and third" can be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0020] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which the embodiments of this application pertain. It should also be understood that terms such as those defined in general dictionaries should be understood to have a meaning consistent with their meaning in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as herein.

[0021] The embodiments of this application will be further described below with reference to the accompanying drawings.

[0022] In view of the current problems in the research on key layer identification of aortic dissection CTA images in the field of deep learning-based medical image processing technology, this application provides an automatic key layer identification method based on aortic dissection CTA images.

[0023] The technical solution of this application is described below, starting with the method embodiments.

[0024] Please refer to Figure 1 It illustrates a flowchart of an automatic identification method for key layers based on aortic dissection CTA images provided in this application embodiment, such as... Figure 1 As shown, the method includes at least the following steps S110 to S130.

[0025] Step S110: Establish a classification system, and label the aortic dissection CTA image sequence set based on the classification system. Divide the labeled aortic dissection CTA image sequence set into a training set, a validation set, and a test set. The classification system includes six positive samples and one negative sample.

[0026] In this embodiment, key layers of aortic dissection CTA images are identified based on cross-sectional images of aortic dissection CTA data. For example, please refer to [link / reference]. Figure 2This illustration shows a schematic diagram of key layers in a CTA image of aortic dissection provided in this application embodiment. The key layers include the ascending aorta origin layer, the ascending aorta termination layer, the descending aorta origin layer, the pulmonary artery bifurcation level, the celiac trunk level, the superior mesenteric artery level, the renal artery level (including the left and right renal arteries), and the abdominal aortic bifurcation level. A classification system is established based on predefined anatomical features, thereby dividing all key layers of aortic dissection CTA images into six positive samples and one negative sample. Specifically, the ascending aorta originates at the junction of the sinus ducts. Because the aortic sinus's image features are relatively prominent, images containing the aortic sinus are classified as a single category. The uppermost image of the aortic sinus structure from top to bottom represents the ascending aorta origin, and images containing the aortic sinus structure are classified as the first type of positive samples. The ascending aorta termination level is similar to and sometimes the same level as the descending aorta origin level; therefore, the ascending aorta termination level and the descending aorta origin level are classified as the same level. Since the aortic arch's image features are relatively prominent, images from the appearance to the disappearance of the aortic arch are classified as the same category. The level where the aortic arch just disappears from top to bottom is the key level for the ascending aorta termination and the descending aorta origin; images containing the aortic arch structure are classified as the second type of positive samples. Images at the pulmonary artery bifurcation level include images from the beginning of the "V" shape to its disappearance. The level with the most prominent bifurcation features is the key level, and images at the pulmonary artery bifurcation level are classified as the third type of positive samples. The celiac trunk and superior mesenteric artery levels are close to each other and have similar characteristics. Classifying them using binary classification yields an accuracy of only 0.57–0.62. Therefore, they are grouped into one category. Images from the appearance to disappearance of branches of the celiac trunk and superior mesenteric artery belong to this category, with the middle layer of the branches being the key layer. The celiac trunk and superior mesenteric artery levels are thus classified as the fourth type of positive samples. Images at the renal artery level (including both left and right renal arteries) include images from the appearance to disappearance of branches, with the middle layer of the branches being the key layer. The renal artery level is thus classified as the fifth type of positive samples. Images at the abdominal aortic bifurcation level start from the bifurcation level and stop at the secondary bifurcation, with the uppermost layer being the key layer. The abdominal aortic bifurcation level is thus classified as the sixth type of positive samples. Among all the key layers of aortic dissection CTA images, excluding the first to sixth types of positive samples, the rest are negative samples.

[0027] In this embodiment of the application, an aortic dissection CTA image sequence set is constructed by collecting aortic dissection CTA image sequences from multiple patients in cardiovascular surgery and public datasets. The original three-dimensional aortic dissection CTA images are divided into two-dimensional aortic dissection CTA image sequences, thereby transforming the problem of key layer identification for three-dimensional images into a classification and optimal selection problem for two-dimensional images, which greatly simplifies the process of key layer identification for aortic dissection CTA images. Furthermore, based on the classification system, medical experts, including physicians from the Chest Hospital, annotated the aortic dissection CTA image sequence set. Specifically, for each patient's aortic dissection CTA image sequence in the set, all key layers in all aortic dissection CTA images were annotated. After all patients' aortic dissection CTA image sequences were annotated, the annotated aortic dissection CTA image sequence set was divided into datasets according to a preset ratio. Specifically, each patient's complete aortic dissection CTA image sequence was used as the dataset division unit, and the annotated aortic dissection CTA image sequence set was randomly divided into training, validation, and test sets according to a 7:1:2 dataset division ratio. This ensured that all aortic dissection CTA images of the same patient were assigned to only one dataset, guaranteeing the rigor and accuracy of subsequent key layer evaluation processes based on each dataset.

