Determination of values ​​of physical properties of thrombus

Abstract

A computer-implemented method for determining a value of a physical property of a thrombus is provided, the method comprising the steps of receiving x-ray image data having a temporal sequence of x-ray images representative of an expansion of a stent of an endoluminal stent retriever device on the thrombus, determining a value of the physical property of the thrombus based on an expansion rate of the stent in the temporal sequence of x-ray images, and outputting the value of the physical property. Further provided is a system having one or more processors configured to perform the steps of such a method.

Description

[Technical field]

[0001] SUMMARY The present disclosure relates to determining the value of a physical property of a thrombus. A computer-implemented method, computer program product, and system are disclosed. [Background technology]

[0002] A thrombus or blood clot is an obstruction of a blood vessel. A thrombus can occur in a vein or in an artery. In the former case, venous thrombosis, blood pools, resulting in swelling and fluid pooling. In the latter case, arterial thrombosis, blood supply is restricted, resulting in a condition known as ischemia, risking damage to tissues supplied by the artery. In both cases, a part of the thrombus can break off as an embolus. The embolus can lodge elsewhere in the body, forming an embolus that blocks the blood vessel as well. A thrombus can occur in various parts of the body, including the heart and the brain, and its effects can be severe unless treated promptly. In the brain, for example, a thrombus or embolism can cause conditions such as (ischemic) stroke.

[0003] A variety of endoluminal therapeutic devices are available to treat thrombosis. These include mechanical thrombectomy devices, which are used to remove thrombi from blood vessels. Currently, there are two main groups of mechanical thrombectomy devices: aspiration catheters and stent retrievers. Aspiration catheters typically include a delivery catheter, which is used to deliver irrigation fluid to the thrombus, and an extraction catheter, which is used to extract the irrigation fluid, along with the thrombus debris. In use, the aspiration catheter is placed near the clot and aspiration is applied at that location, resulting in the clot debris being extracted from the body. Stent retrievers typically include an expandable wire mesh tube, or stent, which is designed to remove the clot in one piece. Prior to its deployment, the wire mesh tube is housed within the delivery catheter. The delivery catheter is placed so that it overlaps the clot, and the delivery catheter is then withdrawn so that the wire mesh tube expands out of the distal end of the delivery catheter. As the wire mesh tube expands, it captures the clot. The wire mesh tube is then retracted into a delivery catheter and removed from the body.

[0004] Stent retriever "SR" thrombectomy is currently the first-line treatment for acute stroke resulting from intracranial large vessel occlusion "LVO" in the anterior cerebral circulation. However, stent retriever thrombectomy can fail for a variety of reasons. These include anatomical challenges (e.g., tortuous arterial tree from the aortic arch to the target occlusion site), large volume of clots, tandem occlusion, clot characteristics (e.g., fresh clot vs. organized clot), and different pathomechanisms (embolic vs. nonembolic occlusion). The pathomechanism of acute LVO (embolic vs. nonembolic) or the physical properties of the clot (soft vs. hard / organized) can play an important role in the response to stent retrievers. Organized (hard, fibrin-rich) clots are more elastic and less sticky than fresh (soft, red blood cell-rich) clots, which results in less engagement with the stent retriever and can lead to loss of some or all of the clot during retrieval by the stent retriever, especially in the case of tortuous arterial trees.

[0005] Furthermore, as described, for example, in the article by Choi Dae-Han, "Gradual expansion of stent retriever in mechanical thrombectomy for curved middle cerebral artery ...," Acta Neurochirurgica (2019) 161:2003-2012, organized clots can cause tension in the stent retriever-deployed segment of the parent artery, which can induce arterial spasm. Such effects increase the likelihood of stent retriever breakage.

[0006] To successfully recanalize an LVO with an organized clot, the first and simplest option is the intra-arterial administration of a vasodilator, which releases tension on the arterial segment in which the stent retriever was deployed, thereby increasing the vessel diameter to facilitate clot retrieval. The second option is the intra-arterial administration of a clot dissolving agent, tissue plasminogen activator (tPA), to soften the clot before retrieval. Another option is the simultaneous application of a stent retriever and an aspiration or intermediate catheter. Stent retriever and aspiration thrombectomy can be performed sequentially or simultaneously. This helps to prevent the loss of the engaged clot as the stent retriever is retracted through a tortuous or stenosed segment of the arterial tree. If an organized clot still occludes the vessel after the simultaneous application of a stent retriever and intermediate catheter, permanent stent placement can be considered as the final action.

[0007] Another factor that increases the risk of stent retriever failure is the damage of the atheromatous surface by the stent when acute LVO is due to intracranial atherosclerotic stenosis (ICAS) thrombus occlusion. In this case, the use of a stent retriever may result in further platelet activation, resulting in repeated reocclusion in the future. To prevent repeated reocclusion in cases of LVO due to ICAS, inhibition of platelet function can play an important role. Therefore, the first option is to administer glycoprotein IIb / IIIa inhibitors, which inactivate platelets and thereby prevent repeated reocclusion, as discussed in the literature in Kim, BM. "Causes and Solutions of Endovascular Treatment Failure.", J. stroke, vol. 19, no. 2, pp. 131-142, 2017.

[0008] The current workflow in stent retriever thrombectomy procedures is to retrieve the stent between 3 and 8 minutes after stent deployment. However, the chance of successful clot retrieval in the first pass for hard / organized clots is lower than that for soft clots. Some physicians may perform multiple stent retriever thrombectomy procedures to retrieve the entire clot. Several alternative workflows, such as those mentioned above, have also been proposed to treat organized clots. However, a literature article by Zaidat, OO et al., "First pass effect: A new measure for stroke thrombectomy devices," Stroke, vol. 49, no. 3, pp. 660-666, 2018, reported a correlation between a stent retriever procedure completed in a single pass and positive outcomes. In addition, Ng, PP et al., "Intraprocedural predictors of post-stent retriever thrombectomy subarachnoid hemorrhage in middle cerebral artery stroke," J. Neurointerv. Surg., vol. 11, no. 2, pp. 123-126, 2019, reported that two or more rounds of SR can increase the risk of post-procedure subarachnoid hemorrhage (SAH). The same study shows that the risk of post-procedure SAH is higher in patients who had a stent placed more than 2 cm distal to the middle cerebral artery thrombus. Summary of the Invention

[0009] [Problem to be solved by the invention]

[0010] Thus, from the above discussion, it can be seen that removal of thrombus by mechanical thrombectomy is a challenging procedure and there is a need for improved support for physicians in such procedures.

[0011] [Means for solving the problem]

[0012] According to one aspect of the present disclosure, a computer-implemented method for determining a value of a physical property of a thrombus is provided, the method comprising: receiving x-ray image data having a temporal sequence of x-ray images representative of the expansion of a stent of an endoluminal stent retriever device over the thrombus; determining a value of a physical property of the thrombus based on a rate of expansion of the stent in a time sequence of X-ray images; outputting the value of the physical property; has.

[0013] The inventors have observed that the expansion rate of a stent on a thrombus indicates various physical properties of the thrombus, including the dimensions of the thrombus, e.g., length, diameter, and volume, and the composition of the thrombus, e.g., soft vs. hard, or organized. These physical properties influence the choice made by the physician regarding the method of extracting the thrombus. Thus, by using the above method, the physical properties of the thrombus can be determined during the stent retriever procedure and can then be used by the physician to determine the course of action in extracting the thrombus. Thus, the physician can be better assisted in performing such procedures or interventions.

