Selection of the most relevant X-ray images for hemodynamic simulations

Abstract

A method and apparatus for selecting one or more diagnostic images for generating a physiological model are provided, wherein a set of candidate images is determined for examination by a user, in particular by a physician. Thus, the candidate images are determined using one or more target measures such as density measures, motion measures, deviation measures, etc., which have been derived for each diagnostic image of an X-ray angiography series and by analyzing said target measures. Subsequently, a fitness score based on the requirements of the physiological model that should be generated from the selected candidate images is assigned to each candidate image.

Description

Technical Field

[0001] The present invention relates to a method, a corresponding device and a computer program for image selection. In particular, the present invention relates to a method for automatically selecting one or more images from a set of candidate images using a fitness score, wherein the fitness score depends on the subsequent use of the selected image, in particular on the kind of fluid dynamics simulation of the image to be performed based on the selected image. Background Art

[0002] Diagnostic images acquired using X-ray angiography provide an important tool for obtaining information about the coronary arteries. These images allow accurate assessment of coronary artery disease by means of a variety of different methods.

[0003] One such method is an image-based "virtual" determination of a hemodynamic measure, such as fractional flow reserve (FFR) or instantaneous wave-free ratio (iFR). Both FFR and iFR are measures of the pressure drop of blood along a vessel of interest (e.g., due to a stenosis in the vessel of interest). It can be determined as the pressure distal to the stenosis (P d ) and the pressure in the aorta (P a ) ratio.

[0004] In the past, FFR and / or iFR measurements were usually performed invasively by measuring the pressure at the location distal to the lesion and at the location proximal to the lesion using corresponding intravascular measurement devices including pressure sensors. For FFR, these invasive measurements must be performed during maximum blood flow (i.e., under hyperemia), which can cause discomfort to the patient. In contrast, iFR measurements can be performed at rest during a specific period of diastole, thereby avoiding the necessity of inducing hyperemia in the patient.

[0005] In recent years, efforts have been made to non-invasively determine FFR and / or iFR values ​​by means of the above-mentioned virtual method. According to the virtual method, the fluid dynamics in the coronary arteries of a patient are simulated based on a physiological model including a fluid dynamics model representing the blood flow through the vessel of interest. Summary of the invention

[0006] It is an object of the present invention to provide an image selection method which allows selecting one or more images which are most suitable for such physiological modelling in a reduced amount of time and with a minimum of user interaction.

[0007] More specifically, the object of the invention is to provide a method and an apparatus which allow to automatically preselect a set of candidate images from one or more series of diagnostic images for presentation to a user, in particular a physician, so that the user has to review a reduced number of images. Even more specifically, the object of the invention is to provide a method and an apparatus which allow a user to quickly determine the suitability of a set of candidate images for a particular kind of fluid dynamics simulation.

[0008] The object is achieved by a method for selecting one or more diagnostic images to generate a physiological model, the method being a computer-implemented method comprising the steps of obtaining a plurality of diagnostic images of a target structure, deriving a plurality of target metrics including at least one corresponding target metric for each of the plurality of diagnostic images, analyzing the plurality of target metrics to select a set of candidate images, and assigning a fitness score to each candidate image in the set of candidate images, the fitness score indicating the fitness of the corresponding candidate image for generating a physiological model.

[0009] In particular, one or more of these steps may be implemented as units of the corresponding apparatus. In an example, the obtaining step may be implemented by means of an input unit, the deriving step may be implemented by means of a calculation unit, the analyzing step may be implemented by means of an analysis unit, and the selecting step and the step of assigning a fitness score may be implemented by means of a selection unit.

[0010] The physiological model may be generated based on a plurality of diagnostic images of the coronary vasculature that have been acquired during an X-ray angiography. Preferably, not all diagnostic images that have been acquired are used, but a selection thereof is used. In particular, the selection is performed such that the most suitable diagnostic images are used for generating the physiological model. Therefore, which diagnostic image and how many diagnostic images are best suited for a particular physiological model largely depends on the desired result of the fluid dynamics simulation.

[0011] A plurality of diagnostic images may in particular refer to a series of diagnostic images that have been acquired by means of a medical imaging modality. In some embodiments, a plurality of diagnostic images may refer to a plurality of diagnostic image series acquired for one patient, so that the selection process is performed among the plurality of series. This allows combining the information provided in the different diagnostic image series and may lead to improved accuracy. Any medical imaging modality may be used to acquire one or more diagnostic image series, as long as it allows visualization and identification of the target structure. As examples of suitable medical imaging modalities, computed tomography, magnetic resonance imaging, ultrasound imaging, etc. are mentioned.

[0012] In some embodiments, one or more diagnostic image series may specifically refer to one or more series including multiple X-ray angiography images, particularly two-dimensional X-ray angiography images. For X-ray angiography, a contrast agent may be introduced into a target structure. X-ray angiography imaging may then be performed on the target structure during contrast agent inflow and contrast agent outflow. That is, a series of X-ray angiography images may include a subset of diagnostic images representing contrast agent inflow, a subset of diagnostic images representing contrast agent-filled target structures, and a subset of diagnostic images representing contrast agent outflow. In this case, it may be beneficial to select diagnostic images as candidate images with low shortening and low overlap. In addition, in any case, the contrast agent filling should be sufficient to provide suitable contrast.

[0013] The selected one or more diagnostic images may be used to generate a physiological model.In this context, the term physiological model may particularly refer to a model of the target structure, including a geometric model representing the geometry of the target structure and / or a fluid dynamic model representing the dynamics of a fluid through the target structure.

[0014] The target structure may specifically refer to a vessel of interest or multiple vessels of interest that should be evaluated for a specific patient. In some embodiments, the target structure may specifically refer to the coronary vasculature of the patient, which includes one or more vessels of interest.

[0015] In this case, the physiological model may be generated from the one or more selected diagnostic images by segmenting one or more vessels of interest represented in the selected diagnostic images and generating a geometric model representing the vessel geometry of the vessels of interest and / or a fluid dynamics model representing the fluid dynamics through the vessels of interest. In some embodiments, the geometric model may specifically correspond to a two-dimensional geometric model or a quasi-three-dimensional geometric model in which a third dimension is approximated.

