A method for deblurring an image resulting from an acquisition sequence by a medical imaging device.

Abstract

The invention relates to a method (100) for deblurring an experimental medical image (Mn) obtained from an acquisition sequence by a medical imaging device (1) by: calculating the respective Fourier transforms (TFMn, TFM0) of said experimental image (Mn) and a reference image (M0) of the same resolution; determining the respective magnitudes (YMn, YM0) of the Fourier transforms of the experimental image and the reference image; obtaining a deblurring mask (D) by calculating the ratio of the magnitudes (YMn, YM0); applying said ratio (D) to the Fourier transform of the experimental image (TFMn) by multiplication, resulting in a corrected Fourier transform (TFMnc); obtaining the deblurred experimental image (Mnc) by applying the inverse Fourier transform to said corrected Fourier transform (TFMnc). Figure for the abstract: Fig 4

Description

Title of the invention: Method for deblurring an image resulting from an acquisition sequence by a medical imaging device

[0001] The present invention relates, in general, to image processing in the field of medical imaging. The invention describes more particularly a method for correcting blur or a deblurring method, also known as "deblurring" according to Anglo-Saxon terminology, of a medical image resulting from an acquisition sequence via a medical imaging device.

[0002] In medical imaging, an acquisition sequence allows for the collection and storage of raw data on the anatomy and / or function of an organ, enabling the reconstruction of a three-dimensional visual representation of said organ—in other words, an image of said organ. Such acquisitions may induce, in the frequency domain, a loss or filtering of certain high and / or low frequencies, resulting in blurring of said image.

[0003] Although the invention can be applied to all techniques or modalities based on imaging and of which a resulting image has undergone the virtual application of a filter strongly attenuating or suppressing certain frequencies in the frequency domain, we will distinguish mainly in the remainder of this application, by way of illustration, magnetic resonance imaging, better known by the acronym MRI for "Magnetic Resonance Imaging", and in particular perfusion MRI and more specifically perfusion by arterial spin labeling, known by the acronym ASL for perfusion by "arterial spin labeling", according to Anglo-Saxon terminology.

[0004] Magnetic resonance imaging is based on an analysis of the response of a proton in a water molecule when excited in a magnetic field. This response depends on the environment of such a proton and thus allows differentiation between different types of tissue. A Nuclear Magnetic Resonance imaging device, such as device 1 of a SAIM medical imaging analysis system illustrated by way of non-limiting example in Figures 1 and 2, is generally used. This device delivers a plurality of digital image sequences 12 of one or more parts of a patient's body, by way of non-limiting example, the brain, heart, and lungs. MRI acquisition is performed by sampling a parallelepiped volume in a given slice plane. Figure 3 illustrates the slice planes most commonly used in medical imaging: View A: frontal or coronal (or frontal) projection, the patient's right side is shown on the left of the image: as if the observer were looking simply the patient facing forward; - View B: on an axial (or transverse) view, the patient's right is always located on the left of the image: as if the observer were looking at the patient - lying on his back - from his feet; - View C: on a sagittal or profile view, the image is presented with the anterior part of the patient on the left, as if the observer were looking at the patient from the left profile.

[0005] The device 1 applies a combination of high-frequency electromagnetic waves to the body part under consideration and measures the signal re-emitted by certain atoms, such as, by way of non-limiting example, hydrogen for Nuclear Magnetic Resonance imaging. The device 1 thus makes it possible to determine the magnetic properties and, consequently, the chemical composition of biological tissues and therefore their nature, in each elementary volume of interest, commonly called a voxel, of the imaged volume. As shown in [Fig. 1], a Nuclear Magnetic Resonance imaging device 1 is controlled using a console 2. A user 6, for example an operator, practitioner, or researcher, can thus select commands 11 to operate the device 1, based on parameters or instructions 16 entered via an input human-machine interface 8 of the analysis system.Such a human-machine interface 8 can consist, for example, of a computer keyboard, a pointing device, a touchscreen, a microphone, or, more generally, any interface designed to translate a gesture or instruction issued by a human 6 into control or parameter data. From information 10 produced by said device 1, a plurality of sequences of digital images 12 of a part of a human or animal body are obtained. We will also call such information 10 or images 12 "experimental data".

