Computer-implemented method, computer program product, computer-readable storage medium, use of computer-implemented method and system for performing steps of method

Abstract

The invention relates to a computer-implemented method for determining the position and / or orientation of a valve component or annulus component, preferably the annulus (2) of a heart valve, in particular during the implantation of a heart valve prosthesis (3), comprising the following steps: (i) receiving, in particular real-time receiving, via an input interface; at least one first image (4) and at least one second image (5) of the moving heart; (ii) identifying in the at least one first image (4) at least one first feature (7) and at least one second feature (8) indicative of the position and / or orientation of the valve or annulus assembly using a data processing unit (6); (iii) determining, using the data processing unit (6), at least one static or dynamic geometrical relationship between the at least one first feature (7) and the at least one second feature (8) on the basis of the positioning of the first feature (7) and the second feature (8) in the at least one first image (4), identifying, using the data processing unit (6), the at least one second feature in the at least one second image (5); (iv) determining a corrected position and / or orientation (20) of the valve component or annulus component in the at least one second image (5) based on the at least one second feature (8), the at least one static or dynamic geometry and a static or dynamic geometry model; and (v) transmitting the modified position and / or orientation (20) of the valve component or annulus component to the output interface.

Description

Technical Field

[0001] The present invention relates to a method for determining the position and / or orientation of a valve assembly (preferably a valve annulus of a heart valve), a computer program product, a computer-readable storage medium, the use of the computer-implemented method, and a system for performing the steps of the method. Background Technology

[0002] Cardiac surgery, especially transcatheter aortic valve implantation or transcatheter aortic valve replacement (TAVI / TAVR), requires precise localization of the patient's native valve annulus, which serves as an anchor point, such as for aortic heart valve prostheses.

[0003] Aortic valve implantation carries a serious risk of medical complications, which may result in stroke or even death due to detachment of calcified material or thrombus, acute kidney injury / failure due to contrast agent injection, vascular dissection and / or vascular hematoma or bleeding.

[0004] Incorrect placement of aortic prosthesis valves can lead to a variety of adverse surgical outcomes, including valve displacement, the need for subsequent aortic valve surgery or repositioning of the aortic valve, arrhythmias, aortic regurgitation (e.g., paravalvular leak due to incomplete sealing with native tissue), cardiac injury, and / or even coronary artery obstruction.

[0005] Various medical imaging techniques, such as magnetic resonance imaging (MRI), computed tomography (CT), and fluoroscopy, enable clinicians to identify regions of interest, particularly a patient's native valvular components, especially the tip of the coronary valve, for cardiac surgery, particularly for TAVI / TAVR. However, accurate identification and tracking of native valvular components (such as the annulus, coronary valve, or calcifications), especially in conjunction with medical devices (such as delivery devices and / or prosthetic aortic valves), are affected by cardiac and respiratory cycles, blood flow, and systolic / diastolic blood pressure. Furthermore, tracking radiopaque anatomical structures is challenging because imaging techniques such as fluoroscopy require the injection of contrast agents to accurately identify and track these features.

[0006] EP 2 757 528 B1 relates to a method and apparatus for tracking an object in a target region of a continuously moving organ (such as the heart) based on determining a dynamic geometric relationship between two features.

[0007] However, existing technologies have failed to provide a simple, reliable, and robust method for clinicians to infer the origin of the valve component during the implantation of a heart valve prosthesis.

[0008] Specifically, existing technologies fail to account for the location and / or orientation of the native valve assembly during transient rapid ventricular pacing (RVP). RVP is used to deliver rapid electrical pulses to the heart to increase heart rate, for example via an external pacemaker, thereby creating a brief period of “cardiac arrest” that allows for the positioning and deployment of aortic valve prostheses.

[0009] In addition, there is a need in the prior art for: simplifying the prosthetic implant’s crossing of the valve annulus based on angiographic / non-angiographic image data, and positioning the prosthetic implant relative to the position and / or orientation of the native valve assembly, especially when performing rapid pacing during valve delivery. Summary of the Invention

[0010] The present invention also provides a simple, efficient and convenient method for determining and displaying the location and / or orientation of a patient's native valve components, particularly using imaging devices commonly available to clinicians.

[0011] This invention relates to a computer-implemented method for determining the position and / or orientation of a patient's native valvular assembly (preferably the non-coronary leaflet of the native aortic valve, particularly the tip of the non-coronary leaflet). The method includes receiving (particularly in real-time) at least one first image and at least one second image of a moving heart acquired at two different time points (preferably during implantation of a heart valve prosthesis) via an input interface. The method includes determining a corrected position and / or orientation of the native valvular assembly in at least one second image by applying at least one pre-trained machine learning model to at least one first image and / or at least one second image. The method includes overlaying the corrected position and / or orientation of the native valvular assembly onto at least one second image to generate a synthetic second image, and displaying the synthetic second image on a display.

[0012] This allows the computer-implemented method to provide and display simple synthetic images containing the location and / or orientation of the valve components, enabling the method to be used in conjunction with commonly available imaging techniques, such as fluoroscopy. This further simplifies the learning curve for clinicians who are typically experienced in using and applying commonly used imaging devices, such as fluoroscopy devices. At least the first and second images can be images acquired during implantation, replacement, or repair of a heart valve prosthesis during transapical, transseptal, transarterial, transvenous, transfemoral, transjugular, transsubclavian / transaxillary, transcarotid, transvesical, or transseptal valve interventions. In a preferred embodiment, the method may include at least one first and second image acquired during TAVI / TAVR, particularly during transfemoral TAVI / TAVR.

[0013] In addition to receiving at least one first image and at least one second image, the method may include receiving only at least one second image and determining the corrected location and / or orientation of the native valve component by applying at least one machine learning model to at least one second image.

[0014] This computer-based method can be applied to determine the location and / or orientation of the native valvular components of the native aortic valve, mitral valve, tricuspid valve, or pulmonary valve.

[0015] This pre-trained machine learning model can be configured to provide synthetic second images in real time.

[0016] At least one first image can be a preoperative CT image or a preoperative MRI image. This allows the method to overlay image information (such as first and second features) in three dimensions with a second image. The second image can be an intraoperative (particularly real-time) fluoroscopic image. Fluoroscopic images are typically acquired in different orientations during TAVI / TAVR; for example, the C-arm can be in a cephalic (CRA), caudal (Cau), left anterior oblique (LAO), or right anterior oblique (RAO) orientation. Three-dimensional CT images allow for tracing the native cardiac valve components in fluoroscopic images with different orientations, and further, particularly anatomical or instrumental features, thereby precisely locating these features.

[0017] Preferably, at least one first image and / or at least one second image can be an intraoperative fluoroscopic image.

[0018] The pre-trained machine learning model can be a pre-trained neural network configured to extract hierarchical feature maps from at least one first image using a data processing unit. The pre-trained neural network may include downsampling at least one first image, particularly progressively downsampling at least one first image, extracting abstract feature maps, and upsampling the abstract feature maps to the original resolution of at least one first image.

[0019] The architecture of this pre-trained machine learning model can be called a U-Net pre-trained machine learning model. It can provide reliable and efficient feature mapping for medium and high resolution images, so that the position / orientation of the native valve components can be robustly determined and included in the synthetic second image.

[0020] The U-net architecture performs poorly for small or low-resolution images. The architecture can be modified to reduce or avoid downsampling, thereby preserving the original small / low-resolution images.

[0021] A machine learning model can be a pre-trained residual neural network comprising multiple convolutional layers and at least one residual block for bypassing at least one convolutional layer.

[0022] This residual neural network provides enhanced feature extraction without reducing accuracy and alleviates the problems of vanishing or exploding model gradients.

[0023] The pre-trained residual neural network may include at least one bottleneck residual block. In a preferred embodiment, the pre-trained neural network has at least three convolutional layers, particularly exactly three convolutional layers.

[0024] The "resting" state of the heart induced by rapid pacing results in the position / orientation of the native valvular components being affected differently compared to a normal cardiac cycle. During rapid pacing, the heart beats repeatedly and rapidly, for example, between 180 and 220 beats per minute, and experiences a systolic phase. This significantly affects the position / orientation of the native valvular components (especially the native valves, preferably non-coronary valves) because the stroke volume of blood flow decreases significantly and blood pressure (especially systolic blood pressure) drops briefly.

[0025] At least one first image may be an image of normal blood flow, particularly acquired in the absence of rapid pacing, and / or at least one second image may be an image of at least one image of reduced blood flow, particularly acquired during rapid pacing. Correction of position and / or orientation can be determined by further correcting the position and / or orientation of the patient's native valvular components in response to blood flow (particularly due to rapid pacing).

[0026] Alternative methods for regulating blood flow can be used in conjunction with this method; for example, physiological pacing can regulate blood flow in a way that is more in line with physiological rhythms, or temporary (heparin-coated) balloon occlusion can be used to regulate blood flow.

[0027] At least one second image (preferably acquired during rapid pacing) provides a more precise corrected location and / or orientation of the native valvular assembly. Since heart valve prostheses are typically positioned and deployed during rapid pacing, a more precise determination of the corrected location / orientation of the native valvular assembly allows for more accurate positioning / orientation, thereby avoiding adverse effects due to abnormal heart valve prosthesis placement. At least one first image acquired in the absence of rapid pacing can still be used to predict the location / orientation of the native valvular assembly during rapid pacing, as a pre-trained machine learning model can be configured to determine the impact of rapid pacing based on training data including both pacing and non-pacing images, as well as at least one second image acquired during rapid pacing.

[0028] The method may include determining, in real time, the device position and / or device orientation of a heart valve prosthesis and / or a delivery device for delivering the heart valve prosthesis in at least one of at least one first and / or second image. The method may include determining geometric heart valve prosthesis characteristics based on the device position and / or device orientation. The method may include overlaying the geometric heart valve prosthesis characteristics with at least one synthetic second image.

[0029] This allows clinicians or autonomous systems to correlate the modified position / orientation of the native valve assembly with the device position / orientation and / or the geometric characteristics of the heart valve prosthesis in the synthetic second image, thereby facilitating and simplifying cardiac interventions, particularly simplifying the performance of TAVI / TAVR procedures.

[0030] Geometric heart valve prosthesis characteristics can indicate longitudinal sub-portions of the heart valve prosthesis. The method may include receiving at least one user input defining a longitudinal sub-portion of the heart valve prosthesis. Geometric heart valve prosthesis characteristics may include, or be constituted by, a vertical representation relative to the longitudinal axis of the heart valve prosthesis.

[0031] The computer-implemented method may include determining the length of a heart valve prosthesis in at least one first and / or second image, thereby reliably determining a longitudinal sub-portion of the heart valve prosthesis. This can be achieved by identifying the outer edge of the heart valve prosthesis, which is typically visible in two-dimensional first and / or second images, such as fluoroscopic images. The distance between the proximal edge of the heart valve prosthesis and the distal edge of the heart valve prosthesis may be determined to infer its length. Alternatively, the method may include detecting radiopaque position / orientation markers of the medical device (particularly the heart valve prosthesis) to determine geometric heart valve prosthesis characteristics indicating the longitudinal sub-portion.

[0032] User input specifying a longitudinal sub-section can be provided by a graphical user interface, for example, 20% of the total length of the heart valve prosthesis. Therefore, the geometric characteristics of the heart valve prosthesis can be represented by a longitudinal sub-section starting from the proximal / distal edge of the heart valve prosthesis, for example, 20% of the heart valve prosthesis.

[0033] Geometric features of a heart valve prosthesis may include two-dimensional or three-dimensional representations, or may consist of two-dimensional or three-dimensional representations. To improve the visibility of the underlying second image, geometric features of the heart valve prosthesis may be translucent. A three-dimensional representation of the heart valve prosthesis features can also provide information indicating the orientation of the heart valve prosthesis relative to anatomical structures.

[0034] A vertical representation relative to the longitudinal axis may include at least one line (especially a curve or dashed line), a point, an arrow, a triangle, a cross, a star, a square, or a triangle, or a combination thereof.

[0035] The method may include determining a patient's cardiac and / or respiratory cycles based on at least one first image, at least one second image, and / or additional patient data (particularly cardiac electrocardiogram). The cardiac and / or respiratory cycles may be further taken into consideration to determine the corrected location and / or orientation.

[0036] This enables the determination of the corrected location / orientation of native valve components in a more accurate and reliable manner based on pre-trained machine learning models.

[0037] At least one first image and at least one second image can be used to determine the cardiac cycle and / or respiratory cycle, particularly to differentiate between the cardiac and respiratory cycles. Both the cardiac and respiratory cycles typically affect the motion of the native valvular assembly in a substantially periodic manner at different time scales. Therefore, this method can be adapted to determine the respective effects of both independently of each other based on multiple first and / or second images.

[0038] At least one first image and / or at least one second image may include device features. At least one device feature may include radiopaque location and / or orientation markings of the heart valve prosthesis and / or delivery device for delivering the heart valve prosthesis. At least one device feature may include a pigtail catheter or guidewire located near the non-coronary valve, preferably having a basic predefined posture of contact with the non-coronary valve. At least one device feature may include a coiled catheter or guidewire located within the left ventricle, preferably having a basic predefined posture of contact with the left ventricle near the apex.

[0039] At least one first and / or second image, including at least one instrument feature, enables the pre-trained model to more accurately infer the corrected location / orientation of the native valve assembly, because the instrument feature is typically spatially arranged in a predetermined manner due to surgical and anatomical constraints. For example, instruments surgically positioned in non-coronary flanks (such as pigtail catheters or guidewires) allow the corrected location / orientation of the non-coronary flank to be determined more reliably.

[0040] Contrast agents are typically injected via a pigtail catheter. If the pigtail catheter is placed near or within the non-coronary fossa, it can be more easily identified during contrast agent injection. The contrast agent may dissipate before reaching other anatomical structures, thus requiring more contrast agent or a prolonged injection period to achieve adequate visibility.

[0041] Radiopaque location / orientation markings can allow for more robust determination of the location / orientation of a heart valve prosthesis or a sub-part of a heart valve prosthesis. Instead of radiopaque location / orientation markings, at least one first and / or second image may include radiopaque location / orientation markings (e.g., markings relative to the radiopaqueness of X-rays) or the inherent radiopaqueness of the heart valve prosthesis.