[0016] In some embodiments, the fluid dynamics model may correspond in particular to a lumped parameter fluid dynamics model. In such a lumped parameter fluid dynamics model, the fluid dynamics of a blood vessel is approximated by the topology of discrete entities. As an example, a blood vessel tree may be represented by a topology of resistor elements each having a specific resistance. Therefore, the outlet at the distal end of the blood vessel is also represented by a specific resistor element. The resistor element is then grounded, such as to represent the connection of the blood vessel to the venous system. Similarly, the corresponding resistor element may be connected to a series of resistor elements representing the blood vessel of interest, such as to represent the outflow from the blood vessel of interest at certain bifurcations. These resistor elements may also be typically grounded. Compared to other methods such as Navier-Stokes, the lumped parameter fluid dynamics model reduces the dimensionality. Therefore, the use of a lumped parameter fluid dynamics model may allow the calculation of the fluid dynamics inside the blood vessel to be simplified, and is therefore particularly beneficial in terms of processing time.

[0017] The term target measure may particularly refer to one or more characteristics of a target structure as represented in at least one diagnostic image. In this context, the term target measure may particularly refer to a target structure density measure, i.e., a quantitative measure indicating how much target structure is visible in a corresponding diagnostic image, a motion measure indicating the motion of a target structure (e.g., between two consecutive diagnostic images), an overlap measure indicating the amount of overlap in a diagnostic image indicating the target structure, a visibility measure indicating the visibility of the target structure (e.g., the visibility of an edge), and / or a deviation feature indicating the deviation from the expected target acquisition time of a specific diagnostic image. Each of these target measures may affect the fitness of a specific diagnostic image for generating a physiological model. As indicated, one or more such target measures may be derived for each of the diagnostic images. Subsequently, the target measures may be analyzed to select a set of appropriate candidates that can be used to generate a physiological model from a plurality of diagnostic images. Thus, it should be understood that the effect of each of the target measures on fitness may vary depending on the type of physiological model to be generated.

[0018] As an example, in some embodiments, a physiological model is used to extract geometric parameters of a target structure, such as vessel dimensions in the vasculature. In this case, a single diagnostic image may be sufficient for modeling. Therefore, in order to make the geometric modeling particularly accurate, such as in the case of coronary target structures due to cardiac phase, factors such as segmentation accuracy and motion of the target structure are important requirements. That is, the method will perform a fitness score that takes into account factors such as overlap in the target structure, contrast filling, clarity of individual features of the target structure, visibility of relevant bifurcations, and motion phase (e.g., cardiac phase).

[0019] Based on the fitness score, the method can output one or more candidate images that are considered particularly suitable for geometric modeling. In some embodiments, the output can be provided to a user, and the user can then view the output and determine the candidate images to be used for modeling. In some embodiments, the output can be provided directly to a processor capable of performing modeling. The processor can then randomly use any suitable candidate, or can select one of the candidates based on the fitness score.

[0020] In some embodiments, the physiological model including the fluid dynamics model is used for more complex hemodynamic simulations. In these cases, multiple diagnostic images may be required to generate the physiological model. In some embodiments, at least one diagnostic image representing the inflow of contrast agent into the target structure, at least one diagnostic image representing the outflow of contrast agent from the target structure, and at least one diagnostic image representing the target structure completely filled with contrast agent may be required. These three (or more) specific diagnostic images at these three different phases can be used in particular to determine the measure of the inflow and outflow velocity of the contrast agent (from the diagnostic image representing all three phases), and the measure of myocardial perfusion (from the diagnostic image representing the outflow phase) that can be used as the boundary condition of the hemodynamic simulation in the case of the coronary artery vascular system as the target structure. In addition, the diagnostic image representing the fully filled phase can particularly allow the details of the target structure to be viewed most accurately. In the case of the vascular system, that is, the case of the blood vessel of interest to be modeled, this allows the identification and determination of possible branches from the blood vessel of interest, and therefore the corresponding outlet is integrated into the physiological model. This generally improves the accuracy of the hemodynamic simulation. The above-mentioned requirements for suitable candidate images, such as overlap, contrast filling, definition of individual features of the target structure, and motion phase, also need to be considered.

[0021] Therefore, candidate images are identified from the analysis of the target metrics based on the requirements of the physiological model and, therefore, the complexity of the hemodynamic simulation.

[0022] For this purpose, the term candidate image may refer to a diagnostic image that has been selected as a possible candidate for further physiological modeling. In some embodiments, a set of candidate images is typically identified from which a user or processor may select a desired candidate. In some embodiments, only a single candidate image may be identified.

[0023] Each identified candidate image is then assigned a corresponding fitness score. The term fitness score herein may refer in particular to a quantitative value indicating the fitness of the corresponding candidate image for physiological modeling of the target structure. It should be understood that the fitness score may not necessarily be visualized and may also refer to a score determined solely for the purpose of internal processing by a processor implementing physiological modeling.

[0024] In some embodiments, the fitness score may specifically refer to a score assigned based on the classification of a candidate image representing the target structure during contrast inflow, contrast outflow or during the full filling phase and one or more respectively weighted target measures such as an overlap measure identified for the target structure in that particular image, a motion measure determined for the target structure in said image, etc. The weighting of each of these measures is adjusted based on the influence of said feature on the physiological model, i.e., the weighting may be adjusted based on the specific requirements of the respective physiological model.

[0025] Providing a fitness score based on the requirements of the corresponding physiological model can therefore be used to achieve a direct and less time-consuming image selection, which allows the user to identify suitable images from a large number of images without requiring extensive user interaction.

[0026] In some embodiments, the derivation of the plurality of target measures comprises generating a corresponding processed image for each of the plurality of diagnostic images, the generating comprising: assigning a plurality of quantitative values ​​to a plurality of pixels of the corresponding diagnostic image, the quantitative values ​​indicating a probability that the pixels represent a target structure; and deriving a target structure density measure for each of the plurality of diagnostic images based on the plurality of quantitative values. In some embodiments, the target structure density measure for each of the plurality of images is derived based on a sum of the plurality of quantitative values.

[0027] In some embodiments, deriving the target measure includes deriving a target structure density measure. Thus, the term target structure density measure may particularly refer to a quantitative measure indicating how much of the target structure is visible in the corresponding diagnostic image. That is, the target structure density measure value of a diagnostic image representing a target structure with a high degree of contrast agent filling will be high, and the target structure density measure value of a diagnostic image representing a target structure with a low degree of contrast agent filling will be low.

[0028] It will be appreciated that in the case of a series of diagnostic images acquired from contrast inflow to contrast outflow, a diagnostic image representing a fully contrast filled target structure will have a higher target structure density measure value than a diagnostic image representing the target structure during the inflow and outflow phases. Thus, it will be appreciated that each diagnostic image may specifically have a target structure density measure. In the case of a target structure corresponding to coronary vasculature, the target structure density measure may specifically indicate the amount of blood vessels visible in each particular diagnostic image.