[0028] Step S120: Preprocess the aortic dissection CTA image sequence set to obtain a standardized image set, and perform data augmentation processing on the aortic dissection CTA images in the standardized image set that belong to the training set to obtain an augmented training set; train the model using a classification network architecture based on the augmented training set, and evaluate the performance of the classification model through a validation set to determine the optimal classification model.

[0029] In this embodiment, the aortic dissection CTA image sequence set is preprocessed to obtain a standardized image set. Specifically, the XY plane of the original aortic dissection CTA image sequence is resampled to unify the resolution to a preset size, such as 1mm x 1mm, while keeping the Z-axis direction unchanged. Since different devices and scanning parameters can cause global shifts or scaling of CT values, the gray values ​​of the resampled aortic dissection CTA image sequence are normalized to a preset range, such as [-1, 1] or [0, 1], thereby reducing device dependence, accelerating model convergence, and improving model generalization. After resampling, due to the different spacing of the original data, the aortic dissection CTA images of different patients may still be inconsistent in matrix size. Therefore, based on the image center, the resampled and normalized aortic dissection CTA image sequence is cropped to a fixed pixel size, such as 256*256, to finally obtain the standardized image set. Furthermore, since medical data is usually limited and the number of samples in each category is imbalanced, it can easily lead to model overfitting. Therefore, data augmentation processing is performed on the aortic dissection CTA images in the standardized image set that belong to the training set. Specifically, the aortic dissection CTA images are flipped, rotated, and translated to obtain an augmented training set, thereby increasing the data diversity in the training set. In particular, data augmentation is performed on key layer categories with small data volumes, which can reduce data limitations, prevent the model from being biased towards categories with large data volumes, effectively alleviate class imbalance, improve the model's generalization ability, and enhance the model's stability in practical applications.

[0030] In this embodiment, a classification network architecture is constructed based on deep learning networks. This architecture includes, but is not limited to, Xception, Inception_V3, and ResNet18, as well as optimized classification network architectures that add attention modules or other modules to the above-mentioned traditional classification network architectures. To correspond to six positive samples and one negative sample, the output dimension of the fully connected layers in the classification network architecture is set to 7. After the classification network architecture is constructed, the model is trained using the expanded training set. Due to the significant domain differences between medical images and natural images, the benefits of using pre-trained weights based on natural images are limited, and may even have negative effects. Therefore, pre-trained weights based on natural images are not loaded during model training to ensure that the model is trained from scratch. Furthermore, the classification performance of the classification model is evaluated using a validation set, and the optimal classification model can be determined after comparison.

[0031] Step S130: Classify the aortic dissection CTA image sequences of each patient in the test set according to the optimal classification model, and obtain the classification prediction result through the softmax function. The classification prediction result includes the probability of each aortic dissection CTA image belonging to each category. For each patient's aortic dissection CTA image sequence, extract the probability sequence of each category based on the classification prediction result, perform threshold filtering and outlier removal on the probability sequence of each category to obtain candidate image sequences of each category. According to the preset selection rules of each category, determine the optimal key layer of each category from the candidate image sequences of each category.

[0032] In this embodiment of the application, the aortic dissection CTA image sequence of each patient in the test set is classified based on the trained optimal classification model, and the probability of each aortic dissection CTA image in the aortic dissection CTA image sequence belonging to each category is determined by the softmax function. The sum of the probabilities of each aortic dissection CTA image in all categories is 1. Finally, the classification prediction result is obtained and saved. The classification prediction result includes the probability of all aortic dissection CTA images of each patient belonging to each category.

[0033] Furthermore, for each patient's aortic dissection CTA image sequence, a probability sequence for each category is extracted based on the classification prediction results. Threshold filtering and outlier removal are then performed on the probability sequences for each category to obtain candidate image sequences for each category. Specifically, based on the classification prediction results, the probability sequence for the current category is extracted. Based on a preset probability threshold, aortic dissection CTA images with probability values ​​greater than the preset threshold are selected from the probability sequences of the current category. The location indices of the selected aortic dissection CTA images are recorded. Outlier removal is performed on the image location indices to obtain at least one continuous location index sequence. Candidate image sequences for the current category are then determined based on the location index sequences. Further, according to the preset selection rules for the current category, one or more aortic dissection CTA images are selected from the candidate image sequences of the current category as the optimal key layer for the current category.

