Method for post-processing a perfusion acquisition sequence by a medical imaging device

Abstract

The present invention relates to a method (100) for post-processing sampled time-dependent experimental perfusion signals (S, 10, 12) in order to generate (140) pharmacokinetic parameters (QI). Such a method is implemented by a processing unit (4) of a medical imaging analysis system, the unit being pre-trained in a process (101) that enables learning of the destructive influence of the acquisition of a perfusion sequence on the arterial signal and the redundancy of information related to the arterial input function (AIF, AIF´) shared by at least two sets of tissue signals. Such an arterial input function (AIF, AIF´) is thus directly generated (110) in step (120) by the processing unit (4) trained from the first arterial input function (AIF0) and the pre-selected tissue signal (S) as such.

Description

Technical Field

[0001] The present invention relates to a method for post-processing a perfusion acquisition sequence by a medical imaging device.

Background Art

[0002] The medical imaging device can be a magnetic resonance imaging, i.e., MRI device, or an imaging device based on the use of X-rays such as a CT scanner. Such a post-processing method makes it possible to estimate a sample of the arterial input function, also known by the acronym AIF. This sample has the remarkable feature of being very insensitive to the effects of acquisition and thus maintaining linearity with respect to the concentration of the contrast agent, which is essential for the quantitative estimation of pharmacokinetic parameters, which ultimately makes it possible to characterize lesions such as tumors or ischemic tissue.

[0003] Magnetic resonance imaging is based on the analysis of the response of the protons of water molecules when the protons of water molecules are excited in a magnetic field. This response depends on the environment of such protons and thus makes it possible to distinguish different types of tissues. A nuclear magnetic resonance imaging device such as device 1 of the medical imaging analysis system SAIM shown as a non-limiting example in FIGS. 1 and 2 is generally used. This provides a plurality of digital image sequences 12 of one or more parts of the patient's body, non-limiting examples being the brain, the heart, the lungs. For this purpose, the device applies a combination of high-frequency electromagnetic waves to the part of the body in question and measures the signal re-emitted by certain atoms such as hydrogen for nuclear magnetic resonance imaging, as a non-limiting example. Thus, this device makes it possible to determine the magnetic properties, and as a result the chemical composition and thus the nature of the biological tissue, in each elementary volume, generally called a voxel, of the imaged volume. As shown in FIG. 1, the nuclear magnetic resonance imaging device 1 is controlled by a console 2. Thus, a user 6, for example, an operator, a doctor, or a researcher, can select a command 11 to control the device 1 based on parameters or instructions 16 input by the input human-machine interface 8 of the analysis system. Such a human-machine interface 8 can be composed of, for example, a computer keyboard, a pointing device, a touch screen, a microphone, or, more generally, any interface provided to represent gesture commands or instructions issued by a human 6 as control or setting data. Based on the information 10 generated by the device 1, a plurality of digital image sequences 12 of a part of the human or animal body are obtained. Such information 10 or images 12 will also be referred to as "experimental data".

[0004] CT scanners in modern radiology share some usage similarities with magnetic resonance imaging devices as shown in FIGS. 1 and 2. These similarities can illustrate a variant of the medical imaging analysis system SAIM incorporating an imaging device 1 in the form of a CT scanner instead of a magnetic resonance imaging device. The present invention will be described below mainly for the purpose of solving the influence of acquisition by magnetic resonance imaging devices. However, the specific description may also apply to other imaging devices (such as CT scanners) as required.

[0005] The image sequence 12 can optionally be stored in a server 3, i.e., a computer equipped with its own storage means, and can constitute the patient's medical file 13. Such a file 13 can include different types of images, such as functional images showing the activity of tissues or anatomical images reflecting the characteristics of tissues. The image sequence 12 or more generally the experimental data is analyzed by the processing unit 4, which is provided for this purpose. Such a processing unit 4 can be composed of, for example, one or more microprocessors or microcontrollers that execute appropriate application program instructions placed in the storage means of the imaging analysis system. "Storage means" means any volatile or preferably non-volatile computer memory. Non-volatile memory is computer memory by a technology that enables it to retain its data without a supply of electrical energy. It can include data resulting from inputs, calculations, measurements, and / or program instructions. The main currently available non-volatile memories are of the electrically writable type such as EPROM (Erasable Programmable Read-Only Memory), or of the more electrically writable and erasable type such as EEPROM (Electrically Erasable Programmable Read-Only Memory), flash, SSD (Solid State Drive), etc. Non-volatile memory is distinguished from memory known as "volatile" where data is lost in the absence of a power supply. The main currently available volatile memories are of types such as RAM (Random Access Memory, also called "Read-Write Memory"), DRAM (Dynamic Random Access Memory which requires periodic refreshing), SRAM (Static Random Access Memory which requires such refreshing in the event of a power drop), DPRAM or VRAM (especially suitable for video), etc. In the remainder of this document, "data memory" can be volatile or non-volatile depending on the application in question.

