Method and device for determining fractional flow reserve, computing equipment and storage medium

Abstract

The invention provides a method and device for determining fractional flow reserve, computing equipment and a storage medium. The method comprises the following steps: acquiring blood vessel segmentation data and lesion segmentation data of a target blood vessel; determining fusion features based on the blood vessel segmentation data and the lesion segmentation data through a feature extraction network; and determining a fractional flow reserve (FFR) value at at least one target position in the target blood vessel through a prediction network based on the fusion feature.

Description

Technical Field

[0001] This disclosure relates to the field of data processing, and in particular to a method, apparatus, computing device, and storage medium for determining fractional blood flow reserve. Background Technology

[0002] Fractional flow reserve (FFR) refers to the ratio of the maximum blood flow available to the myocardial region supplied by a coronary artery in the presence of stenosis to the theoretically maximum blood flow available to the same region under normal conditions. It reflects the health of the artery. A non-invasive method for measuring FFR is desired.

[0003] The methods described in this section are not necessarily methods that had been previously conceived or adopted. Unless otherwise specified, no method described in this section should be assumed to be prior art simply because it is included in this section. Similarly, unless otherwise specified, the issues mentioned in this section should not be considered to be accepted in any prior art. Summary of the Invention

[0004] According to one aspect of this disclosure, a method for determining the fractional flow reserve (FFR) is provided, comprising: acquiring vessel segmentation data and lesion segmentation data of a target vessel; determining fusion features based on the vessel segmentation data and the lesion segmentation data using a feature extraction network; and determining the fractional flow reserve (FFR) value at at least one target location in the target vessel based on the fusion features using a prediction network.

[0005] According to another aspect of this disclosure, an apparatus for determining the fractional flow reserve is provided, comprising: a data acquisition unit for acquiring vascular segmentation data and lesion segmentation data of a target blood vessel; a feature extraction unit for determining fusion features based on the vascular segmentation data and the lesion segmentation data through a feature extraction network; and a prediction unit for determining an FFR value at at least one target location in the target blood vessel based on the fusion features through a prediction network.

[0006] According to another aspect of this disclosure, a computing device is provided, comprising: a memory, a processor, and a computer program stored on the memory, wherein the processor is configured to execute the computer program to implement a method for determining a fraction of blood flow reserve according to one or more embodiments of this disclosure.

[0007] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a method for determining a fraction of blood flow reserve according to one or more embodiments of this disclosure.

[0008] According to another aspect of this disclosure, a computer program product is provided, including a computer program, wherein the computer program, when executed by a processor, implements a method for determining a fractional blood flow reserve according to one or more embodiments of this disclosure.

[0009] These and other aspects of this disclosure will be apparent from the embodiments described below, and will be elucidated with reference to the embodiments described below. Attached Figure Description

[0010] Further details, features, and advantages of this disclosure are disclosed in the following description of exemplary embodiments in conjunction with the accompanying drawings, in which: Figure 1 This is a schematic diagram illustrating an example system in which various methods described herein may be implemented according to exemplary embodiments; Figure 2 This is a flowchart illustrating a method for determining fractional blood flow reserve according to an exemplary embodiment; Figure 3 This is a schematic block diagram illustrating an apparatus for determining a fraction of blood flow reserve according to an exemplary embodiment; Figure 4 This is a block diagram illustrating an exemplary computer device that can be applied to an exemplary embodiment. Detailed Implementation

[0011] In this disclosure, unless otherwise stated, the use of terms such as "first," "second," etc., to describe various elements is not intended to limit the positional, temporal, or importance relationships of these elements; such terms are merely used to distinguish one element from another. In some examples, the first element and the second element may refer to the same instance of that element, while in other cases, based on the context, they may refer to different instances.

[0012] The terminology used in the description of the various examples described in this disclosure is for the purpose of describing particular examples only and is not intended to be limiting. Unless the context explicitly indicates otherwise, an element may be one or more unless the number of elements is specifically limited. As used herein, the term "multiple" means two or more, and the term "based on" should be interpreted as "at least partially based on". Furthermore, the terms "and / or" and "at least one of..." cover any one of the listed items and all possible combinations thereof.

[0013] Exemplary embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0014] Figure 1This is a schematic diagram illustrating an example system 100 in which various methods described herein may be implemented according to exemplary embodiments.

[0015] refer to Figure 1 The system 100 includes a client device 110, a server 120, and a network 130 that communicatively couples the client device 110 and the server 120.

[0016] Client device 110 includes a display 114 and a client application (APP) 112 that can be displayed on the display 114. Client application 112 can be an application that needs to be downloaded and installed before running, or a lightweight application (liteapp). If client application 112 is an application that needs to be downloaded and installed before running, client application 112 can be pre-installed on client device 110 and activated. If client application 112 is a mini-app, user 102 can run client application 112 directly on client device 110 without installing it, by searching for client application 112 in the host application (e.g., by the name of client application 112) or by scanning the graphic code of client application 112 (e.g., barcode, QR code, etc.). In some embodiments, client device 110 can be any type of mobile computing device, including mobile computers, mobile phones, wearable computing devices (e.g., smartwatches, head-mounted devices including smart glasses, etc.), or other types of mobile devices. In some embodiments, the client device 110 may alternatively be a stationary computer device, such as a desktop computer, server computer, or other type of stationary computer device. In some alternative embodiments, the client device 110 may also be or may include a medical image printing device.