[0042] The method may include determining the location and / or orientation of at least one anatomical feature based on at least one first image and / or at least one second image. The anatomical feature may be selected from at least one of the following: (i) at least one of the right and left coronary lobes, particularly the tip of at least one coronary lobe; (ii) at least one coronary artery, particularly two coronary arteries; (iii) the outline of the left ventricle; (iv) the interventricular septum separating the right and left ventricles, particularly a membranous septum; and (v) calcifications near the original heart valves. The method may also include displaying the location and / or orientation of at least one anatomical feature in at least one synthetic second image.

[0043] Displaying at least one anatomical feature, particularly one or more coronary arteries, the left ventricular ventricular contour, and / or membranous septum, provides valuable information to clinicians, especially relevant to performing cardiac interventions, particularly TAVI / TAVR. For example, it allows for adjustments to the placement of prosthetic heart valves to ensure they are deployed in the target location / orientation without obstructing coronary arteries, that conduction pathways within the membranous septum are not blocked, and / or that the coiled guidewire can be optimally positioned within the left ventricular contour.

[0044] The method may include identifying the presence or absence of a native valvular component in at least one second image. If the native valvular component is identified as present, the corrected location and / or orientation of the native valvular component in the at least one second image may be determined solely by applying at least one pre-trained learning model to the at least one second image. If the native valvular component is identified as absent, the corrected location and / or orientation of the native valvular component in the at least one second image may be determined solely by applying a pre-trained machine learning model to at least one first image (and optionally at least one second image).

[0045] This allows for the determination of the corrected location and / or orientation of the native valve component in at least one second image, based on whether the native valve component is present or absent, relying on more relevant images. This also enables the corrected location / orientation of the native valve component to be displayed even in images where the native valve component is not visible (e.g., due to contrast agent dissipation or radiopaque blocking caused by medical devices).

[0046] The method may include classifying at least one first and / or at least one second image to determine whether the corresponding image was acquired under the action of a contrast agent.

[0047] The method may include, if the image is classified as not acquired under contrast agent treatment, determining the corrected location and / or orientation of the native valve assembly, and / or the location and / or orientation of at least one anatomical feature visible under contrast agent treatment, based on at least one pre-trained machine learning model and at least one other image classified as acquired under contrast agent treatment. Alternatively or additionally, the method may include, if the image is acquired under contrast agent treatment, determining the location and / or orientation of instrument features invisible under contrast agent treatment, based on at least one pre-trained machine learning model and at least one image classified as not acquired under contrast agent treatment.

[0048] This can provide valuable classification data for the pre-trained machine learning model, or different pre-trained machine learning models, to determine location / or orientation. Specifically, this can allow for the determination of corrected location / or orientation and / or position / or orientation of instrument features / anatomical features that are only visible under contrast agent conditions, based on images that are actually classified as acquired under contrast agent conditions. Correspondingly, based on non-contrast images, corrected location / or orientation and / or orientation / position of instrument features / anatomical features that are not visible under contrast agent conditions can be determined more reliably (e.g., because they are radiopaque and their visibility is blocked by the contrast agent).

[0049] The method may include applying a first pre-trained neural network to images classified as acquired under contrast agent treatment, and applying a second pre-trained neural network, different from the first pre-trained neural network, to images classified as not acquired under contrast agent treatment.

[0050] This allows for more precise and reliable determination of the corrected location and / or orientation of anatomical / instrument features. Pre-trained machine learning models can be switched in real-time to provide clinicians with a concise display of the synthesized second image. These models can be specifically adapted and trained using imaging data relevant to the target intervention.

[0051] Images classified as acquired under contrast agent conditions can be used for segmentation of the aortic root region. Based on this segmentation, the corrected location and / or orientation of the native valvular components and / or the location and / or orientation of at least one anatomical feature can be determined.

[0052] This can provide additional segmentation data for pre-trained machine learning models, thereby improving prediction performance regarding location / orientation by using a smaller amount of training data to provide stronger discriminative power.

[0053] The aortic root portion may include at least one of the aforementioned anatomical features.

[0054] At least one first image and at least one second image can be obtained in the absence of a contrast agent.

[0055] This enables specially adapted pre-trained machine learning models to infer the corrected location / orientation of the native valve component based on features / markers visible in non-contrast images, thus determining the corrected location / orientation of the native valve component, which is typically invisible in the absence of contrast agents.

[0056] Another aspect of the invention relates to a computer program product that, when executed by a computer, causes the computer to perform the steps of the previously described computer-implemented method.

[0057] Another aspect of the invention relates to a computer-readable storage medium containing instructions that, when executed by a computer, cause the computer to perform the steps of the previously described computer-implemented method.

[0058] Another aspect of the invention relates to a computer implementation method for training a machine learning model (particularly the previously described machine learning model) for determining the position and / or orientation of the native valvular components (preferably non-coronary valves, particularly the tips of non-coronary valves) of an aortic heart valve, particularly for the previously described computer implementation method. The training method includes receiving an input training dataset comprising at least one image of a moving heart acquired during heart valve prosthesis implantation, the image being labeled with a tag including the non-coronary valve (preferably including tags for all three coronary valves), and preferably including a tag for a pigtail catheter or guidewire located near the non-coronary valve. The at least one image may include at least one tag for the heart valve prosthesis, optionally including a tag for the edge of the heart valve prosthesis. The training method includes performing a machine learning model training process by predicting the output position and / or orientation of the non-coronary valve, comparing the output position and / or orientation with the non-coronary valve tags, and adjusting model parameters using a loss function.

[0059] The training dataset for this training method can include or consist of contrast images, allowing the trained machine learning model to be applied to contrast images, and in particular, only to contrast images.

[0060] The first pre-trained machine learning model, previously described, applied to images classified as acquired under contrast agent treatment, can be obtained through a computer-based method for training such a machine learning model.

[0061] At least one image may include at least one first image and at least one second image. The at least one first image may be a normal blood flow image, particularly acquired in the absence of rapid pacing, and the at least one second image may be a reduced blood flow image, particularly acquired during rapid pacing.

[0062] Another aspect of the invention relates to a computer-implemented method for training a machine learning model (particularly as described above) for determining the location and / or orientation of the native valvular components (preferably non-coronary valves, particularly the tips of non-coronary valves) of an aortic heart valve. The machine learning model can be configured to use the previously described method. The training method includes receiving an input training dataset consisting of at least one image of a moving heart acquired during heart valve prosthesis implantation, the image labeled to include at least one location and / or orientation of the native valvular components (particularly non-coronary valves). This at least one image was not acquired under contrast agent exposure. The training method includes performing a machine learning model training process by predicting the output location and / or orientation of the native valvular components; comparing the output location and / or orientation with the location and / or orientation of the native valvular components; and adjusting model parameters using a loss function.

[0063] This method enables the training of machine learning models based solely on non-contrast images to predict the location and / or orientation of native valvular components, particularly those that are essentially invisible in non-contrast images. Therefore, valvular intervention can be performed without the need for contrast agent injection. This can expedite valvular intervention and avoid adverse reactions associated with contrast agent injection, such as allergic reactions, contrast-induced nephropathy, cardiovascular effects, extravasation, and / or prolonged radiation exposure.

[0064] The previously described second pre-trained machine learning model applied to images classified as not acquired under contrast agent conditions can be obtained through a computer implementation method for training such a machine learning model.

[0065] This training method based on non-angiography images requires a large amount of labeled training dataset, particularly labeled by clinicians or machine learning models configured to perform such labeling. Clinicians or machine learning models can label non-angiography images based on at least one pre-defined relationship relating native valvular components (especially the NCC) to specific instrument features (especially the pigtail catheter typically placed near the NCC) and / or anatomical features (such as ribs, vertebrae, or calcifications, particularly calcifications on the aortic heart valve).

[0066] Non-contrast images can be labeled based on previously / subsequently acquired contrast images.

[0067] Different imaging devices, such as echocardiography (especially transthoracic or transesophageal echocardiography), MRI, or CT (especially ECG-gated CT), can be used to label the training dataset.

[0068] Alternatively, the training method can be trained using at least one image acquired under contrast agent conditions and subsequently processed (particularly using a machine learning model) to render the contrast agent-induced radiopaqueness essentially invisible. However, since the at least one image was initially acquired under contrast agent conditions, annotation is simplified.

[0069] At least one first and / or second image can be preprocessed.

[0070] The computer implementation method may include preprocessing at least one first image and / or at least one second image of a moving heart received from an input interface, such that at least one anatomical landmark is arranged in at least one first and / or second image in a first predetermined spatial relationship.

[0071] The method may include preprocessing at least one first and / or second image such that first and second anatomical landmarks in the first and / or second image are arranged relative to each other in a second predetermined spatial relationship in at least one first and / or second image. These anatomical landmarks remain substantially stationary relative to each other regardless of patient movement.

[0072] Anatomical landmarks may consist of calcifications, reference medical devices, or bones (particularly the patient’s ribs and / or spine), which can serve as reference locations regardless of the orientation in which at least one first and / or second image is acquired.

[0073] This enables the computer-implemented method to more reliably determine the corrected location and / or orientation of the native valvular components without taking into account the patient's relative motion during image acquisition. In particular, by preprocessing at least one first and / or second image, the influence of respiratory and cardiac cycles can be reduced, thereby enabling the pre-determined spatial relationships.

[0074] Preprocessing of at least a first image and / or at least a second image can be achieved by applying an image stabilization algorithm. The image stabilization algorithm can be adapted to stabilize / center at least one first / second image by (i) cropping or (ii) padding at least one first / second image.

[0075] Cropping at least one first / second image may include adjusting the boundaries of the first and / or second images according to a predefined template or dynamically, such that at least one first feature and / or at least one second feature are arranged according to a predetermined spatial relationship.

[0076] Filling at least one first / second image may include extending the boundary of at least one first / second image, particularly by adding pixels with predetermined pixel attributes such as color or transparency.

[0077] Preprocessing at least one first image and / or at least one second image to achieve a first / second pre-determined spatial relationship based on anatomical features can improve the ability to determine reliable corrected locations and / or orientations.

[0078] The aforementioned prior art EP 2 757 528 B1 also lacks a reliable computer-implemented method, computer program product, computer-readable storage medium, and system for performing the method for determining the orientation and / or location of a valve assembly or valve annulus assembly (particularly the valve annulus). Specifically, the prior art cannot accurately track the position / orientation of the valve annulus and / or requires repeated injections of contrast agent.

[0079] Specifically, existing technologies fail to account for the position and / or orientation of the valve assembly or valve annulus assembly during transient rapid ventricular pacing (RVP). RVP is used to deliver rapid electrical pulses to the heart to increase heart rate, for example via an external pacemaker, thereby creating a brief period of “cardiac arrest” that allows for the positioning and deployment of aortic valve prostheses.

[0080] In addition, there is a need in the prior art to simplify the passage of prosthetic implants through the valve annulus and to position them by providing the position and / or orientation of the valve annulus, especially during rapid pacing during valve delivery.

[0081] Another aspect of the present invention ingeniously solves and provides a solution to the aforementioned technical problems, particularly by providing a reliable computer-implemented method, a computer program product, a computer-readable storage medium, the use of the method, and a system for performing the steps of the method, for determining and transmitting the corrected position / orientation of a valve assembly or valve annulus assembly (preferably a valve annulus). The valve assembly or valve annulus assembly may include a right coronary cusp, a left coronary cusp, a non-coronary cusp, an aortic valve annulus, a right aortic sinus, a left aortic sinus, a posterior aortic sinus, and / or Arndius's node, or a combination thereof.

[0082] In a preferred embodiment, the position / orientation of the aortic valve annulus is determined and transmitted. However, alternatively, for example, the orientation / position of the mitral, tricuspid, or pulmonary valve annulus or its components (particularly leaflets / valve, valve sinuses, or chordae tendineae) can be determined and transmitted.

[0083] This invention will further improve the placement / positioning accuracy of heart valve prostheses relative to valve assemblies or valve annulus assemblies, with accuracy independent of clinician education and training, and reduce intraoperative and / or postoperative clinical complications such as valve displacement, paravalvular leak, prosthesis embolism, atrioventricular block, coronary artery (ostium) obstruction, hemolysis, valvular dysfunction, turbulent blood flow, hemodynamic instability, or plaque detachment. Furthermore, this invention can reduce operative time and / or contrast agent injection, while allowing for standardized procedures, which will minimize human error and ensure consistent and reproducible results for patients. This invention will also further shorten operative time and reduce radiation exposure by guiding patients and clinicians with fluoroscopy.

[0084] According to a first aspect of the invention, a computer-implemented method is provided for determining the position and / or orientation of a valve assembly or valve annulus assembly (preferably a valve annulus), particularly during the implantation of a heart valve prosthesis. The method includes receiving (particularly in real-time) at least one first image and at least one second image of a moving heart via an input interface. The method includes using a data processing unit to identify at least one first feature and at least one second feature in the at least one first image indicating the position and / or orientation of the valve assembly or valve annulus assembly. The method further includes using the data processing unit to determine at least one static or dynamic geometric relationship between the at least one first feature and the at least one second feature based on the positioning of the first and second features in the at least one first image. The method includes using the data processing unit to identify at least one second feature in the at least one second image. The method further includes determining a modified position and / or orientation of the valve assembly or valve annulus assembly in the at least one second image based on the at least one second feature, at least one modified static or dynamic geometric relationship, and a static or dynamic geometric relationship model for considering (particularly during rapid pacing of the heart) the position and / or orientation of the valve assembly or valve annulus assembly. The method also includes transmitting the corrected position and / or orientation of the valve assembly or valve annulus assembly to the output interface.

[0085] This enables the calculation and integration of the determination of the position and / or orientation of the valve assembly or annulus assembly, for example, for the heart valve prosthesis to cross the annulus (especially during rapid pacing), and subsequently for precise positioning / orientation of the heart valve prosthesis during deployment. In addition to correcting the position, determining the corrected orientation of the valve assembly or annulus assembly allows for more precise positioning of the valve assembly or annulus assembly, thereby avoiding potential risks when deploying the heart valve prosthesis within the annulus.