[0034] In specific embodiments, the optimal key layers for each category are selected through the following process: Please refer to Figure 3The document illustrates a flowchart of the optimal key layer selection process for the ascending aortic origin layer provided in this application embodiment. For the ascending aortic origin layer, the probability sequence corresponding to the image of this category is extracted from the classification prediction results and analyzed. The location index of the aortic dissection CTA image with a probability value greater than a preset probability threshold in the probability sequence is saved, and isolated points in the image location index are removed to reduce evaluation error. Multiple consecutive images containing the aortic sinus are obtained, which are candidate image sequences corresponding to the category of ascending aortic origin layer. The uppermost layer in the candidate image sequence is selected as the optimal key layer for the category of ascending aortic origin layer according to the top-to-bottom direction of the aortic structure.

[0035] For specific embodiments, please refer to Figure 4 This document illustrates a flowchart of the optimal key layer selection process for the ascending aorta termination and descending aorta initiation layers provided in this application embodiment. For the ascending aorta termination and descending aorta initiation layers, the probability sequence corresponding to the image of this category is extracted from the classification prediction results and analyzed. The location index of aortic dissection CTA images with probability values ​​greater than a preset probability threshold in the probability sequence is saved, and isolated points in the image location index are removed to reduce evaluation errors. Multiple consecutive images containing the aortic arch are obtained, which are candidate image sequences corresponding to the category of ascending aorta termination and descending aorta initiation layers. The lowest layer in the candidate image sequence is selected as the optimal key layer for the category of ascending aorta termination and descending aorta initiation layers according to the direction of the aortic structure from top to bottom.

[0036] For specific embodiments, please refer to Figure 5 The document illustrates a flowchart of the optimal key layer selection process for pulmonary artery bifurcation level provided in an embodiment of this application. For pulmonary artery bifurcation level, the probability sequence corresponding to the image of that category is extracted from the classification prediction results and analyzed to determine the maximum probability value in the probability sequence. If the maximum probability value is unique, the aortic dissection CTA image layer corresponding to the maximum probability value is selected as the optimal key layer for the pulmonary artery bifurcation level category. If the maximum probability value is not unique, multiple consecutive images of pulmonary artery bifurcation layers containing "V" shaped features are obtained, which are candidate image sequences corresponding to the pulmonary artery bifurcation level category. The middle layer in the candidate image sequence is selected as the optimal key layer for the pulmonary artery bifurcation level category.

[0037] For specific embodiments, please refer to Figure 6This document illustrates a flowchart of the optimal key layer selection process for the celiac trunk and superior mesenteric artery levels provided in this application embodiment. For the celiac trunk and superior mesenteric artery levels, probability sequences corresponding to the images of that category are extracted from the classification prediction results and analyzed. The location indexes of aortic dissection CTA images with probability values ​​greater than a preset probability threshold in the probability sequence are saved, and isolated points in the image location indexes are removed to reduce evaluation errors. Continuous candidate image sequences are saved. The number of candidate image sequences is usually one or two. If there is only one candidate image sequence, it is divided into two groups from the middle position to obtain two groups of multiple continuous images containing branches of the celiac trunk and superior mesenteric artery. The middle layer of each group of candidate image sequences is selected as the optimal key layer for the category of celiac trunk and superior mesenteric artery levels according to the direction of the aortic structure from top to bottom.

[0038] For specific embodiments, please refer to Figure 7 This document illustrates a flowchart of the optimal key layer selection process for the renal artery level, as provided in an embodiment of this application. For the renal artery level, which includes both left and right renal arteries, the number of optimal key layers for this category is not fixed because the relative positions of the branches of the left and right renal arteries are not fixed, and sometimes there are two renal arteries or accessory renal arteries. The probability sequence corresponding to the image of this category is extracted from the classification prediction results and analyzed. The location index of aortic dissection CTA images with probability values ​​greater than a preset probability threshold is saved, and isolated points in the image location index are removed to reduce evaluation errors. Continuous candidate image sequences are saved, and probability peaks (probability values ​​greater than or equal to the probabilities of adjacent layers) are searched in the probability sequence. For each probability peak, the middle layer of the candidate image sequence corresponding to each probability peak is selected as the optimal key layer for the renal artery level category.