[0029] The derivation of the target structure density measure is achieved by processing each of the plurality of diagnostic images using an image processing algorithm specifically tuned to the target structure. The output of this processing step is a processed image, ie one processed image is generated for each diagnostic image.

[0030] In some embodiments, each pixel of the corresponding diagnostic image may be assigned a quantitative value between 0 and 1, which indicates the chance or probability that the pixel belongs to the target structure, to generate a processed image. That is, pixels in the diagnostic image that have a high probability of belonging to the target structure will be assigned a value close to 1, while pixels that have a high probability of belonging to the background will be assigned a value close to 0. Therefore, the processed image corresponds to a pixel map with values ​​between 0 and 1.

[0031] A target structure density measure can then be derived based on the pixel map by considering the quantitative value assigned to each of the pixels. In some embodiments, this can be achieved by determining the target structure density measure as the sum of all quantitative values ​​within a processed image. Given that pixels representing target structures have higher values ​​than pixels representing background, and given that target structures become more visible, providing more contrast agent filling in the target structure, adding the quantitative values ​​to obtain the target structure density measure will produce a target structure density measure with a higher value for the contrast-filled target structure. Therefore, a high value of the target structure density measure is directly proportional to a diagnostic image in which the target structure is clearly visible. In addition, given that less contrast agent will result in more background, the target structure density measure obtained as the sum of the quantitative values ​​from the pixel map will be lower in the contrast inflow and contrast outflow phases and in cases where the target structure is poorly visible for other reasons.

[0032] Given that the processed images are derived from the diagnostic images, the target structure density measure values ​​determined therefrom can be directly related to the corresponding diagnostic images. As a result, the value of the target structure density measure obtained for each diagnostic image can be determined by first image processing the diagnostic images to obtain a processed image comprising a pixel map, and then summing the quantitative values ​​of each pixel to obtain the target measure.

[0033] According to some embodiments, selecting a set of candidate images includes: analyzing the target structure density measure derived according to the measurement time; and obtaining a first subset of candidate images representing a contrast agent inflow phase, a second subset of candidate images representing a contrast agent completely filled phase, and a third subset of candidate images representing a contrast agent outflow phase based on the analysis.

[0034] In some embodiments, the selection of candidate images may include analyzing the target structure density measure according to the measurement time. Hereby, the term measurement time may refer in particular to the time between the start of contrast agent inflow and the contrast agent outflow, during which a plurality of diagnostic images are acquired. The amount of data points used for this analysis depends on the timing of the diagnostic imaging, i.e., there is one data point indicating the target structure density measure for each diagnostic image acquired.

[0035] The resulting curve of the target structure density measure over time allows determining for each of the diagnostic images whether the diagnostic image represents a contrast agent inflow phase, a complete filling phase or a contrast agent outflow phase. That is, when any of these images is selected as a candidate, the candidate image thus selected can be subdivided into three subsets corresponding to the three phases.

[0036] As mentioned above, the amount of images that need to be selected for generating a physiological model may depend on the kind of physiological model that should be generated, ie on the kind and amount of information to be derived from the physiological model.

[0037] In the case of a more complex physiological model, it may be necessary to select at least three - or more - diagnostic images. In some embodiments, each of the at least three diagnostic images should correspond to a different phase of contrast agent flow, i.e., one diagnostic image has been obtained during contrast agent inflow, one image has been obtained since the target structure has been completely filled with contrast agent, and one diagnostic image has been obtained during contrast agent outflow. That is, in some embodiments, a set of candidate objects, i.e., a set of diagnostic images selected as suitable, can be subdivided into three subsets, wherein each subset corresponds to the above-mentioned contrast agent time phase, i.e., one of the designated contrast agent inflow phase, the complete filling phase, and the contrast agent outflow phase.

[0038] Thus, the contrast inflow phase corresponds to the time from injection of contrast until the contrast has completely flowed into the target structure. The full fill phase corresponds to the time the contrast remains in the target structure. The contrast outflow phase corresponds to the time from the end of contrast injection until the contrast has completely left the target structure.

[0039] In particular, the identification of whether a particular diagnostic image - i.e., a potential candidate image - corresponds to a particular phase can be performed by considering curves representing a plurality of target structure density measures as a function of measurement time covering all three phases. Thus, images belonging to the contrast agent inflow phase typically have low to medium target structure density measures that increase with increasing measurement time. Images belonging to the fully filled phase correspond to target structure density measures having a peak of the curve. Finally, images belonging to the contrast agent outflow phase correspond to target structure density measures having low to medium values ​​that decrease with increasing measurement time.

[0040] It will be appreciated that the curve representing the target structure density measure as a function of measurement time may be displayed to a user, but may also correspond to a "virtual" curve processed internally in the device for analysis purposes only.

[0041] In the above-described embodiment, at least three diagnostic images selected as candidates for generating a physiological model each correspond to a different filling phase of the contrast agent flowing into the target structure. However, it should be understood that in other embodiments, the candidates may be selected differently. In some embodiments, for example, where the target structure corresponds to the coronary vasculature, at least three diagnostic images representing different cardiac phases may be selected. This may also be achieved by analyzing the target structure density measure according to the measurement time.

[0042] In particular, in the case of the coronary vasculature, the curve representing the target structure density measure as a function of the measurement time may vary over time according to periodic variations caused by coronary contraction and dilation. These periodic variations also allow a (consistent) cardiac phase to be determined for all diagnostic images, and in this way, allow diagnostic images to be selected as candidates based on the determined cardiac phase. Given that the cardiac phase may be an important factor in the accuracy of physiological modeling, the separately determined cardiac phase may be considered in the fitness score for each candidate image. However, it should be understood that, while in some embodiments, the cardiac phase may be derived solely from the periodic variations of the target structure density measure over time, in other embodiments, the derivation may be supported by electrocardiogram (ECG) or intra-coronary pressure data.

[0043] In some embodiments, the method may further include analyzing the first subset of candidate images, and / or analyzing the third set of candidate images, and determining a visibility metric indicating visibility of the target structure for each of the first subset of candidate images and / or the third subset of candidate images.

[0044] In some embodiments, it is beneficial to select at least three consecutive images or images representing a contrast agent inflow phase or a contrast agent outflow phase obtained within a very short time frame as possible candidates, i.e., further analyze the potential candidate images belonging to the first subset and the third subset. Therefore, an analysis should be performed to identify three or more candidate images, which have low perspective contraction and low overlap, i.e., are suitable with respect to the visibility of the target structure. Thereby, this visibility of the target structure can be expressed in particular according to a so-called visibility measure.