[0017] Server 120 is typically a server deployed by an Internet Service Provider (ISP) or Internet Content Provider (ICP). Server 120 can represent a single server, a cluster of multiple servers, a distributed system, or a cloud server providing basic cloud services such as cloud databases, cloud computing, cloud storage, and cloud communications. It will be understood that, although... Figure 1 The diagram shows that server 120 communicates with only one client device 110, but server 120 can provide background services to multiple client devices simultaneously.

[0018] Examples of network 130 include combinations of local area networks (LANs), wide area networks (WANs), personal area networks (PANs), and / or communication networks such as the Internet. Network 130 can be wired or wireless. In some embodiments, technologies and / or formats including Hypertext Markup Language (HTML), Extensible Markup Language (XML), etc., are used to process data exchanged through network 130. Furthermore, encryption technologies such as Secure Sockets Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), and Internet Protocol Security (IPsec) can be used to encrypt all or some of the links. In some embodiments, custom and / or dedicated data communication technologies can be used to replace or supplement the aforementioned data communication technologies.

[0019] System 100 may further include image acquisition device 140. In some embodiments, Figure 1 The image acquisition device 140 shown may be a medical scanning device, including but not limited to scanning or imaging devices used in positron emission tomography (PET), positron emission tomography with computerized tomography (PET / CT), single photon emission computed tomography with computerized tomography (SPECT / CT), computed tomography (CT), medical ultrasonography, nuclear magnetic resonance imaging (NMRI), magnetic resonance imaging (MRI), cardiovascular angiography (CA), digital radiography (DR), etc. For example, image acquisition device 140 may include digital subtraction angiography scanner, magnetic resonance angiography scanner, computed tomography angiography scanner, positron emission tomography scanner, positron emission tomography (PET) scanner, single photon emission computed tomography (SPT) scanner, computed tomography scanner, medical ultrasound examination equipment, magnetic resonance imaging (MRI) scanner, digital radiography scanner, etc. Image acquisition device 140 may be connected to a server (e.g., Figure 1The system connects to server 120 (or a separate server of the imaging system, not shown in the figure) to process image data, including but not limited to converting scan data (e.g., converting it into a medical image sequence), compressing it, correcting pixels, and reconstructing it in three dimensions.

[0020] Image acquisition device 140 may be connected to client device 110, for example, via network 130, or otherwise directly connected to client device to communicate with client device.

[0021] Optionally, the system may also include an intelligent computing device or a computing card 150. The image acquisition device 140 may include or be connected (e.g., detachably connected) to such a computing card 150. As an example, the computing card 150 can perform image data processing, including but not limited to conversion, compression, pixel correction, reconstruction, etc. As another example, the computing card 150 can implement a method for determining the fractional blood flow reserve according to embodiments of this disclosure.

[0022] The system may also include other components not shown, such as a data storage unit. The data storage unit may be a database, data repository, or other form of device for data storage; it may be a conventional database, or it may include a cloud database, a distributed database, etc. For example, direct image data generated by the image acquisition device 140, or medical image sequences or three-dimensional image data obtained through image processing, may be stored in the data storage unit for subsequent retrieval by the server 120 and client device 110. Furthermore, the image acquisition device 140 may also directly provide direct image data or medical image sequences or three-dimensional image data obtained through image processing to the server 120 or client device 110, etc.

[0023] Users can use client device 110 to control the acquisition of images or videos, view the acquired images or videos (including preliminary image data or images after analysis and processing), view analysis results, interact with the acquired images or analysis results, input acquisition commands, configure data, etc. Client device 110 can send configuration data, commands, or other information to image acquisition device 140 to control image acquisition device acquisition, data processing, etc.

[0024] For the purposes of this disclosure's embodiments, Figure 1In the example, client application 112 can be an image sequence management application that provides various functions, such as storage management, indexing, sorting, and classification of acquired image sequences. Correspondingly, server 120 can be a server used in conjunction with the image sequence management application. Server 120 can provide image sequence management services to client application 112 running on client device 110 based on user requests or instructions generated according to embodiments of this disclosure. For example, it can manage image sequence storage in the cloud, store and classify image sequences according to specified indexes (including, but not limited to, sequence type, patient identifier, body part, acquisition target, acquisition stage, acquisition machine, presence of lesions, severity, etc.), and retrieve and provide image sequences to client devices according to specified indexes, etc. Alternatively, server 120 can also provide or allocate such service capabilities or storage space to client device 110, whereby client application 112 running on client device 110 provides corresponding image sequence management services based on user requests or instructions generated according to embodiments of this disclosure, etc. It is understood that the above is only one example, and this disclosure is not limited thereto.

[0025] Figure 2 This is a flowchart illustrating a method 200 for determining a fractional blood flow reserve according to an exemplary embodiment. Method 200 can be implemented on a client device (e.g., Figure 1 The execution is performed at the client device 110 shown, that is, the execution entity of each step of method 200 can be... Figure 1 The client device 110 shown. In some embodiments, method 200 can be performed on a server (e.g., Figure 1 The method 200 is executed at server 120 (as shown in the figure). In some embodiments, the method 200 may be executed in combination by a client device (e.g., client device 110) and a server (e.g., server 120).

[0026] The steps of method 200 are described in detail below.

[0027] refer to Figure 2 In step 210, the vessel segmentation data and lesion segmentation data of the target vessel are obtained.

[0028] In step 220, a feature extraction network determines fusion features based on the blood vessel segmentation data and the lesion segmentation data.

[0029] At step 230, based on the fusion features, the fractional flow reserve (FFR) value at at least one target location in the target blood vessel is determined by a prediction network.