[0006] The image sequences 12 can optionally be stored within a server 3, i.e., a computer equipped with its own storage means, and constitute a patient's medical record 13. Such a record 13 can include images of different types, such as structural images highlighting tissue activity or anatomical images reflecting tissue properties. The image sequences 12, or more generally the experimental data, are analyzed by a processing unit 4 arranged for this purpose. Such a processing unit 4 can, for example, consist of one or more microprocessors or microcontrollers implementing instructions from suitable application programs loaded into storage means of said image analysis system. "Storage means" means any volatile computer memory or, advantageously, non-volatile computer memory. Non-volatile memory is computer memory whose technology allows data to be retained even without a power supply. It can contain data resulting from input, calculations, measurements, and / or program instructions. The main non-volatile memories currently available are electrically writable, such as EPROM (Erasable Programmable Read-Only Memory), or electrically writable and erasable, such as EEPROM (Electrically Erasable Programmable Read-Only Memory), flash memory, SSD (Solid-State Drive), etc. Non-volatile memories are distinct from so-called "volatile" memories, whose data is lost without a power supply.The main volatile memories currently available are of the RAM type (Random Access Memory, also known as "live memory"), DRAM (dynamic random access memory, requiring regular updates), SRAM (static random access memory requiring such updates during power failure), DPRAM or VRAM (particularly suited to video), etc. A "data memory," in the rest of this document, can be volatile or non-volatile depending on the intended application.

[0007] Said processing unit 4 includes means for communicating with the outside world to collect images. Said means for communicating further enable the processing unit 4 to deliver or output, ultimately, a rendering, for example graphic and / or audio, of an estimate or quantification of a biomarker or a pharmacokinetic parameter QI developed by said processing unit 4 from the experimental data 10 and / or 12 obtained by Magnetic Resonance Imaging, to a user 6 of the imaging analysis system via an output human-machine interface 5.Throughout this document, the term "output human-machine interface" refers to any device, used alone or in combination, that outputs or delivers a graphical, haptic, auditory, or, more generally, human-perceptible representation of a reconstructed physiological signal—in this case, a biomarker—to a user of a Magnetic Resonance Imaging (MRI) analysis system. Such an output human-machine interface may consist of, but is not limited to, one or more screens, speakers, or other suitable alternative means. The user of the analysis system can thus confirm or refute a diagnosis, decide on a therapeutic course of action deemed appropriate, conduct further research, fine-tune the settings of measuring equipment, and so on.Optionally, this user 6 can also configure the operation of the processing unit 4 or the output human-machine interface 5, using operating and / or acquisition parameters 16. For example, they can define display thresholds or choose the estimated or quantified biomarkers, indicators, or parameters for which they wish to have a . representation. The user utilizes the previously mentioned input human-machine interface 8 or a second input interface provided for this purpose. Advantageously, the input human-machine interface 8 and output human-machine interface 5 can constitute a single physical unit. These input human-machine interfaces 8 and output human-machine interfaces 5 of the image analysis system can also be integrated into the acquisition console 2. There is a variant, described in connection with [Fig. 2], in which an imaging system, as described above, further includes a preprocessing unit 7 to analyze the image sequences 12, derive experimental signals 15, and deliver these signals to the processing unit 4, thus relieving it of this task.

[0008] More specifically, perfusion MRI provides information on the capillary microcirculation of tissues by analyzing their microcirculation. The essential quantitative parameters evaluated are blood volumes and temporal data (such as transit time, time to peak contrast, etc.). The ultimate goal of perfusion MRI is to measure, or estimate, the blood flow irrigating the organ being examined, expressed in milliliters / minute / 100 grams of tissue, corresponding to tissue perfusion. The differentiation between perfused and non-perfused tissues relies on the use of an intravascular marker: - exogenous in the first case of application: injected contrast product; - endogenous in the second case of application: by labeling the hydrogen nuclei of the water in the bloodstream.