[0086] The "resting" state of the heart induced by rapid pacing results in the position / orientation of the valvular components or annular components being affected differently compared to a normal cardiac cycle. During rapid pacing, the heart beats repeatedly and rapidly, for example, between 180 and 220 beats per minute, and experiences a systolic phase. This significantly affects the position / orientation of the valvular components or annular components (especially the annulus) because the stroke volume of blood flow decreases significantly and blood pressure (especially systolic blood pressure) drops briefly.

[0087] The position / or orientation of the valve assembly or valve annulus assembly is constantly in motion based on the respiratory and cardiac cycles. For example, during rapid pacing, the aortic valve is primarily in the open state, and the position of the valve annulus corresponds to the systolic phase of a normal cardiac cycle, while the respiratory cycle persistently influences the position and / or orientation of the valve annulus.

[0088] The computer implementation method may include processing at least one first and / or at least one second image of a moving heart received from an input interface, such that at least one first and / or at least one second feature identified by the processing unit is arranged relative to at least one first and / or second image in a predetermined spatial relationship.

[0089] This enables the computer-implemented method to more reliably determine the corrected location and / or orientation of the valvular components without taking into account the patient's relative motion during image acquisition. In particular, by processing at least one first and / or second image, the influence of respiratory and cardiac cycles can be reduced, thereby enabling the pre-determined spatial relationships.

[0090] Processing of at least one first image and / or at least one second image can be achieved by applying an image stabilization algorithm. The image stabilization algorithm can be adapted to stabilize / center at least one first / second image by (i) cropping or (ii) padding at least one first / second image.

[0091] Cropping at least one first / second image may include adjusting the boundaries of the first and / or second images according to a predefined template or dynamically, such that at least one first feature and / or at least one second feature are arranged according to a predetermined spatial relationship.

[0092] Filling at least one first / second image may include extending the boundary of at least one first / second image, particularly by adding pixels with predetermined pixel attributes such as color or transparency.

[0093] Alternatively, at least one third feature can be identified by the processing unit, and a predetermined spatial relationship of the third feature relative to the first / second image can be determined. The third feature may consist of calcifications, reference medical devices, or bones (particularly the patient's ribs and / or spine), which can serve as a reference location and are independent of at least one first / second feature.

[0094] Processing at least one first and / or at least one second image to achieve a predetermined spatial relationship based on multiple third features can improve the ability to determine reliable corrected locations and / or orientations.

[0095] Heart valve prostheses can be aortic heart valve prostheses, which are commonly used in TAVI / TAVR procedures.

[0096] The at least one first image can be a preoperative CT image or a preoperative MRI image. This allows the method to overlay image information (such as first and second features) in three dimensions with a second image. The second image can be an intraoperative (particularly real-time) fluoroscopic image. Fluoroscopic images acquired during TAVI / TAVR are typically in different orientations; for example, the C-arm can be in a cephalic (CRA), caudal (Cau), left anterior oblique (LAO), or right anterior oblique (RAO) orientation. Three-dimensional CT images allow for the tracking of anatomical structures or surgical second and / or first features in fluoroscopic images with different orientations, thereby precisely locating these features.

[0097] Alternatively, the first image can also be an intraoperative fluoroscopic image.

[0098] The first feature indicating the location / or orientation of the valve assembly or valve annulus assembly may be formed by calcification and / or vascular plaque, surgical features (e.g., pigtail guidewire / catheter), or the valve assembly or valve annulus assembly itself, which may be visible upon injection of contrast agent. In a preferred embodiment, calcification and / or vascular plaque on the native valve may constitute the first feature.

[0099] The at least one second feature can be a low-density or high-density marker. A low-density marker can be a fat deposit, cyst or fluid accumulation, or an air sac or gas sac. A high-density marker can be calcification and / or vascular plaque, implants or medical devices, bone structures and / or contrast agents.

[0100] The first image may be a normal blood flow image acquired in the absence of rapid pacing, and the second image may be a blood flow reduction image acquired during rapid pacing.

[0101] The second image acquired during rapid pacing preferably provides a more accurate location of the second feature during the actual rapid pacing process, thereby allowing for a more precise determination of the corrected location and / or orientation of the valve assembly or valve annulus assembly (especially the valve annulus). Heart valve prostheses are typically positioned and deployed during rapid pacing, making more precise determination of the corrected location / or orientation a way to avoid adverse effects caused by incorrect positioning / orientation of the heart valve prosthesis.

[0102] The method may include identifying the presence or absence of at least one first feature in at least one second image that indicates the position and / or orientation of a valve assembly or valve annulus assembly. If the first feature is identified as present, a corrected position and / or orientation of the valve assembly or valve annulus assembly is determined in at least one second image based on the first feature. If the first feature is identified as absent, a corrected position and / or orientation of the valve assembly or valve annulus assembly is determined in at least one second image based solely on at least one second feature and at least one static or dynamic geometric relationship and a static or dynamic geometric relationship model.

[0103] This allows the method to selectively optimize the corrected position and / or orientation of a determined valve component or valve annulus component based on whether a first feature is visible in at least one second image. Therefore, if the presence of the first feature can be determined (e.g., based on fluoroscopic imaging combined with contrast agent injection), the first feature indicating the position and / or orientation of the valve component or valve annulus component (e.g., during pacing) can be identified more precisely.

[0104] Alternatively, the first feature can consist of anatomical landmarks, such as calcifications that are at least partially visible without the use of contrast agents. The first feature can also consist of a pre-implanted, existing cardiac valve prosthesis that is visible without contrast agents under fluoroscopic imaging.

[0105] Conversely, even if the first feature is identified as absent, the method still provides a reliable corrected location and / or orientation for the valve assembly or valve annulus assembly. This combination ensures consistent and accurate determination of the corrected location / or orientation regardless of the presence or absence of the first feature.

[0106] The method may include determining in real time, in at least one of a first image and / or a second image, the position and / or orientation of a heart valve prosthesis and / or a delivery device for delivering the heart valve prosthesis, particularly a corrected position and / or orientation relative to a valve assembly or valve annulus assembly.

[0107] By precisely determining the position / orientation of the prosthesis or delivery device, the positional deviation relative to the valve assembly or valve annulus assembly can be determined. This provides valuable data for clinicians to manually or via partially or fully automated robotic systems to navigate and deploy the heart valve prosthesis.

[0108] A first feature in at least one first image can indicate the maximum radial size of the valve annulus. The method may include the step of determining a corrected radial size of the valve annulus in at least one second image based on the maximum radial size. In addition to the maximum radial size, a target radial size may also be determined based on the correction, taking into account the radial size of the valve annulus, particularly during rapid pacing.

[0109] The maximum radial dimension can be determined based on multiple first images showing different radial dimensions of the valve annulus during normal heart rhythm.

[0110] Because the valve annulus is not a rigid structure, its radial dimensions change during the cardiac cycle (i.e., during systole and diastole). Optimal placement of the heart valve prosthesis can be achieved by expanding the prosthesis to correspond to its maximum radial dimension.

[0111] Reliable determination of the target radial dimension, combined with the corrected position and / or orientation of the valve annulus, facilitates the selection and correction of heart valve prosthesis deployment. Improper deployment of heart valve prostheses, such as inappropriate dilation and / or alignment during deployment (e.g., via balloon catheter), can lead to valvular embolism (e.g., based on valve displacement), conduction disturbances, coronary artery obstruction, paravalvular leak / valvular malposition, stroke, or tearing / rupture of the aortic valve annulus.

[0112] At least one defined static or dynamic geometry can indicate the patient's cardiac and / or respiratory cycles. The first static or dynamic geometry may be based on the cardiac cycle, and the second static or dynamic geometry may be based on the respiratory cycle.

[0113] Multiple first images can be acquired, preferably in a sequential manner, to determine the dynamic geometry of the cardiac and / or respiratory cycles, particularly for distinguishing between the cardiac and respiratory cycles. Both the cardiac and respiratory cycles typically influence the motion of the valvular assembly or valvular annulus assembly in a substantially periodic manner at different time scales; therefore, the method can be adapted to determine their respective effects independently of each other based on multiple first images.

[0114] For example, respiratory cycles can be determined based on periodic movements of anatomical or surgical features, such as the respiratory movements of the patient's diaphragm causing cardiac displacement. Cardiac cycles can be determined based on electrocardiograms (ECGs) and / or by cardiac movements of anatomical or surgical features (e.g., pigtail guidewires / catheters).

[0115] Because the cardiac cycle changes, especially during rapid pacing, while the respiratory cycle remains constant, the corrected position / orientation of the valve assembly or valve annulus assembly can be determined in a more reliable manner based on dynamic geometry.

[0116] The at least one second feature may consist of the patient's anatomical features or surgical features (particularly the medical device or the posture of the medical device). One second feature may be an anatomical feature, and the other second feature may be a surgical feature.

[0117] Using multiple secondary features, particularly multiple anatomical features and / or multiple surgical features, allows for increased accuracy and reliability, and reduces error-proneness, such as reducing misidentification of at least one secondary feature. This can be advantageous if some secondary features become less visible / more visible during contrast agent injection. For example, medical devices are typically configured to be radiopaque and visible under fluoroscopy, and therefore become less visible during the injection of a radiopaque contrast agent. Conversely, many anatomical features become more visible under fluoroscopy during contrast agent injection.

[0118] In particular, the orientation of a medical device may be related to the position / or orientation and / or shape of a curved / arc guidewire / catheter (especially a pigtail catheter / guidewire) that is known to those skilled in the art and has been inserted into a patient’s ventricle or noncoronary valve.

[0119] The orientation of a curved / arc guidewire / catheter can be helical. This orientation can continuously deform during the cardiac and / or respiratory cycles and can be directly correlated with the characteristics of the heart's ECG. Therefore, this orientation is particularly useful for determining the corrected position / orientation of valvular assemblies or valvular annular assemblies. Deformations of the medical device or its orientation during pacing and during normal blood flow can be correlated with different ECG characteristics.

[0120] As an alternative to or supplement to commonly used pigtail catheters, guidewires / catheters with coiled / arc distal tips can be used, which are particularly suitable for TAVI / TAVR. The coiled / arc guidewire / catheter in this application can be constructed from commercially available Safari guidewires.

[0121] In a preferred embodiment, a first medical device comprising a pigtail catheter / guidewire is placed in the non-coronary valve, and / or a second medical device comprising a curved / coiled guidewire / catheter is placed in the left ventricle. Placing the pigtail catheter / guidewire in the non-coronary valve of the aortic valve allows for optimized contrast agent injection and hemodynamic monitoring, deflects the medical device away from vulnerable structures, and can also aid in the positioning of the aortic heart valve prosthesis. Placing the curved / arc-shaped guidewire / catheter in the left ventricle provides support for the heart valve prosthesis positioning system and can serve as a navigation aid to navigate the heart's anatomy, while also acting as an anatomical reference point.

[0122] This method can be further adapted to determine the corrected location and / or orientation of the valve assembly or valve annulus assembly without injecting contrast agent at any step of the procedure. This contrast agent-free approach can reduce surgical costs and optimize surgical workflow, making the approach more economical and operationally efficient.

[0123] The method may include processing (preferably in real time) at least one first and / or second image, and including at least one of (i) the corrected position and / or orientation of the valve assembly or valve annulus assembly and (ii) the position and / or orientation of the heart valve prosthesis in (I) preoperative visualization of the heart or (II) a computational model (particularly a finite element model) of the valve assembly or valve annulus assembly.

[0124] This allows for enhanced patient-specific visualization and more spatial information, particularly three-dimensional information, of the valve assembly or annulus assembly and the heart valve prosthesis to be provided to clinicians or robotic control systems, thereby facilitating the execution of surgical procedures, especially TAVI / TAVR. Furthermore, by providing information on the spatial relationship between the valve assembly or annulus assembly and the heart valve prosthesis, it enables preoperative planning, which is crucial for complex procedures such as TAVI / TAVR, and minimizes potential human error, ensuring consistent performance regardless of the clinician's level of expertise.

[0125] The computational model may include at least one static or dynamic geometric relationship, and in particular, may include multiple static or dynamic geometric relationships. The at least one static or dynamic geometric relationship may be determined based on a defined respiratory cycle or cardiac cycle, specifically, a first static or dynamic geometric relationship may be determined based on the respiratory cycle and another static or dynamic geometric relationship may be determined based on the cardiac cycle. These static or dynamic geometric relationships may be used as boundary conditions for the computational model.

[0126] The method may include modified position and / or orientation of the valve assembly or valve annulus assembly and / or cardiac valve prosthesis shown in: (i) in at least one first or second image; (ii) in preoperative visualization; and / or (iii) in a computational model.

[0127] This method enhances visualization by providing dynamics of the valve assembly or valve annulus assembly and / or cardiac valve prosthesis, thereby helping clinicians perform surgery in a precise and accurate manner, resulting in safer surgeries, improved patient outcomes, and reduced operation time.

[0128] The method may include receiving user-generated input, such as via a user interface, to manually adjust the corrected position and / or orientation of a valve assembly or valve annulus assembly (especially the valve annulus), and displaying (preferably in real time) the adjusted corrected position and / or orientation in place of the previous corrected position and / or orientation.

[0129] This method prevents potential computational or procedural errors by enabling clinicians to adjust the results—for example, by interactively selecting locations or regions in the image—allowing them to modify established positions / orientations. Clinicians typically possess tacit knowledge and experience in anatomy and physiology, which can allow them to identify outliers / errors that are not easily detected from computational data.

[0130] The method may include classifying at least one first and / or second image to determine whether the corresponding image was acquired under the action of a contrast agent. The method may further include: if the image was not acquired under the action of a contrast agent, determining at least one feature that is not visible without a contrast agent, based on another feature of at least one first feature and one second feature that is visible without a contrast agent.

[0131] This allows for the reliable identification of features that are not visible or difficult to see under the action of contrast agents, such as radiopaque medical devices or calcifications, or features that are not visible or difficult to see without the action of contrast agents, such as many native anatomical structures, such as the aortic arch or valve annulus.