[0045] Determining such visibility measures of images from the inflow and / or outflow phases of the contrast agent allows selecting candidate images that are particularly suitable for performing flow velocity assessment through the target structure, thereby allowing the derivation of important boundary conditions for fluid dynamics modeling, such as flow velocity, vessel wall resistance, etc. Similar to the above case, this can be achieved in particular by analyzing the target structure density measure (indicative of the amount of contrast agent inside the target structure) as a function of the measurement time.

[0046] In some embodiments, deriving the plurality of target measures comprises identifying, for each of the plurality of diagnostic images, a motion measure indicative of motion of the target structure. In some embodiments, the motion measure is identified by determining, for each of the plurality of diagnostic images, a corresponding processed image and analyzing the processed image according to a measurement time, wherein the analyzing comprises subtracting two consecutive processed images from each other to determine a motion signature.

[0047] In some embodiments, the processed images can also be used to determine the degree of motion of the target structure. This allows avoiding the selection of candidate images that represent large motion of the target structure, and therefore can even include motion blur. For this purpose, the processed images are considered as a function of the measurement time. This is possible by mapping each processed image to a corresponding diagnostic image obtained at a specific measurement time point. Therefore, the processed images are sorted based on their measurement time, i.e., a series of processed images are obtained. Then, the pixel maps represented in adjacent (continuous) processed images can be subtracted from each other. Therefore, a large mean absolute difference can indicate that a large motion has occurred between two corresponding diagnostic images. The large mean absolute difference can be used as a motion measure indicating the motion of the target structure. The larger the value of the motion measure, the more motion is expected.

[0048] This concept is particularly important in the case of coronary target structures. The cardiac cycle introduces coronary motion in the coronary vessels, which may affect the fit of a particular diagnostic image to the physiological model. Therefore, a diagnostic image that is identified as belonging to a particular phase of the cardiac cycle (usually end-diastole) may be superior to other diagnostic images with respect to motion.

[0049] According to some embodiments, the deriving of the plurality of target measures includes identifying, for each of the plurality of diagnostic images, an overlap measure indicative of overlap in the target structure.

[0050] Another factor that may affect the accuracy of physiological modeling is the overlap of the target structure. As an example, in the case where the target structure belongs to the (coronary) vasculature, multiple blood vessels visible in the (two-dimensional) diagnostic image may be only partially visible due to the overlap of the blood vessels from the specific direction from which the diagnostic image was acquired.

[0051] Therefore, the amount of overlap in a particular diagnostic image can also be taken into account in the fitness score based on a so-called overlap measure. Overlap can be detected automatically by image processing using target structure-specific algorithms, and / or manually. In the case of the (coronary) vasculature as target structure, overlap can be detected, for example, by detecting closed loops in the pixel map represented in the processed image. This is the case because the pixel map only distinguishes between background and structure. Closed loops with higher quantitative values ​​indicating the presence of a structure allow to conclude that there are at least two vessels overlapping (because blood vessels usually have an inflow and an outflow). In some embodiments, in particular, the closest loop can be identified by performing contour tracking within the target structure map in the processed image. In some embodiments, the orientation score can be used to determine the closest loop directly from the diagnostic image.

[0052] In some embodiments, deriving multiple target measurements includes receiving additional procedural information for each of the multiple diagnostic images. The additional procedural information may include information about administered medications that may affect the diagnostic images or physiological modeling. Examples of such medications are adenosine, nitroglycerin, or acetylcholine. The additional procedural information may also include sensor readouts, such as information about pressure, (blood) flow, ECG, or other information that may be acquired together with the diagnostic images, at some time before or after the diagnostic images are acquired. Based on the additional procedural information, other target measurements may be derived, such as cardiac phase, coronary resting state, coronary hyperemia state, or other.

[0053] It should be understood that the additional procedural information itself may also be considered as an additional objective measure. As an example, since the dilation of the coronary arteries is limited and may not be consistent, diagnostic images acquired without the administration of nitroglycerin or too long after said administration may have their fitness scores reduced based on this. In some embodiments, diagnostic images may only receive fitness scores above 0 if they are acquired directly after intracoronary adenosine injection, as this may be a necessary requirement for the corresponding physiological modeling.

[0054] In some embodiments, the derivation of the plurality of target metrics includes identifying, for each of the plurality of diagnostic images, a deviation metric indicative of a deviation from a desired target acquisition time.

[0055] In some embodiments, it may be beneficial to select diagnostic images that have been acquired at a particular target acquisition time, particularly when the target structure shows time-varying motion. As an example, if the target structure includes coronary vasculature, the cardiac cycle induces periodic variations in the motion of the vasculature. In this case, a particular target acquisition time may be determined that takes into account this periodic variation. That is, diagnostic images acquired at this target acquisition time should be prioritized over diagnostic images acquired at different measurement times.

[0056] The target acquisition time may thus correspond to an optimal phase in the cardiac cycle. In some specific embodiments, the optimal phase may correspond, for example, to the end diastole. In other embodiments, other phases may also be used, in particular when previously used diagnostic images have been used for a phase different from the end diastole. In this case, the phase to be used for the diagnostic image should correspond to the previously used phase. That is, phase consistency of all images should be maintained.

[0057] For this purpose, diagnostic images not obtained at the target acquisition time are not necessarily less suitable for physiological modeling. As an example, they may present better contrast, etc., which makes them more suitable than diagnostic images obtained at the target acquisition time. In this way, a corresponding image-specific value for the deviation measure is introduced into the fitness score. The deviation measure is an indicator of the distance from a selected optimal phase, i.e. one or more target acquisition times at which the optimal phase is achieved. In the case of coronary physiology, the deviation measure is an indicator of the distance from the optimal phase in the cardiac cycle.

[0058] According to some embodiments, the fitness score is based on a weighted sum of one or more target metrics. In some specific embodiments, the fitness score may be based in particular on a weighted sum of a motion metric and / or an overlap metric and / or a deviation metric. In some embodiments, the respective weighting factors are adjusted based on one or more hemodynamic parameters to be modeled using a physiological model to be generated based on one or more images to be selected from a set of candidate images.

[0059] In some embodiments, a fitness score indicating whether a candidate image is particularly suitable for generating a physiological model can be derived from a weighted sum of some or all of the above metrics. In some embodiments, the fitness score can be determined according to the following formula:

[0060] S 流入 =χ·(1-w 1 ·Ow 2 ·Mw 3 ·θ)

[0061] Therefore, the term χ∈(0,1) corresponds to an indicator determined based on the target structure density measure, whereby the indicator indicates whether the target structure density measure has identified the corresponding diagnostic image as an image of the first subset representing the contrast agent inflow phase, an image of the second subset representing the contrast agent completely filled phase, or an image of the third subset representing the contrast agent outflow phase. 1 O corresponds to the weighted overlap measure, the term w 2 M corresponds to the weighted motion measure, and the term w 3 · θ corresponds to the weighted deviation measure.