[0009] In this respect, ASL perfusion imaging falls under this second application case for measuring cerebral perfusion, that is, the blood supply to brain tissues, without the use of contrast agents or potential pharmacological side effects. This technique is therefore non-invasive, which makes its acquisition faster and allows it to be repeated. ASL makes it possible to highlight perfusion abnormalities in many pathologies, particularly those affecting the brain, in a completely non-invasive manner, making it a particularly interesting technique from a clinical point of view. It is a differential technique during which two acquisitions are performed: an acquisition with arterial proton labeling and a control acquisition. ASL perfusion imaging thus quantifies perfusion in the brain by labeling the hydrogen nuclei of arterial intravascular water with one or more radiofrequency pulses.These pulses serve to reverse the longitudinal magnetization of arterial blood as it flows through a single plane (or tissue plate) just upstream of the region of interest. This alters the measurable magnetization and the proton relaxation time Tl (the time it takes for the protons to return to equilibrium). This process is called "tagging" and involves converting the water present in the blood circulating in the arteries and through the neck into a [missing word - likely "magnetism"]. endogenous tracer. Thus, after a delay allowing blood to circulate from the labeling site to its final point of exchange with the brain, called arterial transit time (ATT), images are acquired containing signals from both the labeled water and the water from static tissues. Associated with this process is a post-labeling delay (PLD), defining the time between labeling and image acquisition. The second control acquisition, called the "control" image, is performed without prior arterial labeling: the arterial protons in the region of interest are then in equilibrium, completely relaxed. Subtracting the labeling acquisition from the control acquisition cancels the signal from static tissues and yields a perfusion-weighted image.It should be noted that the signal difference is on the order of a few percent, resulting in a low signal-to-noise ratio. Therefore, it is necessary to acquire several dozen pairs of labeled images and then average them to obtain a satisfactory signal-to-noise ratio.

[0010] Thus, data acquisition using ASL technology requires two main steps: labeling of circulating blood protons and image acquisition. Regarding proton labeling, there are three main types: - Continuous labeling (known by the English acronym CASL), the labeling is carried out continuously at the level of the vessels of the neck, this technique allows to obtain a good signal / noise ratio but generates a high energy deposition in the tissues; - Pulsed labeling (known by the English acronym PASL) uses brief pulses of radio frequencies limiting the deposition of energy in tissues; - and pseudo-continuous marking (known by the English acronym pCASL) uses very short-duration radio frequency pulse trains.

[0011] Regarding image acquisition, as previously mentioned, it is necessary to acquire numerous pairs of labeled and control images. Due to a satisfactory signal-to-noise ratio and rapid acquisition that limits motion artifacts, the echo-planar imaging technique, known in English as EPI, is primarily used. However, to improve image quality, 3D sequences have been developed. It is possible to use ultrafast, single-shot 3D sequences combining gradient echo and spin-echo acquisition (known in English as GRASE). The use of such sequences facilitates the suppression of static tissue signals.

[0012] However, due to an echo time, representing the time interval between proton excitation and signal occurrence, that is too long relative to the decrease in the relaxation time T2, i.e., the proton phase shift time, ASL images can be affected by significant blurring, visible both in and through the plane, a phenomenon that is even more pronounced with the GRASE sequence, a phenomenon mentioned, for example, in the scientific article “Reducing blurring artifacts in 3D-GRASE ASL by integrating new acquisition and analysis strategies” published in Proceedings of the International Society for Magnetic Resonance in Medicine volume 22 (2014), page 2704. Such blurring acts as a low-pass filter in the frequency domain, allowing low-frequency values ​​to pass through and blocking all high-frequency values.Thus, distinguishing between different brain tissues, such as white matter, gray matter, a tumor, and cerebrospinal fluid (CSF), becomes difficult. The blurring is particularly problematic at the boundary between gray and white matter, which exhibit different perfusion levels (a long transit time in white matter significantly affecting labeling) and where white matter signals are affected by gray matter signals.

[0013] To address this blurring issue in a medical experimental image, blur suppression methods are known to be used. Such methods generally perform deconvolution in the image domain after estimating the deconvolution kernel in the frequency domain. We can cite, in particular, the work of Chappell et al., with reference to the scientific article “Reducing blurring artifacts in 3D-GRASE ASL by integrating new acquisition and analysis strategies” published in Proceedings of the International Society for Magnetic Resonance in Medicine, volume 22 (2014), page 2704. Furthermore, any two-dimensional image blur suppression method could potentially be applied to said experimental image by applying said method to each slice of the image independently.In this regard, the work of Wang and Tao presents a series of deblurring techniques in the context of natural image deblurring in the scientific article "Recent progress in image deblurring" published in arXiv: 1409.6838.