[0132] The step of using a data processing unit to determine the corrected location and / or orientation of a valve assembly or valve annulus assembly can be based on cardiac ECG.

[0133] In addition to static or dynamic geometric relationships, using ECG to determine the corrected location and / or orientation of valve components or valve annulus components allows for deviation correction, thereby determining the corrected location / or orientation in a more accurate manner.

[0134] Another aspect of the invention relates to a computer program product including instructions that, when executed by a computer, cause the computer to perform the steps of the previously described method.

[0135] Another aspect of the invention relates to a computer-readable storage medium including instructions that, when executed by a computer, cause the computer to perform the steps of the previously described method.

[0136] Another aspect of the invention relates to using the aforementioned computer-implemented method for at least one of the following: (i) planning transcatheter aortic valve implantation or transcatheter aortic valve replacement; (ii) translating, rotating, and / or tilting at least one medical device (particularly a heart valve prosthesis delivery system for navigation and / or deployment of a heart valve prosthesis) along a longitudinal axis; and (iii) optimizing the dosage of contrast agent and / or the time interval for injecting contrast agent into a body catheter.

[0137] Translational, rotational, and tilting of medical devices along the longitudinal axis enables multidimensional optimization of the device's position / orientation relative to the valve assembly or annulus assembly. Therefore, clinicians can adjust the device to entirely different positions / orientations of the valve assembly or annulus assembly during pacing to account for the patient's specific anatomy. This provides additional utility for asymmetric heart valve prostheses and for alignment with typically spherical or oval asymmetric valve or annulus assemblies. Furthermore, this allows for optimized navigation / deployment to prevent potential obstruction of the left or right coronary artery by the prosthesis.

[0138] Optimize contrast agent dosage / injection intervals to reduce costs, improve workflow efficiency, and further prevent adverse reactions to contrast agent injections, such as contrast agent-induced nephropathy.

[0139] Another aspect of the invention relates to a system including means adapted to perform the steps of the previously described computer-implemented method, particularly a data processing unit.

[0140] The data processing unit can be configured to determine the dynamic radial dimensions of a heart valve prosthesis or an expansion unit (particularly including a balloon) for expanding a heart valve prosthesis in real time. The data processing unit is configured to (i) provide expansion control commands for operating the expansion unit, or (ii) provide expansion control instructions to a user to operate the expansion unit for radially expanding the heart valve prosthesis within the valve annulus, particularly until at least the corrected radial dimension of the valve annulus is reached.

[0141] Clinicians can monitor / manipulate the expansion of heart valve prostheses efficiently and in a time-saving manner without experiencing a steep learning curve. Alternatively, expansion can be performed partially or fully automatically. By making the expansion of the prosthesis more controllable, the adverse effects caused by improper deployment of heart valve prostheses, as previously mentioned, can be avoided.

[0142] The system, particularly the system's data processing unit, may include a user interface that allows users to input user input. User input may include: (i) the type of the selected heart valve prosthesis; (ii) the target position and / or orientation of the selected heart valve prosthesis relative to the valve assembly or valve annulus assembly (particularly the valve annulus); and / or (iii) the adjusted corrected position / orientation of the valve assembly or valve annulus assembly (particularly the valve annulus). The system, particularly the data processing unit, may be adapted to determine the target position / orientation based on the user input and the adjusted position / orientation of the valve assembly or valve annulus assembly, and preferably provides data to display the target position / orientation.

[0143] The data processing unit can be configured to determine deployment control instructions based on the corrected position and / or orientation of the valve assembly or valve annulus assembly and the position and / or orientation of the heart valve prosthesis, for operating the delivery device to deploy the heart valve prosthesis in the valve annulus. The control instructions provide instructions regarding at least one (especially all) of the following: (i) adjusting the tilt of the heart valve prosthesis relative to the valve annulus; (ii) adjusting the circumferential orientation of the heart valve prosthesis relative to the valve annulus; and (iii) adjusting the longitudinal position of the heart valve prosthesis.

[0144] This provides guidance for the precise navigation and deployment of heart valve prostheses and minimizes the risk of human error.

[0145] The system may include a pacing unit configured to deliver pacing signals to a target region of the patient's heart near the valve assembly or valve annulus assembly to reduce flow through the valve. The pacing unit may be operated by a data processing unit.

[0146] This could provide an integrated system that incorporates rapid pacing into a single medical system, thereby simplifying surgical time, such as the time required to switch between different medical devices.

[0147] The pacing unit can be positioned on a catheter / guidewire, particularly on a curved / arc-shaped catheter / guidewire, such as a Safari guidewire / catheter. This allows the pacing unit to be inserted at least partially into the heart, particularly the left ventricle, for rapid pacing of the heart, such as at a rate of 180-220 beats per minute or higher.

[0148] However, the pacing unit can also be mounted on a separate medical device adapted for insertion into the right ventricle, for example, via the femoral vein in the groin through the right atrium, to achieve rapid pacing as known to those skilled in the art.

[0149] The pacing unit can be connected to or is compatible with an external pacemaker. This allows the pacing unit to be easily discarded after medical procedures.

[0150] The data processing unit may be configured to receive at least one post-deployment image after the heart valve prosthesis has been deployed, and to identify and verify in at least one first image the position and / or orientation of the heart valve prosthesis relative to a first feature indicating the position and / or orientation of the valve assembly or valve annulus assembly.

[0151] If the heart valve prosthesis is abnormally positioned, this method allows for immediate corrective action to reduce the risk of complications. By verifying the position / orientation of the heart valve prosthesis obtained pre-implantation relative to the primary characteristic, particularly during normal flow (i.e., without rapid pacing), discrepancies indicating abnormal positioning can be identified that may lead to the previously mentioned adverse reactions, such as valve displacement or paravalvular leak.

[0152] The data processing unit may be configured to: (i) control the imaging device positioning unit for positioning and / or orienting the imaging device, or (ii) provide user instructions on how to operate the imaging device positioning unit. The imaging device positioning unit may include or be composed of a C-arm.

[0153] The integrated collaboration between the data processing unit and the imaging device positioning unit helps to execute procedure steps in a precise and efficient manner, thereby optimizing surgical outcomes and simplifying the procedure.

[0154] The imaging device can be an external imaging device, particularly an external X-ray imaging device, such as a C-arm. Alternatively, the imaging device can be an in vivo imaging device, particularly an in vivo imaging device adapted for intrathoracic imaging, especially a delivery device attached to a heart valve prosthesis. Alternatively, the imaging device can also be a transesophageal echocardiography (TEE) imaging device.

[0155] Another aspect of the invention relates to a system for guiding the placement of a heart valve prosthesis within the valve annulus, particularly the system as described above. The system includes a data processing unit adapted to display in real-time on a display device the position and / or orientation of the heart valve prosthesis relative to the reduced blood flow image of the valve assembly or valve annulus assembly.

[0156] The corrected position and / or orientation of the valve assembly or valve annulus assembly can be based on the already corrected annulus orientation to take into account the position and / or orientation of the annulus during rapid pacing.

[0157] The system (in particular the system’s data processing unit) can be adapted in the manner previously described to determine the corrected position and / or orientation, i.e., based on static or dynamic geometric relationships and second features identified in the second image.

[0158] The data processing unit can be adapted to generate in real time visual markers (especially color-coded markers) indicating the relative position and / or orientation of the heart valve prosthesis, and / or instructions regarding the desired orientation and / or position of the heart valve prosthesis within the valve annulus. The visual markers or instructions may be based on at least one, and especially all of the following: (i) the relative tilt of the heart valve prosthesis's orientation relative to the modified orientation of the valve annulus; (ii) the relative circumferential orientation of the heart valve prosthesis relative to the modified orientation of the valve annulus; and (iii) the relative longitudinal position of the heart valve prosthesis relative to the modified position of the valve annulus.

[0159] Other embodiments of the present invention and improvements to the described embodiments will become apparent in the following description of the embodiments. Attached Figure Description

[0160] The present invention will now be described with reference to certain embodiments and accompanying drawings, which illustrate:

[0161] Figure 1A and 1B Fluorescent fluoroscopic images of the heart during systole and diastole in the absence of rapid pacing during TAVI / TAVR procedures;

[0162] Figure 2A and 2B Fluorescent fluoroscopic images of the heart during rapid pacing in TAVI / TAVR procedures;

[0163] Figure 3A and 3B CT and fluoroscopic images of the heart during TAVI / TAVR surgery;

[0164] Figure 4A and 4B Two enhanced fluoroscopic images of the heart during TAVI / TAVR surgery;

[0165] Figure 5A and 5B Fluorescent fluoroscopic images of the heart acquired during TAVI / TAVR procedures in the absence of rapid pacing, and magnified portions of those images (including visual indicators).

[0166] Figure 6A and 6B Fluorescent images of the heart during rapid pacing in TAVI / TAVR surgery, showing the heart valve prosthesis in a contracted state and in a partially dilated state, respectively.

[0167] Figure 7 : A three-dimensional model of the aortic arch, including the valve annulus registered with fluoroscopic images during TAVI / TAVR;

[0168] Figure 8A The first CT image includes a schematic representation of a pigtail catheter deployed within the patient's noncoronary valve.

[0169] Figure 8B : A second CT image showing a schematic representation of a coiled / arc-shaped guidewire deployed in the patient's left ventricle;

[0170] Figure 8C The following is a display based on: Figure 8A and 8B Fluorescent images of deployed pigtail catheters and deployed curved / arc guidewires;

[0171] Figure 9A and 9B A combined three-dimensional computational model based on CT images and at least two fluorescence fluoroscopic images acquired in different orientations;

[0172] Figure 9C :Include Figure 9A and Figure 9B Fluorescent perspective images of three-dimensional computational model information;

[0173] Figure 9D : A schematic representation of a system including a data processing unit for performing the method according to the invention;

[0174] Figure 10A and 10B A two-dimensional virtual schematic representation of a heart valve prosthesis and an aortic valve annulus, the aortic valve annulus having visual markers indicating the circumferential orientation of the heart valve prosthesis and visual markers indicating the circumferential orientation of the valve annulus.

[0175] Figure 10C A three-dimensional schematic representation of the heart valve prosthesis and the aortic root, indicating the tilt of the heart valve prosthesis relative to the valve annulus;

[0176] Figure 11 : A schematic side view of the aortic valve complex, which includes the three valves of the native aortic heart valve;

[0177] Figure 12A and 12D : An example training image of the input training dataset, which shows the location and deployment of a heart valve prosthesis in the aortic valve during a TAVI intervention;

[0178] Figure 13A The corrected position of the noncoronary flap, determined by a pre-trained machine learning model, is superimposed on the fluorescence fluoroscopic image.

[0179] Figure 13BThe corrected location of the noncoronary lobes and the schematic segmentation of the aortic root, determined by a pre-trained machine learning model, are superimposed on fluorescence fluoroscopic images.

[0180] Figure 13C and 13D : Shows the way Figure 13A and 13B The corrected position of the noncoronary valve predicted by the pre-trained machine learning model is displayed as a histogram relative to the labeled aortic valve annulus.

[0181] Figure 14A and 14B During TAVI intervention, the corrected position of the noncoronary flap, the position and orientation of anatomical features, and the position and orientation of instrument features superimposed on fluoroscopic images; and

[0182] Figure 14C A schematic diagram illustrating how heart and respiratory rhythms change over time. Detailed Implementation

[0183] Figure 1A and 1B First fluorescence fluoroscopic images of the heart during diastole and systole during TAVI / TAVR surgery are shown 4.

[0184] The average heart rate is approximately 60 to 80 beats per minute. During normal cardiac function, the sinoatrial (SA) node initiates atrial contractions via electrical signals, which in turn initiates ventricular contractions for normal blood flow. Therefore, an ECG typically displays a predictable pattern consisting of P waves, QRS complexes, and T waves. The P wave corresponds to atrial contraction, the QRS complex represents ventricular contraction, and the T wave indicates ventricular recovery. This cardiac cycle can be monitored via ECG using this system.

[0185] Figure 1A and 1B A medical device is shown, comprising a Safari guidewire 101 and a pigtail catheter 102. The pigtail catheter 102 is advanced through the aorta and positioned in the non-coronary valve of the patient's aortic arch. The Safari guidewire 101 is controlled to pass through the valve into the left ventricle to support TAVI / TAVR procedures, i.e., advancing a heart valve prosthesis along the Safari guidewire 101.

[0186] Figure 1A and 1B It is further shown that the Safari guidewire 101, which is in contact with or near the apex of the left ventricle of the heart during normal cardiac function, often deforms. Figure 1A The image shows the heart in diastole, when the left ventricle relaxes and fills with blood. Figure 1B The image shows a heart in systole, where the ventricular walls contract and exert force on the Safari guidewire.

[0187] During a normal cardiac cycle, deformation of the medical device (particularly the Safari guidewire 101) and the first feature (e.g., the pigtail catheter 102) can be monitored to determine the dynamic geometry between the Safari guidewire and the first feature. The position and / or orientation of the pigtail catheter 102 directly indicates the position / orientation of the annulus near the non-coronary valve. Alternatively or additionally, the dynamic geometry can be determined by monitoring the movement of the Safari guidewire 101 and a first feature different from the pigtail catheter 102 (e.g., calcification near the annulus). Furthermore, contrast agents can be injected to directly monitor the dynamic behavior of the annulus and the Safari guidewire or anatomical features to determine static or dynamic geometry.

[0188] Based on mathematical models (such as pre-trained machine learning models), static or dynamic geometric relationships can be determined, taking into account the position and / or orientation of the valve annulus during rapid cardiac pacing. Therefore, this model is named the static or dynamic geometric relationship model.

[0189] Machine learning models, particularly convolutional neural networks, are preferably trained using fluoroscopic images of a patient population acquired during normal cardiac cycles and fluoroscopic images acquired during rapid pacing. These training / validation images can be acquired during contrast agent injection to accurately determine the position and / or orientation of the valve annulus in both images, thus obtaining the most reliable results. To label baseline truths, the position / or orientation of the valve annulus and the positions / or orientations of first and second features can be labeled, for example, manually or automatically by a clinician, so that these baseline truths can be used during training and validation. The correct position / or orientation of the valve annulus in the absence of rapid pacing and during rapid pacing, as well as the positions / or orientations of the first and second features, are used for training and validating the model in a manner known to those skilled in the art.