[0014] Further aspects, features, and advantages of the present disclosure will become apparent from the following description of examples which proceeds with reference to the accompanying drawings. [Brief description of the drawings]

[0015] [Figure 1] 1 is a flow chart illustrating an example of a method for determining a value of a physical property of a thrombus, according to some aspects of the present disclosure. [Diagram 2] FIG. 2 is a schematic diagram illustrating an example of a system 200 for determining a value of a physical property of a thrombus, according to some embodiments of the present disclosure. [Diagram 3]1A-1C are schematic diagrams illustrating examples of various stages in the expansion of a stent 130 of an endoluminal stent retriever device over a thrombus 110, according to some embodiments of the present disclosure. [Figure 4] 1A-1C are schematic diagrams illustrating an example of A) training a neural network 140 to generate a latent space encoding of a temporal sequence of X-ray images, and B) performing inference using the trained neural network 140 to generate a latent space encoding of a temporal sequence of X-ray images 120, in accordance with some embodiments of the present disclosure. [Diagram 5] 1 is a schematic diagram illustrating an example of A) training a neural network 140 to generate a latent space encoding of a temporal sequence of X-ray images, and B) performing inference using the trained neural network 140 to generate a latent space encoding of a temporal sequence of X-ray images 120, where the training data also includes workflow data, in accordance with some embodiments of the present disclosure. [Figure 6] 1A-1C are schematic diagrams illustrating an example of A) training a neural network 140 to predict a shape of a stent from a temporal sequence of X-ray images, and B) performing inference by a trained neural network 140 to predict a shape of a stent from a temporal sequence of X-ray images, according to some embodiments of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] Examples of the present disclosure are provided with reference to the following description and drawings. In this description, for purposes of explanation, many specific details of particular examples are described. Reference to an "example," "embodiment," or similar language means that the characteristics, structures, or features described in connection with that example are included in at least that one example. It should also be understood that features described in connection with one example may also be used in another example, and that not all features are necessarily repeated in each example for the sake of brevity. For example, features described in connection with a computer-implemented method may be implemented in a computer program product and in a system in a corresponding manner.

[0017] In the following description, reference is made to a computer-implemented method that includes determining a value of a physical property of a thrombus. Reference is made to an example in which a thrombus is located in a blood vessel. In general, the thrombus may be located in a vein or in an artery. In other words, the thrombus may be a venous thrombus or alternatively an arterial thrombus. It should be understood that the thrombus may generally be located in any part of the body. For example, the thrombus may be located in the brain, the heart, the lungs, or in a limb, such as a leg. In some examples, the thrombus may be located in the brain and the purpose of the treatment procedure may be for the treatment of an (ischemic) stroke.

[0018] It should be noted that the computer-implemented methods disclosed herein may be provided as a non-transitory computer-readable storage medium storing computer-readable instructions that, when executed by at least one processor, cause the at least one processor to perform the method. In other words, the computer-implemented methods may be implemented in a computer program product. The computer program product may be provided by dedicated hardware, or by hardware capable of executing software in association with appropriate software. When provided by a processor, the functions of the method features may be provided by a single dedicated processor, or by a single shared processor, or by multiple individual processors, some of which may be shared. One or more functions of the method features may be provided by a processor shared within a networked processing architecture, such as, for example, a client / server architecture, a peer-to-peer architecture, the Internet, or a cloud.

[0019] Explicit use of the terms "processor" or "controller" should not be construed as exclusively referring to hardware capable of executing software, but may implicitly include, but is not limited to, digital signal processor "DSP" hardware, read only memory "ROM" for storing software, random access memory "RAM", non-volatile storage, and the like. Furthermore, examples of the present disclosure may take the form of a computer usable storage medium, or a computer program product accessible from a computer readable storage medium, the computer program product providing program code for use by or in connection with a computer or any instruction execution system. For purposes of this description, a computer usable storage medium or a computer readable storage medium may be any device that can have, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. The medium may be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system or device or a propagation medium. Examples of a computer-readable medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory "RAM", a read-only memory "ROM", a rigid magnetic disk, and an optical disk. Current examples of optical disks include Compact Disk-Read Only Memory "CD-ROM", Compact Disk-Read / Write "CD-R / W", Blu-Ray, and DVD.

[0020] As mentioned above, removal of a thrombus by mechanical thrombectomy is a difficult procedure, and knowledge of the physical characteristics of the thrombus is useful in helping the physician determine the course of action in its removal.

[0021] 1 is a flow chart illustrating an example of a method for determining a value of a physical property of a thrombus according to some embodiments of the present disclosure. receiving X-ray image data S110 having a temporal sequence of X-ray images 120 representing the expansion of a stent 130 of an intraluminal stent retriever device over a thrombus 110; determining a value of a physical property of the thrombus 110 based on the expansion rate of the stent 130 in a time sequence of X-ray images 120; A step S130 of outputting the value of the physical property; has.

[0022] The inventors have observed that the expansion rate of a stent on a thrombus is indicative of various physical properties of the thrombus, including the dimensions of the thrombus, e.g., length, diameter, and volume, and the composition of the thrombus, e.g., soft vs. hard, or organized. These physical properties influence the choice made by the physician regarding the method of extracting the thrombus. Thus, by using the above method, the physical properties of the thrombus can be determined during the stent retriever procedure and can then be used by the physician to determine a course of action in extracting the thrombus.

[0023] The method described above with reference to Figure 1 may be implemented by the system shown in Figure 2, which is a schematic diagram illustrating an example of a system 200 for determining a value of a physical property of a thrombus, according to some embodiments of the present disclosure. Thus, the operations described in connection with the flowchart shown in Figure 1 may also be performed by the system 200 shown in Figure 2, and vice versa.

[0024] 1, in act S110, x-ray image data is received. The x-ray image data includes a temporal sequence of x-ray images 120 depicting the expansion of a stent 130 of an endoluminal stent retriever device over a thrombus 110.

[0025] In general, the X-ray image data received in operation S110 may be generated by a projection X-ray imaging system. A projection X-ray imaging system typically includes a support arm, such as a so-called "C-arm" or "O-arm", that supports an X-ray source-detector arrangement. A projection X-ray imaging system may alternatively include a support arm having a different shape than these examples. A projection X-ray imaging system typically generates the projection X-ray images with the support arm held in a stationary position relative to the imaging area during acquisition of the image data. The temporal sequence of X-ray images 120 received in operation S110 may be acquired periodically, i.e. at regular intervals, or intermittently, i.e. at irregular intervals. The temporal sequence of X-ray images 120 may be fluoroscopic, i.e. live images. This allows operation S120, which determines the value of the physical property of the thrombus 110, to be performed in real time. The X-ray image data may be generated, for example, by a projection X-ray imager 220 shown in FIG. 2. As an example, the x-ray image data may be generated by a Philips Azurion 7 x-ray imaging system commercially available from Philips Healthcare of Best, The Netherlands.

[0026] The x-ray image data received in operation S110 includes a temporal sequence 120 of x-ray images representing the expansion of a stent 130 of an endoluminal stent retriever device over the thrombus 110. In this regard, a temporal sequence may be generated during a stent retriever procedure to extract a thrombus from a vein. As described above, a stent retriever typically includes an expandable wire mesh tube, or stent, that is designed to remove a clot in one piece. Prior to its deployment, the wire mesh tube is housed within a delivery catheter. The delivery catheter is positioned such that it overlaps the clot, and then the delivery catheter is withdrawn such that the wire mesh tube expands out of the distal end of the delivery catheter. As the wire mesh tube expands, it captures the clot. The wire mesh tube is then retracted into the delivery catheter and removed from the body. The temporal sequence of x-ray images 120 represents the stent and thrombus during at least some of the time periods during which the wire mesh tube or stent is expanded.

[0027] The X-ray image data received in operation S110 may be received, for example, from a projection X-ray imaging system, such as the projection X-ray imager 220 shown in FIG. 2, or from a computer-readable storage medium, or from the Internet or the cloud. The X-ray image data may be received by one or more processors 210 shown in FIG. 2. The X-ray image data may be received via any form of data communication, including wired communication, optical communication, and wireless communication. As some examples, when wired or optical communication is used, the communication may be via signals transmitted over electrical or optical cables, and when wireless communication is used, the communication may be via, for example, RF or optical signals.

