ANGIOGRAPHY DERIVED CORONARY FLOW

Abstract

An apparatus and a method for assessing a vasculature is provided in which a time series of diagnostic images is used in combination with at least one boundary parameter associated with said time series to determine a quantitative fluid dynamics parameter indicative of the fluid flow through the vasculature using a trained classifier. By providing both, the time series of diagnostic images and the at least one boundary parameter to the determination, it is ensured that the classifier is provided with consistent data allowing for a more accurate determination of the quantitative fluid dynamics parameter.

Claims

1. An apparatus for assessing a vasculature, comprising: an input unit configured to receive a time series of diagnostic images of the vasculature, and at least one boundary parameter associated with said time series of diagnostic images; a computation unit comprising a trained classifier device, the computation unit configured to generate a combination result based on the time series of diagnostic images and the at least one boundary parameter, and determine, using the trained classifier device, a quantitative fluid dynamics parameter indicative of the fluid flow through the vasculature based on the combination result.

2. The apparatus according to claim 1, wherein the computation unit further comprises a processing unit, wherein the trained classifier device is configured to receive the time series of diagnostic images, classify the time series of diagnostic images based on a trained ground truth to generate a classification result, and provide the classification result to the processing unit, wherein the processing unit is configured to receive the classification result, generate the combination result based on the classification result and the at least one boundary parameter, and determine the quantitative fluid dynamics parameter based on the combination result.

3. The apparatus according to claim 1, wherein the computation unit further comprises a processing unit, wherein the processing unit is configured to generate the combination result based on the time series of diagnostic images and the at least one boundary parameter, and provide the combination result to the trained classifier device, wherein the trained classifier device is configured to receive the combination result, classify the combination result based on a trained ground truth to generate a classification result, and provide the classification result to the processing unit, wherein the processing unit is further configured to receive the classification result based on the combination result, and determine the quantitative fluid dynamics parameter based on the classification result.

4. The apparatus according to claim 1, wherein the trained classifier device is trained with a ground truth for the quantitative fluid dynamics parameter, wherein the trained classifier device is trained using a virtual time series of diagnostic images indicative of a contrast agent dynamic through the vasculature.

5. The apparatus according to claim 4, wherein the virtual time series of diagnostic images is generated by defining at least one virtual vessel tree, defining a virtual contrast agent injection rate, and modelling the flow speed through the least one vessel tree based on a fluid dynamics model.

6. The apparatus according to claim 1, wherein the combination result is generated by using the at least one boundary parameter associated with said time series of diagnostic images to perform an adjustment of the time series of diagnostic images.

7. The apparatus according to claim 6, wherein the adjustment comprises one or more of: a normalization of a frame rate, an adjustment of an image contrast, a normalization of an image resolution, an adjustment of a sequence length, a selection of projection angles.

8. The apparatus according to claim 1, wherein the at least one boundary parameter comprises at least one system parameter and/or at least one measurement boundary parameter.

9. The apparatus according to claim 8, wherein the at least one boundary parameter comprises one or more of: a frame rate, a projection angle, a projection resolution, a contrast agent injection rate, a contrast agent volume, a contrast agent dilution, an injection pressure, an injection timing.

10. The apparatus according to claim 1, wherein the computation unit comprises a processing unit, wherein the processing unit comprises a second trained classifier device.

11. A method for assessing a vasculature, comprising the steps of receiving a time series of diagnostic images of the vasculature, receiving at least one boundary parameter associated with said time series of diagnostic images, generating a combination result based on the time series of diagnostic images and the at least one boundary parameter, and determining, using a trained classifier device, a quantitative fluid dynamics parameter indicative of the fluid flow through the vasculature based on the combination result.

12. The method according to claim 11, further comprising generating, by the trained classifier device, a classification result by receiving the time series of diagnostic images and classifying the time series of diagnostic images based on a trained ground truth, generating the combination result based on the classification result and the at least one boundary parameter, and determining the quantitative fluid dynamics parameter based on the combination result.