[0190] In a preferred embodiment, the training images include anatomical and / or surgical features as first and second characteristics. The model is preferably adapted to track static or dynamic geometric relationships between these features so that the corrected position / orientation of the valve annulus can be determined based solely on these static or dynamic geometric relationships. This allows the model to be trained to correlate static / dynamic geometric relationships acquired during a normal cardiac cycle and to determine the corrected position / orientation of the valve annulus that it presents / will present during rapid pacing.

[0191] Based on multiple first fluoroscopic images acquired during a normal cardiac cycle, the dynamic geometric relationship between the first feature / valve annulus and the second feature is determined. This dynamic geometric relationship, along with the position / orientation of the valve annulus, is then input into the model to determine the corrected position and / or orientation of the valve annulus during rapid pacing, even if the first feature indicating the position / or orientation of the valve annulus is not visible.

[0192] Figure 2A and 2B Fluorescent fluoroscopic images of the heart during rapid pacing in a TAVI / TAVR procedure are shown 5. Figure 2A and 2B A delivery device with an implanted heart valve prosthesis 3 is also shown, which is advanced through the valve annulus 2 for deployment during rapid pacing. A pigtail catheter 102 and a Safari guidewire 101 are used in conjunction with... Figure 1A and 1B Similar placement methods are used. Rapid pacing is typically performed at a rate of 180 to 220 beats per minute, which reduces blood flow and pressure, providing a more controlled environment for the deployment of the aortic valve prosthesis 3. This environment reduces the risk of valve displacement, embolism, or abnormal placement during deployment. During rapid pacing, the diastolic phase is significantly shortened, preventing the ventricles from fully filling, which differs from the situation under normal heart rhythm / blood flow. Figure 2A An example is shown with a slightly shortened contraction period. Figure 2B This shows a significantly shortened diastolic period. Figure 2A In the image, the position of valve annulus 2 is visible due to the injection of contrast agent. The injection of contrast agent allows the position of valve annulus 2 to be directly tracked as a first feature 7, indicated by the black dashed line. As a second feature, the attitude of Safari guidewire 101, which continuously deforms in the absence of rapid pacing and during rapid pacing, is monitored.

[0193] During rapid pacing, the position of valve annulus 2 is more frequently and to varying degrees in a contracted state, thus exhibiting a different position / orientation compared to that under normal cardiac cycles based on reduced blood flow and cardiac pressure. This can be achieved using... Figure 2A and 2B Multiple similar fluorescence imaging images were used to train / validate the previously described machine learning model.

[0194] Figure 3A and 3B CT images 42 and fluoroscopic images 5 of the heart during TAVI / TAVR surgery are shown, in which the second feature 8 and the first feature 7 valve annulus indicating the location and / or orientation of the valve annulus 2 are highlighted.

[0195] Figure 3A The second feature 8 and the first feature 7 identified in the CT image 42 are shown, thereby allowing the determination of static geometric relationships via the identified features 7 and 8. Figure 3A and 3BThe second feature 8 is marked with a white box, and the first feature 7 is marked with a black dashed line. The second feature 8 consists of low-density markers in CT image 4. The first feature 7 consists of the valve ring 2 visible in CT image 42. Figure 3A The CT image 42 was acquired under normal blood flow. The CT image 42 allows for the localization of features 7 and 8, and thus the location and orientation of the valve annulus 2 in three dimensions.

[0196] Figure 3B Feature matching based on the second feature 8 in the first CT image 42 is shown, and the same second feature 8 is also identified in the second fluoroscopic image 5. The first feature 7, indicating the location of the valve annulus 2, is not visible in the second fluoroscopic image 5. The absence of the first feature 7 is detected, and its position / orientation is determined and projected into the second image 5 based on the static geometric relationship between the identified features 7 and 8. Figure 3B The first feature 7 is marked with a black dashed line.

[0197] The second feature 8 is used to change the position / orientation of the imaging device (e.g., a C-arm) for the deployment of a heart valve prosthesis.

[0198] However, during rapid pacing that reduces blood flow, the position / orientation of valve annulus 2 deviates from the position of the valve annulus that should be determined during normal blood flow.

[0199] To counteract this bias, a machine learning model was used, trained on multiple images acquired during rapid pacing, both during normal blood flow and during periods of reduced blood flow. This model was configured to correct for established static geometry, taking into account the position / orientation of the valve annulus during rapid pacing, and based on… Figure 3B The static geometric relationship shown determines the corrected position and / or orientation 20 of the lobe ring 2.

[0200] Figure 3A and 3B The corrected position and / or orientation 20 shown in CT image 4 or fluoroscopic image 5 are illustrated to provide guidance for the navigation and deployment of the heart valve prosthesis via pigtail catheter 102 and Safari catheter 101.

[0201] Figure 4A and 4B Two fluoroscopic images of the heart during TAVI / TAVR surgery are shown 4 and 5, in which the second feature 8 and the first feature 7, indicating the position / orientation of the valve annulus 2, are highlighted. Figure 3A and 3B Similarly, the first fluoroscopic image 4 was acquired without rapid pacing, while the second fluoroscopic image 5 was acquired during rapid pacing. Figure 4A and 4BThis illustrates how the corrected position / orientation 20 of the valve annulus 2 is determined, taking into account rapid pacing, based on the static geometric relationship between the first feature 7 and two second features 8 that indicate the position / orientation of the valve annulus 2. To show that the corrected position / orientation 20 of the valve annulus differs in the absence of pacing and during pacing, Figure 4A and 4B The corrected position / orientation 20 is included in all of them.

[0202] Figure 4A and 4B The second feature 8, marked by a larger white rectangle, consists of low-density anatomical landmarks. Preferably, the first feature 7 and / or the second feature 8 are selected in a partially or fully autonomous manner to reduce the workload of clinicians. Figure 4A and 4B Another second feature 8, marked by a smaller white rectangle, is a surgical feature consisting of a pigtail catheter 102. A Safari guidewire 101 is further deployed in the left ventricle to support and guide the passage of the heart valve prosthesis 3 through the annulus and to deploy the prosthesis 3.

[0203] In a preferred embodiment, multiple first fluoroscopic images 4 are used to determine the dynamic geometric relationship between the first feature 7 and the second feature 8. Preferably, multiple first images 4 are used to distinguish the effects of the respiratory cycle and the cardiac cycle on the motion of features 7 and 8. This improves the accuracy and reliability of the corrected position / orientation 20 of the valve annulus 2 during rapid pacing. During rapid pacing or in the absence of rapid pacing, the effect of the respiratory cycle remains substantially the same, while the effect of the cardiac cycle changes significantly. Preferably, the corrected position / orientation 20 is directly displayed in real time in the intraoperative fluoroscopic image 5.

[0204] Alternatively, tracking similar to Figure 3A and 3B The dynamic / static geometric relationship between two or more anatomical features, or the dynamic / static geometric relationship between two or more surgical features, is used to determine the correction location / orientation20.

[0205] In a preferred embodiment, the dynamic geometry of the position of the Safari guidewire 101 introduced into the left ventricle relative to the pigtail catheter 102 placed on the non-coronary valve is tracked over time, particularly in conjunction with ECG. Because the medical device is very close to the valve annulus 2 and directly related to cardiac function, these surgical features formed by the medical device are particularly useful for predicting the motion of the valve annulus 2.

[0206] Figure 4B Further illustration shows the delivery device for delivering the heart valve prosthesis 3, which has been introduced via the aortic arch. In a preferred embodiment, the position / orientation of the heart valve prosthesis 3 in images 4 and 5 (see...) Figure 4A The corrected position / orientation 20 is simultaneously tracked to support the navigation / deployment of the heart valve prosthesis 3 and reduce the workload of clinicians (see [link]). Figures 5A-6B ).

[0207] Figure 5A and 5B The image shown is a fluorescence fluoroscopic image 4 during TAVI / TAVR surgery in the absence of rapid pacing, and a magnified portion of the image, which includes the corrected position / orientation of valve annulus 20 and the position / orientation of valve annulus 2.

[0208] Figure 5A and 5B The image shows a delivery device carrying an aortic heart valve prosthesis 3 being advanced through the aortic arch, with a pigtail catheter 102 placed in the noncoronary valve and a Safari guidewire 101 placed in the left ventricle.

[0209] Figure 5A and 5B The position of the first feature 7, which indicates the location / orientation of the valve annulus 2, is determined in a manner previously described, for example, by tracing a second feature consisting of anatomical and / or surgical features. Similarly, a machine learning model is used to determine the corrected location / orientation 20 of the valve annulus 2.

[0210] Figure 5A and 5B The corrected position / orientation 20 of the valve annulus 2, adjusted to account for rapid pacing, is shown displaced in the anterograde direction relative to the position / orientation 20 of the valve annulus 2 during normal flow in the absence of rapid pacing. Fluorescence imaging 4 was acquired during the injection of a radiopaque contrast agent via a pigtail catheter 102, thus the valve annulus 2 and the aortic arch are visible and can be reliably traced. However, by providing the corrected position / orientation 20 of the valve annulus 2 during pacing, the additional injection of contrast agent during rapid pacing can be avoided.

[0211] In a preferred embodiment, the method includes determining whether a first feature 7 formed by the valve annulus 2 is directly visible (e.g., due to the injection of contrast agent) or whether the first feature 7 is invisible. If the first feature is identified as present, the method includes determining a correction position / orientation 20 based on the identified valve annulus 2. Otherwise, the method determines this via a static or geometric relationship between the first feature 7 and at least one second feature as described above.

[0212] Figure 5A and 5B Furthermore, in addition to correcting the position / orientation 20, the method also includes determining the maximum radial dimension of the valve annulus 2 in the absence of rapid pacing. Figure 5A and 5BThe maximum radial dimension is depicted as the length of the dashed line, which indicates the first feature 7 of the indicator lobe 2.

[0213] Alternatively, the method can also determine the position / orientation of the valve annulus 2 and the corrected position / orientation 20 of the valve annulus 2 solely based on tracking the pigtail catheter 102 and the Safari guidewire 101 as first and second features. Both the pigtail catheter 102 and the Safari guidewire 101 are visible without the need for contrast agents. This reduces the need for any contrast agent injection.

[0214] Figure 6A and 6B Fluorescent images of the heart during rapid pacing in TAVI / TAVR surgery are shown 4, in which the heart valve prosthesis 3 is in a contracted state 31 and the heart valve prosthesis is in a partially expanded state 32.

[0215] The delivery device, equipped with the artificial heart valve prosthesis 3, was guided via fluoroscopy through the femoral artery and across the aortic arch to reach the aortic valve annulus. The pigtail catheter 102 was placed in the non-coronary valve, and the Safari guidewire was placed in the left ventricle. The pacing lead was inserted through the right femoral vein and advanced into the right ventricle of the heart to deliver rapid pacing (…). Figure 6A and 6B (not shown in the image), allowing the heart valve prosthesis 3 to pass through the valve annulus.

[0216] Once the delivery device is introduced, the boot software used to execute the computer-implemented method is activated.

[0217] The distal tip of the pigtail catheter 102 was detected and tracked in multiple fluoroscopic images in the absence of rapid pacing. The valve annulus was also detected and tracked in these images, and the dynamic geometric relationship between the pigtail catheter 102 constituting the second feature and the valve annulus 2 constituting the first feature was determined. Based on a dynamic geometric relationship model, this relationship was modified to account for valve annulus displacement due to pressure changes and contraction during rapid pacing. This modification was performed by a pre-trained machine learning model that received the position / orientation of the second feature and the dynamic geometric relationship as input. The C-arm used to acquire the fluoroscopic images was positioned in a single known orientation (e.g., a standard bilobal view) during the detection and tracking of these features.

[0218] The method includes a fluoroscopic image 5 displayed during pacing and a corrected position / orientation 20 of the valve annulus, which in... Figure 6A and 6B It is not visible in the fluorescence imaging 5.

[0219] The system used to perform this method also provides a user interface in which the user can select the type of valve and determine the target implantation depth 18 of the selected heart valve prosthesis 3, for example, in a fully automatic manner based on a lookup table.

[0220] The user interface is adapted to receive user input to correct the target valve implantation depth, for example, by selecting the spatial location / orientation on a display device.

[0221] Figure 6A and 6B Further illustration shows the detection and tracking of the prosthetic heart valve 3 in fluoroscopic image 5. The valve-specific target implantation depth 18, determined by the valve prosthesis's position / orientation, is indicated by a white dashed line and displayed in real-time in fluoroscopic image 5. This assists clinicians, or can be performed fully automatically, to align the valve implantation depth 18 with the corrected position / orientation 20 of the valve annulus. The system can optionally include instructions on how to best achieve alignment.

[0222] exist Figure 6B In the preferred embodiment shown, the delivery device has an expansion unit consisting of a nylon balloon catheter 12. The balloon catheter 12 is radiopaque and therefore invisible under fluorescence fluoroscopy. However, partially expanded balloon catheter 12 is marked with a light gray elliptical area to indicate that the balloon catheter 12 is partially expanded.

[0223] The system detects and displays the balloon's expansion status on the monitor (e.g., Figure 6B (75% of the balloon's capacity), allowing clinicians or robotic controllers to precisely control the expansion state. The balloon's expansion state can be detected by using radiopaque fluid for balloon expansion or by monitoring the expansion of the normally radiopaque heart valve prosthesis 3. To further guide the expansion of the heart valve prosthesis 3, Figure 6B The black dashed line in the figure shows the corrected position / orientation 20 of the valve annulus, and also shows the target radial dimension of the valve annulus indicated by its longitudinal length. The target radial dimension of the valve annulus is determined based on the maximum radial dimension of the valve annulus monitored in fluoroscopic images during normal cardiac blood flow (see [reference]). Figure 5A and 5B ).