[0028] In act S120, a value of a physical property of the thrombus 110 is determined based on the expansion rate of the stent 130 in the time sequence of X-ray images 120. In this regard, the physical property may be, for example, a dimension of the thrombus, e.g., a length or diameter, or a volume, or a composition of the thrombus, e.g., whether it is soft, hard, organized, etc.

[0029] FIG. 3 is a schematic diagram showing an example of various stages in the expansion of a stent 130 of an intraluminal stent retriever device on a thrombus 110 according to some embodiments of the present disclosure. View A of FIG. 3 represents a thrombus prior to introduction of a stent retriever "SR" into a blood vessel. In view A, the thrombus 110 completely blocks the blood vessel. View B of FIG. 3 represents a thrombus and a stent of a stent retriever device immediately after the stent is deployed from the stent retriever. In view B, the stent is positioned to overlap the thrombus and the stent begins to expand onto the thrombus. The stent compresses the thrombus laterally to the vessel axis and the outer surface of the stent conforms to the outer surface of the thrombus. View C represents the stent and thrombus against a soft clot, approximately 3-5 minutes after the stent is deployed. In this example, the stent cuts through and completely encapsulates the soft clot. View D represents the stent and thrombus against a hard or organized clot, approximately 3-5 minutes after the stent is deployed. In this example, the stent has further compressed the thrombus laterally to the vessel axis compared to FIG. B, and the outer surface of the stent still conforms to the outer surface of the thrombus. After a further period of time, depending on the physical properties of the thrombus, such as its size and hardness, the stent for the hard clot may remain in the same or similar expanded state as in FIG. D, or may cut open some or all of the clot, thereby partially or completely encapsulating the clot. Thus, as can be seen particularly in FIG. C and D, a distinction between soft and hard clots may be made based on the expansion rate of the stent, which is higher for the soft clot in FIG. C than for the hard clot as in FIG. D.

[0030] The expansion rate of the stent also provides information about other physical properties of the clot besides hardness. For example, the length of the stent can be determined based on the expansion rate of the stent at multiple locations along its length over time. During the initial period of stent expansion, i.e., the time interval between the deployment of the stent and the stage represented by diagram B, the stent compresses the thrombus in a direction perpendicular to the axis of the vessel, and the outer surface of the stent conforms to the outer surface of the thrombus. During this initial period, the length of the thrombus can be determined based on the locations along the length of the stent where the expansion rate of the stent is significantly reduced. These locations may correspond to areas where the outer surface of the stent contacts the outer surface of the thrombus. Thus, the length of the clot can be determined based on the expansion rate of the stent along its length. Because the stent is more easily identifiable in an x-ray image than the thrombus, this technique may provide an accurate method of determining the length of the clot.

[0031] The diameter and volume of the thrombus may also be determined based on the expansion rate of the stent along its length. For example, a model representing the expansion rate of the stent along its length may be generated for thrombi having different diameters or volumes. The model may then be used to relate the diameter and volume of the thrombus to the expansion rate of the stent along its length. The model may also be used to generate a look-up table of values ​​of the expansion rate of the stent along its length for different values ​​of diameter or volume. In use, the diameter and volume of the thrombus may be determined from the look-up table using the detected value of the expansion rate of the stent along its length.

[0032] The expansion rate of the stent 130 may be determined in operation S120 using a variety of techniques. For example, image analysis techniques may be used to determine the expansion rate of the stent at one or more locations along its length in a temporal sequence of X-ray images, and the physical properties of the thrombus may be determined from the expansion rate as described above. For example, by registering the images to each other using image intensity or the ends of the stent, changes in the shape of the stent in the images may be determined, and these changes are used to calculate the expansion rate using the time interval between images. The expansion rate can then be used to determine the value of the physical property of the thrombus.

[0033] In another example, a neural network is used to determine the value of the physical property of the thrombus 110 in operation S120, as described in more detail below. In particular, in a particular example, the determining step includes a neural network trained using a training sequence of X-ray images, each representing the expansion of a stent on a thrombus having known value(s) of one or more physical properties of the thrombus. It has thus been found that the neural network can be trained to learn the physical properties or derived values ​​of the physical properties from such X-ray image sequences. During inference, when the trained network is used with a live X-ray image sequence of the thrombus under examination as an input, correspondingly, one or more current values ​​of the physical properties of the thrombus under examination can be provided as an output of the network.

[0034] In act S130, the value of the physical property is output. The value of the physical property may be output in a variety of ways, including to a computer readable storage medium or a display device such as monitor 240 shown in Figure 2. By outputting the value of the physical property, a physician may use this information to determine a course of action in treating the thrombus.

[0035] As mentioned above, in some examples, a neural network is used to determine values ​​of physical properties of the thrombus 110 in operation S120 and provide such values ​​as outputs of the network. Examples of neural networks for this purpose are described with reference to Figures 4 and 5.

[0036] FIG. 4 is a schematic diagram illustrating an example of A) training a neural network 140 to generate a latent space encoding of a temporal sequence of X-ray images, and B) performing inference using the neural network 140 trained to generate a latent space encoding of a temporal sequence of X-ray images 120, according to some aspects of the disclosure. The exemplary neural network 140 shown in FIG. 4 has a variational autoencoder architecture. The neural network 140 may alternatively have another architecture, such as a convolutional neural network "CNN" architecture, or a recurrent neural network "RNN" architecture with a unidirectional or bidirectional long-short-term memory "LSTM" architecture. With reference to FIG. 4A, during training, training X-ray images 120 depicting the expansion of a stent of an endoluminal stent retriever device over a thrombus are generated. T A temporal sequence of x-ray images 120 input to the variational autoencoder neural network 140 is trained to generate a latent space encoding z that represents the input temporal sequence. The variational autoencoder derives a latent space encoding z from the training x-ray images 120 input to the encoder. T The variational autoencoder is trained by forcing its decoder to reproduce the temporal sequence of training X-ray images 120. A plurality of temporal sequences of training X-ray images are used to train the neural network 140. During training, the variational autoencoder also T Distribution of latent space coding of temporal sequences of 160 T This distribution is shown in FIG. 4 as a normal distribution 160 with mean m and variance s. TThe training data used to train the neural network 140 is selected to represent thrombi having physical properties with known values. The training data may represent, for example, soft clots. In doing so, the neural network 140 may generate a latent space encoding z of the input temporal sequence of images, as well as a distribution 160 of the latent space encodings, for thrombi having known values ​​of physical properties, for example, for soft thrombi. T It learns to generate

[0037] 4B, in inference, a new temporal sequence of X-ray images 120 is input to a trained neural network 140. The trained neural network 140 generates a latent space encoding 150, z, of the input image in each new sequence. The position of the latent space encoding 150 with respect to the new temporal sequence is then multiplied by the position of the training X-ray images 120. T Distribution of latent space coding generated from temporal sequences of 160 T The clots depicted in the new temporal sequence of X-ray images 120 are determined with respect to the training X-ray images 120. T If the blood clots in the time sequence are of a similar type, for example if the blood clots are soft clots, then the latent space encoding 150 will T Distribution of latent space coding generated from temporal sequences of 160 T In other words, the latent space coding 150 is close to the mean of the distribution 160 T In contrast, a blood clot depicted in the new temporal sequence of X-ray images 120 is an inlier of the training X-ray images 120. T If the clot is a different type from the clot in the time sequence of x-ray images 120, e.g., if the clot is a hard clot, the latent space encoding 150 may T Distribution of latent space coding generated from temporal sequences of 160 T In other words, the latent space coding 150 is TIn the example shown in FIG. 4, the latent space coding 150 generates a distribution 160 using the time sequence for the soft clot. T , and it is therefore concluded that the clot in the new temporal sequence of X-ray images 120 is a hard clot.

[0038] More generally, the value of the physical property of the thrombus 110 may be determined based on training x-ray images 120 that represent the expansion of a stent over a thrombus having a known value of the physical property. T Distribution of latent space coding generated from temporal sequences of 160 T A latent space encoding 150 of the input image for x, z, can be determined using the neural network shown in FIG. 4 based on the location of z.