[0030] The above method allows for the reception of both vessel segmentation data and lesion segmentation data as input, and the determination of fusion features based on these two data. This ensures that the model, when making predictions, must simultaneously consider both the global morphology of the vessels and the local pathological state. According to embodiments of this disclosure, it is ensured that lesion information crucial for FFR prediction is not ignored during the input stage.

[0031] For example, vessel segmentation data can refer to three-dimensional binary volume data or probabilistic volume data, which can be used to represent the complete three-dimensional morphology of the coronary artery tree. Lesion segmentation data can refer to another three-dimensional binary volume data or probabilistic volume data, which can be used solely to mark the location of atherosclerotic plaques or stenotic regions. The feature extraction network can be any convolutional neural network capable of processing three-dimensional data, such as 3D-U-Net or V-Net. Determining fusion features, in the broadest sense, can refer to the network ultimately integrating information from two (or more) input data sources into one or a set of high-dimensional feature vectors or feature maps through its internal convolutions and nonlinear transformations.

[0032] In one alternative embodiment, the vessel segmentation data and lesion segmentation data may not be three-dimensional volumetric data, but rather represented as three-dimensional mesh data or point cloud data. In another embodiment, the lesion segmentation data may not be a separate data entity, but rather implemented by modifying the vessel segmentation data (e.g., labeling healthy vessel voxels as 1 and lesion vessel voxels as 2). In an additional embodiment, in addition to the vessel and lesion segmentation data, the original CTA grayscale image data may also be acquired simultaneously and used as additional input information to determine the fusion features.

[0033] According to some embodiments, determining fusion features based on the blood vessel segmentation data and the lesion segmentation data through a feature extraction network may include extracting a first feature based on the blood vessel segmentation data and extracting a second feature based on the lesion segmentation data respectively through the feature extraction network; and fusing the first feature and the second feature to obtain fusion features.

[0034] In such an embodiment, the network can be forced to extract features from blood vessels and lesions separately. By establishing independent feature representations for blood vessels (first feature) and lesions (second feature), critical but often smaller lesion features can be prevented from being overwhelmed or averaged out when mixed with the large global vascular features. This greatly enhances the model's ability to perceive individual lesions, making it more sensitive to key factors that cause changes in FFR.

[0035] Understandably, the first feature can aim to capture macroscopic, contextual information about blood vessels, such as their course, diameter variations, and branching relationships, while the second feature can aim to capture microscopic, pathological information about lesions, such as their volume, length, and degree of eccentricity within the lumen. The fusion step can be performed in the middle or deep layers of the network, for example, by concatenating the first and second feature tensors along the channel dimension, or by element-wise addition or multiplication.

[0036] In some alternative and / or additional embodiments, the fusion can be a more sophisticated gated fusion. For example, a weight can be learned via a small neural network to dynamically determine the proportion of the first and second features in the fused features. In another embodiment, the second feature (lesion feature) can be used as a modifier to act on the first feature (vascular feature) in a non-linear manner (e.g., spatial transformation), rather than a simple concatenation or addition. It is understood that the above are merely examples, and this disclosure is not limited thereto.

[0037] In some examples, feature extraction is achieved through structural separation. For instance, a two-subnetwork architecture can be used: blood vessel segmentation data enters subnetwork A, which outputs the first feature (e.g., a feature tensor of [H, W, C1]), and lesion segmentation data enters subnetwork B, which outputs the second feature (e.g., a feature tensor of [H, W, C2]).

[0038] According to some embodiments, extracting a first feature based on the blood vessel segmentation data and extracting a second feature based on the lesion segmentation data through the feature extraction network includes extracting the first feature through a first sub-network of the feature extraction network and extracting the second feature through a second sub-network of the feature extraction network.

[0039] According to this embodiment, dedicated computational resources are provided for the extraction of vascular and lesion features by using two physically independent sub-networks (dual-branch networks). This separation ensures complete decoupling of the two features during the extraction phase, and each sub-network can be optimized to learn the optimal parameters for its specific task, thereby maximizing the design goal of enhancing the lesion as an individual entity.

[0040] Understandably, the first and second sub-networks can be two parallel branches of the feature extraction network. For example, the first sub-network could be a deeper 3D-ResNet for processing global vessel segmentation data, while the second sub-network could be a shallower 3DCNN with a smaller receptive field for processing local lesion segmentation data. After extracting their respective first and second features, they output the features to the subsequent fusion layer.

[0041] In some alternative and / or additional embodiments, the first and second subnetworks may share weights in the initial few layers (e.g., the first two convolutional blocks) to learn common low-level visual features (such as edges and corners) before separating into two independent branches to learn their respective high-level semantic features. In another embodiment, the two subnetworks may have the exact same network structure (i.e., Siamese networks) but be trained and inferred on different data.

[0042] According to some embodiments, the neural network includes two sub-networks corresponding to a blood vessel segmentation map and a lesion segmentation map, respectively, and the outputs of the sub-networks are fused during the prediction phase. According to some embodiments, the neural network employs an attention mechanism during feature fusion to highlight features of the lesion region.

[0043] According to some embodiments, extracting a first feature based on the blood vessel segmentation data and extracting a second feature based on the lesion segmentation data can be performed through an attention mechanism.

[0044] In this embodiment, the attention mechanism allows the network to dynamically learn to focus on lesion regions while sharing a backbone network. It uses lesion data as a cue to guide the network to concentrate its computational resources (attention) on the regions in the feature map corresponding to the lesions. A typical implementation involves inputting vessel segmentation data into a backbone network to extract a first feature (global feature map). Simultaneously, lesion segmentation data is input into a smaller attention network to generate a second feature (i.e., an attention map). Then, by element-wise multiplying the global feature map and the attention map, a weighted feature map (i.e., a fused feature) is obtained, which amplifies the feature responses of lesion regions and suppresses features of non-lesion regions.