[0062] In some embodiments, the weighting factor w is adjusted according to the requirements of the physiological model, i.e., the purpose of the physiological model should be used. 1 , w 2 and w 3 As an example, when selecting an image from the contrast agent inflow phase as a reference for the inflow velocity, the weighting factor w can be 1 is set relatively small, since vessel overlap will not be significant in the early contrast inflow phase. Therefore, the weight of the overlap measure may be less than the weight of the motion measure and / or the deviation measure.

[0063] In some embodiments, the (pre)selected candidate images selected by the method may be presented to a user, in particular a physician, for final selection. To simplify the selection, the candidate images may in particular be presented together with a graphical representation of their fitness scores. It should be understood that the fitness scores presented are always specific to the respective physiological model. That is, if the physiological model is used for a different purpose, the fitness scores may change.

[0064] The suitability score may correspond to a graphical representation of a numerical value shown in or next to the corresponding candidate image. In some embodiments, the suitability score may also correspond to a color-coded scale indicating suitability. Other ways of representing different suitability scores for each diagnostic image may also be envisioned. By presenting the score in a visual manner, a user may more easily select an image from the candidate images.

[0065] In another aspect, an apparatus for selecting one or more diagnostic images to generate a fluid dynamic model comprises: an input unit configured to obtain a plurality of diagnostic images of a target structure; a computing unit configured to derive a plurality of target metrics including at least one corresponding target metric for each of the plurality of diagnostic images; an analyzing unit configured to analyze the plurality of target metrics; and a selecting unit configured to select a set of candidate images based on the analysis of the plurality of target metrics and to assign a fitness score to each candidate image in the set of candidate images, the fitness score indicating the fitness of the corresponding candidate image for generating a fluid dynamic model. According to yet another embodiment, the selecting unit comprises a classifier that has been trained using a training data set that relates the one or more diagnostic images to correspondingly measured hemodynamic parameter data.

[0066] In some embodiments, a device is provided, which is configured to perform the method as described above. For this purpose, the device may include: an input unit, a calculation unit, an analysis unit and a selection unit. In addition, in some embodiments, the device may include a display unit, such as a liquid crystal display, etc., to display a graphical representation of candidate images and / or their corresponding fitness scores to a user. In some embodiments, the device may also include a modeling unit or be communicatively connected to the modeling unit, which is used to generate a physiological model from one or more diagnostic images, which have been presented as candidate images that are particularly suitable for the physiological model and have therefore been selected by the user. Based on the physiological model, one or more hemodynamic parameters, such as pressure, flow rate, etc., can be simulated, i.e., modeled, to perform virtual FFR or iFR.

[0067] To this end, the modeling unit may generate the physiological model by segmenting one or more vessels of interest represented in the selected diagnostic image and generating a geometric model representing vessel geometry of the vessels of interest and a fluid dynamics model representing fluid dynamics through the vessels of interest.

[0068] Furthermore, in addition to using a programmed algorithm to select candidate images, it is also possible to provide the device, and in particular the selection unit, with a trained classifier and use machine learning for image selection. For this purpose, the classifier implementing the machine learning algorithm can be trained based on user behavior, i.e. can be trained by tracking preferred user selections of candidate images.

[0069] Alternatively or additionally, the classifier may be trained using a training data set. The training data set may thus in particular comprise one or more values ​​for measured hemodynamic parameter data, such as pressure values ​​or FFR values ​​etc., associated with the corresponding diagnostic images. This combination of diagnostic images and corresponding hemodynamic parameter values ​​allows training of optimal diagnostic images with respect to contrast agent filling, cardiac phase etc. Ie, these factors may be trained by training on the measured data.

[0070] In some embodiments, the classifier can also be trained based on training against a measured data set to derive a target structure density measure and / or a motion measure and / or an overlap measure and / or a deviation measure. In some embodiments, the trained classifier can be specifically trained for overlap detection based on a simulated training data set to determine an overlap measure.

[0071] In another aspect, there is provided a computer program for performing the above method, said computer program, when executed by a processing unit, being adapted to control an apparatus as described above. In yet another aspect, there is provided a computer readable medium having stored thereon the above computer program.

[0072] It shall be understood that the method according to claim 1, the apparatus according to claim 12, the computer program according to claim 14 and the computer readable medium according to claim 15 have similar and / or identical preferred embodiments, in particular as defined in the dependent claims.

[0073] It shall be understood that a preferred embodiment of the invention may also be any combination of the dependent claims or the above embodiments with the corresponding independent claim.

[0074] These and other aspects of the invention will be apparent from and elucidated with reference to the embodiments described hereinafter. BRIEF DESCRIPTION OF THE DRAWINGS

[0075] In the following figures:

[0076] Figure 1 An apparatus for image selection according to an exemplary embodiment is schematically illustrated.

[0077] Figure 2 An exemplary graphical representation of analysis of a target metric as a function of time is illustrated in accordance with an embodiment.

[0078] Figure 3 An exemplary selection method according to an embodiment is illustrated. DETAILED DESCRIPTION

[0079] The illustrations in the drawings are schematically. In different drawings, similar or identical elements are provided with the same reference signs.

[0080] Figure 1 An exemplary embodiment of an apparatus 1 for performing image selection from a plurality of diagnostic images to obtain one or more candidate images for generating a physiological model is schematically represented. The apparatus 1 comprises an input unit 100, a calculation unit 200, an analysis unit 300 and a selection unit 400. In addition, the apparatus 1 is communicatively connected to a display unit 500. The display unit 500 is connected to an input module 501 and is also in communication with a modeling unit 2.

[0081] In accordance with Figure 1 In the exemplary embodiment of the present invention, the input unit 100 is configured to receive a plurality of diagnostic images 10 and provide the diagnostic images 10 to the computing unit 200. The diagnostic images 10 correspond to X-ray angiography images. However, it should be understood that different imaging modalities may also be used to acquire the diagnostic images 10.

[0082] The computing unit 200 is then configured to perform image processing on the diagnostic image 10. Thereby, the image processing algorithm used by the computing unit 200 is specifically tuned for the target structure, which is located in accordance with Figure 1In an exemplary embodiment of the invention, the image corresponds to the coronary vasculature. The processed image includes a vascular map of the coronary vasculature with the vascular structures highlighted. This is achieved by assigning to each pixel in the diagnostic image a pixel value indicating the probability that the pixel belongs to a vessel. That is, a high pixel value (close to 1) indicates that the pixel most likely belongs to a vessel representation, and a low pixel value (close to 0) indicates that the pixel most likely belongs to a background pixel.