[0039] Note that in the above example, the decoder portion of the neural network that reconstructs the input image during training, i.e., the portion on the right side of FIG. 4A and FIG. 4B, is not required to perform inference. This is because in this example, the value of the physical property of the thrombus 110 is determined using the latent space encoding 150,z of the input image, which does not depend on the presence of a reconstructed image. In this example, the decoder portion is used simply to train the neural network. In another example, described below, the decoder portion of the neural network is used during inference, and the neural network determines the value of the physical property of the thrombus 110 based on the difference DE between the predicted shape of the stent and the actual shape of the stent in the input image.

[0040] Note that in the above example, "clot hardness" was used as an example of a physical property of a clot for which training data was selected, with the known value being "soft." However, alternatively, or in addition, neural network 140 may be trained using training data representing other known values ​​of the physical property of a clot, and during inference, neural network 140 may also be used to determine how close the latent space encoding 150 of a new temporal sequence of x-ray images 120 is to the mean value of the distributions for these other known values ​​of the physical property, by determining the z-positions for each of these distributions. In doing so, neural network 140 may be trained and used to determine values ​​of other physical properties of clot 110.

[0041] Thus, with reference to the method illustrated in FIG. 1, in one example, operation S120 of determining a value of a physical property of the thrombus 110 includes: inputting a plurality of images from a temporal sequence into a neural network 140; generating a latent space encoding 150, z, of the input image using a neural network 140; and A distribution of latent space codings 160 generated from a temporal sequence of training x-ray images 120 representing the expansion of a stent over a thrombus with known values ​​for physical properties T determining a value of the physical property of the latent space encoding 110 of the input image based on the z position of the latent space encoding 150 of the input image; It may include, The neural network 140 uses the training x-ray images 120 T is trained to generate a latent space encoding 150 of an input image, z, using training data having a temporal sequence of Each sequence represents the expansion of a stent of an endoluminal stent retriever device over a thrombus having a known value for a physical property.

[0042] In the above method, 120 training X-ray images TDistribution of latent space coding generated from temporal sequences of 160 T The latent space encoding 150 of the input image with respect to z, the location of the z, can be determined by calculating the distance between the location of the z, of the input image and the mean or center of gravity of the distribution. This distance can be measured using a function such as, for example, Euclidean distance or geodesic distance. This distance provides an analog measure of how close the physical property of the thrombus represented in the input image is to a known value of the physical property. This measure may be expressed on a continuous scale as a ratio or percentage representing how close the physical property of the thrombus represented in the input image is to a known value of the physical property. Alternatively, a binary classification of the physical property, e.g., hard vs. soft composition, may be provided by digitizing this distance using a threshold. The threshold may be, for example, a function that is used to determine the distribution of the latent space encoding 160 T For example, if the distance is less than 1 s for the new temporal sequence, the new temporal sequence may be classified with the known value of the physical property from the training data.

[0043] The neural network described with reference to FIG.

[0044] Training X-ray images 120 T receiving training data including a plurality of time sequences of a physical characteristic of the endoluminal stent retriever device, each time sequence representing an expansion of a stent of the endoluminal stent retriever device over a thrombus having a known value for the physical characteristic; Input training data to the neural network 140, and For each of the input multiple training X-ray images in the temporal sequence, Using a neural network 140 to generate a latent space encoding z of the input training x-ray images; Utilizing a neural network 140 to reconstruct the input training X-ray images from the latent space encoding z; Adjusting parameters of the neural network 140 based on the difference between the input training x-ray image and the reconstructed input training x-ray image; Repeat the generation, reconstruction, and refinement until a stopping criterion is met. The input image may be trained to generate a latent space encoding 150,z by

[0045] Generally, training a neural network involves inputting a training data set into the neural network and iteratively adjusting the parameters of the neural network until the trained neural network provides accurate outputs. Training is often performed using dedicated neural processors such as graphics processing units "GPUs" or neural processing units "NPUs" or tensor processing units "TPUs". Training often employs a centralized approach, where cloud- or mainframe-based neural processors are used to train the neural network. Following training with the training data set, the trained neural network may be deployed to a device for analyzing new input data during inference. The processing requirements during inference are significantly lower than those required during training, allowing the neural network to be deployed to a variety of systems such as laptop computers, tablets, mobile phones, etc. Inference may be performed, for example, by a central processing unit "CPU", GPU, NPU, TPU, on a server, or in the cloud.

[0046] Thus, the process of training the neural network 140 described above involves adjusting its parameters. The parameters, more specifically the weights and biases, control the operation of the activation functions in the neural network. In supervised learning, the training process automatically adjusts the weights and biases so that when presented with input data, the neural network accurately provides the corresponding expected output data. To do this, a loss function or error value is calculated based on the difference between the predicted output data and the expected output data. The loss function value may be calculated using functions such as the negative log-likelihood loss, the mean absolute error (or L1 norm), the mean squared error, the root mean squared error (or L2 norm), the Huber loss, or the (binary) cross-entropy loss. Other loss functions, such as the Kullback-Leibler divergence, are used to calculate the average number of training x-ray images 120. T Distribution of latent space coding generated from temporal sequences of 160 T It may further be used when training the variational autoencoder to ensure that σ resembles a standard Gaussian distribution with mean 0 and standard deviation 1. During training, the value of the loss function is typically minimized, and training is terminated when the value of the loss function meets a stopping criterion. In some cases, training is terminated when the value of the loss function meets one or more of several criteria.

[0047] Various methods are known for solving loss minimization problems, such as gradient descent and quasi-Newton methods. Various algorithms have been developed to implement these methods and their variations, including, but not limited to, stochastic gradient descent (SGD), batch gradient descent, mini-batch gradient descent, Gauss-Newton, Levenberg-Marquardt, momentum methods, Adam, Nadam, Adagrad, Adadelta, RMSProp, and Adamax "optimizers". These algorithms use chaining to calculate the derivatives of the loss function with respect to the model parameters. This process is called backpropagation because the derivatives are calculated starting at the last layer or output layer and moving towards the first layer or input layer. These derivatives inform the algorithm how the model parameters must be adjusted to minimize the error function. That is, adjustments to the model parameters are made starting at the output layer and working backwards in the network until the input layer is reached. In the first training iteration, the initial weights and biases are often randomized. The neural network then predicts output data, which is also random. Backpropagation is then used to adjust the weights and biases. The training process is performed iteratively by adjusting the weights and biases at each iteration. Training is terminated when the error, or difference between the predicted output data and the expected output data, is within an acceptable range for the training or validation data. The neural network can then be deployed, and the trained neural network makes predictions for new input data using the trained values ​​of its parameters. If the training process is successful, the trained neural network accurately predicts the expected output data from the new input data.

[0048] In general, the training data used in the training method described above with reference to FIG. 4 may be extracted from procedure records of previous mechanical thrombectomy procedures in which an endoluminal stent retriever device was used. TA temporal sequence of x-ray images typically forms part of these treatment records. Training x-ray images 120 from tens, hundreds, or thousands, or more, of such treatments. T The temporal sequence of may be used to train the neural network 140. The training data may represent patients with different ages, genders, body mass indexes, etc. The known physical properties of the thrombus represented in the training data may be known values ​​of the dimensions of the thrombus, or known composition of the thrombus, as described above. The dimensions of the thrombus may be known by calculating its value from training x-ray images using a magnification determined based on the geometric configuration of the x-ray source, the x-ray detector, and the relative positions of the x-ray source and the x-ray detector relative to the thrombus, or alternatively may be determined by measuring the extracted clot during post-procedure histological analysis. Such information is typically stored in the procedure record. The composition of the thrombus may similarly be determined from histological analysis of the extracted clot, which information is also typically stored in the procedure record. Alternatively, the physical properties of the thrombus may be determined from image data, such as magnetic resonance image data representative of the thrombus, or from CT image data representative of the thrombus, or from intravascular ultrasound "IVUS" image data representative of the thrombus.