[0045] As a concrete, non-restrictive example, using an attention-based architecture, vascular and lesion data can be merged at the input (e.g., as two channels) into a shared backbone network, outputting a shared feature map F. shared At this point, the features are mixed. Next, the attention module begins its work, generating an attention map A. The values ​​of this attention map A (typically between 0 and 1) represent the importance of each location in the space. The process of generating attention map A can itself be viewed as a separate feature extraction of lesion information, extracting features that represent the location of the lesion and the degree of attention it deserves. Subsequently, the shared feature map F... shared Element-wise multiplication with attention map A yields the final weighted feature map F. attended = F shared* A. In this step, features from the lesion area are amplified, while features from other areas are suppressed. This weighting process can be viewed as a fusion. Therefore, under the attention mechanism, the first feature can be understood as an unweighted shared feature map F containing global information. shared The second feature can be understood as the attention map A itself, which specifically reflects the importance of the lesion, and the final F is obtained. attended This can be understood as a result that has already incorporated the first and second features.

[0046] In some alternative and / or additional embodiments, the attention mechanism can be implemented as cross-modal attention, where lesion features are used as queries and vascular features are used as keys and values. The network retrieves the most relevant information about the lesion from the vascular features in this way. In another embodiment, a self-attention mechanism can be employed, where vascular and lesion data are concatenated along the channel dimension and input together. The network learns the interdependencies between all locations, thereby implicitly learning the importance of the lesion region.

[0047] According to some embodiments, the blood vessel segmentation data and the lesion segmentation data can be input into the feature extraction network through different channels.

[0048] According to this embodiment, by placing blood vessel segmentation data and lesion segmentation data in different channels, the network can clearly distinguish between these two types of information from the first convolutional layer. This provides the necessary data foundation for all subsequent separately extracted strategies (whether subnetworks or attention mechanisms), ensuring that the network can immediately differentiate between the two types of information.

[0049] Assuming the input medical image data is a three-dimensional volumetric data of size D*H*W, different channels can refer to a four-dimensional input tensor of size D*H*W*C, where C is the number of channels. For example, setting C=2, channel C1 stores vessel segmentation data, such as a mask of the form 0 or 1, and channel C2 stores lesion segmentation data, such as a mask of the form 0 or 1. In some alternative and / or additional embodiments, the number of channels can be greater than 2, for example, C=3, where C1 stores the original CTA grayscale image, C2 stores the vessel segmentation mask, and C3 stores the lesion segmentation mask. In another embodiment, if a dual-sub-network architecture is used, the different channels can be understood as two completely independent input streams, i.e., one D*H*W*1 tensor enters the first sub-network, and another D*H*W*1 tensor enters the second sub-network. It is understood that the above are merely examples, and this disclosure is not limited thereto.

[0050] According to some embodiments, determining the FFR value at at least one target location in the target blood vessel by means of a prediction network includes obtaining the FFR value by means of the prediction network based on a sequence of centerline sampling points labeled at the at least one target location.

[0051] According to this embodiment, the FFR prediction result can be the FFR value of each sampling point along the centerline of the blood vessel. By reducing the three-dimensional (3D) prediction problem to a one-dimensional (1D) sequence of centerline sampling points, the prediction task can be greatly simplified, and the prediction results can be highly clinically intuitive and comparable.

[0052] The central line sampling point sequence can refer to a series of three-dimensional coordinate points (x, y, z) selected at certain intervals along the central line of the blood vessel (usually from proximal to distal). i , y i , z i In one implementation, the feature extraction network can output a three-dimensional fused feature map, and then the system samples or interpolates a one-dimensional feature sequence from this map based on the coordinates of the centerline sequence. This one-dimensional feature sequence is then fed into a prediction network (e.g., a recurrent neural network (RNN) or a one-dimensional convolutional network (1D-CNN), ultimately outputting a sequence of FFR values ​​that correspond one-to-one with the centerline points.

[0053] In some alternative and / or additional embodiments, the centerline itself can be used as one of the inputs to the prediction network, rather than just for marking output locations. For example, the coordinates and tangent vector of the centerline can be encoded as features and input into the prediction network along with fused features extracted from the image. In another embodiment, the prediction network can directly output a three-dimensional FFR volume data (i.e., voxel-level FFR prediction), and then the system can sample from this three-dimensional volume data based on the centerline sequence to obtain the final FFR value sequence.

[0054] According to some embodiments, the centerline sampling point sequence can be sparse sampling or variable step-size sampling. Such embodiments allow for intelligent allocation of computational resources by employing sparse or variable step-size sampling. For example, fewer sampling points (sparse) can be used in healthy vessel segments with gentle FFR changes, while denser sampling points (variable step-size) can be used in lesions or bifurcation regions with drastic FFR changes. This can significantly reduce the number of points requiring prediction without sacrificing prediction accuracy in critical areas, thus lowering the computational burden.

[0055] It is understood that variable step size sampling can be predetermined based on geometric or lesion information. For example, sampling points can be more densely packed near the lesion area. As a non-limiting concrete example, the centerline sampling step size can be set to, for example, 0.5 mm within a certain distance from the lesion segmentation data markers; while in other areas, the step size can be set to 2 mm. Sparse sampling can be performed by adding a skip sampling step on top of fixed step size sampling (e.g., one point every 1 mm), for example, retaining only one of every five points. It is understood that the above values ​​are examples and this disclosure is not limited thereto.