13. The method according to claim 11, further comprising generating the combination result based on the time series of diagnostic images and the at least one boundary parameter, classifying, by the trained classifier device, the combination result based on a trained ground truth to generate the classification result, and determining the quantitative fluid dynamics parameter based on the classification result.

14. A computer program for controlling an apparatus, which, when executed by a processing device, is adapted to perform the method according to claim 11.

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

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0087] In the following drawings:

[0088] FIG. 1 schematically illustrates an apparatus for assessing a vasculature according to a first exemplary embodiment,

[0089] FIG. 2 represents a flow chart for a method for assessing a vasculature according to the first exemplary embodiment,

[0090] FIG. 3 schematically illustrates an apparatus for assessing a vasculature according to a second exemplary embodiment,

[0091] FIG. 4 represents a flow chart for a method for assessing a vasculature according to the second exemplary embodiment,

[0092] FIG. 5 represents a flow chart for a method for training a classifier device based on a virtual training dataset in accordance with the invention,

[0093] FIG. 6 represents a flow chart for a method for classifying an input dataset in accordance with the invention, and

[0094] FIG. 7 schematically illustrates an exemplary embodiment for a neural network that may be used as a classifier device according to the invention.

DETAILED DESCRIPTION OF EMBODIMENTS

[0095] The illustration in the drawings is schematically. In different drawings, similar or identical elements are provided with the same reference numerals.

[0096] FIG. 1 represents schematically a first exemplary embodiment of an apparatus 1 for assessing a vasculature, in particular a coronary vasculature, based on at least one quantitative fluid dynamics parameter, such as a coronary flow reserve, that has been derived on the basis of a time series of diagnostic images and a set of associated boundary parameters.

[0097] The apparatus 1 comprises an input unit 100 and a computation unit 2. In the specific embodiment of FIG. 1, the computation unit 2 comprises a processing unit 200 and a trained classifier device 300. The trained classifier device 300 is, in the exemplary embodiment according to FIG. 1, implemented as a neural network comprising a plurality of nodes that are interconnected with one another, such that information input to the neural network can be communicated amongst the individual nodes.

[0098] The apparatus 1 may further comprise or be communicatively connected to a display device 400 which is configured to generate a graphical representation of the results of the assessment performed by apparatus 1 and present the graphical representation to a user, such as a physician, e.g. for potential further analysis and/or therapy planning.

[0099] Input unit 100 is configured to receive a time series of diagnostic images 10. In the specific exemplary embodiment according to FIG. 1, these diagnostic images have been obtained using X-ray angiography of a coronary vasculature upon contrast agent injection. As such, the diagnostic images in the time series of diagnostic images are indicative of the progression of the contrast agent through the vasculature over time. This is the case since, by using X-ray angiography, the contrast agent is visible in the diagnostic images and, accordingly, it can be tracked, where in the vasculature the contrast agent has already progressed and where no contrast agent may be found. Since the time series of diagnostic images allows tracking the progression of the contrast agent through the vasculature, they may be used to derive information on the fluid flow properties through the vasculature.

[0100] Input unit 100 is further configured to receive a dataset 20 specifying at least one boundary parameter. In the exemplary embodiment of FIG. 1, the dataset 20 specifies multiple boundary parameters comprising both, system parameters and measurement boundary parameters. More particularly, in the specific embodiment according to FIG. 2, the input unit receives a dataset 20 comprising an indication about the frame rate, the projection angle and the projection resolution for the time series of the diagnostic images received at the input unit as system parameters as well as a contrast agent injection rate, a contrast agent volume, a contrast agent dilution, and injection pressure and an injection timing for the time series of the diagnostic images as measurement boundary parameters. Accordingly, the system and measurement boundary parameters received are associated with the time series of diagnostic images.

[0101] It shall be noted that, although the dataset 20 comprising the at least one boundary parameter comprises all these different parameters, also only a subset thereof may be considered to assess the vasculature.