[0014] However, such blur suppression methods focus on correcting blur that only passes through the plane. Therefore, the development of a method for suppressing and / or correcting three-dimensional blur (simultaneously correcting blur within and through the plane) of an image resulting from an acquisition sequence via a medical imaging device is of great interest. As such, such a method makes it possible to remove blurry artifacts from images in order to improve their readability and facilitate their interpretation for the purpose of using such images in the field clinical to ultimately characterize different lesions with altered metabolic properties such as tumors, ischemic tissues, or multiple sclerosis.

[0015] To this end, the invention provides a method for deblurring an experimental medical image resulting from an acquisition sequence by an imaging device describing a plurality of elementary volumes of interest of a patient's organ, called "voxels," said method being implemented by a processing unit of a medical imaging analysis system. Such a method comprises: - a step of calculating the Fourier transform of said experimental image; - a step of calculating the Fourier transform of a reference image with the same resolution as the experimental image; - a step of determining a first modulus of the Fourier transform of the experimental image and a second modulus of the Fourier transform of the reference image; - a step to obtain a deblurring mask consisting of calculating the ratio between the first and second modules; - a step of applying said ratio to the Fourier transform of the experimental image by multiplication to obtain a corrected Fourier transform; - a step of obtaining a deblurred experimental image consisting of applying the inverse Fourier transform on said corrected Fourier transform.

[0016] In a particular embodiment of the invention, said reference image may originate from an atlas, a control patient or an artificial anatomical reference.

[0017] In a preferred embodiment of the invention, said method may include a step prior to the Fourier transform calculation steps, consisting of a step of acquiring a structural image of said patient with the same resolution as said experimental image. Said structural image then corresponds to said reference image.

[0018] Preferably, in order to ensure the voxel-by-voxel correspondence of the reference image with respect to the experimental image, the method according to the invention may include a step prior to the Fourier transform calculation steps consisting of a step of registration of the experimental image with respect to the reference image.

[0019] According to an advantageous embodiment, said process may include a subsidiary step prior to the step of obtaining the deblurred experimental image and subsequent to the step of obtaining a deblurring mask consisting of applying a smoothing filter on said deblurring mask.

[0020]

[0021]

[0022]

[0023]

[0024]

[0025]

[0026]

[0027] According to the previous embodiment, the smoothing filter can be a Gaussian filter. Preferably, the Fourier transforms of the experimental image and the reference image are respectively calculated using the fast Fourier transform algorithm. According to a preferred embodiment, the experimental image acquisition sequence is a perfusion acquisition sequence by a magnetic resonance medical imaging device during which an endogenous tracer circulates within said plurality of elementary volumes of interest. According to a second object, the invention relates to a computer program product comprising one or more program instructions executable by the processing unit of a computer, said program instructions being loadable into a non-volatile memory of said computer and whose execution by said processing unit causes the implementation of a post-processing method according to the invention. According to a third object, the invention relates to a computer-readable storage medium containing the instructions for such a computer program product. Finally, the invention further relates to a medical imaging analysis system comprising a processing unit arranged to deblur an experimental medical image resulting from an acquisition sequence by a medical imaging device in relation to a plurality of elementary volumes of interest, called voxels, said processing unit being configured to: - calculate the Fourier transform of said experimental image; - calculate the Fourier transform of a reference image of the same resolution solution than the experimental image; - determine a first magnitude of the Fourier transform of the experimental image and a second magnitude of the Fourier transform of the reference image; - obtain a deblurring mask by calculating the ratio between the first and second modules; - apply said ratio to the Fourier transform of the experimental image by multiplication to obtain a corrected Fourier transform; - obtain the deblurred experimental image by applying the inverse Fourier transform to said corrected Fourier transform. The invention will be better understood and other features and advantages thereof will become apparent from the following description of particular embodiments of the invention, given by way of illustrative and non-limiting examples, and with reference to the accompanying drawings, among which: [Fig. 1], already described, illustrates a simplified description of a system analysis of images obtained by magnetic resonance imaging,

[0028] [Fig.2], already described, illustrates a simplified description of a variant of a system for analyzing images obtained by magnetic resonance,

[0029] [Fig.3] already described, illustrates the most commonly used section planes in medical imaging,

[0030] [Fig.4] illustrates an example of a functional algorithm for a deblurring process according to the invention,

[0031] [Fig.5] illustrates variants of the implementation of a functional algorithm for a deblurring process according to the invention,

[0032] [Fig.6] illustrates a gain resulting from the implementation of a deblouting process 100 according to the invention,

[0033] [Fig.7] illustrates a gain resulting from the implementation of a deblouting process 100 in accordance with the invention.