[0049] Alternatively, the neural network 140 may be trained using training data in which thrombi are represented with known values ​​of other physical properties, so that the neural network can be used to determine how close the physical properties of a thrombi in a new time sequence are to other types of physical properties other than its size or its composition, as described in the examples above.

[0050] In one example, the neural network described with reference to FIG. 4 is trained to emphasize more data within a defined region of interest than outside the region of interest. In this example, the region of interest includes a stent and a thrombus. The region of interest is defined in the training data, i.e., in the training x-ray images 120. T The time sequence of the training x-ray images 120 may be defined manually or automatically. TA region of interest may be manually defined in the training x-ray images 120 by providing user input via a user input device, such as a mouse, operating in combination with a graphical user interface "GUI" that displays a region of interest. The region of interest may be defined by the user using, for example, a bounding box. Alternatively, the region of interest may be determined automatically. For example, the region of interest may be determined using a trained neural network or feature detection algorithm to locate the region of interest in the training x-ray images 120. T Thus, in one example, the neural network 140 may further be defined as:

[0051] The system is trained to generate a latent space encoding 150,z of an input image by receiving an input indicating a region of interest in a temporal sequence of training x-ray images, the region of interest comprising a stent and a thrombus; Adjusting the parameters of the neural network based on the difference between the input training X-ray images and the reconstructed training X-ray images within the region of interest is performed by applying weightings to the difference, with a relatively high weighting being applied to portions of the temporal sequence of the training X-ray images within the region of interest and a relatively low weighting being applied to portions of the temporal sequence of the training X-ray images outside the region of interest.

[0052] By using the region of interest to train the neural network in this manner, the neural network may be trained to generate a latent space encoding that is more specific to the region of interest and thus may provide a more accurate determination of the value of the physical property of the thrombus 110.

[0053] As will be explained in the examples below, various additional data may also be used to train the neural network 140.

[0054] As outlined above, the process of extracting a thrombus is complex, and the workflow steps performed by the physician during its extraction depend in part on the physical properties of the thrombus. For example, organized (hard, fibrin-rich) clots are more elastic and less sticky than fresh (soft, red blood cell-rich) clots, which may lead to less engagement with the stent retriever and loss of part or all of the clot during retrieval by the stent retriever, especially in the case of tortuous arterial trees. As a result, the delay between the deployment and retrieval of the stent of the stent retriever device may vary depending on its composition. As another example, organized clots may cause tension in the stent retriever-deployed segment of the parent artery, which may induce arterial spasm. In this case, intra-arterial vasodilators may be administered, or a stent retriever and aspiration or intermediate catheters may be used. Permanent stent placement, for example using a balloon angioplasty device, may also be considered as a final action. Glycoprotein IIb / IIIa inhibitors may also be administered to inactivate platelets, thereby preventing repeated reocclusion of blood vessels.

[0055] Thus, physicians performing mechanical thrombectomy procedures are faced with a multitude of workflow step choices, such as "wait X minutes between stent deployment and retrieval," "inject vasodilator," "use stent retriever in combination with aspiration catheter / intermediate catheter," "insert permanent stent using balloon angioplasty procedure," "administer glycoprotein IIb / IIIa inhibitor," etc. Each of these steps may impact the success of the procedure. Other workflow steps that may also impact the success of the procedure and are also performed during mechanical thrombectomy procedures include steps such as "move distal end of stent retriever / aspiration catheter X centimeters from clot," "increase suction by aspiration catheter," and "generate 3D CT images to detect risk of SAH." These latter workflow steps may or may not depend on the physical characteristics of the clot. However, they may also impact the success of the procedure.

[0056] In some examples, the training data used to train the neural network described above with reference to FIG. 4 also includes the workflow steps that were used to successfully treat the thrombus in the temporal sequence of training x-ray images. This data may be extracted from treatment records of past mechanical thrombectomy procedures. In inference, these workflow steps are used to provide recommendations for subsequent workflow step(s) during the mechanical thrombectomy procedure.

[0057] In these examples, the training data used to train the neural network described with reference to FIG. T The method also includes workflow data for each temporal sequence of training X-ray images. The workflow data defines one or more subsequent workflow steps that are used to successfully treat the thrombus in the temporal sequence of training X-ray images. In inference, the method described with reference to FIG. Distribution of latent space encoding with common workflow data160 T determining the location of z, the latent space encoding 150 of the input image, relative to outputting the common workflow data to provide one or more recommended subsequent workflow steps for treating the thrombus; Also includes.

[0058] By providing such recommendations, physicians performing mechanical thrombectomy procedures are provided with additional guidance that can be used to improve the success of the procedure.

[0059] The above method of recommending workflow steps stems from the insight that if the latent spatial encoding of a sequence of images from the current procedure is similar to the latent spatial encoding of a sequence of images from the training data, then successful workflow steps taken in the procedure represented in the training data may also result in a similar level of success in the current procedure. The above method may be visualized as follows: T The temporal sequences of x-ray images 120 are each labeled with one or more subsequent steps that were used to successfully treat the thrombus in the sequence. T The latent space encodings of the temporal sequences of x-ray images form clusters in the latent space according to the subsequent steps that were used to successfully treat the thrombus in each sequence. For example, temporal sequences that share a common workflow step "wait 3 minutes between stent deployment and retrieval" may form one cluster in the latent space. Temporal sequences that share a common workflow step "wait 5 minutes between stent deployment and retrieval" may form another cluster in the latent space. In inference, the location of the latent space encodings of a new temporal sequence of x-ray images is determined for each of these clusters. If the latent space encodings of the new temporal sequence are close to the cluster "wait 5 minutes between stent deployment and retrieval", then this workflow step is output as a recommended subsequent workflow step for treating the thrombus. In contrast, if the latent space encodings of the new temporal sequence are close to the cluster "wait 3 minutes between stent deployment and retrieval", then this workflow step is output. The location of the latent space encodings for the new temporal sequence of images may be determined for the clusters by determining the distance between the centroids of the distributions of the latent space encodings that form the clusters. The distance may be calculated in the same way as described above, ie, by using functions such as Euclidean distance or geodesic distance.

[0060] A related example is described with reference to Fig. 5, which is a schematic diagram illustrating an example of A) training a neural network 140 to generate a latent spatial coding of a temporal sequence of X-ray images, and B) performing inference using the trained neural network 140 to generate a latent spatial coding of a temporal sequence of X-ray images 120, where the training data also includes workflow data, according to some aspects of the disclosure. Fig. 5 corresponds to Fig. 4, and additionally illustrates multiple distributions 160 of the latent spatial coding with common workflow data. T 0..n 5. Features in FIG. 5 that share the same labels as features in FIG. 4 perform the same functions as described with reference to FIG. 4. In the example shown in FIG. 5, common workflow step 170 T 0..n The latent space coding of a temporal sequence with T 0..n , where x is the number of workflow steps in the workflow step sequence 170 and 171. As an example, the workflow step “Typical Workflow” distribution may represent a cluster of latent space encodings for successful procedures where the workflow step “Inject vasodilator” was performed. The workflow step “Workflow Variant 1” distribution may represent a cluster of latent space encodings for successful procedures where a different workflow step “Use a stent retriever in combination with an aspiration catheter” was performed, for example. The workflow step “Workflow Variant 2” distribution may represent a cluster of latent space encodings for successful procedures where a different workflow step “Insert a permanent stent using a balloon angioplasty procedure” was performed, for example. In inference, the location of the latent space encodings for the new temporal sequence may be calculated based on the location of the corresponding workflow step 170. T 0..n Distribution 160 T 0..n Then, the distribution 160 with the closest centroid is determined. T 0..n The workflow step corresponding to is output as a recommended subsequent step.