[0102] The input unit 100 then provides the time series of diagnostic images 10 and the dataset 20 specifying the associated set of boundary parameters to computation unit 2. In the exemplary embodiment according to FIG. 1, the time series of diagnostic images 10 may particularly be provided to processing unit 200. Processing unit 200 receives the time series of diagnostic images 10 and the dataset 20 and processes them in order to generate a combination result based thereon.

[0103] In the specific exemplary embodiment of FIG. 1, this processing, by the processing unit 200, particularly encompasses adjusting the diagnostic images in the time series using the dataset 20 of boundary parameters.

[0104] In order to perform such an adjustment, the processing unit 200, in the specific embodiment according to FIG. 1, normalizes the frame rate of the time series of diagnostic images. This normalization may be performed by leaving out particular frames of the time series of diagnostic images. Alternatively or additionally, the normalization may be performed by interpolating individual frames of the time series of diagnostic images. The normalization may hereby particularly be performed using respective system parameters as boundary parameters, such as the frame rate of the time series of diagnostic images.

[0105] In the specific embodiment according to FIG. 1, the adjustment further comprises an image contrast adjustment of the diagnostic images in the time series of diagnostic images. Such image contrast adjustment may particularly be performed, by the processing unit 200, based on measurement boundary parameters as boundary parameters, in particular measurement boundary parameters relating to the contrast agent properties. These measurement boundary parameters may hereby particularly include parameters such as the concentration of the contrast agent upon injection, the contrast agent volume and/or the contrast agent injection rate.

[0106] Further, in the specific embodiment according to FIG. 1, the adjustment comprises a normalizing of the image resolution based on respective system parameters as boundary parameters, such as the projection resolution. Subsequently, the image sequence length may be adjusted to the contrast timing based on one or more measurement parameters as boundary parameters, such as the contrast agent volume or the like. This processing results in the most meaningful images being passed to the trained classifier device.

[0107] In the specific embodiment of FIG. 1, the adjustment of the time series of diagnostic images further comprises selecting particular projection angles to generate a combined stack of projection images from the diagnostic images. Hereby, some projection angles may be excluded, while others may be more preferred. This adjustment once more may make use of respective system parameters and/or measurement boundary parameters, such as the projection angle and/or a projection resolution or the like.

[0108] In the exemplary embodiment according to FIG. 1, the processing unit 200 thus uses the dataset 20 indicative of the boundary parameters to adjust the time series of diagnostic images 10. In doing so, the processing unit 200 generates a combination result 30 based on the time series of diagnostic images 10 and the dataset 20. The processing unit 200 then provides the combination result 30 to the trained classifier device 300.

[0109] In the exemplary embodiment according to FIG. 1, the trained classifier device 300 corresponds to a neural network, in particular a 2.5 D encoder network architecture. The trained classifier device 300 has been trained with a ground truth for classification of the combination result 30 as explained further below with reference to FIG. 6.

[0110] The trained classifier device 300 receives the combination result 30 and classifies the combination result 30 based on the trained ground truth to generate a classification result 40. The trained classifier device 300 then outputs the classification result 40 and provides said classification result 40 to the processing unit 200 for further processing.

[0111] The processing unit 200 receives the classification result 40 which has been generated based on the combination result 30. In the specific exemplary embodiment according to FIG. 1, the processing unit 200 determines, based on the classification result, a quantitative fluid dynamics parameter for the vasculature that has been shown in the time series of diagnostic images.

[0112] In some embodiments, the classification result may particularly provide the quantitative fluid dynamics parameter directly. In these embodiments, the trained classifier device may particularly use the already normalized input to the trained classifier device to predict said quantitative fluid dynamics parameter, such as, for example, a quantitative flow velocity value measured in mm/s.

[0113] In some embodiments, the trained classifier may not provide the quantitative fluid dynamics parameter directly, but rather further processing on the classification result is done. As an example, when determining the coronary flow reserve, a first classification result may be provided representative of the patient under rest and a second classification result may be provided representative of the patient under hyperemia. The ratio of these classification results may then be used to determine a CFR value.