[0039] For specific embodiments, please refer to Figure 3 This document illustrates a flowchart of the optimal key layer selection process for the abdominal aortic bifurcation level, as provided in an embodiment of this application. For the abdominal aortic bifurcation level, the probability sequence corresponding to the image of that category is extracted from the classification prediction results and analyzed. The location indices of aortic dissection CTA images with probability values ​​greater than a preset probability threshold are saved, and isolated points in the image location indices are removed to reduce evaluation errors. This results in multiple consecutive images containing the abdominal aortic bifurcation, i.e., a candidate image sequence corresponding to the abdominal aortic bifurcation level category. The uppermost layer in the candidate image sequence is selected as the optimal key layer for the abdominal aortic bifurcation level category, following the top-to-bottom direction of the aortic structure. It is important to note that... Figures 3 to 7 slices and Figures 3 to 7The aortic dissection CTA images described in the specific embodiments all convey the same meaning. Furthermore, the optimal classification model trained using the Xception architecture identified the following levels on the test set: the starting level of the ascending aorta, the ending level of the ascending aorta and the starting level of the descending aorta, the pulmonary bifurcation level, the celiac trunk level, the superior mesenteric artery level, the renal artery level, and the abdominal aortic bifurcation level. The recognition accuracies for each category were 0.9079, 0.9605, 0.9211, 0.9737, 0.9605, 0.8640, and 0.8684, respectively, with an average processing time of 0.0138 seconds per image. This indicates that the automatic key level identification method based on aortic dissection CTA images provided in this application can efficiently, directly, and accurately complete the key level identification task.

[0040] In summary, the present application provides an automatic identification method for key layers based on aortic dissection CTA images. This method establishes a classification system, annotates a set of aortic dissection CTA image sequences based on this system, and divides the annotated set into training, validation, and test sets. This provides a high-quality and reliable data foundation for subsequent classification models, effectively preventing overfitting and ensuring the generalization ability of the identification results. Furthermore, the annotated data based on the classification system enables the model to learn more robust anatomical features, ensuring the accuracy and clinical reliability of subsequent automatic identification results. Most importantly, it divides the original three-dimensional aortic dissection CTA images into two-dimensional aortic dissection CTA image sequences. This approach transforms the problem of key-level identification in 3D images into a classification and optimal selection problem in 2D images, greatly simplifying the process of key-level identification in aortic dissection CTA images and thus improving the accuracy and real-time performance of the final identification results. Preprocessing the aortic dissection CTA image sequence set to obtain a standardized image set eliminates differences in spatial resolution, grayscale values, and size among aortic dissection CTA images from different sources, reducing the model's dependence on specific scanning equipment and improving its generalization performance. Furthermore, data augmentation is performed on the aortic dissection CTA images in the standardized image set that belong to the training set, resulting in an augmented training set, effectively increasing the diversity of training samples. This study addresses the class imbalance problem in the dataset, further improving the model's robustness and recognition accuracy. Based on an expanded training set, a classification network architecture is used for model training, and the performance of the classification model is evaluated using a validation set to determine the optimal model. The optimal classification model is then used to classify the aortic dissection CTA image sequences of each patient in the test set, and the classification prediction results are obtained using a softmax function. These prediction results include the probability of each aortic dissection CTA image belonging to each category. For each patient's aortic dissection CTA image sequence, probability sequences for each category are extracted based on the classification prediction results. Threshold filtering and outlier removal are then applied to these probability sequences to optimize the model's initial prediction results, yielding the final category values. The candidate image sequence effectively eliminates unreliable predictions, significantly improving the accuracy and robustness of the final key layer localization. Based on preset selection rules for each category, the optimal key layer for each category is determined from the candidate image sequences, thus achieving precise identification of the unique optimal key layer from a series of similar images. The identification results fully conform to the anatomical standards for clinical diagnosis, ultimately achieving a balance between high automation, high real-time performance, and high accuracy in the identification process. This saves doctors' image reading time, assists them in making faster diagnoses, and improves work efficiency in surgical planning and long-term follow-up monitoring. It also contributes to the intelligent information feature extraction of CTA images, accelerating the development and application of artificial intelligence technology in the field of aortic disease.