[0056] In some alternative and / or additional embodiments, the sampling density may not be predetermined but dynamically determined by the model. For example, the first prediction uses very sparse points, and if the model detects that the FFR gradient (rate of change) between two sparse points is greater than a certain threshold, more sampling points are automatically added in that interval for denser prediction. In another embodiment, the sampling points may be determined based on the geometric features of the blood vessel (such as curvature), automatically increasing sampling density in regions with high vessel tortuosity, and this disclosure is not limited thereto.

[0057] For example, the FFR prediction result can be a prediction value covering the entire map or a local prediction value for a selected vessel segment.

[0058] As an example, the centerline sampling point sequence can cover the complete topology tree of the target vessel. This enables a "global input, global output" approach. The complete topology tree can refer to the centerline starting from the coronary artery ostium and extending along all major branches and their first- and second-order sub-branches. In implementation, the feature extraction network takes the three-dimensional segmentation data of the entire coronary tree as input, while the centerline sampling point sequence contains the point set of all these branches. The final output of the prediction network can be a sequence (or graph) structured data containing the FFR values ​​of all branches.

[0059] According to other embodiments, the centerline sampling point sequence may correspond to at least one predetermined segment of the target blood vessel. Such applications allow for refined analysis of a specific segment of blood vessel—e.g., a segment extending from one bifurcation point to the next. The predetermined segment can be manually selected by the user on a 3D model via a human-computer interface (e.g., clicking on a lesion, and the system automatically extracts segments of a certain length or range before and after the lesion), or the system can automatically detect all areas where the stenosis exceeds a specific threshold or percentage, define each such area as a predetermined segment, and then perform independent FFR prediction for each segment. It is understood that the above are merely examples, and this disclosure is not limited thereto.

[0060] According to some embodiments, the feature extraction network and the prediction network can be trained end-to-end. End-to-end training means treating the feature extraction network and the prediction network as a single, complete system for joint optimization. This means that the gradient of the loss function (i.e., the difference between the predicted FFR and the true FFR) will propagate back to the front end of the feature extraction network. This forces the feature extraction network to learn the features most useful for FFR prediction, rather than learning some general features that may be irrelevant to FFR prediction, thereby maximizing the performance of the entire model.

[0061] During training, the system can take vessel and lesion segmentation data as input and the ground truth FFR value of the centerline as a supervision signal. The loss (e.g., mean squared error) can be calculated at the output of the prediction network. Then, an optimizer (such as Adam or SGD) uses backpropagation to simultaneously update the weight parameters of both the prediction network and the feature extraction network (including all its subnetworks or attention modules). The first and second features output by the feature extraction network are intermediate quantities in this end-to-end process.

[0062] In some alternative and / or additional embodiments, staged training can be employed. For example, in the first stage, a feature extraction network is pre-trained using an independent loss function (such as an autoencoder loss) to learn how to reconstruct vascular morphology; in the second stage, the parameters of the feature extraction network are frozen, and only the prediction network is trained to learn how to predict FFR from the fixed features. In additional embodiments, an auxiliary loss can be added on top of end-to-end training; for example, a supervisory signal can be extracted from the intermediate layers of the feature extraction network, making it responsible not only for the final FFR but also for the accurate localization of the lesion area.

[0063] According to some embodiments, the method may further include normalizing at least one of the vessel segmentation data and the lesion segmentation data before determining the fusion features.

[0064] Such embodiments can improve the stability and generalization ability of the model through data preprocessing steps. This ensures that medical images from different sources and with different imaging parameters (e.g., different hospitals, different CT scanners) are scaled to a uniform numerical range (e.g., 0 to 1 or -1 to 1). This helps prevent gradient explosion or vanishing during network training, accelerates model convergence, and improves the model's adaptability to different data sources. In some alternative and / or additional embodiments, normalization can be instance normalization or batch normalization, which are performed inside the network's convolutional layers, rather than during the preprocessing stage. In additional embodiments, in addition to normalization, other data preprocessing steps may be included, such as data augmentation, which increases the diversity of training samples and improves the model's robustness by randomly rotating, scaling, cropping, or elastically deforming the input segmented data.

[0065] According to some optional, non-limiting embodiments, the input to the model may additionally include an index or information on the amount of FFR change. For example, pre-calculated FFR change in a local area may be used as input. The local area may be a lesion area, a stenosis area, or a bifurcation area, and these areas may have some known or influencing factors on FFR changes. The FFR change may be estimated by another simplified model, clinically measured, or derived in other ways that can be understood by those skilled in the art. Optionally, physical characteristics may include physiological parameters such as downstream myocardial mass / vascular bed volume extrapolated based on empirical models, and this disclosure is not limited thereto. In such embodiments, it is understood that the method may additionally include identifying at least one region of interest capable of causing pressure changes based on medical imaging data of the target vessel, and determining at least one local pressure change index associated with each of the identified at least one region of interest. The method may include determining the fractional flow reserve (FFR) value at at least one target location in the target vessel based on the at least one local pressure change index. This allows the model to focus its attention on key areas that have a significant impact on hemodynamics (such as lesions or specific morphologies), improving computational efficiency and prediction accuracy. It also makes the model's output (pressure drop values ​​at various localities) more clinically interpretable.

[0066] According to some embodiments, the region of interest may include at least one of a lesion region, a stenosis region, and a bifurcation region. In one embodiment, the lesion region may be further subdivided into calcified plaque regions and non-calcified (soft) plaque regions. Exemplarily, a stenosis region may be classified according to its degree of stenosis (e.g., mild, moderate, severe). Exemplarily, a bifurcation region may be further subdivided according to the angle of its bifurcation and the ratio of the diameter of the main branch to the branch at the bifurcation. In addition, the region of interest may also include other known hemodynamically relevant regions, such as tortuous segments of the vessel (e.g., >90 degrees) or areas where endovascular stents are implanted, and this disclosure is not limited thereto.