[0061] Thus, in this example, the training data used to train the neural network 140 described with reference to FIG. T 0..n 160, a distribution of latent space encodings with common workflow data, the distribution 160 comprising a plurality of training X-ray image temporal sequences having T 0..n The operation of determining the location of z in the latent space encoding of the input image 150 is carried out by dividing each distribution 160 of the latent space encoding with the common workflow data. T 0..n Further, the operation of outputting the common workflow data to provide one or more recommended subsequent workflow steps for treating the thrombus 110 includes determining a location of the latent space encoding 150, z of the input image relative to a center of gravity of the distribution 160 having a center of gravity closest to the location of the latent space encoding 150, z of the input image. T 0..n The method includes outputting common workflow data for the

[0062] In these examples, the workflow steps may include a success metric that represents the probability of one or more subsequent workflow steps resulting in successful treatment of the thrombus 110. In the inference, the success metric for the output workflow step(s) may also be output. The success metric may also be calculated from the procedure records of past mechanical thrombectomy procedures. For example, the success metric may be calculated based on the distribution 160 with the closest centroid among all successful procedures in a neighborhood of the input latent space encoding 150 of the input image, z location. T 0..n The success metric may be estimated by calculating the percentage of successful actions in which the common workflow steps corresponding to the latent space encoding 150 of the input image were executed. Thus, if the z location of the input image is close to the centroid, the calculated success metric may be higher. T 0..nIf the input image is close to an edge of the latent space encoding 150 of the input image, the success metric may be lower because the neighborhood of the z location may overlap with a different distribution than the distribution with the closest centroid. Alternatively, the success metric may be output directly by the neural network 140 or may be based on the reconstruction error DE between the predicted image and the input image. By outputting the success metric, the method facilitates the physician to execute the recommended workflow steps with confidence in the expected outcome. Thus, in some examples, the workflow data may also include a success metric corresponding to one or more subsequent workflow steps, the success metric representing a probability of the one or more subsequent workflow steps resulting in a successful treatment of the thrombus 110, and the operation of outputting the common workflow data further includes outputting the corresponding success metric.

[0063] In some examples, the operation of determining the z-position of the latent space encoding 150 of the input image relative to the distribution 160 of latent space encodings with common workflow data may be performed using a neural network. T The operation of determining a location of the latent space encoding 150, z of the input image relative to the thrombus includes inputting the latent space encoding 150, z of the input image into a second neural network. The second neural network is trained to determine one or more subsequent workflow steps for successfully treating the thrombus based on the latent space encoding 150, z of the input image using training data having a plurality of temporal sequences of training x-ray images, each sequence representing the expansion of a stent of an endoluminal stent retriever device on a thrombus having a known value for a physical property, the training data having workflow data for each temporal sequence of training x-ray images, the workflow data defining one or more subsequent workflow steps used to successfully treat the thrombus in the temporal sequences of training x-ray images.

[0064] The distribution shown in Fig. 5T 0..n By using a neural network to perform this operation, we can obtain the latent space encoding of new temporal sequences and multiple distributions. T 0..n The task of determining the distance between the centroids of the two images is reduced to this second neural network. This second neural network can be trained in a similar manner as described above.

[0065] In some examples, the neural network described with reference to Figs. 4 and 5 may be trained using additional data in the form of patient data. By providing this patient data as a separate input to the neural network, the neural network may learn dependencies between this data and the generated latent space encoding 150. Patient data such as age, sex, etc. may be input to the neural network during training and used to train the neural network 140. During inference, the patient data may be similarly input to the trained neural network and used to generate the latent space encoding 150,z of the input image. The patient data may be extracted from health records of patients who have undergone a past mechanical thrombectomy procedure. Similarly, volumetric images representing the vasculature surrounding the thrombus may be available for the patient, and these volumetric images may also be used during training and inference. The volumetric images may include, for example, CT images, or MRI or MRA images of the vasculature. Thus, in some examples, the neural network 140 is trained to generate the latent space encoding 150,z of the input image further based on the patient data, and the method described with reference to Fig. 4 may further include: receiving patient data relating to a thrombus 110; inputting patient data into a neural network; generating a latent space encoding 150z of the input image using a neural network further based on the input patient data; Including, The patient data includes one or more of: a volumetric image representative of the vasculature surrounding the thrombus 110; electronic health record data relating to the thrombus 110;

[0066] In some examples, the neural network described with reference to Figures 4 and 5 may be trained using additional data in the form of extracted stent data and / or thrombus data. The stent data and / or thrombus data may be extracted from a temporal sequence of X-ray images. By providing this data as a separate input to the neural network, similar to the additional patient data, the neural network may learn dependencies between this data and the generated latent space encoding 150. Thus, in some examples, the neural network 140 is trained to generate the latent space encoding 150, z, of the input image further based on the stent data and / or thrombus data, and the method described with reference to Figure 4 further includes: Extracting stent and / or thrombus data from a temporal sequence of X-ray images; inputting stent data and / or thrombus data into a neural network; generating a latent space encoding 150, z, of the input image using a neural network further based on the input stent data and / or thrombus data; Including, the stent data includes one or more of a size, a shape, and an expansion rate of the stent over time in a temporal sequence of x-ray images; The thrombus data includes the size and / or composition of the thrombus 110.

[0067] In these examples, the data may be extracted manually or automatically from the X-ray images. For example, the thrombus data may be determined using image processing or computer vision algorithms to segment the clot. A neural network may also be trained to extract these features from the X-ray images. The stent data may be determined using computer vision techniques or neural networks, such as, for example, thresholding, region growing, template matching, level sets, active contour modeling, and neural networks such as U-Nets.

[0068] In the example described above with reference to FIGS. 4 and 5, the neural network uses training x-ray images 120 representing the expansion of a stent over a thrombus having known values ​​for the physical properties. T The value of a physical property of the thrombus 110 is determined based on the z-position of the latent space encoding 150 of the input image relative to the distribution 160 of latent space encodings generated from the temporal sequence of the thrombus 110.

[0069] An alternative example of this technique is to predict the shape of the stent in one or more images in the time sequence from the input images and determine the value of the physical property of the thrombus 110 based on the difference DE between the predicted shape of the stent in one or more images in the time sequence and the actual shape of the stent in one or more images in the time sequence.

[0070] Thus, in one example, with reference to the method shown in FIG. 1, operation S120 of determining a value of a physical property of the thrombus 110 includes: inputting a plurality of images from a time sequence into a neural network 140; predicting from the input images a shape of the stent in one or more images in a time sequence; determining a value of a physical property of the thrombus 110 based on a difference DE between a predicted shape of the stent in one or more images in the time sequence and an actual shape of the stent in one or more images in the time sequence; Including, The neural network 140 uses the training x-ray images 120 T and trained to predict a shape of a stent in one or more images in the time sequence using training data having a plurality of time sequences of Each sequence represents the expansion of a stent of an endoluminal stent retriever device over a thrombus having a known value for a physical property.

[0071] This method is illustrated diagrammatically in FIG. 6, which is a schematic diagram illustrating an example of A) training a neural network 140 to predict the shape of a stent from a temporal sequence of X-ray images, and B) performing inference using a neural network 140 trained to predict the shape of a stent from a temporal sequence of X-ray images, according to some aspects of the disclosure. With reference to FIG. 6B, in inference, a number of images from a temporal sequence 120 are input to the neural network. In the illustrated example, the neural network predicts the shape of a stent in a future image after a period Dt from one or more images in the temporal sequence. The shape of the stent may alternatively be predicted at the same input image or images, i.e., when Dt=0. A difference DE between the shape of the stent in the input image and the predicted shape of the stent is then calculated. The difference DE represents how close the value of the physical property of the thrombus 110 in the input image is to the value of the physical property of the thrombus 110 in the training images used to train the neural network.

[0072] This method of Fig. 6 may be implemented by the neural networks shown in Fig. 4 and Fig. 5, with the difference that during inference, the decoder part of these neural networks is required to calculate the difference DE, which is shown at the bottom of Fig. 4 and Fig. 5. This difference DE may be calculated using a loss function such as L1 loss, L2 loss, etc.