[0041] It should be noted that, in the embodiments of this application, if the above-mentioned method for automatic identification of key layers based on aortic dissection CTA images is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of this application, or the part that contributes to related technologies, 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 an electronic device to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), magnetic disks, or optical disks. Thus, the embodiments of this application are not limited to any specific hardware and software combination.

[0042] Correspondingly, embodiments of this application provide a computer-readable storage medium storing a computer program thereon. When executed by a processor, this computer program implements the steps in the method for automatic identification of key layers based on aortic dissection CTA images described in any of the above embodiments. Correspondingly, embodiments of this application also provide a computer program product, which, when executed by a processor of an electronic device, is used to implement the steps in the method for automatic identification of key layers based on aortic dissection CTA images described in any of the above embodiments.

[0043] Based on the same technical concept, this application provides an electronic device for implementing the method described in the above-described method embodiments, which is an automatic identification method for key layers based on aortic dissection CTA images. Figure 8 This is a hardware entity diagram of an electronic device provided in an embodiment of this application, such as... Figure 8 As shown, the electronic device 800 includes a memory 810 and a processor 820. The memory 810 stores a computer program that can run on the processor 820. When the processor 820 executes the program, it implements the steps in the method for automatic identification of key layers based on aortic dissection CTA images according to any embodiment of this application.

[0044] The memory 810 is configured to store instructions and applications executable by the processor 820, and can also cache data to be processed or already processed by the processor 820 and various modules in the electronic device (e.g., image data, audio data, voice communication data and video communication data), which can be implemented by flash memory or random access memory (RAM).

[0045] When the processor 820 executes a program, it implements the steps of an automatic identification method for key layers based on aortic dissection CTA images, as described above. The processor 820 typically controls the overall operation of the electronic device 800.

[0046] The aforementioned processor can be at least one of the following: Application Specific Integrated Circuit (ASIC), Digital Signal Processor (DSP), Digital Signal Processing Device (DSPD), Programmable Logic Device (PLD), Field Programmable Gate Array (FPGA), Central Processing Unit (CPU), Controller, Microcontroller, and Microprocessor. It is understood that other electronic devices can also implement the functions of the aforementioned processor, and this application does not specifically limit the specific implementation.

[0047] The aforementioned computer storage media / memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), magnetic random access memory (FRAM), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM), etc.; or it can be various electronic devices that include one or any combination of the above-mentioned memories, such as mobile phones, computers, tablet devices, personal digital assistants, etc.

[0048] It should be noted that the descriptions of the storage medium and device embodiments above are similar to those of the method embodiments above, and have similar beneficial effects. For technical details not disclosed in the storage medium and device embodiments of this application, please refer to the descriptions of the method embodiments of this application for understanding.

[0049] It should be understood that the phrase "one embodiment" or "an embodiment" throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, "in one embodiment" or "in an embodiment" appearing throughout the specification does not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. It should be understood that in the various embodiments of this application, the sequence numbers of the above-described processes do not imply a sequential order of execution; the execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application. The sequence numbers of the above-described embodiments are merely descriptive and do not represent the superiority or inferiority of the embodiments.

[0050] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0051] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components can be combined, or integrated into another system, or some features can be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed can be through some interfaces, and the indirect coupling or communication connection between devices or units can be electrical, mechanical, or other forms.

[0052] The units described above as separate components may or may not be physically separate. The components shown as units may or may not be physical units. They may be located in one place or distributed across multiple network units. Some or all of the units may be selected to achieve the purpose of the embodiments of this application, depending on actual needs.

[0053] In addition, each functional unit in the various embodiments of this application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in hardware or in the form of hardware plus software functional units.

[0054] Alternatively, if the integrated units described above are implemented as software functional modules and sold or used as independent products, they can also be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, or the parts that contribute to related technologies, 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 the device automatic test line to execute all or part of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROMs, magnetic disks, or optical disks.

[0055] The methods disclosed in the several method embodiments provided in this application can be arbitrarily combined without conflict to obtain new method embodiments.

[0056] The features disclosed in the several method or device embodiments provided in this application can be arbitrarily combined without conflict to obtain new method or device embodiments.