[0034] To simplify the description, the same reference is used in different figures to designate the same object, element, or step. Thus, when the description cites a referenced object, element, or step, this object, element, or step can be identified in several figures. Furthermore, the figures and the description are given as non-limiting examples of embodiments.

[0035] As a preliminary point, for the purposes of this invention, an image is a BGR matrix digital representation of elements called "voxels," of thickness k (along a transverse axis), BGR(i,j), where i and j are integer indices used to identify the voxel located in row i and column j of the BGR matrix. It is a graphical representation in the form of a table or matrix of BGR(i,j) elements, each encoding a triplet of integer values ​​between zero and 255, according to the RGB color coding system. RGB stands for "Red Green Blue." Such a computer-based color coding system is the closest to the hardware currently available for creating output human-machine interfaces such as computer screens.In general, these systems reconstruct a color through additive synthesis from three primary colors: red, green, and blue, forming a mosaic on the screen that is usually too small to be distinguished by the human eye. RGB encoding indicates a light intensity value for each of these primary colors. Such a value is generally encoded using one byte and therefore belongs to a range of integer values ​​between zero and 255. Alternative color encodings could be used. In this case, each voxel, or element of the BGR array, would describe a set of numerical values ​​adapted to that encoding instead of the triplet mentioned above for RGB encoding.

[0036] A method for deblurring a medical image, according to the invention and illustrated in [Fig.4], advantageously translates into a computer program product whose program instructions are intended to be implanted in the program memory of an element of a medical imaging system, such as the S system according to Figures 1 and 2, for example a computer or computer server or, more generally, any electronic object with sufficient computing power.

[0037] Figure 4 thus illustrates a method 100, according to the invention, for deblurring an experimental image Mn resulting from an acquisition sequence by an imaging device, such as device 1 of the medical image analysis system S illustrated in Figures 1 and 2, in relation to a plurality of n voxels. Such an experimental image Mn can be considered as all or part of the "total" image resulting from said acquisition sequence. The image Mn can therefore be the total image or a portion of the total image representing an area of ​​interest. Such a method 100 can be implemented by the processing unit 4 of such a medical image analysis system S.

[0038] For illustrative but not limiting purposes, such a method 100 according to the invention is advantageously described in the remainder of this document and implemented for the processing of a medical image, representing the brain of a patient, resulting from a perfusion acquisition sequence by a magnetic resonance medical imaging device 1 during which an endogenous tracer passes within said plurality of n voxels.

[0039] Such a deblurring method 100 according to the invention consists of analyzing an experimental medical image Mn in the frequency domain and comparing said image Mn to a reference image MO of the same resolution, namely the same number of voxels per unit length, as said image Mn in order to recover the frequencies lost or truncated from the experimental image Mn. Thus, said reference image MO represents a region of interest similar to said image Mn, for example a brain as illustrated in [Fig. 4], according to the same plane of section (the planes of section have been explained previously and illustrated in [Fig. 3]) for example in coronal section as illustrated in [Fig. 4].

[0040] In a preferred embodiment to ensure that the deblurring process 100 is as efficient as possible, the reference image MO may be a structural image of the patient. Such a structural image MO represents the same area of ​​interest and has the same resolution as the experimental image Mn. In this case illustrated in [Fig. 5], the process 100 may include a preliminary step 101. Such a step 101 consists of acquiring a structural image of the patient from an acquisition sequence using a medical imaging device. The acquisition of the reference structural image MO may be performed simultaneously with the acquisition of the experimental image Mn using the same imaging device. medical imaging, such as, for example, magnetic resonance imaging (MRI) equipment. By way of illustration, but not limitation, the patient's structural MRI image may be: - a so-called "Tl-weighted" image, often used for anatomy, since such an image results from an acquisition sequence, called "anatomical," which favors the detection of relatively immobile, i.e., intracellular water. For such an image, fat appears hyperintense, light in color, and water hypointense, dark in color; - a so-called "T2-weighted" image, often used as a functional image, since such an image results from an acquisition sequence that favors the detection of mobile water, i.e., extracellular or intravascular. For such an image, water appears hyper-intense, light in color, and fat appears slightly darker than water; - a so-called "FLAIR" image, for "Fluid Attenuated Inversion Recovery" (using Anglo-Saxon terminology), is very well suited to brain imaging, as such an image results from an acquisition sequence that suppresses the signal from the cerebrospinal fluid. With this type of image, white matter lesions, softening (stroke), and inflammatory demyelination (multiple sclerosis) appear hyperintense and are particularly well highlighted. The invention is not limited to the use of this type of structural image. Any other structural image may be used to implement method 100 according to the invention.