[0067] According to some embodiments, determining at least one local pressure change index associated with at least one identified region of interest includes determining the at least one local pressure change index by a prediction model, wherein the type of the at least one region of interest is input into the prediction model.

[0068] In some additional or alternative implementations, the type of region of interest (ROI) can be used as a one-hot encoded vector, concatenated with the image data, and input into the prediction model. In another implementation, a dedicated subnetwork can be trained for each ROI type. When determining local pressure change indicators, the system dynamically selects and calls the corresponding subnetwork for calculation based on the input ROI type.

[0069] According to some embodiments, determining at least one local pressure change index associated with at least one identified region of interest includes determining the at least one local pressure change index based on medical imaging data of the local region corresponding to the at least one region of interest.

[0070] This implementation enhances the model's ability to learn local features. By inputting only the local region image data corresponding to the ROI into the model, instead of the complete, massive whole-vascular tree image, the amount of data that needs to be processed is greatly reduced. This allows the model to be designed to be more lightweight, while also forcing the model to learn more deeply the direct relationship between local morphology and pressure drop, thus improving the model's robustness.

[0071] In some additional or alternative embodiments, the medical imaging data of a local region can be a three-dimensional voxel cube centered at the centroid of the ROI. In another embodiment, it can be a three-dimensional model of a vascular segment extending a certain length within and around the ROI along the vessel's centerline. In yet another embodiment, it can be a stack of multiple representative two-dimensional (2D) cross-sectional images within the ROI region. Exemplarily, at least one local pressure drop value can be determined based on a subset of local image data corresponding to the at least one region of interest. Also exemplaryly, at least one local pressure drop value can be determined based on local three-dimensional volume data containing the at least one region of interest.

[0072] According to some embodiments, the size of the local region is determined based on the diameter of the vessel containing the corresponding region of interest. This embodiment can further address the scale dependence problem of lesion morphology in vessels of different diameters. For example, a 50% stenosis in a proximal vessel with a diameter of 4 mm and a distal vessel with a diameter of 2 mm will have drastically different absolute sizes and hemodynamic effects. By dynamically determining the size of the local region based on the vessel diameter (e.g., always truncating twice the lesion length), the model can focus more on the local region rather than irrelevant regions, thus obtaining more accurate calculation results. Furthermore, the model can learn a scale-invariant law regarding the relative degree of stenosis, thereby enhancing the model's generalization ability.

[0073] In some additional or alternative implementations, the size of the local region can be determined as N times the reference diameter of the vessel at the location of the corresponding ROI (e.g., N=3). In another implementation, all captured local regions, regardless of their physical size, can be rescaled to a uniform standard size through interpolation or downsampling. Exemplarily, the original reference diameter of the vessel can also be input into the model as an additional numerical feature.

[0074] It is understood that although the various operations are depicted in the accompanying drawings in a specific order, this should not be construed as requiring that these operations must be performed in the specific order shown or in chronological order, nor should it be construed as requiring that all the operations shown must be performed to obtain the desired result. For example, two steps described in order herein may be performed in reverse order or may be performed concurrently. As another example, one or more steps in the various embodiments of this disclosure may be omitted.

[0075] Furthermore, it is understood that the methods for predicting or determining data according to one or more embodiments of this disclosure are not methods for doctors to directly determine diagnostic results, but rather involve data processing or information processing processes during the medical process. The data processing results can be used for doctors' reference, thereby assisting doctors in their medical operations. It is understood that the information processing methods, data prediction methods, determination methods, decision-making methods, etc., according to one or more embodiments of this disclosure are executed by a computer or a device containing a computer.

[0076] It is understood that throughout this disclosure, images, image sequences, or images may be or may include two-dimensional image data, or may be or may include three-dimensional image data. Images, image sequences, or images may be image data that is directly acquired and stored or otherwise transmitted to a terminal device for user use. Images, image sequences, or images may also be processed image data after various image processing steps. Images, image sequences, or images may undergo other analytical processes (e.g., analysis of the presence of lesion features or lesions) and include analytical results (e.g., delineation of regions of interest, tissue segmentation results, etc.). It is understood that this disclosure is not limited thereto.

[0077] Figure 3This is a schematic block diagram illustrating an apparatus 300 for determining the fractional flow reserve (FFR) according to an exemplary embodiment. The apparatus 300 for determining the FFR may include a data acquisition unit 310, a feature extraction unit 320, and a prediction unit 330. The data acquisition unit 310 is used to acquire vessel segmentation data and lesion segmentation data of a target vessel. The feature extraction unit 320 can be used to determine fusion features based on the vessel segmentation data and the lesion segmentation data using a feature extraction network. The prediction unit 330 can be used to determine the FFR value at at least one target location in the target vessel based on the fusion features using a prediction network.

[0078] It should be understood that Figure 3 The various modules of the device 300 shown can be connected to the reference. Figure 2 The steps in method 200 described correspond to each other. Therefore, the operations, features, and advantages described above for method 200 and its variations also apply to apparatus 300 and its included modules. For the sake of brevity, some operations, features, and advantages will not be repeated here.

[0079] According to embodiments of the present disclosure, a computing device is also disclosed, including a memory, a processor, and a computer program stored on the memory, wherein the processor is configured to execute the computer program to implement the steps of the method for determining a fraction of blood flow reserve according to embodiments of the present disclosure and variations thereof.