[0006] The processing unit 4 includes means for communicating with the outside world in order to collect images. Moreover, the communication means enables the processing unit 4 to ultimately provide or output to the user 6 of the imaging analysis system, via the output human-machine interface 5, graphic and / or acoustic representations, for example, of estimated values or defined amounts of biomarkers or pharmacokinetic parameter QIs created by the processing unit 4 based on the experimental data 10 and / or 12 obtained by magnetic resonance imaging. Throughout this document, "output human-machine interface" means any device used alone or in combination that enables the output or provision to the user 6 of the magnetic resonance imaging analysis system of a reconstructed physiological signal, in this case a biomarker, in graphic, haptic, acoustic, or more generally perceptible form to humans. Such an output human-machine interface 5 can be composed, without limitation, of one or more screens, speakers, or other suitable alternative means. Thus, the user 6 of the analysis system can confirm that the diagnosis is correct or invalidate the diagnosis, determine a treatment action considered appropriate, initiate further research, make the adjustment parameters of the measuring device more accurate, etc. Optionally, this user 6 can further set the operation of the processing unit 4 or the output human-machine interface 5 by means of the operation and / or acquisition parameters 16. Thus, for example, the user 6 can define a display threshold or select an estimated or quantified parameter for which a biomarker, indicator, or display is desired. For this purpose, the user utilizes the input human-machine interface 8 described above or a second input interface provided for this purpose. Advantageously, the input human-machine interface 8 and the output human-machine interface 5 can constitute just one and the same physical entity. The input human-machine interface 8 and the output human-machine interface 5 of the imaging analysis system can further be integrated into the acquisition console 2. There are variants that will be described in relation to Figure 2.In contrast, the imaging system as described above further includes a preprocessing unit 7 that analyzes the image sequence 12, estimates the experimental signals 15 therefrom, and thus sends these experimental signals 15 to the processing unit 4 that is released from this task.

[0007] Among the techniques or methods based on magnetic resonance imaging, perfusion magnetic resonance imaging can be distinguished. Such techniques are illustrated in FIG. 3 showing a cardiac perfusion acquisition sequence, and are composed of applying a high-frequency pulse called a magnetization preparation high-frequency pulse to move the magnetization to its steady state (thereby eliminating the magnetization history as well), and subsequently applying a series of acquisition high-frequency pulses to spatially sample the volume of interest. The set of parameters selected by the technician aims to maximize or minimize various effects. Among the most important ones, one can find the selection of the type of pulse that enables obtaining a sequence called T1-weighted enhancement or T2 * weighted sensitivity. During acquisition, the technician injects a tracer in the form of a contrast agent into the venous system of the patient being examined. Thus, the magnetization of the volume of interest changes during the movement of the contrast agent distributed from corner to corner of the body by the circulatory system, and this change depends on the amount of the tracer. Thus, FIG. 3 shows such changes in the magnetization of the volume of interest at three separate instants t0, t2, t3 before and after the arrival of the tracer at the instant t1 in the volume of interest respectively. Thus, the perfusion sequence is composed of periodically obtaining the magnetization state of the volume of interest by repeatedly applying the acquisition high-frequency pulses that may be preceded by a preparation pulse in a specific case. Thus, the perfusion sequence enables obtaining a time sample S(t) of the change in the magnetization state due to the acquisition parameters selected by the technician for each basic volume or voxel.

[0008] In the case of a CT scanner, it differs from MRI in terms of its basic principle of higher or lower opacity of anatomical structures through which X-rays pass. Therefore, the principle of the perfusion sequence consists of injecting a contrast agent that enables changing this opacity. The main destructive effect that can be mentioned is the subsampling of the AIF.