[0224] After successful deployment of the heart valve prosthesis 3 using balloon catheter 12, the system verifies the deployment position / orientation of the heart valve prosthesis 3. The valve's position / orientation can be verified by detecting and tracking the heart valve prosthesis in the absence of rapid pacing in post-deployment fluoroscopic imaging, for example, by comparing its position / orientation with that of the valve annulus. The delivery device can then be removed.

[0225] Figure 7A three-dimensional model 19 of the aortic arch is shown, including the valve annulus 2 registered with two-dimensional fluoroscopic images during TAVI / TAVR. The fluoroscopic images include a delivery device (on which a heart valve prosthesis 3 is mounted), a deployed pigtail catheter 102, and a deployed Safari guidewire 101. A pacing lead 22 is introduced into the left ventricle to provide rapid pacing during passage through the valve annulus 2. Contrast agent can be injected via the pigtail catheter 102 for detection and tracking of the valve annulus during the procedure.

[0226] Then, the software used for bootstrapping and tracking can be activated. This generates... Figure 7 A three-dimensional model 19 of the patient is shown, and then registered with two-dimensional fluoroscopic images acquired in real time during contrast agent injection intraoperatively. The three-dimensional model 19 includes the valve annulus 2, hinge point, and commissure, and their corresponding three-dimensional positions / or orientations. In a preferred embodiment, the three-dimensional model 19 includes the positions of the right coronary artery ostium, the left coronary artery ostium, and / or the positions of the left coronary-noncoronary, right coronary-left coronary, and noncoronary-right coronary commissures between the valves. Instructions are provided to the user via software on how to optimize the projected view of the fluoroscopic images, and the user or robot controller operates a C-arm to acquire the fluoroscopic images for optimal projection.

[0227] Figure 7 This further demonstrates that the position / orientation of the heart valve prosthesis 3 is detected and tracked in real time and included in model 19. Users can adjust the target implantation depth of the heart valve prosthesis 3 via a user interface (see [link to model 19]). Figure 6A and 6B The software identifies and displays the 3D model 19 as described above (see [link]). Figure 6A and 6B The corrected position / orientation of the valve annulus and the target implantation depth during pacing. Figure 7 (Not shown in the image). Alternatively, the 3D model 19 is corrected to account for rapid pacing. In another alternative approach, the 3D model 19 is registered with fluorescence fluoroscopy images acquired during rapid pacing of the heart. Based on the registration of the 3D model 19 and the fluorescence fluoroscopy images, the system determines a more precise position / orientation of the valve annulus relative to the target implantation depth of the heart valve prosthesis 3. In particular, the circumferential position of the heart valve prosthesis 3 relative to the corrected position / orientation of the valve annulus during rapid pacing can be determined. The 3D model 19 may also include the tilt of the longitudinal axis of the heart valve prosthesis 3 relative to the radial orientation of the valve annulus, thereby allowing for more precise deployment.

[0228] The user can further adjust the corrected position / orientation of the valve annulus 2 via user input. This allows potential calculation or surgical errors to be overcome / resolved based on the clinician's expertise. The system then adjusts the target implantation depth accordingly and displays it on the monitor. In a preferred embodiment, the system is also adapted to detect the expansion status of the heart valve prosthesis 3, such as... Figure 6B The described procedure provides guidance during the deployment of the heart valve prosthesis 3 via a balloon catheter and validates the deployment of the heart valve prosthesis 3.

[0229] Figure 8A The first CT image shows a schematic representation of a pigtail catheter 102 deployed within the noncoronary valve of a patient. Figure 8B A second CT image showing a schematic representation of an arc-shaped / coiled guidewire 101 deployed in the left ventricle of a patient is shown. Figure 8C The display shows according to Figure 8A and 8B Fluorescent imaging of the deployed pigtail catheter 102 and the deployed curved / arc guidewire 101. For ease of illustration, in Figures 8A to 8C The location of the 2nd valve ring is included. Figure 8A A pigtail catheter 102 is shown positioned directly adjacent to the valve annulus 2. A computer-implemented method can be configured to identify and track the pigtail catheter 102, particularly its shape, as a first feature to indirectly determine the position / orientation of the valve annulus 2. Alternatively, calcifications on the aortic valve / valve annulus 2 can be identified and tracked as a first feature. The computer-implemented method can also track the shape of the arcuate / coiled guidewire 101, particularly the arcuate / coiled guidewire 101 within the left ventricle, as a second feature. In the absence of rapid pacing, the dynamic geometric relationship between these two features can be utilized to determine the position of the valve annulus 2. Based on a machine learning model, this dynamic geometry can be adapted to account for the reduction in pressure / blood flow during rapid pacing to reliably determine the corrected position / orientation of the valve annulus. Execution is based on... Figures 8A-8C The computer-based implementation method does not require the injection of contrast agents because the aforementioned features are visible under fluorescence fluoroscopy.

[0230] Figure 9A and 9B A combined three-dimensional computational model 191 based on CT images and at least two fluoroscopic images with different orientations is shown. This three-dimensional computational model 191 is generated based on the segmentation of rigid anatomical structures (e.g., bones) of the patient in the CT and fluoroscopic images. Subsequently, the two-dimensional and three-dimensional segmentation information using the bones is projected to register the two-dimensional fluoroscopic images and the CT images, thereby achieving precise alignment. The accuracy of the three-dimensional computational model 191 and the alignment can be improved by using two fluoroscopic images with different orientations.

[0231] Figure 9C It shows that it contains Figure 9A and 9B The three-dimensional computational model 191 information of fluorescent perspective image. Figure 9C A pigtail catheter 102 is shown positioned within a noncoronary valve. The position / orientation and shape of the pigtail catheter 102 are detected and tracked in multiple fluoroscopic images as a first feature indicating the valve annulus position / orientation. Segmented skeletal or other anatomical features or surgical features in the images or 3D model 191 can be used as a second feature to determine the dynamic geometry as previously described, thereby allowing for the determination of the corrected position / orientation of the valve annulus during rapid pacing.

[0232] Based on these characteristics, a C-arm can be operated (see...) Figure 9D Alternatively, it can provide operational instructions to clinicians. The position / orientation of the heart valve prosthesis 3, particularly its three-dimensional representation, can then be identified and tracked, and included in the three-dimensional computational model 191. Subsequently, using the corrected position / orientation of the valve annulus, the heart valve prosthesis 3 is deployed to the target implantation depth. The presence of the three-dimensional model 191 provides additional information on the orientation of the heart valve prosthesis 3, allowing for correction of the tilt and circumferential alignment of the heart valve prosthesis 3 relative to the valve annulus.

[0233] Figure 9D A schematic representation of system 201 is shown, which includes a data processing unit 6 for performing a method to determine the corrected position / orientation of the valve annulus during rapid pacing. System 201 is connected to a display device 17, which displays in real time the corrected position / orientation of the heart valve prosthesis 3 relative to the valve annulus.

[0234] like Figures 10A-10C As further described in detail, system 201 is also adapted to display the tilt and circumferential orientation of the prosthesis 3 relative to the valve annulus correction position / orientation 20. System 201 is adapted to provide control commands to clinicians or robot controllers to adjust the tilt, circumferential orientation, and longitudinal position of the prosthesis 3 so that the prosthesis 3 is aligned with the valve annulus 2 during rapid pacing.

[0235] The data processing unit 6 is also adapted to operate the C-arm 16 for positioning the imaging unit to acquire fluorescence fluoroscopic images, thereby placing the C-arm in an optimal projection position and allowing it to move between forward / backward, right-oblique, or left-oblique positions, as well as tilting towards the head or tail in a manner known to those skilled in the art. The C-arm is movable in four or more degrees of freedom. Alternatively, the data processing unit 6 may provide user instructions on the display 17 only regarding how to obtain the optimal projection view and how to operate the C-arm.

[0236] Figure 9DThe data processing unit 6 further illustrates the expansion states 31 and 32 of the heart valve prosthesis 3 displayed on the display 17 (see [reference]). Figure 6A and 6B The balloon catheter 12 is displayed on the monitor 17 in real time with instructions on how to operate it for deploying the prosthesis 3. These instructions are based on the target radial dimension determined in a manner previously described (see [link to previous description]). Figure 6A and 6B ).

[0237] Figure 10A and 10B A two-dimensional virtual schematic representation of a heart valve prosthesis and annulus is shown, having a visual marker 33 indicating the circumferential orientation of the heart valve prosthesis 3 and a visual marker 21 indicating the circumferential orientation of the aortic annulus.

[0238] The circumferential orientation of a heart valve prosthesis can be determined by the non-rotationally symmetric radiopaque design of the medical device, particularly by the radiopaque indicator of the heart valve prosthesis or by the design of the heart valve prosthesis itself.

[0239] The circumferential orientation of the valve annulus can be determined based on the valves of the native aortic valve annulus in fluorescence fluoroscopy images, for example, based on a double-valve view.

[0240] The visual marker 33 indicating the circumferential orientation of the heart valve prosthesis 3 was determined based on non-angiographic images using the previously described method. The visual marker 21 indicating the circumferential orientation of the valve annulus was determined based on angiographic images using the previously described method.

[0241] The orientation of these visual markers 33, 21 relative to each other allows clinicians or robot controllers to simplify the circumferential alignment of the heart valve prosthesis relative to the valve annulus (especially relative to the aortic valve). The method may include the following steps: calculating these visual markers 33, 21 via the previously described processing unit, and displaying these visual markers 33, 21 as visual indicators on the display device 17 (see [link to documentation]). Figure 9D ).

[0242] like Figure 10A As shown, visual markers 33 and 21 are misaligned, while Figure 10B Visual markers 33 and 21 are shown to be substantially perfectly aligned, indicating the target orientation / target location of the heart valve prosthesis. These visual markers 33 and 21 can aid in the deployment of the heart valve prosthesis within the aortic valve annulus. Visual markers 33 and 21 may be color-coded or have specific shapes to facilitate the alignment of the heart valve prosthesis with the annulus.

[0243] Figure 10C Three-dimensional schematic computational models of the heart valve prosthesis 3 and the aortic root are shown 19 and 191 (see Figure 7 or Figures 9A-9C The visual markings 33 and 21 of the heart valve prosthesis 3, which indicate the tilt of the heart valve prosthesis 3 relative to the valve annulus 2, are used to indicate the circumferential orientation of the heart valve prosthesis 3 relative to the valve annulus 2. In addition to the previously described visual markings 33 and 21 of the heart valve prosthesis 3 (indicating the circumferential orientation of the heart valve prosthesis 3 relative to the valve annulus 2), this three-dimensional representation also includes a color coding scheme.

[0244] In current medical imaging, the color coding scheme indicates the tilt T of the heart valve prosthesis 3 relative to its target position within the annulus 2. This color coding scheme also indicates the longitudinal position of the heart valve prosthesis 3 relative to the annulus 2. In addition to the color coding scheme, arrows can also indicate directional information. This allows for reliable placement of the aortic valve prosthesis within the annulus 2 and reduces the demands on clinicians during TAVI / TAVR procedures, eliminating the need for extensive learning. Correct placement can further improve hemodynamics, prevent paravalvular leaks, valve displacement, or embolism, maximize implant lifespan, and optimize overall patient outcomes.

[0245] Figure 11 A side view schematic diagram of the aortic valve complex is shown, which includes three valves (cusp): the noncoronary valve 55 of the native aortic heart valve, the right coronary valve 56, and the left coronary valve 57. Figure 11 The aortic valve complex is shown in a three-valve view, making all three valves 55, 56, and 57 visible. Figure 11 The tips of valves 55, 56, and 57 are highlighted with an asterisk. During TAVI / TAVR procedures, a pigtail catheter, typically used for injecting contrast agents, can be placed in non-coronary valve 55 without obstructing the coronary artery. Therefore, non-coronary valve 55 may be easily identifiable due to the higher concentration of contrast agent.

[0246] Figure 12A-12D Training image 51 from the input training dataset is shown, illustrating the location and deployment of the heart valve prosthesis 3 within the aortic valve during a TAVI intervention. Training image 51 is used to train a machine learning model to predict and output the locations of non-coronary valves, specifically the right coronary fossa 56 and the left coronary fossa 57.

[0247] Training image 51 was labeled to include the positions of the upper edge 34 and lower edge 35 of the medical device comprising the heart valve prosthesis 3. Because the stent has high radiopaqueness, the heart valve prosthesis 3 is visible without the need for contrast agent injection. Therefore, the length of the heart valve prosthesis 3 can be provided when training the machine learning model. Figure 12A-12D The distance between the top edge 34 and the bottom edge 35, indicated by the circle and rectangle respectively.

[0248] Figures 12A-12CThe training image 51 also includes a label that identifies the pigtail duct 54 and highlights it with a triangle. The pigtail duct 54 is typically positioned near the non-coronary lobule 55, so that its position / orientation in the training image 51 can indicate the non-coronary lobule 55.

[0249] Training images 51 are further labeled by clinicians based on contrast agent injection, echocardiography (especially transthoracic or transesophageal echocardiography), cardiac MRI, or CT (especially ECG-gated CT). In this way, features (such as valves 55, 56, and 57 of the aortic valves) can be labeled for training a machine learning model to reliably identify locations throughout the various steps of the interventional procedure. Figure 11 similar, Figure 12B-12C In the aortic valve, valves 55, 56, and 57 are marked with an asterisk.

[0250] The first machine learning model can be specifically trained on angiographic images, making it more accurate to determine the location / orientation of non-coronary flaps (especially additional anatomical / instrumental features) when the model is applied to angiographic images.

[0251] The second machine learning model can be specifically trained on non-contrast images, making it more accurate to determine the location / orientation of the non-coronary flap (especially additional anatomical / instrumental features) when the model is applied to non-contrast images.

[0252] Additional instrument features and / or anatomical features can be marked in training image 51, especially Figure 12A and 12B The pre-shaped guidewires shown are, for example, the SAFARI guidewire (which is typically introduced into the left ventricle of the heart during TAVI / TAVR), and / or the outlines of the coronary arteries, the left ventricle, and the septa between the ventricles.