[0080] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium is also disclosed, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the method for determining the fractional blood flow reserve and variations thereof according to embodiments of the present disclosure.

[0081] According to embodiments of the present disclosure, a computer program product is also disclosed, comprising a computer program, wherein when executed by a processor, the computer program implements the steps of the method for determining a fraction of blood flow reserve and variations thereof according to embodiments of the present disclosure.

[0082] While specific functions have been discussed above with reference to specific modules, it should be noted that the functions of the various modules discussed herein may be divided into multiple modules, and / or at least some functions of multiple modules may be combined into a single module. The specific module discussed herein performing an action includes the specific module itself performing the action, or alternatively, the specific module calling or otherwise accessing another component or module that performs the action (or performs the action in conjunction with the specific module). Therefore, a specific module performing an action may include the specific module performing the action itself and / or another module that the specific module calls or otherwise accesses to perform the action. For example, the various modules or units described according to one or more embodiments of this disclosure may be combined into a single module or unit in some embodiments. As another example, two or more modules or units may be described in parallel in one or more embodiments of this disclosure, while in other embodiments, these modules and units may have one or more inclusion relationships. As used herein, the phrase "entity A initiates action B" or "entity A causes action B to be performed" may refer to entity A issuing an instruction to perform action B, but entity A itself does not necessarily perform action B. For example, the phrase "display module causes display..." could mean that the display module instructs a display (not shown) or other possible display device to display, without the display module itself needing to perform the "display" action.

[0083] It should also be understood that this article can describe various technologies in the general context of software and hardware components or program modules. The above regarding... Figure 3 The various modules described may be implemented in hardware or in hardware in combination with software and / or firmware. For example, these modules may be implemented as computer program code / instructions configured to execute in one or more processors and stored in a computer-readable storage medium. Alternatively, these modules may be implemented as hardware logic / circuit. For example, in some embodiments, one or more of the modules or units described according to one or more embodiments of this disclosure may be implemented together in a System on Chip (SoC). The SoC may include an integrated circuit chip (which includes a processor (e.g., a Central Processing Unit (CPU), microcontroller, microprocessor, digital signal processor (DSP), etc.), memory, one or more communication interfaces, and / or one or more other components of circuitry) and may optionally execute received program code and / or include embedded firmware to perform functions.

[0084] According to one aspect of this disclosure, a computing device is provided, including a memory, a processor, and a computer program stored in the memory. The processor is configured to execute the computer program to implement the steps of any of the method embodiments described above.

[0085] According to one aspect of this disclosure, a non-transitory computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the method embodiments described above.

[0086] According to one aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the steps of any of the method embodiments described above.

[0087] In the following text, combined with Figure 4 Illustrative examples describing such computer devices, non-transitory computer-readable storage media, and computer program products.

[0088] Figure 4 An example configuration of a computer device 400 that can be used to implement the methods described herein is shown. For example, Figure 1 The server 120 and / or client device 110 shown may include an architecture similar to computer device 400. The aforementioned device / apparatus for determining fractional blood flow reserve may also be implemented wholly or at least partially by computer device 400 or similar device or system.

[0089] Computer device 400 can be a variety of different types of devices, such as a service provider's server, a device associated with a client (e.g., a client device), a system-on-a-chip, and / or any other suitable computer device or computing system. Examples of computer device 400 include, but are not limited to: desktop computers, server computers, laptop or netbook computers, mobile devices (e.g., tablets, cellular or other wireless phones (e.g., smartphones), notebook computers, mobile stations), wearable devices (e.g., glasses, watches), entertainment devices (e.g., entertainment appliances, set-top boxes communicatively coupled to a display device, game consoles), televisions or other display devices, automotive computers, and so on. Therefore, the range of computer device 400 can be from full-resource devices with large amounts of memory and processor resources (e.g., personal computers, game consoles) to low-resource devices with limited memory and / or processing resources (e.g., traditional set-top boxes, handheld game consoles).

[0090] Computer device 400 may include at least one processor 402, memory 404, multiple communication interfaces 406, display device 408, other input / output (I / O) devices 410, and one or more mass storage devices 412 capable of communicating with each other, such as via system bus 414 or other suitable connections.

[0091] Processor 402 may be a single processing unit or multiple processing units, and all processing units may include single or multiple computing units or multiple cores. Processor 402 may be implemented as one or more microprocessors, microcomputers, microcontrollers, digital signal processors, central processing units, state machines, logic circuits, and / or any device that manipulates signals based on operating instructions. Among other capabilities, processor 402 may be configured to acquire and execute computer-readable instructions stored in memory 404, mass storage device 412, or other computer-readable media, such as program code of operating system 416, program code of application program 418, program code of other program 420, etc.

[0092] Memory 404 and mass storage device 412 are examples of computer-readable storage media for storing instructions executed by processor 402 to perform the various functions described above. For example, memory 404 may generally include both volatile and non-volatile memory (e.g., RAM, ROM, etc.). Furthermore, mass storage device 412 may generally include hard disk drives, solid-state drives, removable media, including external and removable drives, memory cards, flash memory, floppy disks, optical disks (e.g., CDs, DVDs), storage arrays, network-attached storage, storage area networks, etc. Both memory 404 and mass storage device 412 may be collectively referred to herein as memory or computer-readable storage media, and may be non-transitory media capable of storing computer-readable, processor-executable program instructions as computer program code, which may be executed by processor 402 as a specific machine configured to perform the operations and functions described in the examples herein.