[0009] As shown in Figure 4, the estimation of pharmacokinetic parameters is based on the theory of tracer dilution that enables the association as S = AIF * IRF by means of the convolution of the AIF with the concentration S of the contrast agent in the tissue (for five regions of interest A1 to A5 as shown in Figure 4) and a filter called "IRF" that represents the "impulse response" or "impulse response function" related to the recirculation of the contrast agent. Therefore, Figure 5 shows five time signals S respectively related to the regions A1 to A5.

[0010] By directly collecting samples of the concentration of the contrast agent in the tissue in the acquired images and selecting a proper sample of the AIF, the IRF, and thus the pharmacokinetic parameters, can be estimated.

[0011] The perfusion sequence makes it possible to expose lesions that are not visible in their original state, which is very useful for a physician who endeavors to establish a diagnosis and make a treatment decision in the treatment of the lesion. However, in order to complete the decision-making, whether in the clinical or research field, it is necessary to obtain an AIF that is hardly affected by acquisition. Figure 5 shows the signal S obtained by a perfusion sequence that depends on the concentration C of a contrast agent or tracer according to a plurality of acquisition parameters that explain a plurality of curves S. In fact, a magnetic resonance image captures the magnetization state of tissue, and the presence of a contrast agent acts indirectly on the signal to be acquired. The contrast agent will affect the rate at which the magnetization returns to the steady state. In other words, for one and the same set of acquisition parameters, the magnetization will vary according to the concentration of the contrast agent injected into the patient. This relationship is neither linear nor bijective. However, as shown by the figure on the right side represented in Figure 5, in the case of a low amount of contrast agent, it is usually considered that the relationship between the signal S acquired and the concentration C of the contrast agent is linear. Such a figure on the right side shows an enlarged view when the concentration C of the contrast agent is low. Regardless of the acquisition parameters, the curves S and C merge into a straight line identified as DI represented by a dashed line within a region ZL called the gray linearity region in Figure 5. Therefore, when the concentration C is low, the directly acquired signal S can be used without correcting the influence of acquisition, which is generally accepted for tissue signals. However, as shown by the curve in the figure on the left side of Figure 5, the higher the concentration C, the greater the influence of acquisition. Therefore, the approximation that is accepted for tissue signals may not apply to the arterial signal characterized by the maximum concentration of the contrast agent. Therefore, it is necessary to correct the influence of acquisition in order to estimate a proper AIF sample.

[0012] At present, several post - processing methods as disclosed in the paper "Physics - informed neural networks for myocardial perfusion MRI quantification", van Herten et al. 2022 have been proposed in the literature to correct the influence of this acquisition. Among them, some are related to the use of an AIF called population AIF (using an AIF database corresponding to a given injection protocol and patient type), or some that are hardly used, such as those that require blood samples taken simultaneously with the acquisition to truly know the amount of contrast agent present at a given instant. On the other hand, two methods that the present invention directly competes with, and a third method that the present invention proposes improvements to, are considered.

[0013] The first known method consists of digitally simulating the influence of the acquisition by a physical model derived from the Bloch equations. These models require very detailed knowledge of the perfusion sequence and should therefore be re - evaluated for each individual case. In addition, the signal S acquired in one and the same perfusion examination will depend on parameters specific to the tissue that generated them, in particular T1 and T2 * T1 and T2 * will therefore need to be estimated by other acquisition sequences and necessarily be evaluated by post - processing methods that have biases. Thus, the acquisition of additional parameters whose values vary for each voxel of the image brings about a cause of error, as well as a new problem to be solved, namely the multimodal registration between the perfusion image and the T1 and T2 maps. Finally, the set of parameters required for this method is often lost because it is not necessarily saved in daily clinical practice, thereby limiting the future reuse of perfusion acquisitions.

[0014] The second method proposes using a set of acquisition parameters that significantly limit the impact of the acquisition and thus allow the arterial signal to be considered a realistic AIF. However, such a method degrades the resolution of the image. As a result, it is not possible to very accurately estimate pharmacokinetic parameters using such images. In fact, this acquisition is combined with a conventional perfusion acquisition that provides sufficient resolution for the evaluation of these parameters by using the AIF obtained from an arterial signal of the acquisition optimized for this purpose. This method has the advantage of allowing the direct use of signals resulting from different acquisitions. However, it assumes that there is a technician trained in this type of setting and that this type of operation is available on the MRI device, which is still rare at present. In addition, this method assumes that two types of acquisitions are obtained simultaneously. In other words, this method cannot include any past perfusion acquisitions. However, over 30 years, many databases have been formed using conventional acquisition methods. This method would exclude them.