[0083] In accordance with Figure 1 In an exemplary embodiment of the present invention, the computing unit 200 then analyzes the received processed images by determining the sum of the pixel values ​​of each processed image. The sum corresponds to a target structure density measure. Thus, a processed image having many pixels with higher values ​​(i.e., many pixels belonging to a blood vessel representation) has a higher target structure density measure than the remaining pixels. Typically, a higher target structure density measure also corresponds to a higher contrast agent filling degree, because at a lower contrast agent filling degree, many blood vessel structures will not be visible. Correspondingly, an image having a lower contrast agent filling degree of blood vessels will have a lower target structure density measure.

[0084] Then, according to Figure 1 In an exemplary embodiment of the present invention, the computing unit 200 may provide the target structure density measure, optionally together with the diagnostic image and / or the processed image, to the analyzing unit 300 .

[0085] The analysis unit 300 then analyzes the target structure density measure as a function of the measurement time. In other words, the analysis unit 300 determines for each diagnostic image the point in time at which the specific diagnostic image was obtained and relates the point in time to the corresponding target structure density measure of the specific diagnostic image. In this context, the term measurement time corresponds in particular to the time from the early influx of contrast agent to the late outflow of contrast agent from the vascular system.

[0086] Then, based on the curve of the target structure density measure, the analysis unit 300 can distribute the diagnostic images into three subsets of images, ie, images belonging to the early contrast agent inflow phase, images belonging to the contrast agent complete filling phase, and images belonging to the contrast agent outflow phase can be identified.

[0087] To this end, Figure 2 A graph 20 of a target structure density measure D as a function of the measurement time t is schematically illustrated. Figure 2 Below, a plurality of diagnostic images 10 are shown, and the plurality of diagnostic images 10 include diagnostic images 11, 12, 13 and 14. Figure 2 In a particular embodiment of the present invention, the diagnostic images 11, 12, 13 and 14 correspond to two-dimensional X-ray angiography images. The target structure density measure D corresponds to a density measure indicative of the density of blood vessels in each of the diagnostic images.

[0088] In graph 20, the target structural density measure D for diagnostic image 11 is indicated as 21, the target structural density measure D for diagnostic image 12 is indicated as 22, the target structural density measure D for diagnostic image 13 is indicated as 23, and the target structural density measure D for diagnostic image 14 is indicated as 24. Based on these target structural density measures, diagnostic images 11, 12, 13, and 14 can be assigned to one of the three different phases indicated herein above.

[0089] In accordance with Figure 2 In the exemplary embodiment of the present invention, the target structure density measure 21 for the diagnostic image 11 is in the low to medium range and is increasing. Therefore, the diagnostic image 11 belongs to the early contrast agent inflow phase. Similarly, the target structure density measure 22 appears to indicate the contrast agent inflow phase for the diagnostic image 12. The target structure density measures 23 and 24 are in the high range. Therefore, the corresponding diagnostic images 13 and 14 can be regarded as belonging to the contrast agent complete filling phase.

[0090] It can also be appreciated from the graph 20 that the value of the target structure density measure D shows a periodic variation over time. This is due to the cardiac phase. In some embodiments, the analysis unit 300 can use the periodic variation to derive the cardiac phase. This allows a consistent cardiac phase to be determined for all candidate images.

[0091] Return to Figure 1 , the analysis unit 300 can therefore use the curve representing the target structure density measure D according to time t to determine whether the corresponding image belongs to the contrast agent inflow phase, the contrast agent complete filling phase or the contrast agent outflow phase for the diagnostic image 10. In addition, the analysis unit 300 can use the periodic changes of the curve to determine a consistent cardiac phase for all diagnostic images.

[0092] In accordance with Figure 1 In an exemplary embodiment of the present invention, the analysis unit 300 may also be configured to determine a motion measure M. For this purpose, the analysis unit 300 may be configured to use a processed image comprising a vessel map by subtracting two adjacent vessel maps, i.e. vessel maps represented in two processed images derived from two successively acquired diagnostic images. By means of subtracting two adjacent images, a mean absolute difference of adjacent vessel maps may be determined, which may be used as the motion measure M. If the value of the mean absolute difference is large, the (coronary) motion is large, whereas if the value is small, it may also be assumed that the motion is small. That is, a higher value of the motion measure M indicates more motion than a smaller value.

[0093] The analysis unit 300 may also be configured to determine an overlap measure O. Figure 1In an exemplary embodiment of the present invention, the analysis unit 300 can be configured to identify closed loops in each image as represented in the processed images. These closed loops can be considered to indicate overlapping blood vessels. Based on these closed loops, the analysis unit 300 can then determine an overlap measure O for each processed image, which indicates the amount of overlap in the corresponding diagnostic image 10.

[0094] Then, according to Figure 1 In an exemplary embodiment of the present invention, the analysis unit 300 may provide a plurality of diagnostic images, their corresponding target structure density measures and motion and overlap measures, and optionally a plurality of processed images to the selection unit 400. The selection unit 400 may receive the plurality of diagnostic images, the target measures, the motion measures, and the overlap measures, and the plurality of processed images (if provided), and may use this information to determine, for each of the plurality of diagnostic images, a respective fitness score indicating the fitness of the diagnostic image for a particular purpose. Figure 1 In an embodiment, those diagnostic images should be selected by the selection unit 400, which are most suitable for generating a physiological model including a geometric model and a fluid dynamic model of the coronary vasculature for the purpose of deriving one or more hemodynamic parameters of the coronary vasculature as a target structure.

[0095] In accordance with Figure 1 In the exemplary embodiment of the present invention, the selection unit 400 calculates the fitness score S according to the following equation:

[0096] S = x·(1-w 1 ·Ow 2 ·Mw 3 ·θ)

[0097] Wherein, the term χ∈(0,1) corresponds to an indicator determined based on the target structure density measure, which indicates whether the target structure density measure has identified the corresponding diagnostic image as an image of the first subset representing the contrast agent inflow phase, an image of the second subset representing the contrast agent completely filled phase, or an image of the third subset representing the contrast agent outflow phase. 1 The O term corresponds to the product of the superimposed feature and the corresponding weighting factor. In addition, the term w 2 M corresponds to the product of the motion measure and the corresponding weighting factor.

[0098] In accordance with Figure 1 In an exemplary embodiment of the invention, wherein the target structure is the coronary vasculature, the term w is also considered. 3 θ. θ corresponds to a deviation measure that indicates how far a particular image is from the optimal phase in the cardiac cycle. Figure 1In the embodiment of , the optimal phase corresponds to the end diastole. Therefore, the deviation measure θ indicates the distance from the end diastole. The factor w 3 A weighting factor corresponding to the deviation measure.