[0041] As an alternative or in addition, such a MO reference image may originate from: - an atlas, namely a typical image resulting from an average of images acquired from various patient populations or from the same patient; - of a healthy subject otherwise known as a "control" patient; - or even from an artificial anatomical reference frame. The use of such images can be of great interest, for example, when the structural image of the patient is lost or accidentally erased, or when problems in acquiring the structural image of the patient are encountered, due, for example, to patient movements during acquisition, these movements being able to be voluntary (for example in the case of claustrophobia) or involuntary (for example in the case of certain diseases such as Parkinson's disease).

[0042] In addition, in order to guarantee perfect alignment between the reference image MO and the experimental image Mn, optionally, as illustrated in [Fig. 5], said method 100 may include a preliminary step 102 of registering the experimental image Mn with respect to the reference image MO or vice versa of the image of The MO reference image is referenced to the experimental image Mn to ensure that the two images MO and Mn correspond voxel by voxel. Indeed, the closer the MO reference image is to the experimental image Mn, the better the deblurring result will be. After such optional registration, the two images, being approximately the same size, are superimposable because they share common orientations.

[0043] As mentioned previously, the invention consists of analyzing an experimental medical image Mn in the frequency domain. Thus, as illustrated in Figures 4 and 5, said method 100 comprises first 110 and second 120 steps for calculating the Fourier transforms TFMn and TFM0, respectively, of the experimental image Mn and the reference image MO. In this respect, the Fourier transform is a mathematical technique for determining the frequency spectrum of a signal resulting from an acquisition sequence. Since said signal resulting from an acquisition sequence is a discrete signal, the discrete Fourier transform may be preferred. Indeed, the analysis of discrete signals by discrete Fourier transform makes it possible to represent a discrete signal in the frequency domain: it is then possible to directly process the image frequencies.Therefore, to obtain the frequencies contained in said signal, an operation called transformation is applied, and the result of this transformation is called the transform.

[0044] In practice, to calculate a discrete Fourier transform, an algorithm called the "Fast Fourier Transform" (FFT) is advantageously used, allowing for faster calculation. However, the invention is not limited to the use of this algorithm. Any other algorithm for calculating a Fourier transform may be used to implement method 100 according to the invention.

[0045] Images obtained by applying a Fourier transform produce a complex image. Thus, in general, the magnitude of the Fourier transform is calculated and represented. Such a magnitude is calculated as the square root of the sum of the square of the imaginary part and the square of the real part. In this respect, as illustrated in [Fig. 4], said method 100 according to the invention includes a step 130 of determining the magnitude YMn of the Fourier transform TFMn of the experimental image Mn and the magnitude YM0 of the Fourier transform TFM0 of the experimental image MO. Advantageously, such a step 130 may include a normalization process for said magnitudes YMn and YM0 so that said magnitudes share a common scale of values. Such normalization is performed for each magnitude YMn and YM0 and consists, for example, of dividing the values ​​of each magnitude YMn and YM0 respectively by their center frequency value, in other words, the "zero frequency".Alternatively, this normalization may consist of using any normalization system that allows both modules YMn and YM0 to share the same scale of values. In addition, each pixel of said module, acting as a divisor, is preferably non-zero to prevent division by zero. To achieve this, a constant, such as a tenth or a hundredth, could be added to avoid unbalancing the scale.