[0114] In some embodiments, densitometry may be used. In these embodiments, the classification result may comprise an indication about a contrast agent volume in one or more particular vessels of interest in the vasculature. Hereby, the sum of all contrast agent volumes in each of the vessels may, for example, be compared to the amount of contrast agent injected. This may then be normalized using the contrast agent dilution. In case the comparison shows that the summed up amount of the contrast agent in the entire vasculature is smaller than the amount of contrast agent injected, a volumetric flow rate and/or a relative flow rate may be calculated for particular different vessels in the vasculature. Alternatively or additionally, the determination may allow to derive the flow speed of the fluid flow through a particular vessel in the vasculature by using an approximation for the cross sectional area of said vessel. Further possibilities of deriving quantitative fluid dynamics parameters are also foreseen.

[0115] An information about the thus determined quantitative fluid dynamics parameter is then provided to display unit 400. Display unit 400 receives the information about the quantitative fluid dynamics parameter and generates a graphical representation thereof. The display unit 400 then displays the graphical representation of the information about the quantitative fluid dynamics parameter on a respective display device for a user to visually acknowledge the information. In some embodiments, the graphical representation provided to the user may further comprise a graphical presentation of the vasculature represented in the time series of diagnostic images. In some embodiments, the graphical representation may comprise a graphical representation of one or more selected diagnostic images from the time series. In some embodiments, the graphical representation may comprise further information, such as information related to the boundary parameters used for that particular dataset.

[0116] FIG. 2 schematically represents a method for assessing a vasculature to be performed by the apparatus 1 according to the first exemplary embodiment. It is noted that the flow chart of the method according to FIG. 2 is to be understood as an exemplary embodiment and that the invention is not limited to this exemplary embodiment.

[0117] In the exemplary embodiment according to FIG. 2, input unit 100, in step S101, receives a time series of diagnostic images 10 upon contrast agent injection. These diagnostic images may, for example, have been acquired using X-ray angiography, and may, as an example, represent a coronary vasculature.

[0118] In the exemplary embodiment of FIG. 2, the diagnostic images in the time series of diagnostic images may particularly be considered as being indicative of the progression of the contrast agent through the coronary vasculature over time. This allows drawing conclusions with respect to the fluid flow properties inside the vasculature.

[0119] In step S102, input unit 100 further receives a dataset 20 specifying at least one boundary parameter. The dataset 20 is associated with the time series in that it specifies one or more boundary parameters that relate to the acquisition of the time series of diagnostic images. Specifically, in the exemplary embodiment of FIG. 2, the dataset 20 specifies a plurality of boundary parameters, in particular system parameters, such as frame rate, projection angle, projection resolution or the like and measurement boundary parameter, such as a contrast agent injection rate, a contrast agent volume, a contrast agent dilution, and injection pressure, an injection timing for the time series of the diagnostic images or the like.

[0120] In step S103, the input unit 100 then provides the time series of diagnostic images 10 and the dataset 20 specifying the associated set of boundary parameters to processing unit 200 of computation unit 2.

[0121] In step S201, processing unit 200 receives the time series of diagnostic images 10 and the dataset 20 and, in step S202, adjusts the diagnostic images in the time series 10 using the dataset 20 of boundary parameters. In the specific embodiment of FIG. 2, this adjustment comprises normalizing the frame rate of the time series of diagnostic images based on the system parameters by leaving out particular frames of the time series of diagnostic images and/or by interpolating individual frames of the time series of diagnostic images. The adjustment may further comprise an image contrast adjustment of the diagnostic images in the time series of diagnostic images based on measurement boundary parameters. Such image contrast adjustment may particularly be performed, by the processing unit 200, based on measurement boundary parameters, in particular measurement boundary parameters relating to the contrast agent properties. These measurement boundary parameters may hereby particularly include parameters such as the concentration of the contrast agent upon injection, the contrast agent volume and/or the contrast agent injection rate. The adjustment of step S202 may further comprise a normalizing of the image resolution based on respective system parameters as boundary parameters, such as the projection resolution. In the specific embodiment of FIG. 2, the image sequence length is further adjusted to the contrast timing based on one or more measurement parameters as boundary parameters, such as the contrast agent volume or the like.