[0015] Finally, the third method shown in FIG. 6 proposes to estimate the AIF samples using all the redundant information shared by the tissue signals. The signal of each voxel representing a region of interest of the tissue, such as regions A1 to A5 shown as dashed circles in FIG. 6 as an example, is the convolution product between the IRF belonging to each voxel and the AIF shared by all the tissue voxels, and it is possible to simultaneously estimate the IRF of each voxel and the AIF of all the voxels. Since neither the IRF nor the AIF is known in this method, it is denoted by the term blind deconvolution. Such a method 100 is implemented, for example, by the processing unit 4 of the medical analysis system shown in FIG. 1 or 2 supplied with the experimental data 10 and / or 12. First, step 110 of selecting the tissue signal S and the first arterial input function AIF0 in the arterial region A0 generally represented by a dashed square in FIG. 6, and then step 120 of iteratively minimizing the residual between the measured tissue signal S and the one reconstructed based on the parametric model AIFm of the arterial input function AIF. The purpose is to adjust the parameters of the arterial input function model AIFm to minimize the residual. A very diverse number of methods for performing this minimization 120 are described in the literature. They are generally distinguished by the constraints imposed on the arterial input function model. Some methods are said to be "unconstrained" because the samples of the arterial input function can take any form. This first type of method is rarely used because it is very susceptible to the influence of acquisition noise. Other methods are parameterized by a given model. However, this second type of method is limited by the nature of the model of the arterial input function used, thus limiting the forms that the arterial input function can take.

[0016] The selection and number of tissue signals selected in step 110 vary according to the method selected. Even if it has been shown that, theoretically, those numbers can be reduced to two, including methods that do not require any constraints on the form of the arterial input function sought, restricting the number of signals to two makes a very large number of assumptions that are not really very realistic in practice, particularly with regard to the influence of noise and the type of perfusion model representing these two signals. Generally, the number of tissue signals used in the literature varies from six to twelve.

[0017] The second step 120 of minimizing the error generally consists of minimizing the sum of squared residuals between the signal selected in step 110 and that reconstructed based on the adjustment of the parameters of the arterial input function model and the pharmacokinetic parameters. Therefore, the iterative step 120 consists of searching for this set of parameters that enables the reduction of this total error. The adjustment is customarily performed by the gradient descent method, and each iteration involves a large number of mathematical operations that are costly in terms of computation time. Thus, the stopping criterion for determining the final number of iterations varies from study to study and may relate to the minimum change in error from one iteration to another or directly to the value of the error. In the literature, methods proposing 50 to several hundred iterations can be found, which results in implementation times of several minutes. When this second step is performed, the method enables the generation of a second arterial input function AIF that is considered to be "realistic" for the selected set of tissue signals. However, precisely due to the nature of the method, the method only shows the correct form of the required second arterial input function AIF. Therefore, it is necessary to perform step 130, which consists of adjusting the scale (height of the sample) of the AIF. There are multiple methods in the literature for performing this step of scaling or normalization. For example, it can be assumed that the required AIF shares the area under the curve of AIF0 or at least a portion that is hardly affected by acquisition. Next, the thus-obtained second corrected arterial input function AIF' can be used to estimate one or more of its pharmacokinetic parameters QI over all or part of the voxels of the perfusion image and thus create information that is appropriate for making treatment decisions. Such pharmacokinetic parameters QI can be the subject of any form of one or more graphical representations and can be the subject of output via an appropriate human-machine interface such as interface 5 related to the medical imaging analysis system according to FIG. 1 or 2, so that a member of the medical staff can, for example, make a diagnosis. SUMMARY OF THE INVENTION

[0018] The invention makes it possible to overcome all or part of the drawbacks raised by known or aforementioned solutions.

[0019] Among the many advantages provided by the invention, - a free form, i.e. an unconstrained form, for the arterial input function that is not very affected by acquisition noise, - a reduction in the number of tissue signals required for the implementation of the method, - the non-iterative implementation of the estimation of the arterial input function that makes it possible to obtain an execution time on the order of seconds with a standard computer for the general public, as opposed to several minutes according to the prior art, can be mentioned for the implementation of the post-processing method.