[0046] Once said moduli YMn and YMO are determined, said method 100 includes a step 140 for obtaining a deblurring mask D, as illustrated in [Fig. 4]. Such step 140 consists of calculating the ratio D between the moduli YMn and YMO. Preferably, but not exclusively, the deblurring mask D or ratio D is equal to the division of the modulus YMO of the Fourier transform TFM0 of the reference image MO by the modulus YMn of the Fourier transform TFMn of the experimental image Mn. Advantageously, said step 140 may include a subsidiary processing step for limiting the values ​​of said ratio or said deblurring mask D, consisting of clipping the values ​​of D to a maximum amplitude. Preferably, such a maximum amplitude is between 2 and 100, ideally 5 to 20 to avoid any risk of noise amplification. It should be noted that an amplitude of value 10 seems to be the ideal value.

[0047] Next, said method 100 includes a step 150 of applying said deblurring mask or said ratio D to the Fourier transform TFMn of the experimental image Mn by multiplication. At the end of such step 150, a corrected Fourier transform TFMnc is obtained. Thus, said deblurring mask D is applied to the entire Fourier transform TFMn, that is, to all frequencies and therefore to the entire experimental image Mn. Said step 150 could also concern only certain frequencies.

[0048] Finally, said method 100 includes a final step 160 for obtaining the deblurred experimental image Mnc. Such a step 160 consists of applying the inverse Fourier transform to said corrected Fourier transform TFMnc. Indeed, the inverse Fourier transform is a mathematical operation that allows the reconstruction of the original image (the original signal) from its frequency representation. In the case of the invention, a deblurred experimental image Mnc is thus obtained.

[0049] In addition, as illustrated in [Fig. 5], said process 100 may include an optional step 142, prior to step 150 of applying said ratio D and subsequent to step 140 of obtaining a deblurring mask. Such a step 142 consists of applying a smoothing filter to said deblurring mask D (also called said ratio D). At the end of such a step 142, a smoothed deblurring mask D1 is thus obtained, making it possible to reduce artifacts. Such a smoothing operation 142 may advantageously consist of applying a Gaussian filter. Such a Gaussian filter preferably has a standard deviation between 1 and 50, beyond a standard deviation A standard deviation of 50 would render the smoothing ineffective, ideally with a standard deviation of 10. The invention is not limited to the use of this type of smoothing filter, nor even to the values ​​stated as ideal. Any other smoothing filter could be used to implement the method according to the invention. Furthermore, such a step may advantageously consist of normalizing said ratio D. In this case, such normalization may consist, for example, of dividing the values ​​of said ratio D by the value of its center frequency, called the zero frequency, or consist of any other normalization technique.

[0050] Alternatively, if several experimental images Mnl, Mn2, ..., Mnx have been acquired and are therefore available for processing, steps 110, 130, and 140 (as well as optional steps 101, 102, and / or 142 if necessary) are performed independently for each experimental image in order to calculate an individual deblurring mask Dl, D2, ..., Dx for each. The final deblurring mask Df is then equal to the mean, advantageously arithmetic, or even the median, of all the calculated individual deblurring masks Dl, D2, ..., Dx.

[0051] Figures 6 and 7 illustrate the benefit resulting from the implementation of a deblurring process 100 according to the invention. Indeed, to measure and illustrate the impact of such a deblurring process 100, [Fig. 6] shows the difference in contrast between an experimental image Mn and the deblurred experimental image Mnc. It is also possible, as illustrated in [Fig. 7], to calculate the respective magnitudes YMn and YMnc of the Fourier transforms of the experimental image Mn and the deblurred experimental image Mnc. Such magnitudes YMn and YMnc are shown in [Fig. 7], and the white dashed circles clearly highlight that frequencies lost in the case of the experimental image Mn have been restored in the case of the deblurred experimental image Mnc, the image representing the magnitude YMnc appearing more contrasted than that representing the magnitude YMn.

[0052] It will be appreciated by those skilled in the art that this disclosure is not limited to what is particularly shown and described above. Other modifications may be envisaged without departing from the scope of the present invention as defined by the attached claims. In particular, in the preferred example described above, the method 100 is implemented to deblur an image, representing the brain of a patient, resulting from an ASL perfusion acquisition sequence by a magnetic resonance imaging device.Now, such a 100% compliant process of the invention could be used in the context of all medical imaging techniques, as soon as an image presents a blur, for example for the PET technique, an Anglo-Saxon acronym for positron emission tomography, for which the quality of the acquired image can be limited, for example, by the non-collinearity of the emitted photon pairs, intercrystalline scattering, or even the penetration of crystals.