[0122] The adjustment of step S202 may optionally comprise selecting particular projection angles to generate a combined stack of projection images from the diagnostic images based on respective system and/or measurement boundary parameters. Hereby, some projection angles may be excluded, while others may be more preferred.

[0123] The output of step S202 in the method according to FIG. 2 corresponds to the combination result 30 comprising the diagnostic images of the time series of diagnostic images 10 adjusted based on the dataset 20. In step S203, the processing unit 200 provides the combination result 30 to the trained classifier device 300, which in the exemplary embodiment according to FIG. 2, corresponds to a neural network.

[0124] In step S301, the trained classifier device 300 receives the combination result 30 and, in step S302, classifies the combination result 30 based on the trained ground truth to generate a classification result 40.

[0125] In step S303, the trained classifier device 300 provides the thus generated classification result 40 to the processing unit 200 for further processing.

[0126] In step S204, the processing unit 200 receives the classification result 40 and determines in step S205, based on the classification result, a quantitative fluid dynamics parameter for the vasculature as represented in the time series of diagnostic images.

[0127] In step S206, the processing unit 200 then provides an information about the quantitative fluid dynamics parameter to display unit 400.

[0128] In step S401, display unit 400 receives the information about the quantitative fluid dynamics parameter and, in step S402, generates a graphical representation thereof. In step S403, the display unit 400 then displays the graphical representation of the information about the quantitative fluid dynamics parameter on a respective display device. This allows a user to review the information.

[0129] FIG. 3 represents schematically a second exemplary embodiment of an apparatus 1′ for assessing a vasculature, in particular a coronary vasculature, based on at least one quantitative fluid dynamics parameter, such as a coronary flow reserve, that has been derived on the basis of a time series of diagnostic images and a set of associated boundary parameters. The apparatus 1′ according to FIG. 3 largely corresponds to the apparatus according to FIG. 1. Hereby, similar components are specified using like reference numerals. That is, apparatus 1′ also comprises an input unit 100 and a computation unit 2′. The computation unit of FIG. 3 also comprises a processing unit 200 and a trained classifier device 300′, which, in the exemplary embodiment of FIG. 3, is implemented as a neural network comprising a plurality of nodes that are interconnected with one another.

[0130] Apparatus 1′ according to FIG. 3 also comprises or is communicatively connected to a display device 400 which is configured to generate a graphical representation of the results of the assessment performed by apparatus 1′.

[0131] The procedures described in relation to the first embodiment according to FIG. 1 largely equally apply for the second embodiment according to FIG. 3. That is, input unit 100 is configured to receive a time series of diagnostic images 10 and a dataset 20 specifying at least one boundary parameter.

[0132] In the specific exemplary embodiment according to FIG. 3, the diagnostic images may particularly correspond to diagnostic images acquired using X-ray angiography upon contrast agent injection. Since the contrast agent is visible in X-ray angiography images, the diagnostic images in the time series of diagnostic images are indicative of the progression of the contrast agent through the vasculature over time. This means that, also in the exemplary embodiment of FIG. 3, the time series of diagnostic images 10 allows to track the progression of the contrast agent through the vasculature, they may be used to derive information on the fluid flow properties through the vasculature.

[0133] The dataset 20 specifying the at least one boundary parameter specifies, in the exemplary embodiment of FIG. 3, a plurality of boundary parameters which allow to adjust the diagnostic images in the time series of diagnostic images 10 as described in detail herein above.