[0020] For this purpose, the invention relates to a method for post-processing sampled time-experimental signals resulting from a perfusion acquisition sequence by a medical imaging device and resulting from the movement of a tracer within the basic volume of an organ, said method being implemented by a processing unit of a medical imaging analysis system. Such a method - comprises a step of selecting a first arterial input function related to the arterial region of the organ and a set of tissue signals respectively related to separate tissue regions of said organ, - a step of generating a second arterial input function based on said first arterial input function and the selected set of tissue signals, - a step of creating pharmacokinetic parameters based on said second arterial input function and said experimental signal.

[0021] In order to increase the implementation performance of such a method with respect to known techniques, the step of generating the second arterial input function according to the invention consists of the implementation of basic operations by said processing unit pre-trained according to a process of learning the destructive influence of the perfusion sequence acquisition on the arterial signal and the redundancy of the information related to such a second arterial input function shared by at least two sets of tissue signals.

[0022] According to certain embodiments, such a method may include, before performing the step of creating pharmacokinetic parameters based on the thus corrected second arterial input function and the experimental signal, correcting the generated second arterial input function by scaling it.

[0023] Advantageously, when the medical imaging system comprises an output human-machine interface, the method according to the invention may include creating a graphic representation of the second arterial input function and outputting the graphic representation by the output human-machine interface, in order to provide a member of the medical staff with an opportunity to verify the validity of the generated second arterial input function.

[0024] According to a preferred embodiment, the learning process may consist of deep learning based on minimizing the average value of the squared error between the actual samples of the arterial input function that enabled the generation of the tissue signal and the estimated values of these same samples made by the learning process, and such a learning process may be performed by an Adam optimizer.

[0025] According to a second subject, the invention relates to a computer program comprising one or more program instructions that can be executed by a processing unit of a computer, the program instructions can be placed in a non-volatile memory of the computer, and by executing the program instructions by the processing unit, the post-processing method according to the invention is implemented.

[0026] Similarly, according to a third subject, the invention relates to a computer-readable storage medium containing instructions of such a computer program.

[0027] Finally, the present invention further relates to a medical imaging analysis system including a processing unit provided to communicate with the outside world and receive a set of samples of a time-experimental signal resulting from a perfusion acquisition sequence by a medical imaging device and resulting from the movement of a tracer within the basic volume of an organ. Such a processing unit is pre-trained according to the aforementioned learning process and includes storage means containing program instructions of a computer program according to the present invention.

[0028] Other features and advantages will become clearer upon reading the following description and examining the accompanying drawings.

Brief Description of the Drawings

[0029]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Embodiments for Carrying Out the Invention

[0030] As shown in FIG. 7, the present invention relates to a method 100 for generating a pharmacokinetic parameter of interest (QI) based on one or more sampled time-experimental signals (10, 12) resulting from acquisition by a perfusion medical imaging device (1). Such a method (100) can be implemented by a processing unit (4) of such a medical imaging analysis system. Such a method (100) depends on a process (101) of learning the destructive influence of the perfusion sequence acquisition on the arterial signal, combined with learning the redundancy of the information related to the arterial input function shared by at least two sets of tissue signals. Accordingly, the processing unit 4 is provided to include a convolutional neural network preconfigured by the implementation of the learning process 101 or any other equivalent solution.

[0031] In fact, all of the tissue signals result from the convolution product of one and the same arterial input function AIF and their respective impulse responses IRF. Therefore, it is possible to extract from them the shared information that is the actually sought arterial input function AIF. The learning of the destructive effects of acquisition is itself done by supplying an arterial signal AIF0 that is a degraded version of the actually sought AIF. Therefore, it would be a matter of "deforming" such a selected arterial signal AIF0 so that it can correspond to the information shared by the tissue signals while avoiding proposing solutions that are not very likely. In other words, this point promotes the stability of the algorithm. This learning process 101 can be carried out using a combination of signals physically acquired from volunteers, patients, or using physical phantoms and / or simulated signals that make it possible to generate more diverse cases and thus promote the ability to estimate a realistic arterial input function AIF sought in this way. By using the database ADB generated in this way, the learning 101 can be carried out, for example, using a deep learning algorithm. The estimation of the learning algorithm parameters can be derived by minimizing the estimation error between the sought AIF and those used to generate the tissue signals of the database ADB. The said estimation can also be directly derived by minimizing the error between the tissue signals from the database ADB and the tissue signals estimated by the algorithm when a step of reconstructing the said signals using the AIF' generated by the algorithm is proposed. Therefore, the database ADB can optionally, - the sought AIF, their deformed AIF0 affected by the acquisition, and the signals they generate, - the AIF0 affected by the acquisition, and the signals generated by the sought AIF, can be composed of.