[0099] In accordance with Figure 1 In the embodiment of the invention, the weighting factor w is adjusted according to the requirements of the image-based physiological model. 1 、w 2 and w 3 As an example, a reference inflow velocity is required for a fluid dynamics model. In this case, the weighting factor w of the overlap measure can be 1 is set fairly small. This is so because in the early contrast inflow phase, there is only a little contrast inside the vessels and therefore the overlap of the vessels is negligible. In contrast, when selecting images from the contrast-completely filled phase, the overlap measure should be weighted higher because at this stage the vessel overlap is more significant.

[0100] Thus, the fitness score S can be determined as the product of an indicator of the target structure density measure multiplied by a factor determined to be 1 minus a weighted sum of the motion measure, the overlap measure, and the deviation measure. Thus, the smaller the impact of the feature, the higher the score obtained. Thus, a high fitness score indicates a high chance that the diagnostic image may result in a promising result.

[0101] In accordance with Figure 1 In an embodiment of the present invention, the computing unit 400 determines a fitness score S to derive a set of candidate images, and provides these candidate images together with their corresponding fitness scores to the display unit 500. The display unit 500 generates a graphical representation of each of the received candidate images and displays the graphical representation to the user. Optionally, the fitness score can be displayed next to the corresponding candidate image.

[0102] The user can thus browse a set of candidate images that are automatically pre-selected based on their suitability for physiological modeling. This allows the user to view only those candidate images because they are objectively the best images available.

[0103] The user may then select one or more diagnostic images (based on visual inspection and / or fitness scoring) via the user interface 501 and prompt the display device to provide the (finally) selected diagnostic images to the modeling unit 2. The modeling unit 2 then uses the one or more selected images to generate a physiological model including a geometric model and a fluid dynamics model. The generated model may then be provided to the display device 500 again and a graphical representation thereof may be presented to the user.

[0104] Figure 2Schematically represents a flow chart of a method for image selection according to an embodiment. In step S101 , a plurality of diagnostic images 10 are received at the input unit 100. In step S102, the input unit 100 provides these diagnostic images 10 to the calculation unit 200.

[0105] In step S201, a diagnostic image 10 is received at the computing unit 200. In step S202, the computing unit 200 processes the diagnostic image using an image processing algorithm specifically tuned for a target structure, which in the exemplary embodiment presented herein corresponds to the coronary vasculature. By means of said processing, a processed image is obtained for each diagnostic image, which comprises a vessel map for the coronary vasculature. This is achieved by assigning to each pixel in the diagnostic image a pixel value indicating the probability that said pixel belongs to a vessel. In step S203, the sum of the pixel values ​​of the pixels in each processed image is calculated to determine a target structure density measure. In step S204, the target structure density measure for each image, for a plurality of processed images derived from a plurality of diagnostic images and for a plurality of diagnostic images is provided to the analysis unit 300.

[0106] In step S301, the analysis unit receives a plurality of processed images, a plurality of diagnostic images, and target structure density measurements. In step S302, the analysis unit 300 plots the target structure density measurements according to the measurement time for further analysis. In step S303, the analysis unit 300 analyzes the curve of the target structure density measurements according to the time, and identifies based on the analysis whether the corresponding image belongs to the early contrast agent inflow phase, the contrast agent complete filling phase, or the contrast agent outflow phase, such as relative to the image. Figure 2 described.

[0107] In step S304, the analysis unit 300 also uses the curve of the target structure density measure to determine the cardiac phase at which each diagnostic image was acquired. This allows a consistent cardiac phase to be provided for all candidate images that may be suggested to the user.

[0108] In step S305, the analysis unit 300 determines the motion measure M. Figure 3 In an exemplary embodiment of , this is achieved by subtracting the vessel maps of two adjacent processed images to obtain the mean absolute difference of the adjacent vessel maps. The mean absolute difference can then be used as a motion measure M, where a larger value indicates a large coronary artery motion and a smaller value indicates a smaller coronary artery motion. It should be understood that the output of step S305 should include a motion measure value for each diagnostic image that has been considered.

[0109] In step S306, the analysis unit 300 also determines an overlap measure O by identifying closed loops that appear in the corresponding blood vessel map as shown in each of the plurality of processed images. Depending on how many closed loops are identified for each processed image, the overlap measure O is set to indicate the amount of overlap in a particular diagnostic image corresponding to the corresponding processed image. It should be understood that the output of step S306 typically includes an overlap measure value for each diagnostic image that has been considered.

[0110] In step S307 , the analyzing unit 300 provides the plurality of diagnostic images, their target structure density measures and their determined motion and overlap measures and optionally processed images to the selecting unit 400 .

[0111] In step S401 , the selection unit 400 receives a plurality of diagnostic images, their corresponding target structure density measures, motion measures and overlap measures, and a plurality of processed images (if provided).

[0112] In step S402, the selection unit 400 considers the requirements of the processing required for the image. Figure 2 In a specific embodiment of the present invention, the diagnostic images should be used to generate a physiological model including a geometric model and a fluid dynamics model. Based on these requirements, the selection unit 400 sets a corresponding weighting factor for each of the motion measure and the overlap measure. In addition, the selection unit derives a deviation measure for each diagnostic image, which indicates the distance from the optimal phase in the cardiac cycle, and a corresponding weighting factor, which also depends on the requirements imposed on the diagnostic image by the modeling process.

[0113] In step S403, the selection unit 400 determines the above relative Figure 1 The fitness score S described herein is used to (pre-)select a set of candidate images. In step S404, the selection unit 400 provides these candidate images together with their corresponding fitness scores to the display unit 500.

[0114] In step S501, the display unit 500 receives a set of candidate images that have been pre-selected and their corresponding fitness scores, and in step S502, generates a graphical representation of each candidate image in the set of candidate images. In step S503, the display unit 500 displays the graphical representation of each candidate image from the set of candidate images to the user, which may optionally include the corresponding fitness score.

[0115] In step S504, the user views the presented set of candidate images and selects one or more diagnoses represented in the set of candidate images. The selection prompts the selected one or more diagnostic images to be provided to the modeling unit 2 in step S601. In response to receiving the one or more selected diagnostic images, the modeling unit 2 generates a physiological model including a geometric model and a fluid dynamics model for hemodynamic simulation. Thus, a method is enabled that allows the selection of diagnostic images in an efficient and fast manner without requiring much user interaction.