[0073] This method of FIG. 6 utilizes the same insight as described above for the neural networks shown in FIGS. 4 and 5, i.e., the neural network 140 uses training x-ray images 120 representing the expansion of a stent of an endoluminal stent retriever device over a thrombus having known values ​​for physical properties. T The neural network is trained using temporal sequences of images of the stent retriever device on a thrombus having the same values ​​for the physical properties. As a result, in inference, if the trained neural network is input with a new temporal sequence representing an endoluminal stent retriever device on a thrombus having the same values ​​for the physical properties, the neural network can accurately reconstruct the input images, in which case there is a small difference DE between the predicted shape of the stent in one or more images in the temporal sequence and the actual shape of the stent in one or more images in the temporal sequence. In contrast, in inference, if the trained neural network is input with a new temporal sequence representing an endoluminal stent retriever device on a thrombus having different values ​​for the physical properties, the neural network cannot accurately reconstruct the input images, in which case there is a large difference DE.

[0074] The neural networks shown in Figures 4 and 5 may also be trained to perform the method of Figure 6 in a manner similar to that described above with reference to Figure 4, i.e., by having the neural network 140 reconstruct input images during inference. Referring to Figure 6A, during training, training x-ray images 120 Tare input to the neural network 140. The neural network predicts the shape of the stent in the image or in subsequent images. A loss L is calculated between the predicted shape of the stent and the actual shape of the stent in the training images. The loss L is used to train the neural network. The neural network thus Training X-ray images 120 T receiving training x-ray images including a plurality of time sequences of x-ray images, each time sequence representing an expansion of a stent of an endoluminal stent retriever device over a thrombus having a known value for a physical property; Input the training data into the neural network, For each of a plurality of input training X-ray images in a temporal sequence, Using a neural network to predict a shape of the stent in one or more images in the temporal sequence; adjusting parameters of the neural network based on a difference L between a predicted shape of the stent in one or more images in the temporal sequence and an actual shape of the stent in one or more images in the temporal sequence; Repeat the prediction and adjustment until a stopping criterion is met. By this, it can be trained to predict the shape of a stent in one or more images in a time sequence.

[0075] The training data may be provided in the same manner as described above with reference to Figures 4 and 5. Additionally, as described in connection with training these neural networks, the neural networks may be trained to place a higher emphasis on data within a defined region of interest than outside the region of interest. Thus, in one example, the neural network may be trained to: trained to predict the shape of a stent in one or more images in the temporal sequence of training x-ray images by identifying a region of interest in the temporal sequence of training x-ray images, the region of interest including the stent and the thrombus; Adjusting the parameters of the neural network based on a difference L between the predicted shape of the stent in one or more images in the temporal sequence and the actual shape of the stent in one or more images in the temporal sequence is performed by applying a weight to the difference; A relatively high weighting is applied to portions of the temporal sequence of training X-ray images that are within the region of interest; A relatively low weighting is applied to portions of the temporal sequence of training X-ray images outside the region of interest.

[0076] In the inference, in addition to being used to determine the value of the physical property of the thrombus 110, the difference DE may also be used to recommend one or more subsequent workflow steps for treating the thrombus 110. This may be performed by using training data including workflow data that defines one or more subsequent workflow steps used to successfully treat the thrombus in the temporal sequence of training X-ray images. If the difference DE is small, the successful workflow step used in the training data may be recommended. The neural network may also be trained to learn an association between the difference DE and the subsequent workflow step used to successfully treat the thrombus in the temporal sequence of training X-ray images. Alternatively, the neural network shown in FIG. 6 may additionally be trained to generate a distribution of latent space encoding 150, z, of the input image, and one or more subsequent workflow steps for treating the thrombus 110 may be recommended by determining the position of the latent space encoding of the input temporal sequence of X-ray images relative to the distribution 160 of latent space encoding with the common workflow data, in the same manner as described above with reference to FIGS. 4 and 5.

[0077] The neural network may also be trained to perform this method using additional data, such as patient data, stent data, and thrombus data, as described above in connection with FIGS.

[0078] Thus, in one example, the neural network 140 is trained to predict the shape of the stent in one or more images in the temporal sequence further based on the patient data, and the method further comprises: receiving patient data relating to a thrombus 110; inputting patient data into a neural network; predicting from the input images a shape of the stent in one or more images in the temporal sequence further based on patient data; having The patient data includes one or more of a volumetric image representative of the vasculature surrounding the thrombus 110 and electronic health record data related to the thrombus 110.

[0079] In another example, the neural network is trained to predict a shape of the stent in one or more images in the temporal sequence further based on the stent data and / or the thrombus data, and the method further comprises: Extracting stent and / or thrombus data from a temporal sequence of X-ray images; inputting stent data and / or thrombus data into a neural network; predicting from the input images a shape of the stent in one or more images in the time sequence further based on the input stent data and / or thrombus data; Including, the stent data includes one or more of a size, a shape, and an expansion rate of the stent over time in a temporal sequence of x-ray images; The thrombus data includes the size and / or composition of the thrombus 110.

[0080] In another example, a confidence value of the latent space encoding 150, z, or predicted shape of the stent may be calculated. The confidence value represents the confidence of the prediction made by the neural network 140 and allows a decision to be made based on its output. For example, if the confidence is low, it may be decided not to rely on the physical characteristics of the thrombus to be determined. In this regard, if the confidence value is below a predefined threshold, a warning may be output.

[0081] Thus, the method described with reference to FIG. 4 or FIG. 5 may additionally include: Calculating a confidence value for the latent space encoding 150, z, or the predicted shape of the stent, respectively; and An operation of outputting a confidence value; may include:

[0082] As an example, the neural network 140 may generate the confidence value using a dropout technique. The dropout technique involves repeatedly inputting the same data into the neural network and randomly excluding a percentage of neurons from the neural network at each iteration while determining the output of the neural network. The output of the neural network is then analyzed to provide a mean and variance value. The mean represents the final output, and the magnitude of the variance indicates whether the neural network is consistent in its predictions, in which case the variance is small and the confidence value is relatively high, or whether the neural network is inconsistent in its predictions, in which case the variance is larger and the confidence value is relatively low. The confidence may alternatively be calculated using, for example, the Kullback-Leibler "KL" divergence between the distribution from which the input image representation is sampled and the distribution over the current trained encoding. This divergence indicates how well the input sequence is represented by the learned encoding.

[0083] In another example, a computer program product is provided having instructions that, when executed by one or more processors, cause the one or more processors to perform a computer-implemented method for determining a value of a physical property of a thrombus 110 according to any of the example embodiments described above. The method includes: receiving X-ray image data S110 having a temporal sequence of X-ray images 120 representing the expansion of a stent 130 of an intraluminal stent retriever device over a thrombus 110; determining a value of a physical property of the thrombus 110 based on the expansion rate of the stent 130 in a time sequence of X-ray images 120; A step S130 of outputting the value of the physical property; has.

[0084] In another example, a system 200 for determining a value of a physical property of a thrombus 110 is provided. The system has one or more processors 210 configured to perform steps of a computer-implemented method for determining a value of a physical property of a thrombus according to any of the above embodiments or examples. That is, the processor 210 includes: receiving (S110) x-ray image data having a temporal sequence of x-ray images (120) depicting the expansion of a stent (130) of an intraluminal stent retriever device over a thrombus (110); determining (S120) a value of a physical property of the thrombus (110) based on the expansion rate of the stent (130) in the time sequence of the X-ray images (120); outputting (S130) the value of the physical property; It can be configured as follows.

[0085] For example, the one or more processors 210, in operation, may execute instructions of the computer program product as described above.