[0032] Once learning step 101 enables sufficient performance to be achieved, post - processing method 100 may use a convolutional neural network (or any other equivalent solution), hereinafter referred to as the "network", thus trained within processing unit 4 implementing said method 100. Method 100 includes step 110 of selecting, in the same way as post - processing method 100 described in relation to a third method according to the prior art, a first version AIF0 of the arterial input function and a set of tissue signals S not corresponding to vessels or voids. Such step 110 may be carried out according to various techniques. Thus, such selection may consist of selecting an average arterial input function derived from a known population. It may further result from a manual selection by an engineer on an image resulting from the acquisition of a sampled time - experiment signal of some regions of the samples from which this arterial input function is derived, or from the implementation of any post - processing method enabling the automatic estimation of this arterial input function and the set of tissue signals from the image. Said method 100 includes step 120 of generating an estimated value of a second realistic arterial input function AIF for a set of input signals not affected by the influence of the acquisition of the first version AIF0 of the arterial input function. Different from the previously known method 100, step 120 of method 100 according to the present invention consists of directly generating, based on the basic operations carried out by a network trained according to process 101, a second realistic arterial input function AIF, i.e., without carrying out an iterative process 120 with a large loss. Thus, the implementation of step 120 is significantly faster and simpler than the iterative solutions according to the prior art. Method 100 according to the present invention may further include step 130 of scaling the generated second arterial input function AIF so that it is finally used after correction in step 140 of generating one or more pharmacokinetic parameters QI of interest for all or some of the voxels of the perfusion image, and these pharmacokinetic parameters QI may be the subject of a graphic - format output by an output human - machine interface such as interface 5 of medical imaging analysis system SAIM according to, for example, Figure 1 or 2.However, the present invention provides that the step 130 of normalizing the generated AIF can be directly integrated into the step 120 of directly generating an AIF' that is available for carrying out a step 140 of generating one or more pharmacokinetic parameters QI of interest for all or part of the voxels of the perfusion image that are realistic, normalized, i.e., scaled. A suitable scaling factor can be calculated in a way that conserves mass between the generated arterial input function and the experimental signal, for example by matching low-distortion regions in two signals.

[0033] An embodiment of the learning process 101 of the network according to the present invention that enables verification of the validity of the present invention is considered.

[0034] As described above, the database ADB can be composed of more than 3 million AIF samples simulated by random changes in the parameters of a model called Parker's model. Thus, each AIF sample can be used to generate five tissue signals having different pharmacokinetic parameters. The model used to generate these signals can be a model called the Toft-Ketty model extended by the addition of a delay between the AIF sample and that of the tissue.

[0035] To generate a second realistic arterial input function AIF in step 120 of the post-processing method 100 according to the present invention, a deep learning algorithm is selected. Such a deep learning algorithm is advantageously divided into two branches. The first branch processes samples of the arterial input function AIF0. It is composed of a three-layer one-dimensional convolutional network. The second branch processes a set of tissue signal samples. It is composed of a three-layer two-dimensional convolutional network. The two branches each end with a simple neuron network having only one layer and are then combined by a simple neuron network having three layers. The last layer outputs a sample of the arterial input function AIF having the same size as the size of the first proposal AIF0 of the arterial input function.

[0036] Advantageously, the learning process 101 can be performed by an optimizer called the Adam optimizer based on minimizing the average value of the mean squared error between the actual samples of the arterial input function that enabled the generation of the tissue signal and the estimated values of these same samples made by the learning algorithm. The number of iterations spent on this training was 30 for 80% of the generated database, and thus the validation was performed on the remaining 20%. The results obtained from the validation were 0.69% of the peak value of the true AIF samples with respect to the mean squared error, both for the training database and for what was used for validation. This method is generally described in the literature as yielding the best results. However, this deep learning can be replaced by a tabulation method that enables the association between the input represented by the initial set of arterial input functions and tissue signals and the output representing the second arterial input function not affected by the acquisition. In this case, the learning consists of setting such a table.