[0116] Although in the above-described embodiments, diagnostic images have been acquired using X-ray angiography, it should be understood that other imaging modalities may also be used, such as computed tomography, ultrasound imaging, magnetic resonance imaging, etc.

[0117] Furthermore, although in the above embodiments the method is applied to the coronary vasculature, it should be understood that the method may be equally used for image selection of images of different target structures, in particular image selection of images representing target structures of different parts of the human and / or animal body.

[0118] Furthermore, although in the above-described embodiments the selection has been based on target structure density measures and (weighted) motion measures, overlap characteristics and deviation measures, it will be appreciated that other factors may be included in the selection, such as C-arm angle, aortic pressure values, absence of intravascular devices (e.g. IVUS or guidewires), and / or frame rate.

[0119] Additionally, although in the above-described embodiments the selection has been performed by a manual algorithm, it will be appreciated that machine learning based methods may also be used for image selection.

[0120] Other variations to the disclosed embodiments can be understood and effected by those skilled in the art in practicing the claimed invention, from a study of the drawings, the disclosure, and the appended claims.

[0121] In the claims, the word "comprising" does not exclude other elements or steps, and the word "a" or "an" does not exclude a plurality.

[0122] A single unit or device may fulfill the functions of several items recited in the claims.The mere fact that certain measures are recited in mutually different dependent claims does not indicate that a combination of these measures cannot be used to advantage.

[0123] The processes performed by one unit or several units or devices, such as generating processed images, deriving multiple target metrics, analyzing multiple target metrics, selecting candidate images and / or determining and assigning fitness scores, etc., can be performed by any other number of units or devices. Therefore, these processes according to the present invention can be implemented as program code modules of a computer program and / or as dedicated hardware.

[0124] The computer program may be stored / distributed on appropriate media, such as optical storage media or solid-state media provided together with or as part of other hardware, but the computer program may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunications systems.

[0125] Any reference signs in the claims should not be construed as limiting the scope.

[0126] The present invention relates to a method for selecting one or more diagnostic images to generate a fluid dynamics model, the method comprising the following steps: obtaining multiple diagnostic images of a target structure, deriving multiple target measurements including at least one corresponding target measurement for each of the multiple diagnostic images, analyzing the multiple target measurements to select a set of candidate images, and assigning a fitness score to each candidate image in the set of candidate images, the fitness score indicating the fitness of the corresponding candidate image for generating a fluid dynamics model.

[0127] By means of the method and apparatus for image selection, an automatic image selection process may be established allowing a set of candidate images to be pre-selected so that a user may find the most suitable diagnostic image faster and with a reduced amount of user interaction.

Claims

1. A computer-implemented method for selecting one or more diagnostic images to generate a physiological model, the method comprising the steps of: Obtain multiple diagnostic images of the target structure, deriving a plurality of target metrics including at least one corresponding target metric for each of the plurality of diagnostic images, wherein the target metric indicates one or more characteristics of the target structure in the corresponding diagnostic image, analyzing the plurality of target metrics to select a set of candidate images from the plurality of diagnostic images for generating the physiological model of the target structure, and A fitness score is assigned to each candidate image in the set of candidate images, the fitness score indicating the suitability of the corresponding candidate image for generating the physiological model.

2. The method according to claim 1, wherein: The deriving of the plurality of objective metrics comprises: generating a corresponding processed image for each of the plurality of diagnostic images, the generating comprising assigning a plurality of quantitative values ​​to a plurality of pixels of the corresponding diagnostic image, the quantitative values ​​indicating a probability that the pixels represent the target structure, and A target structure density measure is derived for each diagnostic image of the plurality of diagnostic images based on the plurality of quantitative values.

3. The method according to claim 2, wherein: The target structure density measure for each diagnostic image of the plurality of diagnostic images is derived based on a sum of the plurality of quantitative values.

4. The method according to claim 2 or 3, wherein: Selecting the set of candidate images comprises: analyzing the derived target structure density measure according to the measurement time, and Based on the analysis, we obtain: a first subset of candidate images representing a contrast agent inflow phase; a second subset of candidate images representing a completely contrast agent filled phase; and A third subset of candidate images represents a contrast agent outflow phase.

5. The method according to claim 4, further comprising: analyzing said first subset of candidate images, and / or analyzing said third subset of candidate images, For each candidate image of the first subset of candidate images and / or the third subset of candidate images, a visibility measure indicative of visibility of the target structure is determined.

6. The method according to claim 2 or 3, wherein: The deriving of the plurality of objective metrics comprises: For each diagnostic image of the plurality of diagnostic images, a motion measure indicative of motion of the target structure is identified.

7. The method according to claim 6, wherein: The motion measure is identified by: for each diagnostic image in the plurality of diagnostic images, determining a corresponding processed image, and The processed images according to the measured time are analyzed, wherein the analysis comprises subtracting two consecutive processed images from each other to determine the motion measure.

8. The method according to any one of claims 1 to 3, wherein: The deriving of the plurality of objective metrics comprises: For each diagnostic image of the plurality of diagnostic images, an overlap measure indicative of overlap in the target structure is identified.

9. The method according to any one of claims 1 to 3, wherein: The deriving of the plurality of objective metrics comprises: For each diagnostic image of the plurality of diagnostic images, a deviation measure indicative of a deviation from a desired target acquisition time is identified.

10. The method according to any one of claims 1 to 3, wherein: The fitness score is based on a weighted sum of one or more derived objective metrics.

11. The method according to claim 10, wherein: The respective weighting factors are adjusted based on one or more hemodynamic parameters to be modeled using the physiological model to be generated based on one or more images to be selected from the set of candidate images.

12. An apparatus for selecting one or more diagnostic images to generate a physiological model, comprising: an input unit configured to obtain a plurality of diagnostic images of a target structure, a computing unit configured to derive a plurality of target metrics including at least one corresponding target metric for each of the plurality of diagnostic images, wherein the target metric indicates one or more characteristics of the target structure in the corresponding diagnostic image, an analyzing unit configured to analyze the plurality of target metrics, and A selection unit is configured to select a set of candidate images for generating the physiological model of the target structure from the multiple diagnostic images based on the analysis of the multiple target measurements, and to assign a fitness score to each candidate image in the set of candidate images, the fitness score indicating the suitability of the corresponding candidate image for generating the physiological model.

13. The device according to claim 12, wherein: The selection unit comprises a classifier that has been trained using a training data set that relates one or more diagnostic images to correspondingly measured hemodynamic parameter data.

14. A computer program product for controlling an apparatus according to claim 12 or 13, the computer program product being adapted to perform the method according to any one of claims 1 to 10 when executed by a processing unit.

15. A computer readable medium having stored thereon a computer program according to claim 14.

Images