[0253] Figure 13A The corrected location 59 of the non-coronary fossa, determined by a pre-trained machine learning model (marked with an asterisk), is shown and overlaid with fluoroscopic images to generate a composite image 58. This composite image 58 is received in real-time via an input interface and acquired during the implantation of the heart valve prosthesis 3. Prior to this fluoroscopic image, multiple fluoroscopic images were acquired at different time points. These multiple fluoroscopic images contain only non-angiographic images. This model can be named the non-angiographic pre-trained model. This pre-trained machine learning model is specifically configured to determine the corrected location 59 based on non-angiographic images. For this reason, the pre-trained machine learning model is trained only on non-angiographic images labeled to include at least the location of the non-coronary fossa, particularly the locations of the left and / or right coronary fossa, and preferably at least one other anatomical / instrumental feature.

[0254] The corrected position 59 for the non-coronary lobes was determined using fluorescence fluoroscopic images of previous and synthetic images via a pre-trained machine learning model. This pre-trained machine learning model can be a residual neural network, specifically a residual neural network containing at least three or more layers.

[0255] To ensure sufficient accuracy and reliability of the pre-trained machine learning model, even if it relies solely on non-contrast fluoroscopic images, the model is trained on a dataset of more than 300 patients, which includes fluoroscopic images acquired during TAVI / TAVR interventions.

[0256] Figure 13B A schematic segmentation 59 of the non-coronary fossa, determined by a different pre-trained machine learning model, is shown, which is overlaid with fluoroscopic images to generate a synthetic image 581. This pre-trained machine learning model is trained on multiple angiographic images (particularly based solely on angiographic images) to generate... Figure 13B Segment 59 in the text. This pre-trained machine learning model can be called an imaging pre-trained machine learning model.

[0257] Figure 13B The pre-trained machine learning network used can be a residual neural network with a U-Net architecture, used to reliably generate segmentation 59. Then, Figure 13B The composite image 581 shown may be displayed on a monitor with or without segmentation 59 to provide information to clinicians.

[0258] Figure 13A and 13B The previously received fluorescence fluoroscopic images were acquired without the application of rapid pacing, while the fluorescence fluoroscopic image of composite image 58 was acquired during rapid pacing. Figure 13A and 13B The pre-trained machine learning model is specifically adapted to provide a correction location 59 for the non-coronary lobes, which has been corrected for the effects of rapid pacing.

[0259] Similarly, this can be achieved by providing a training dataset containing previous images acquired without rapid pacing and at least one subsequently acquired image acquired during rapid pacing. The training dataset may include labels indicating whether the images were acquired during rapid pacing or in the absence of rapid pacing, to aid in differentiation via a pre-trained machine learning model.

[0260] The training images are correspondingly labeled / marked to provide the location / orientation of the native heart valve components, and in particular at least one additional anatomical / instrumental feature for training purposes.

[0261] Figure 13C and 13D A histogram of distance D (measured in millimeters) is shown, where D is Figure 13A and 13B The corrected location of the non-coronary valve predicted by the pre-trained machine learning model is shown relative to the distance between the labeled aortic valve annulus. Both histograms illustrate the validation results of applying the pre-trained machine learning model to fluorescein imaging data from 29 patients.

[0262] Figure 13C It was shown that even a small training dataset of non-angiographic training images enabled a pre-trained machine learning model to determine the corrected position / orientation of the native valvular components (i.e., non-coronary valvular vasculature), with an accuracy of less than 1 mm in 80% of patients.

[0263] Figure 13D The study demonstrated that a pre-trained machine learning model trained on angiographic training images was able to more accurately determine the corrected position / orientation of the native valvular components (i.e., non-coronary valvular vasculature), with an accuracy of less than 1 mm in 100% of patients.

[0264] Although Figure 13A and 13C The results from the pre-trained model showed lower accuracy, but these results still provided a more accurate correction for the original valve component position / orientation compared to relying solely on the clinician's specialized skills to analyze the images.

[0265] Petronio, AS, Sinning, JM, Van Mieghem, N. et al. (2015). Optimal implantation depth and adherence to guidelines on permanent pacing to improve the results of transcatheter aortic valve replacement with the Medtronic CoreValve System: The CoreValve prospective, international, post-market AD-VANCE-II study (JACC Cardiovasc Interv, 8(6), 837-846) disclosed that the mean target implantation depth relative to the annulus plane had a considerable margin of error: 6.9 ± 4.3 mm. Therefore, even with a small training dataset, the non-angiographic pre-trained model has already provided significant improvements over routine clinical practice.

[0266] Figure 13A and 13C The improved predictive power of the pre-trained machine learning model for angiography in China and Africa further indicates that, for large training datasets, this pre-trained machine learning model can provide reliable results without relying on contrast agents. This is beneficial for heart valve replacement and implantation, improving patient outcomes and shortening interventional time.

[0267] Despite the above Figures 13A-13D Although not shown, the orientation of the native valve assembly (particularly the non-coronary valve, preferably its tip) can be determined relative to the anatomy. In this way, clinicians or automated systems used to place the prosthetic heart valve 3 can easily infer the commissural alignment of the prosthetic heart valve 3 relative to the native heart valve.

[0268] Figure 14A and 14B The synthetic image 40 is shown, which includes a first modified position of geometric heart valve prosthesis features 41 and a first anatomical feature 42, and a second position and orientation of a second anatomical feature 43.

[0269] Similarly, such as Figures 13A-13DAs described, while acquiring the current fluoroscopic image during rapid pacing, previous fluoroscopic images acquired when rapid pacing was not present may have already been received. A pre-trained machine learning model can be specifically adapted to correct the position / or orientation and / or location / or orientation of anatomical features to address the effects of rapid pacing. This can be achieved by providing corresponding training data images for training the model, preferably including labels indicating whether the images were acquired during rapid pacing.

[0270] For illustrative purposes, both synthetic images 40 are generated based on fluorescence fluoroscopic images acquired under contrast agent treatment, thereby making the patient's anatomy discernible. These fluorescence fluoroscopic images were acquired during TAVI / TAVR surgery. Any of the previously described pre-trained machine learning models, particularly non-contrast machine learning models, can be used to generate synthetic images 40.

[0271] Figure 14A and 14B The medical image 40 also shows a catheter 46 with a curved distal tip (e.g., a pigtail catheter) placed near the non-coronary valve, and a pre-formed guidewire 44 (e.g., a SAFARI guidewire) placed in the left ventricle of the heart. These medical images 40, including the geometric characteristics 41 of the heart valve prosthesis and the location / orientation of anatomical features 42, 43, are displayed on a monitor to the clinician to facilitate placement of the heart valve prosthesis 3 within its native anatomical structure.

[0272] The selection of geometric heart valve prosthesis characteristics 41 is based on the position / orientation of the heart valve prosthesis 3 identified in medical image 40 and user input. This method can be configured to determine the total length of the heart valve prosthesis 3. The length of the heart valve prosthesis 3 can be included in the model by providing labels that indicate, for example... Figure 12A-12D The described heart valve prosthesis 3 has a top edge 34 and a bottom edge 35.

[0273] User input can specify a longitudinal sub-portion of the heart valve prosthesis 3, for example, 20% of the total length of the heart valve prosthesis 3. Therefore, the geometric heart valve prosthesis characteristic 41 can represent a longitudinal sub-portion starting from the bottom edge 35 of the heart valve prosthesis 17, for example, 20% of the total length of the heart valve prosthesis 17.

[0274] Due to the radiopaque markings on the heart valve prosthesis 3, the center of the heart valve prosthesis 3 can be easily identified by this system / method. These radiopaque markings can also indicate specific portions of the heart valve prosthesis 3 or its orientation. Geometric heart valve prosthesis characteristics 41 Figure 14A and 14B The line shown is perpendicular to the longitudinal axis of the heart valve prosthesis 3.

[0275] The first position of the first anatomical feature 42 indicates the tip of the non-coronary valve, which is shown as a black dot in the medical image 40. Figure 14A and 14B The location of the first anatomical feature 42 is also shown connected to the geometric heart valve prosthesis feature 41 via a dashed line, which aids in the alignment of the heart valve prosthesis 3. The second location and orientation of the second anatomical feature 43 are shown in circles, indicating the location and orientation of the coronary arteries 431 and 432. From this, clinicians / software can infer the orientation of the coronary arteries 431 and 432 for placing the heart valve prosthesis 3 and / or the pigtail catheter 46 near the non-coronary valve, thereby avoiding obstruction.

[0276] Medical images 40 can be processed to include the location / orientation of other anatomical features, particularly the location / orientation of the ventricular septum. Figure 14A and 14B (not shown in the image), thus allowing the determination of prosthetic valve regurgitation in the form of a risk of obstruction of the His bundle conduction pathway in the heart. The risk of obstruction of the His bundle conduction pathway can be determined based on the position / orientation of the prosthetic heart valve 3 and the position / orientation of the interventricular septum in medical image 40.

[0277] In addition, the medical image 40 can be processed to further include anatomical features representing the outline of the left ventricle, which can help to position the coiled guidewire (such as the SAFARI guidewire) in the left ventricle during medical procedures (especially TAVI / TAVR procedures).

[0278] In one embodiment, the received images are categorized based on whether they were acquired during contrast agent injection or in the absence of contrast agent. Figures 13A-13D and / or Figure 14A and 14B The pre-trained machine learning models described herein are applied. These machine learning models can be switched so that non-contrast machine learning models are applied to non-contrast images, and contrast pre-trained machine learning models are applied to contrast images.

[0279] In one embodiment, a pre-trained machine learning model is configured to detect the presence or absence of contrast agent in received fluoroscopic images. If the presence of contrast agent is detected, a corrected position / orientation of the native valve assembly is determined based on the received fluoroscopic images. If the absence of contrast agent is detected, the pre-trained machine learning model is configured to determine / predict the corrected position / orientation of the native valve assembly based on previously received images acquired with contrast agent. This corrected position / orientation can be overlaid onto received medical images acquired without contrast agent to generate a synthetic image. Similarly, this can be further performed on additional positions / orientations of anatomical / instrumental features. For example, in this way, even if the fluoroscopic images were not acquired during contrast agent injection, the corrected position / orientation can be determined. Figure 13A The location 59 of the non-coronary lobes shown in 13B is included in the composite image 58.

[0280] Figure 14C A schematic diagram illustrating the changes in heart rhythm and respiratory rhythm over time T is shown. Heart rhythm can be measured via ECG, pulse oximeter, or other heart rate monitors. Respiratory rhythm can be measured via spirometer, carbon dioxide monitor, or other respiratory rate monitors. The previously described images, particularly... Figures 13A-13B as well as Figure 14A and 14B The synthetic images generated in the process can be derived from previously received images, and alternatively, can also be used to determine the cardiac and / or respiratory cycles.

[0281] When predicting the location / orientation of native valvular components, anatomical features, and / or device features, the influence of either or both of respiratory rhythm and cardiac rhythm can be considered. This can be achieved by training a pre-trained machine learning model based on multiple images containing these cycles. The training images preferably contain location / numerical labels associated with different phases of these cycles.

Claims

1. A computer-implemented method for determining the location and / or orientation of a patient's native valve assembly, preferably the noncoronary valve (55) of a native aortic heart valve, particularly the tip of the noncoronary valve, the method comprising the steps of: - Received via an input interface, particularly in real time, at least one first image (4) and at least one second image (5) of the moving heart at two different time points, preferably during the implantation of the heart valve prosthesis (3). - By applying at least one pre-trained machine learning model to the at least one first image (4) and / or the at least one second image (5), the corrected location and / or orientation of the native valve component is determined in the at least one second image (5). - The corrected position and / or orientation (59) of the original valve assembly is superimposed on the at least one second image (5) to generate a synthetic second image (58). - Display the composite second image (58) on the monitor.

2. The computer implementation method according to claim 1, wherein the pre-trained machine learning model is a pre-trained neural network configured to extract hierarchical feature maps from the at least one first image (4) using a data processing unit (6), wherein, The application of the pre-trained neural network includes: - The at least one first image (4) is progressively downsampled, and an abstract feature map is extracted, and - Upsample the abstract feature map to the original resolution of the at least one first image (4).

3. The computer implementation method according to any one of the preceding claims, wherein the machine learning model is a pre-trained residual neural network comprising a plurality of convolutional layers and at least one residual block for bypassing at least one convolutional layer.

4. The computer-implemented method according to any of the preceding claims, wherein the at least one first image (4) is a normal blood flow image, particularly acquired in the absence of rapid pacing, and / or the at least one second image (5) is at least one image of reduced blood flow, particularly acquired during rapid pacing, and wherein, The corrected position and / or orientation is determined by further correcting the position and / or orientation of the patient's native valve assembly in response to blood flow, particularly blood flow due to rapid pacing.

5. The computer-implemented method according to any one of the preceding claims, wherein the method comprises the following steps: - Real-time determination of the device position and / or device orientation of the heart valve prosthesis (3) and / or the delivery device for delivering the heart valve prosthesis (3) in at least one of the at least one first and / or second images (5), - Determine the geometric characteristics of the heart valve prosthesis based on the device location and / or the device orientation, and - The geometric heart valve prosthesis characteristics () are superimposed on the at least one synthetic second image (5).

6. The computer-implemented method of claim 5, wherein the geometric heart valve prosthesis characteristics indicate a longitudinal sub-portion of the heart valve prosthesis, wherein the method particularly includes receiving at least one user input defining the longitudinal sub-portion of the heart valve prosthesis, and wherein the geometric heart valve prosthesis characteristics preferably include a vertical representation relative to the longitudinal axis of the heart valve prosthesis or constituted by a vertical representation relative to the longitudinal axis of the heart valve prosthesis.

7. The computer-implemented method according to any of the preceding claims, comprising determining the patient's cardiac cycle and / or respiratory cycle based on the at least one first image, the at least one second image, and / or additional patient data, particularly an electrocardiogram of the heart, and wherein the cardiac cycle and / or the respiratory cycle are further taken into consideration to determine a correction location and / or orientation.