[0134] However, contrary to the embodiment according to FIG. 1, in the exemplary embodiment according to FIG. 3, the adjusting of the time series of diagnostic images 10 in order to generate the combination result is performed after the classification by the trained classifier device 300′. Accordingly, in the specific embodiment according to FIG. 3, the trained classifier device has been trained with a ground truth relating to the time series of diagnostic images 10 as acquired rather than a ground truth relating to the combination result comprising the adjusted diagnostic images. In the specific embodiment according to FIG. 3, the time series of diagnostic images 10 is provided to the trained classifier device 300. The trained classifier device 300 classifies the time series of diagnostic images 10 and generates a respective classification result 40′ comprising the classified diagnostic images.

[0135] The trained classifier device 300 provides the classification result 40′ to the processing unit 200. Further, in the exemplary embodiment of FIG. 3, the input unit 100 provides a dataset 20 indicative of a plurality of boundary parameters comprising system parameters as well as measurement boundary parameters to the processing unit 200. In the specific embodiment according to FIG. 3, the dataset 20 particularly comprises an indication about the frame rate, the projection angle and the projection resolution for the time series of diagnostic images 10 as well as a contrast agent injection rate, a contrast agent volume, a contrast agent dilution, and injection pressure and an injection timing for the time series of diagnostic images 10.

[0136] In the embodiment according to FIG. 3, the processing unit 200 uses the dataset 20 indicative of the plurality of boundary parameters to adjust the classification result, in particular the classified diagnostic images. In the specific embodiment according to FIG. 3, this adjustment may comprise the same steps as specified in relation to FIG. 1. That is, the processing unit 200 may particularly perform a normalization of the frame rate and/or the image resolution, an image contrast adjustment, a sequence length adjustment and a selection of particular projection angles in order to generate the combination result 30 that is based on the classification result 40′ and the dataset 20.

[0137] The processing unit then uses the combination result 30 to determine a quantitative fluid dynamics parameter for the vasculature to be assessed. A respective information or indication about the thus determined quantitative fluid dynamics parameter is then provided to display unit 400 for subsequent displaying as described in relation to FIG. 1. FIG. 4 schematically represents a method for assessing a vasculature as performed by the apparatus 1′ according to the second embodiment.

[0138] In step S101, input unit 100 receives a time series of diagnostic images 10, which, in the specific embodiment according to FIG. 4, have been acquired using X-ray angiography. Further, the input unit 100 receives, in step S102, a dataset 20 indicative of a plurality of boundary parameters associated with the time series of diagnostic images 10.

[0139] In step S103, the input unit 100 provides the time series of diagnostic images 10 to the trained classifier device 300 and the dataset 20 specifying the associated set of boundary parameters to processing unit 200.

[0140] In step S301, trained classifier device 300 receives the time series of diagnostic images 10 and, in step S302, classifies the time series of diagnostic images 10 based on the trained ground truth as described herein above to generate the classification result 40′ comprising a plurality of classified diagnostic images. In step S303, the trained classifier device 300 provides the classification result 40′ to the processing unit 200.

[0141] In step S201, processing unit 200 receives the classification result 40′ comprising the plurality of classified diagnostic images from the trained classifier device 300 and the dataset 20 from the input unit.

[0142] Subsequently, in step S202, processing unit 200 uses dataset 20 to adjust the classification result 40′, particularly the plurality of classified images therein, using the dataset 20 of boundary parameters. The adjustment of the classified images in the embodiment of FIG. 4 is hereby performed in the same manner as described in relation to the embodiment of FIG. 1 and may thus comprise a normalization of the frame rate and the image resolution, an adjustment of the image contrast and the image sequence length and a selection of particular projection angles.

[0143] The processing unit then uses, in step S203, the thus generated combination result 30 based on the classification result 40′ and the dataset 20 to determine the quantitative fluid dynamics parameter for the vasculature imaged in the time series of diagnostic images. In step S204, the processing unit 200 provides an information about the quantitative fluid dynamics parameter to display unit 400.

[0144] In the specific embodiment of FIG. 4, the display unit 400 receives, in step S401, the information about the quantitative fluid dynamics parameter and, in step S402, generates a graphical representation thereof. Subsequently, in step S403, the display unit 400 displays the graphical representation of the information about the quantitative fluid dynamics parameter to a user.