[0093] Multiple program modules may be stored on mass storage device 412. These programs include operating system 416, one or more application programs 418, other programs 420, and program data 422, and they may be loaded into memory 404 for execution. Examples of such application programs or program modules may include, for example, computer program logic (e.g., computer program code or instructions) for implementing components / functions including method 200 (including any suitable steps of method 200) and / or other embodiments described herein.

[0094] Although Figure 4The modules 416, 418, 420, and 422, or portions thereof, are illustrated as being stored in memory 404 of computer device 400; however, modules 416, 418, 420, and 422 may be implemented using any form of computer-readable medium accessible by computer device 400. As used herein, “computer-readable medium” includes at least two types of computer-readable media: computer storage media and communication media.

[0095] Computer storage media includes volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, DVD, or other optical storage devices, magnetic cassettes, magnetic tapes, disk storage devices or other magnetic storage devices, or any other non-transfer medium that can be used to store information for access by computer equipment.

[0096] In contrast, communication media can embody computer-readable instructions, data structures, program modules, or other data within modulated data signals such as carrier waves or other transmission mechanisms. Computer storage media as defined herein do not include communication media.

[0097] Computer device 400 may also include one or more communication interfaces 406 for exchanging data with other devices, such as via a network, direct connection, etc., as discussed above. Such communication interfaces can be one or more of the following: any type of network interface (e.g., a network interface card (NIC)), wired or wireless (such as IEEE 802.11 Wireless LAN (WLAN)) wireless interface, Wi-MAX interface, Ethernet interface, Universal Serial Bus (USB) interface, cellular network interface, Bluetooth. TM Interfaces, near field communication (NFC) interfaces, etc. Communication interface 406 can facilitate communication across various network and protocol types, including wired networks (e.g., LAN, cable, etc.) and wireless networks (e.g., WLAN, cellular, satellite, etc.), the Internet, etc. Communication interface 406 can also provide communication with external storage devices (not shown) such as storage arrays, network-attached storage, storage area networks, etc.

[0098] In some examples, a display device 408, such as a monitor, may be included for displaying information and images to the user. Other I / O devices 410 may be devices that receive various inputs from the user and provide various outputs to the user, and may include touch input devices, gesture input devices, cameras, keyboards, remote controls, mice, printers, audio input / output devices, and so on.

[0099] Although this disclosure has been described and illustrated in detail in the accompanying drawings and the foregoing description, such description and illustration should be considered illustrative and suggestive, not restrictive; this disclosure is not limited to the disclosed embodiments. By studying the drawings, the disclosure, and the appended claims, those skilled in the art will be able to understand and implement variations of the disclosed embodiments in practicing the claimed subject matter. In the claims, the word "comprising" does not exclude other elements or steps not listed, and the words "a" or "an" do not exclude a plurality. The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be beneficial.

Claims

1. A method for determining fractional flow reserve, comprising: Acquire the segmentation data of the target blood vessel and the segmentation data of the lesion; A feature extraction network is used to determine fusion features based on the blood vessel segmentation data and the lesion segmentation data; as well as Based on the fusion features, the fractional flow reserve (FFR) value at at least one target location in the target blood vessel is determined by a prediction network.

2. The method according to claim 1, wherein, Determining fusion features based on the blood vessel segmentation data and the lesion segmentation data using a feature extraction network includes: extracting a first feature based on the blood vessel segmentation data and extracting a second feature based on the lesion segmentation data respectively using the feature extraction network; and fusing the first feature and the second feature to obtain fusion features.

3. The method according to claim 2, wherein, The extraction of a first feature based on the blood vessel segmentation data and a second feature based on the lesion segmentation data via the feature extraction network includes extracting the first feature through a first sub-network of the feature extraction network and extracting the second feature through a second sub-network of the feature extraction network.

4. The method according to claim 2, wherein, The extraction of the first feature based on the blood vessel segmentation data and the extraction of the second feature based on the lesion segmentation data are performed through an attention mechanism.

5. The method according to any one of claims 1-4, wherein, The blood vessel segmentation data and the lesion segmentation data are input into the feature extraction network through different channels.

6. The method according to any one of claims 1-5, wherein, Determining the FFR value at at least one target location in the target blood vessel via a prediction network includes obtaining the FFR value based on a sequence of centerline sampling points labeled at the at least one target location using the prediction network.

7. The method according to claim 6, wherein the centerline sampling point sequence is sparse sampling or variable step size sampling.

8. The method according to claim 6 or 7, wherein, The sequence of centerline sampling points covers the complete topology tree of the target blood vessel.

9. The method according to claim 6 or 7, wherein, The centerline sampling point sequence corresponds to at least one predetermined vascular segment of the target blood vessel.

10. The method according to any one of claims 1-9, wherein, The feature extraction network and the prediction network are trained end-to-end.

11. The method according to any one of claims 1-10, further comprising: Before determining the fusion features, at least one of the blood vessel segmentation data and the lesion segmentation data is normalized.

12. An apparatus for determining fractional flow reserve, comprising: The data acquisition unit is used to acquire the vessel segmentation data and lesion segmentation data of the target blood vessel; The feature extraction unit is used to determine fusion features based on the blood vessel segmentation data and the lesion segmentation data through a feature extraction network; as well as A prediction unit is used to determine the FFR value at at least one target location in the target blood vessel based on the fusion features via a prediction network.

13. A computing device, comprising: Memory, processor, and computer program stored on said memory, The processor is configured to execute the computer program to implement the steps of the method according to any one of claims 1-11.

14. A non-transitory computer-readable storage medium having a computer program stored thereon, wherein, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-11.

15. A computer program product comprising a computer program, wherein, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-11.

Images