[0037] To ensure the clinical feasibility of the post - processing method 100 according to the present invention, this same deep - learning algorithm was tested on an MRI cardiac perfusion image database composed of 43 elements. These images have the remarkable feature that they were acquired using the second method of the prior art described above. In doing so, the available arterial input function samples were hardly affected by the acquisition associated with each conventional acquisition. Manually, by selecting the myocardial region and extracting 5 samples of tissue signals therefrom, and selecting the left ventricular region and extracting the case AIF0 of the arterial input function sample that was significantly affected by many acquisitions therefrom, the value of the estimated arterial input function sample performed by the learning algorithm was compared with the sample hardly affected by the acquisition, and the learning algorithm could be evaluated on the actual data. The results were evaluated in terms of the mean squared error and the coefficient of determination indicated by R2 and compared with the values obtained by the case AIF0 of the arterial input function sample significantly affected by many acquisitions. Their median values were less than 0.08 compared to 0.15 of the sample of the arterial input function affected by the acquisition in terms of the mean squared error respectively, and exceeded 0.85 compared to less than 0.3 in terms of R2. Therefore, in the case of actual acquisition, the present invention could obtain a realistic and correct estimated value of the required arterial input function AIF sample.

[0038] For the purpose of verifying the validity of the second arterial input function generated in step 120 by a medical practitioner using the post - processing method 100 according to the present invention, the post - processing method 100 according to the present invention may further include step 150 of creating a graphic representation of the second arterial input function AIF before or after its correction. Such step 150 may further consist of outputting the graphic representation by an output human - machine interface when the medical imaging analysis system implementing the method 100 includes an output interface such as interface 5 of the system shown in FIGS. 1 and 2.

[0039] The present invention has been described in relation to non-limiting examples of signals arising from a cardiac acquisition sequence by a perfusion imaging device. The present invention is not limited to this single organ under consideration and may be used to generate pharmacokinetic parameters for any other organ of interest, such as the brain for example.

Claims

1. A method (100) for post-processing a sampled time-experimental signal (S, 10, 12) resulting from a perfusion acquisition sequence by a medical imaging device (1) and resulting from the movement of a tracer within the basic volume of an organ, said method being implemented by a processing unit (4) of a medical imaging analysis system (SAIM), said method (100) comprising: selecting (110) a first arterial input function (AIF0) related to the arterial region (A0) of said organ and a set (S) of tissue signals respectively related to separate tissue regions (A1, A2, A3, A4, A5) of said organ; generating (120) a second arterial input function (AIF, AIF') based on said first arterial input function (AIF0) and said selected set of tissue signals (S); creating (140) a pharmacokinetic parameter (QI) based on said second arterial input function (AIF') and said experimental signal (S, 10, 12), and wherein the step (120) of generating a second arterial input function (AIF) is composed of the implementation of basic operations by said processing unit (4), said processing unit being pre-trained according to a process (101) of learning the destructive influence on the arterial signal of the acquisition of the perfusion sequence and the redundancy of the information related to such a second arterial input function shared by at least two sets of tissue signals. A method characterized by this.

2. The method (100) according to claim 1, comprising a step (130) of correcting by scaling the generated second arterial input function (AIF) before performing the step (140) of generating a pharmacokinetic parameter (QI) based on said second arterial input function (AIF') thus corrected and said experimental signal (S, 10, 12).

3. The method (100) according to claim 1 or 2, wherein said medical imaging system (SAIM) comprises an output human-machine interface (5), and said method (100) comprises a step (150) of creating a graphic representation of said second arterial input function (AlF, AlF') and outputting said graphic representation by said output human-machine interface (5).

4. The learning process (101) comprises deep learning based on minimizing the mean squared error between an actual sample of an arterial input function that enables the generation of tissue signals and the estimated values of these same samples performed by the learning process, the method according to any one of claims 1 to 3.

5. The learning process (101) is performed by an Adam optimizer, the method according to claim 4.

6. Comprising one or more program instructions executable by the processing unit of a computer, the program instructions may be placed in the non-volatile memory of the computer, and by execution of the program instructions by the processing unit, the method according to any one of claims 1 to 5 is implemented, a computer program.

7. A computer-readable storage medium comprising the instructions of the computer program according to claim 6.

8. A medical imaging analysis system (SAIM) comprising a processing unit (4) provided to receive a set of samples of a time-experimental signal (S, 10, 12) resulting from a perfusion acquisition sequence by a medical imaging device (1) and resulting from the movement of a tracer within a basic volume of an organ, the processing unit (4) being pre-trained according to a process (101) of learning the destructive influence of the acquisition of the perfusion sequence on the arterial signal and the redundancy of information related to the arterial input function shared by at least two sets of tissue signals, and comprising storage means comprising the program instructions of the computer program according to claim 7.

Images