[0145] FIG. 5 schematically represents a flow chart for a method for training a classifier device based on a virtual training dataset. Said training dataset used for training the classifier device is preferably similar to an actual dataset for the use case and further comprises a ground truth for a plurality of fluid dynamics parameter values. The vasculature in the exemplary embodiment according to FIG. 5 corresponds to a coronary vasculature.

[0146] In order to train the classifier device, the embodiment according to FIG. 5 foresees, in step S1000, that a virtual vasculature comprising a set of virtual coronary trees is specified. In step S1100, a virtual contrast agent injection rate and a respective virtual coronary flow speed through the virtual coronary tree a specified.

[0147] Hereby, in the specific embodiment according to FIG. 5, a lumped parameter model is used to define variable flow speeds throughout the coronary tree. Particularly, in order to vary the speed of the fluid through the coronary tree, the microvascular resistance boundary conditions for the lumped parameter model are changed, such as to increase the flow speed (in case of a lower resistance) or reduce the flow speed (in case of a higher resistance).

[0148] In step S1200, the coronary tree or coronary trees in the vasculature are combined with a motion model allowing to introduce movement of the vessels in the vasculature.

[0149] Subsequently, in step S1300, the moving vessels of the coronary tree in the vasculature are forward projected onto an empty clinical background. For each vessel in the vasculature and each coronary tree formed by the vessels, different fluid flow speeds, injection times, contrast agent concentrations and image frames per second are modelled. In some embodiments, up to 200, even up to 500 different values for the above variables may be modelled. During the modelling, different backgrounds may be selected randomly. In the specific embodiment according to FIG. 5, up to 50 or even up to 100 different coronary trees are used in order to account for variability in the coronary vasculature. Using the modelling based on these 200 to 500 different values for the variables and the 50 to 100 different coronary trees, a first training dataset is generated comprising training information about the variable factors in the training.

[0150] In step S1400, a second training dataset is generated. Hereby, the forward projected diagnostic images as also used in step S1300 are used again. However, in this case, the above-indicated variables relating to the constant injection times, contrast agent concentrations, and imaging frames per second are maintained constant. Further, the flow speeds in the coronary trees remain the same as before.

[0151] In step S1500, the second training dataset is provided to the classifier device 300. In the exemplary embodiment according to FIG. 5, the classifier device 300 corresponds to a neural network, particularly a 2.5 D encoder network architecture as shown in FIG. 7. The training of said neural network is performed using known training methods, such as back propagation, an Adam optimizer with batch normalization or the like.

[0152] The neural network having a 2.5 D encoder network architecture comprises seven input channels 51. The first seven diagnostic images of the time series of diagnostic images are provided to the seven input channels with one diagnostic image provided per channel. The corresponding ground truth output corresponds to the average fluid velocity from the lumped model.

[0153] In the specific embodiment according to FIG. 6, the network is trained for about 100 epochs in step S1600 and, subsequently, the network having the best validation error is picked. Both, a loss function cross entropy and an Adam optimizer are used.

[0154] Upon finishing training, the trained classifier device is applied, in step S1700, to the first training dataset. This application gives a plurality of fluid dynamic parameter values. The quantitative values along with the information initially input relating to the injection times, contrast agent concentrations, and imaging frames per second for the first training dataset are taken as input set for a four-dimensional linear regression in step S1800 The quantitative ground truth fluid speed is then divided by the qualitative fluid dynamic values and used as an output set of the regression in step S1900, resulting in a respective correction factor.

[0155] FIG. 6 represents schematically a flow chart for a method for classifying an input dataset using a trained classifier device that has been trained as described herein above.