8. The computer-implemented method according to any one of the preceding claims, wherein the at least one first image () and / or the at least one second image () comprises at least one instrument feature, particularly selected from at least one of the following: - The radiopaque position and / or orientation markings of the heart valve prosthesis and / or the delivery device for delivering the heart valve prosthesis. - A pigtail catheter or guidewire located near the non-coronary valve, preferably having a predefined posture in contact with the non-coronary valve, and / or - A coiled catheter or guidewire located in the left ventricle, preferably having a basic predefined posture that contacts the left ventricle near the top of the left ventricle.

9. The computer-implemented method according to any one of the preceding claims further comprises determining the location and / or orientation of at least one anatomical feature based on the at least one first image (4) and / or the at least one second image (5), wherein the at least one anatomical feature is selected from at least one of the following: - (i) at least one of the right coronal lobes and the left coronal lobes, - (ii) at least one coronary artery, - (iii) Outer contour of the left ventricle, - (iv) The interventricular septum separating the right and left ventricles, especially the membranous septum, and - (v) the calcification near the original heart valve, and The location and / or orientation of the at least one anatomical feature are shown in the at least one synthetic second image (5).

10. The method according to any one of the preceding claims, wherein the method comprises one of the following steps: - Identify the presence or absence of the native valve assembly in at least one second image (5), and - If the presence of the native valve component is identified, the corrected position and / or orientation of the native valve component in the at least one second image (5) is determined (20) solely based on applying the at least one pre-trained machine learning model to the at least one second image (5). - If the absence of the native valve component is identified, the corrected position and / or orientation (20) of the native valve component in the at least one second image (5) is determined solely based on applying the pre-trained machine learning model to the at least one first image and optionally the at least one second image.

11. The method according to any one of the preceding claims, wherein the method comprises the following steps: - The images (4, 5) are classified to determine whether the corresponding images (4, 5) were acquired under the action of a contrast agent.

12. The method of claim 11, comprising: - If the images (4, 5) are classified as not acquired under contrast agent conditions, determine the corrected location and / or orientation of the native valve assembly () based on the at least one pre-trained machine learning model and at least one other image (4, 5) classified as acquired under contrast agent conditions, and / or use the location and / or orientation of the at least one anatomical feature visible under contrast agent conditions, and / or - If the images (4, 5) were acquired under the action of a contrast agent, the location and / or orientation of instrument features that are not visible under the action of a contrast agent are determined based on the at least one pre-trained machine learning model and at least one other image (4, 5) that is classified as not acquired under the action of a contrast agent.

13. The method according to any one of claims 11 or 12, comprising applying a first pre-trained neural network to images (4, 5) classified as acquired under the action of a contrast agent, and applying a second pre-trained neural network to images (4, 5) classified as not acquired under the action of a contrast agent.

14. The method according to any one of claims 11, 12 or 13, wherein, The images (4, 5) classified as acquired under contrast agent are used for segmentation of the aortic root portion, and Based on the segmentation, the corrected position and / or orientation of the original valve assembly and / or the position and / or orientation of at least one anatomical feature are determined.

15. The method according to any one of claims 1-10, wherein the at least one first image (4) and the at least one second image (5) are acquired in the absence of a contrast agent.

16. A computer program product, when executed by a computer, causes the computer to perform the steps of the computer-implemented method according to any one of the preceding claims.

17. A computer-readable storage medium comprising instructions that, when executed by a computer, cause the computer to perform the steps of the computer-implemented method according to any one of claims 1-15.

18. A computer-based method for training a machine learning model to determine the location and / or orientation of a native valvular assembly, preferably a non-coronary valve of an aortic heart valve, particularly the tip of a non-coronary valve, said model being specifically used for the method according to claims 1-15, comprising: - Receive an input training dataset comprising at least one image (4, 5) of a moving heart acquired during the implantation of a heart valve prosthesis (3), the image being labeled with labels including non-coronary lobes, preferably including labels for all three coronary lobes, and preferably including labels for a pigtail catheter () or guidewire located near the non-coronary lobes, wherein the at least one image (4, 5) preferably includes at least one label for the heart valve prosthesis (). - The training process of the machine learning model is performed by predicting the output location and / or orientation of the non-coronary lobes; - Compare the output location and / or orientation with the labels for non-coronary lobes, and - Adjust model parameters using the loss function.

19. The computer implementation method of claim 18, wherein the at least one image comprises at least one first image and at least one second image, wherein the at least one first image is a normal blood flow image, particularly acquired in the absence of rapid pacing, and the at least one second image is a blood flow reduction image, particularly acquired during rapid pacing.

20. A computer-based method for training a machine learning model to determine the location and / or orientation of a native valvular assembly, particularly according to claim 18, wherein the native valvular assembly is preferably a non-coronary valve of an aortic heart valve, particularly the tip of a non-coronary valve, said model being specifically used for the method according to any one of claims 1-14, comprising: - Receive an input training dataset consisting of at least one image (4, 5) of a moving heart acquired during the implantation of a heart valve prosthesis (3), the image being labeled to include at least one location and / or orientation of the native valve assembly, particularly the non-coronary valve, wherein the at least one image (4, 5) was not acquired under the influence of a contrast agent. - The training process of the machine learning model is performed by predicting the output position and / or orientation of the native valve component; - Compare the output position and / or orientation with the stated position and / or orientation, and - Adjust model parameters using the loss function.

21. A training dataset for a computer-implemented method of training a machine learning model according to claim 20, comprising at least one image, particularly a fluorescence fluoroscopic image, labeled to include at least one location and / or orientation of the native valvular components, particularly the non-coronary valvular ...

22. A computer-implemented method for determining the position and / or orientation of a valve assembly or valve annulus assembly, preferably a valve annulus (2) of a heart valve, particularly during the implantation of a heart valve prosthesis (3), said method comprising the following steps: - Receive, particularly in real time, at least one first image (4) and at least one second image (5) of the moving heart via an input interface. - Using the data processing unit (6), identify at least one first feature (7) and at least one second feature (8) in the at least one first image (4) that indicates the position and / or orientation of the valve assembly or the valve annulus assembly. - Based on the positioning of the first feature and the second feature (7, 8) in the at least one first image (4), the data processing unit (6) determines at least one static or dynamic geometric relationship between the at least one first feature (7) and the at least one second feature (8), and the data processing unit (6) identifies the at least one second feature in the at least one second image (5). - Based on the at least one second feature (8), the at least one static or dynamic geometric relationship, and a static or dynamic geometric relationship model for considering, particularly during rapid pacing of the heart, the corrected position and / or orientation of the valve assembly or the valve annulus assembly is determined in the at least one second image (5) (20). - Transmit the corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly to the output interface.

23. The computer implementation method according to claim 22, wherein the at least one first image (4) is a normal blood flow image acquired in the absence of rapid pacing, and the second image is a blood flow weakening image acquired during rapid pacing.

24. The method according to any one of claims 22 or 23, wherein the method comprises one of the following steps: - Identify the presence or absence of at least one first feature (7) in the at least one second image (5) indicating the position and / or orientation of the valve assembly or the valve annulus assembly, and - If the presence of the first feature (7) is identified, the corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly is determined in the at least one second image (5) based on the first feature (7). - If the absence of the first feature (7) is identified, the corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly is determined in the at least one second image (5) based solely on the at least one second feature (8) and at least one corrected static or dynamic geometric relationship and the static or dynamic geometric relationship model.

25. The method according to any one of claims 22-24, wherein the method comprises the following steps: - In at least one of the at least one first image and / or second image (5), determine in real time the position and / or orientation of the heart valve prosthesis (3) and / or the delivery device for delivering the heart valve prosthesis (3), particularly the corrected position and / or orientation (20) relative to the valve assembly or the valve annulus assembly.

26. The method according to any one of claims 22-25, wherein the valve assembly or the valve annulus assembly in at least one of the at least one first image (4) indicates the maximum radial dimension of the valve annulus (2), and the method optionally includes the following further steps: Based on the maximum radial dimension, particularly during rapid pacing, the target radial dimension of the valve annulus (2) in the at least one second image (5) is determined.

27. The method according to any one of claims 22-26, wherein at least one defined static or dynamic geometry indicates the patient’s cardiac cycle and / or respiratory cycle, in particular a first static or dynamic geometry based on the cardiac cycle and a second static or dynamic geometry based on the respiratory cycle.

28. The method according to any one of claims 22-27, wherein the at least one second feature (8) comprises: - The patient's anatomical features, or - Surgical characteristics, particularly the orientation of the medical devices (101, 102) or the medical devices (101, 102), Preferably, one of the second features (8) is an anatomical feature, and the other of the second features (8) is a surgical feature.

29. The method according to any one of claims 22-28, wherein the method comprises the following steps: - Processing, preferably real-time processing, of the at least one first and / or second image (4, 5), and including at least one of (i) the corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly and (ii) the position and / or orientation of the heart valve prosthesis (3): ● Preoperative visualization of the heart, and ● The computational model (2) of the valve assembly or the valve annulus assembly, particularly the finite element model.

30. The method according to any one of claims 22-29, wherein the method comprises the following steps: - The corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly and / or the heart valve prosthesis (3) shown in the following items: ● In either the at least one second image (5) or the at least one first image (4), ● In preoperative visualization, and / or ● In the computational model.

31. The method of claim 30, wherein the method comprises the following steps: - Receive user-generated input to manually adjust the corrected position and / or orientation of the valve assembly or the valve annulus assembly (20), and - Display, preferably in real time, the adjusted corrected position and / or orientation in place of the previous corrected position and / or orientation (20).

32. The method according to any one of claims 22-31, wherein the method comprises the following steps: - Classify the at least one first and / or second image (4, 5) to determine whether the corresponding image (4, 5) was acquired under the action of a contrast agent, and - If the images (4, 5) were not acquired under the action of a contrast agent, then at least one feature that is not visible without a contrast agent is determined based on another feature that is visible without a contrast agent among the at least one first feature (7) and the second feature (8).

33. The method according to any one of claims 22-32, wherein the step of using the data processing unit (6) to determine the corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly is based on cardiac electrocardiogram.

34. A computer program product comprising instructions that, when executed by a computer, cause the computer to perform the steps of the method according to any one of claims 22-33.

35. A computer-readable storage medium comprising instructions, when executed by a computer, to cause the computer to perform the steps of the method of any one of claims 22-33.

36. The use of the computer-implemented method according to any one of claims 22-33, for at least one of the following: - Planning for transcatheter aortic valve implantation or replacement; - A delivery system for at least one medical device (101, 102), particularly a heart valve prosthesis (3), capable of translational movement, rotational movement, and / or tilting about a longitudinal axis, for navigation and / or deployment of the heart valve prosthesis (3); and - Optimize the dosage of contrast agent and / or the time interval for injecting contrast agent into the body catheter (11).

37. A system (201) comprising means adapted to perform the steps of the computer-implemented method according to any one of claims 22-33, particularly a data processing unit (6).

38. The system of claim 37, wherein the data processing unit (6) is configured to determine in real time the dynamic radial dimension of the heart valve prosthesis (3) or an expansion unit (12) for expanding the heart valve prosthesis (3), particularly the expansion unit comprising a balloon, and The data processing unit (6) is configured as follows: - (i) Provide expansion control commands for operating the expansion unit (12), or - (ii) Provide the user with expansion control instructions to operate the expansion unit (12). To be used for radial expansion of the heart valve prosthesis (3) in the valve annulus (2), in particular until at least the target radial dimension of the valve annulus (2) is reached.

39. The system according to any one of claims 37 or 38, wherein the data processing unit (6) is configured to determine deployment control commands based on the corrected position and / or orientation (20) of the valve assembly or the valve annulus assembly and the position and / or orientation of the heart valve prosthesis (3), for operating the delivery device to deploy the heart valve prosthesis (3) in the valve annulus (2), wherein the control commands provide instructions regarding at least one, and in particular all of the following: - Adjust the tilt of the heart valve prosthesis (3) relative to the valve annulus (2), - Adjust the circumferential orientation of the heart valve prosthesis (3) relative to the valve annulus (2), and - Adjust the longitudinal position of the heart valve prosthesis (3).

40. The system according to any one of claims 37 to 39, wherein the system includes a pacing unit configured to deliver a pacing signal to a target region of the patient’s heart near the valve assembly or valve annulus assembly for reducing transvalvular flow, wherein, in particular, the pacing unit is operable by the data processing unit (6).

41. The system according to any one of claims 36-40, wherein the data processing unit (6) is configured to receive at least one post-deployment image (41) after the heart valve prosthesis (3) is deployed, and to identify and verify in the at least one first image (4) the position and / or orientation of the heart valve prosthesis (3) relative to the first feature (7) indicating the position and / or orientation of the valve assembly or the valve annulus assembly.

42. The system according to any one of claims 37-41, wherein the data processing unit (6) is configured to: - The imaging device positioning unit (15) controls the positioning and / or orientation of the imaging unit (16), or - Provide user instructions on how to operate the imaging device positioning unit (15), wherein The imaging device positioning unit (15) may include a C-shaped arm.

43. A system for guiding the placement of a heart valve prosthesis (3) within a valve annulus (2), particularly according to any one of claims 37-42, the system comprising a data processing unit (6) adapted to display in real time on a display device (17) the position and / or orientation of the heart valve prosthesis (3) relative to the valve assembly or the valve annulus assembly, a modified orientation and / or position (20).

44. The system of claim 43, wherein the data processing unit (6) is adapted to generate, in real time, visual markers (33, 21) indicating the relative position and / or orientation of the heart valve prosthesis (3), particularly color-coded visual markers, and / or instructions regarding the heart valve prosthesis achieving the desired orientation and / or position in the valve annulus (2) based on at least one, and in particular all of the following: - The relative tilt (T) of the orientation of the heart valve prosthesis (3) relative to the modified orientation of the valve annulus (2). - The relative circumferential orientation of the heart valve prosthesis (3) relative to the modified orientation of the valve annulus (2), and - The relative longitudinal position of the heart valve prosthesis (3) relative to the modified position of the valve annulus (2).

45. The system according to any one of claims 43 or 44, wherein the data processing unit (6) is adapted to display the expansion state of the heart valve prosthesis (3) or the expansion unit (12) in real time on the display device (17).

Images