[0057] The above description is merely an embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for automatic identification of key layers based on CTA images of aortic dissection, characterized in that, The method includes: A classification system is established, and the aortic dissection CTA image sequence set is labeled based on the classification system. The labeled aortic dissection CTA image sequence set is divided into a training set, a validation set, and a test set. The classification system includes six positive samples and one negative sample. The aortic dissection CTA image sequence set is preprocessed to obtain a standardized image set, and the aortic dissection CTA images belonging to the training set in the standardized image set are subjected to data augmentation processing to obtain an augmented training set; a model is trained using a classification network architecture based on the augmented training set, and the performance of the classification model is evaluated through a validation set to determine the optimal classification model; The aortic dissection CTA image sequences of each patient in the test set are classified according to the optimal classification model, and the classification prediction results are obtained by using the softmax function. The classification prediction results include the probability of each aortic dissection CTA image belonging to each category. For each patient's aortic dissection CTA image sequence, the probability sequences of each category are extracted based on the classification prediction results. Threshold filtering and outlier removal are performed on the probability sequences of each category to obtain candidate image sequences of each category. According to the preset selection rules of each category, the optimal key layer of each category is determined from the candidate image sequences of each category.

2. The method according to claim 1, characterized in that, The six types of positive samples include: the aortic sinus structure at the level of the origin of the ascending aorta, the aortic arch structure at the level of the end of the ascending aorta and the beginning of the descending aorta, the level of the pulmonary artery bifurcation, the level of the celiac trunk and the superior mesenteric artery, the level of the renal artery and the level of the abdominal aortic bifurcation. The aortic dissection CTA image sequence set includes aortic dissection CTA image sequences from multiple patients; the classification system is established based on predefined anatomical features; the dataset of the aortic dissection CTA image sequence set is divided into aortic dissection CTA image sequences from each patient.

3. The method according to claim 1, characterized in that, The preprocessing of the aortic dissection CTA image sequence set to obtain a standardized image set includes: The XY plane of the aortic dissection CTA image sequence is resampled to unify the resolution to a preset size; The gray values ​​of the resampled aortic dissection CTA image sequence are normalized to a preset range; Using the image center as a reference, the normalized aortic dissection CTA image sequence is cropped to a fixed pixel size to obtain the standardized image set.

4. The method according to claim 1, characterized in that, The fully connected layer of the classification model has an output dimension of 7, corresponding to the six positive samples and one negative sample. The classification model does not load pre-trained weights based on natural images during training. The classification network architecture includes, but is not limited to, Xception, Inception_V3, and ResNet18, as well as optimized classification network architectures that add attention modules or other modules to the traditional classification network architecture.

5. The method according to claim 1, characterized in that, The optimal key dimensions for each category are obtained through the following process: Extract the probability sequence of the current category based on the classification prediction results; Based on a preset probability threshold, aortic dissection CTA images with a probability value greater than the preset probability threshold are selected from the probability sequence of the current category, and the location index of the selected aortic dissection CTA images is recorded. Isolated points are removed from the image location indexes to obtain at least one continuous location index sequence, and a candidate image sequence for the current category is determined based on the location index sequence. Based on the preset selection rules of the current category, one or more aortic dissection CTA images are selected from the candidate image sequence of the current category as the optimal key layer of the current category.

6. The method according to claim 5, characterized in that, The preset selection rules for each category include: For the ascending aortic origin level and the abdominal aortic bifurcation level, the uppermost layer in the candidate image sequence of the current category is selected as the optimal key level for the ascending aortic origin level and the abdominal aortic bifurcation level. For the layers at the end of the ascending aorta and the beginning of the descending aorta, the lowest layer in the candidate image sequence of the current category is selected as the optimal key layer for the layers at the end of the ascending aorta and the beginning of the descending aorta. For the pulmonary artery bifurcation level, the aortic dissection CTA image layer corresponding to the maximum probability in the probability sequence of the current category is selected as the optimal key layer for the pulmonary artery bifurcation level. If the maximum probability is not unique, the middle layer in the candidate image sequence of the current category is selected as the optimal key layer for the pulmonary artery bifurcation level. For the celiac trunk level and the superior mesenteric artery level, the candidate image sequences of the current category are divided into two groups, and the middle layer of each group of candidate image sequences is selected as the optimal key layer for the celiac trunk level and the superior mesenteric artery level. For the renal artery level, probability peaks are found in the probability sequence of the current category. For each probability peak, the intermediate layer of the candidate image sequence corresponding to each probability peak is selected as the optimal key layer for the renal artery level.

7. An electronic device comprising a memory and a processor, the memory storing a computer program executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the method according to any one of claims 1 to 6.

Images