[0156] In step S2000 a time series of diagnostic images which is similar to the first training dataset is provided to the trained classifier device. In step S2100, a normalization of the data is performed by first subtracting every next frame from the previous and using a threshold on the absolute difference. This allows identifying the first frame of the contrast injection. Subsequently, starting from the first frame the next seven frames are considered. Before being provided to the trained classifier device, shutters are cropped off manually in the exemplary embodiment according to FIG. 6 and a rescaling to a fixed number of pixels is performed for the diagnostic images. Hereby, the same number of pixels as in the training data is used.

[0157] In step S2200, the resulting seven normalized frames are provided to the trained classifier device which, based thereupon, generates an output in step S2300. This output, together with the injection times, contrast agent concentrations, and imaging frames per second for the angiography data is entered into the equation resulting from the regression analysis in step S2400. Finally, in step S2500, the correction factor is multiplied with the classification result output from the trained classifier device.

[0158] FIG. 7 schematically illustrates an exemplary embodiment for a neural network that may be used as a classifier device. The neural network according to the exemplary embodiment of FIG. 7 has a 2.5 D network architecture having seven input channels 51 that are distributed onto 32 channels 52, distributed, in the next level to 64 channels 53 and 128 channels 54. The output 55 of the neural network is the average fluid speed through the vasculature.

[0159] Although in above described embodiments, the diagnostic images have been obtained using X-ray angiography, it shall be understood that in other embodiments, the diagnostic images may be obtained by other imaging methods, such as helical computed tomography or sequential computed tomography, dual energy X-ray, spectral X-ray, magnetic resonance imaging, ultrasound imaging, or the like.

[0160] Further, it shall be understood that, although in the above embodiments, the input unit and the computation unit are implemented as several separate entities, these units may also correspond to the same entity. More specifically, they may be implemented as respective modules and/or a computer program to be executed by a processing device.

[0161] Further, while in the above embodiments, the assessment has been described in particular in relation to the coronary vasculature, it shall be understood that, in other embodiments, the assessment may likewise be performed on other vascular anatomies, such as peripheral, abdominal or neurovascular. Further kinds of vascular anatomies are also foreseeable.

[0162] It may further be understood that while in the above-embodiments, the training of the classifier device has been performed on the basis of a virtually generated training dataset, the training may likewise be performed on the basis of other kinds of datasets, such as measured data that has been accordingly processed to form training dataset.

[0163] Further, it shall be understood that, although in the above embodiments, the classifier device particularly corresponds to a 2.5 D encoder neural network architecture, other architectures for implementing machine learning and/or deep learning techniques may be used for the purpose of the present invention.

[0164] 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.

[0165] In the claims, the word “comprising” does not exclude other elements or steps, and the indefinite article “a” or “an” does not exclude a plurality.

[0166] 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.

[0167] Procedures like the generating of the combination result, the determining of the quantitative fluid dynamics parameter, the classifying of the data, the adjusting of the data, et cetera, performed by one or several units or devices can be performed by any other number of units or devices. These procedures, particularly the classifying of the data and the processing of the data in order to obtain the quantitative fluid dynamics parameter, as performed by the apparatus in accordance with the assessment method, can be implemented as program code means of a computer program and/or as dedicated hardware.

[0168] A computer program may be stored/distributed on a suitable medium, such as an optical storage medium or a solid-state medium, supplied together with or as part of other hardware, but may also be distributed in other forms, such as via the Internet or other wired or wireless telecommunication systems.

[0169] Any reference signs in the claims should not be construed as limiting the scope.

[0170] The invention relates to an apparatus for assessing a vasculature, comprising an input unit configured to receive a time series of diagnostic images of the vasculature and at least one boundary parameter associated with said time series of diagnostic images, a computation unit comprising a trained classifier device, whereby the computation unit is configured to generate a combination result based on the time series of diagnostic images and the at least one boundary parameter and determine, using the trained classifier device, a quantitative fluid dynamics parameter indicative of the fluid flow through the vasculature based on the combination result.

[0171] By means of this arrangement, an accurate, robust and simple derivation of flow-related indices which are important diagnostic indicators in the assessment of a vasculature, in particular a coronary vasculature, is achieved.