[0086] An example of the system 200 is shown in Figure 2. In some examples, the system 200 may also include one or more of a projection x-ray imager 220 for providing a temporal sequence of x-ray images 120, as shown in Figure 2, an endoluminal stent retriever device 230 for performing a mechanical thrombectomy procedure on a blood vessel, as shown in Figure 2, a display device such as a monitor 240, as shown in Figure 2, for displaying values ​​of a physical property of the thrombus 110, the temporal sequence of x-ray images 120 representing the expansion of a stent 130 of the endoluminal stent retriever device on the thrombus 110, one or more recommended subsequent workflow steps for treating the thrombus 110, a confidence value, or the like, a patient bed 250, and a user input device such as a keyboard, mouse, touch screen, or the like for receiving user input related to operations performed by the system.

[0087] The above examples should be understood as illustrating, but not limiting, the disclosure. Further examples are also contemplated. For example, the examples described in relation to a computer-implemented method may also be provided by a computer program product, or by a computer-readable storage medium, or by the system 200 in a corresponding manner. It should be understood that features described with respect to any one example may be used alone or in combination with other described features, or in combination with one or more other features of the examples, or with combinations of other examples. Moreover, equivalents and modifications not described above may also be used without departing from the scope of the invention as defined in the appended claims. In the claims, the word "comprises" does not exclude other elements or operations, and the indefinite articles "a" or "an" do not exclude a plurality. The mere fact that certain features are recited in mutually different dependent claims does not indicate that a combination of these features cannot be advantageously used. Any reference signs in the claims should not be construed as limiting their scope.

Claims

1. 1. A computer-implemented method for determining a value of a physical property of a thrombus, comprising: receiving x-ray image data having a temporal sequence of x-ray images representative of an expansion of a stent of an endoluminal stent retriever device on the thrombus; determining a value of the physical property of the thrombus based on a rate of expansion of the stent in the time sequence of x-ray images; outputting the value of the physical property; The method comprising:

2. 2. The computer-implemented method of claim 1 , wherein the step of determining the value of the physical property of the thrombus includes inputting a plurality of images from a temporal sequence of training x-ray images into a neural network trained using training data having the temporal sequence of training x-ray images, each such training sequence representing the expansion of a stent of an endoluminal stent retriever device on a thrombus having a known value for the physical property.

3. The determining step further comprises: generating a latent space encoding of the input image using the neural network; determining a value of the physical property of the thrombus based on a position of the latent space encoding of the input image relative to a distribution of latent space encodings generated from the temporal sequence of training x-ray images; having the neural network is trained to generate the latent space encoding of the input image using the training data; The computer-implemented method of claim 2 .

4. the training data further comprising workflow data for each temporal sequence of training x-ray images, the workflow data defining one or more subsequent workflow steps to be used to successfully treat the thrombus in the temporal sequence of training x-ray images; The method further comprises: determining a location of the latent space encoding of the input image relative to a distribution of the latent space encoding with common workflow data; outputting the common workflow data to provide one or more recommended subsequent workflow steps for treating the thrombus; The computer-implemented method of claim 2 , comprising:

5. 5. The computer-implemented method of claim 4, wherein the workflow data further comprises a success metric corresponding to the one or more subsequent workflow steps, the success metric representing a probability of the one or more subsequent workflow steps resulting in a successful treatment of the thrombus, and outputting the common workflow data further comprises outputting the corresponding success metric.

6. determining a location of the latent space encoding of the input image relative to a distribution of the latent space encoding with the common workflow data comprises inputting the latent space encoding of the input image to a second neural network; the second neural network is trained to determine the one or more subsequent workflow steps for successfully treating the thrombus based on the latent space encoding of the input images using training data having a plurality of temporal sequences of training x-ray images, each sequence representing an expansion of a stent of an endoluminal stent retriever device on a thrombus having a known value for the physical property, and to output the one or more subsequent workflow steps, the training data having workflow data for each temporal sequence of training x-ray images, the workflow data defining one or more subsequent workflow steps used to successfully treat the thrombus in the temporal sequences of training x-ray images.

5. The computer-implemented method of claim 4.

7. The neural network is trained to generate the latent space encoding of the input image based on patient data, the method further comprising: receiving patient data relating to the thrombus; inputting said patient data into said neural network; generating the latent space encoding of the input image further based on the input patient data using the neural network; having the patient data comprises one or more of: a volumetric image representative of the vasculature surrounding the thrombus; electronic health record data relating to the thrombus; A computer-implemented method according to any one of claims 3 to 6.

8. The neural network is trained to generate the latent space encoding of the input image based on stent data and / or thrombus data, the method further comprising: extracting the stent data and / or the thrombus data from the temporal sequence of X-ray images; inputting the stent data and / or the thrombus data into the neural network; using the neural network to generate the latent space encoding of the input image further based on the input stent data and / or thrombus data; having the stent data comprises one or more of a size, a shape, and an expansion rate of the stent in the temporal sequence of X-ray images, and the thrombus data comprises a size and / or a composition of the thrombus.

8. A computer-implemented method according to any one of claims 3 to 7.

9. The neural network comprises: receiving training data including a plurality of temporal sequences of training x-ray images, each temporal sequence representing an expansion of a stent of an endoluminal stent retriever device over a thrombus having a known value for said physical property; inputting the training data into the neural network; For each of the input training X-ray images in a time sequence, using the neural network to generate a latent space encoding of the input training x-ray images; reconstructing the input training x-ray images from the latent space encoding using the neural network; adjusting parameters of the neural network based on differences between the input training x-ray images and the reconstructed input training x-ray images; repeating said generating, said reconstructing, and said adjusting until a stopping criterion is met.

4. The computer-implemented method of claim 3, wherein the latent space encoding of the input image is trained by:

10. The step of determining the value of the physical property of the thrombus comprises: predicting a shape of a stent in one or more input images in the temporal sequence using the neural network; and determining a value of the physical property of the thrombus based on a difference between the predicted shape of the stent in the one or more images in the temporal sequence and an actual shape of the stent in the one or more images in the temporal sequence; The computer-implemented method of claim 2 , comprising:

11. The neural network is trained to predict a shape of the stent in the one or more images in the temporal sequence based further on patient data, the method further comprising: receiving patient data relating to the thrombus; inputting said patient data into said neural network; predicting from the input images a shape of the stent in the one or more images in the temporal sequence further based on the patient data; having the patient data comprising one or more of: a volumetric image representative of the vasculature surrounding the thrombus; and electronic health record data relating to the thrombus.

11. The computer-implemented method of claim 10.

12. The neural network is trained to predict a shape of the stent in the one or more images in the temporal sequence based further on stent data and / or thrombus data, the method further comprising: extracting the stent data and / or the thrombus data from the temporal sequence of X-ray images; inputting the stent data and / or the thrombus data into the neural network; predicting, from the input images, a shape of the stent in the one or more images in the temporal sequence further based on the input stent data and / or thrombus data; having the stent data comprises one or more of a size, a shape, and an expansion rate of the stent over time in the temporal sequence of x-ray images; The clot data includes a size and / or a composition of the clot. A computer-implemented method according to any one of claims 10 to 11.

13. The neural network comprises: receiving training data including a plurality of temporal sequences of training x-ray images, each temporal sequence representing an expansion of a stent of an endoluminal stent retriever device over a thrombus having a known value for said physical property; inputting the training data into the neural network; For each of a plurality of input training X-ray images in the temporal sequence, using the neural network to predict a shape of the stent in one or more images in the temporal sequence; adjusting parameters of the neural network based on a difference between the predicted shape of the stent in the one or more images in the temporal sequence and an actual shape of the stent in the one or more images in the temporal sequence; repeating said prediction and said adjustment until a stopping criterion is met; 11. The computer-implemented method of claim 10, wherein the method is trained to predict a shape of the stent in one or more images in the temporal sequence by:

14. A system for determining a value of a physical property of a thrombus, the system comprising one or more processors configured to perform the steps of the computer-implemented method according to any one of claims 1 to 13.

15. The system of claim 14 further comprising a projection X-ray imager for providing the temporal sequence of X-ray images.

Images