Claims

Demands

1. Method (100) for deblurring an experimental medical image (Mn) resulting from an acquisition sequence by a medical imaging device (1) describing a plurality of elementary volumes of interest of an organ of a patient, called "voxels", said method (100) being implemented by a processing unit (4) of a medical imaging analysis system (S), said method (100) comprising: - a step (110) of calculating the Fourier transform (FTMn) of said experimental image (Mn); - a step of calculating the Fourier transform (TFM0) of a reference image (MO) of the same resolution as the experimental image (Mn); - a determination step (130) of a first modulus (YMn) of the Fourier transform (TFMn) of the experimental image (Mn) and of a second modulus (YM0) of the Fourier transform (TFM0) of the reference image (MO);- a step of obtaining (140) a deblurring mask (D) consisting of calculating the ratio (D) between the first and second moduli (YMn, YM0); - a step of applying said ratio (D) to the Fourier transform (TFMn) of the experimental image (Mn) by multiplication to obtain a corrected Fourier transform (TFMnc); - a step of obtaining (160) a deblurred experimental image (Mnc) consisting of applying the inverse Fourier transform to said corrected Fourier transform (TFMnc).

2. Method (100) according to the preceding claim, wherein the reference image (MO) is derived from an atlas, a control patient or an artificial anatomical reference.

3. Method (100) according to claim 1 comprising a step (101) prior to the steps (110, 120) of calculating Fourier transform consisting of a step (101) of acquiring a structural image of said patient of the same resolution as the experimental image (Mn), said structural image corresponding to said reference image (MO).

4. Method (100) according to any one of the preceding claims comprising a step (102) prior to the steps (110, 120) of calculating Fourier transform consisting of a step (102) of registration of the experimental image (Mn) with respect to the reference image (MO) or of the reference image (MO) with respect to the experimental image (Mn) so as to ensure their voxel-by-voxel correspondence.

5. A method (100) according to any one of the preceding claims comprising a step (142) prior to step (150) of applying said ratio (D) and subsequent to step (140) of obtaining a deblurring mask (D) consisting of applying a smoothing filter on said deblurring mask (D) to obtain a smoothed deblurring mask (DL).

6. Method (100) according to the preceding claim, wherein the smoothing filter is a Gaussian filter.

7. A method (100) according to any one of the preceding claims, wherein the Fourier transforms (TFMn, TFM0) of the experimental image (Mn) and of the reference image (MO) are respectively calculated with the fast Fourier transform algorithm.

8. A method (100) according to any one of the preceding claims wherein the experimental image acquisition sequence (Mn) is a perfusion acquisition sequence by a magnetic resonance medical imaging device (1) during which an endogenous tracer passes within said plurality of elementary volume of interest.

9. Product computer program comprising one or more program instructions executable by the processing unit (4) of a computer, said program instructions being loadable into a non-volatile memory of said computer and the execution of which by said processing unit (4) causes the implementation of a method (100) according to any one of the preceding claims.

10. Computer-readable storage medium containing instructions for a computer program product according to the preceding claim.

11. A medical imaging analysis system (S) comprising a processing unit (4) arranged to deblur an experimental medical image (Mn) resulting from an acquisition sequence by a medical imaging device (1) in relation to a plurality of elementary volumes of interest, called voxels, said processing unit (4) being configured to: - calculate the Fourier transform (FTMn) of said experimental image (Mn); - calculate the Fourier transform (FTMO) of a reference image (MO) of the same resolution as the experimental image (Mn); - determine a first modulus (YMn) of the Fourier transform (TFMn) of the experimental image (Mn) and a second modulus (YMO) of the Fourier transform (TFMO) of the reference image (MO); - obtain a deblurring mask (D) by calculating the ratio (D) between the first and second modules (YMn, YMO); - apply said ratio (D) to the Fourier transform (FTMn) of the experimental image (Mn) by multiplication to obtain a corrected Fourier transform (FTMnc); - obtain an experimental deblurred image (Mnc) by applying the inverse Fourier transform to said corrected Fourier transform (TFMnc).

12. Medical imaging analysis system (S) according to the preceding claim comprising a program memory containing the program instructions of a computer program product conforming to claim 9.

Images