Method and System for Generating Biomarkers

Abstract

Method for generating a biomarker, comprising: ■ acquiring (ACQ) an image using an MRI system; ■ processing the MRI image to generate (GEN 3D) a three-dimensional image of the lung; ■ generating (GEN F1) a first function (F1) corresponding to the distribution of different signal intensity values; ■ automatically calculating (DET_S) a filtering threshold (S A , S1, S2) of the first function (F1) according to a second signal intensity value distribution function (F2, F g1 , F g2 ); ■ segmenting (SEG) a lung volume comprising the following volumes: o a main volume (V A P ); o a filtered volume (V) of the volume of voxels quantified by the first function (F1) and filtered by means of at least the calculated filtering threshold (S1); ■ normalizing the values of the three-dimensional image of the lung volume (V F P ); ■ generating a biomarker (B) indicating the ratio of the normalized segmented volumes. IND )

Description

[0001] The field of the present invention relates to the field of methods for generating biomarkers of lung regions. More specifically, the present invention relates to methods for processing images to quantify the presence of features that can be associated with a given pathology. The field of the present invention is more specifically applied to the processing of images acquired by MRI.

[0002] Currently, the quantification of inflammatory phenomena and airway remodeling is mainly done by scanner imaging systems. This technique (more widely known as tomographic densitometry, TDM or CT scan) makes it possible to obtain three-dimensional images providing resolution and a sufficient level of contrast to elaborate quantification strategies. More generally, CT scan imaging makes it possible to obtain an indication of the 3D extent of abnormalities in the lung ducts.

[0003] However, this technique has the drawback of irradiation. In addition, this technique cannot distinguish between inflammation and airway remodeling. There is a need to use non-irradiating imaging to elaborate biomarkers and to enable signals that can characterize active inflammatory phenomena and signals of definitive scar lesions, as allowed by MRI imaging.

[0004] However, to date, it has been difficult to define modalities for processing lung region images by MRI due to the fact that the tissue and cellular structure of these regions is variable and difficult to segment without losing data. In addition, MRI signals are not calibrated like those of scanners, so it is not possible to use the raw data of signal values on a given image, and the true values of T1 and T2 are "artifact-laden" in the parametric mapping on the lung, that is, there are acquisition errors due to the magnetic susceptibility of the normal lung. To date, there is no such acquisition modality that makes it possible to define a suitable segmentation of the airways and thus provides the possibility of generating sufficiently effective biomarkers to replace CT scan acquisitions.

[0005] Therefore, there is a need to define a solution that can respond to this problem. The present invention aims to solve the above-mentioned drawbacks.

[0006] According to a first aspect, the present invention relates to a method for generating biomarkers, comprising:

[0007] ■ Acquiring images using an MRI system;

[0008] ■ Processing the MRI images to generate a three-dimensional image of the lung;

[0009] ■ Generating a first function corresponding to the distribution of different signal intensity values of each voxel of a part of the acquired three-dimensional image;

[0010] ■Automatically calculate at least one filtering threshold of the first function according to a second function of at least one different signal intensity value distribution of each voxel of a part of the acquired three-dimensional image;

[0011] ■Segment a volume including the following volumes:

[0012] ○A main volume corresponding to the lung volume;

[0013] ○A filtered volume of the volume of voxels quantified by the first function and filtered at least by means of the calculated filtering threshold;

[0014] ■Normalize the values of the three-dimensional image of the lung volume according to the absolute value of the signal intensity value of the voxels of the image and at least the calculated filtering threshold;

[0015] ■Generate a biomarker indicating the normalized segmentation volume ratio.

[0016] The lung volume corresponds to the volume delimited by the walls of one or more lungs.

[0017] Advantageously, the present invention enables the automation of the generation of biomarkers while defining an adaptive threshold according to the patient. Another advantage is that it enables the configuration of the acquisition to generate different biomarkers according to the desired usage.

[0018] In other words, the step of generating the first function corresponding to the distribution of different signal intensity values of each voxel of a part of the acquired three-dimensional image corresponds to the application of the first function that results in the generation of the distribution of different signal intensity values of each voxel of a part of the acquired three-dimensional image.

[0019] According to an example, the method is implemented by the functions of the acquired MRI images within a console or by the functions of an MRI system within a device including a physical or wireless interface with the console. To implement the method of the present invention, an interface for receiving MRI images and a memory for storing MRI images can be used.

[0020] The device enabling the implementation of the method of the present invention advantageously includes a display and a graphical interface for parameterization configuration according to the biomarker to be generated.

[0021] According to an embodiment, the method includes:

[0022] ■The three-dimensional image acquired by using an MRI system is configured by the following:

[0023] ○T2 weighting; and

[0024] ○An echo time TE greater than a predetermined threshold;

[0025] ■The automatic calculation of at least one filtering threshold includes:

[0026] ○ Obtain a reference volume;

[0027] ○ Generate a reference function corresponding to the distribution of different signal intensity values of each voxel of the reference volume, wherein the reference function is a second distribution function;

[0028] ○ Calculate the standard deviation of the reference function;

[0029] ○ Determine a filtering threshold according to the calculated standard deviation of the reference function.

[0030] The advantage is that biomarkers can be generated, enabling quantification of the volume of mucus load and / or edema without implementing irradiation techniques.

[0031] According to an example of this embodiment, the acquisition is parameterized, and the acquisition includes:

[0032] ■ The acquisition of the image is reconstructed from multiple images acquired over one or more cycles at at least 4 echo times, wherein the echo times are configured according to increasing durations;

[0033] ■ A spin echo sequence; and

[0034] ■ Parameterization, aiming to emit signals to pre-saturate or saturate the acquired signals.

[0035] The advantage is that the processing of the circulating volume is significantly improved to eliminate them from the image processing.

[0036] According to an example of this same embodiment, a second acquisition of the image is performed using a first configuration to outline the lung volume, and steps of image processing are performed to combine the image acquired using T2-weighted acquisition with the image acquired through the second acquisition. The first configuration defines T1-weighted parameterization or proton density-weighted acquisition. In other words, the first acquisition enables configuration of T1-type weighting or proton density-type weighting.

[0037] The advantage is that a mask is constructed, enabling outlining of the lung volume and organs such as the heart for processing the image acquired using T2-weighted acquisition. In fact, generally, the image acquired using T2-weighted acquisition cannot obtain the segmentation of the lung volume, and the separate segmentation enables outlining of the organ contours.

[0038] According to an example of this embodiment, a merging operation is performed between at least one image acquired using T1-weighted with an ultra-short echo time UTE and at least one image acquired using T2-weighted to generate an image in which the data from the acquired images have been combined to segment the lung volume.

[0039] The advantage is that an image is generated based on the merging of the two images, and the resulting image provides the best isolation ability for the region of interest within the lung volume.

[0040] According to an example of this embodiment, the reference threshold is based on the reference function F A of the reference distribution value and the standard deviation of the reference function F A value between 10 times and 20 times the value of is established by a combination of values.

[0041] The advantage is that the noise level can be eliminated. Another advantage is to define an adaptive threshold to generate a threshold compatible with a large number of patients.

[0042] According to an example of this embodiment, the reference function F A of the reference distribution value is the main mode M of the distribution of the signal intensity values of the images acquired with T2-weighted within the lung volume V P . The advantage is that a filter that can be applied to different types of patients with different profiles can be implemented using the reference value. Therefore, the filter is adaptive and does not require the acquisition configuration or profile of the patient. The selection of the main mode enables an adaptive threshold to be defined for each patient. B

[0043] According to an example of this embodiment, the method includes a normalization step, and the normalization step includes calculating the volume intensity product (VIP) of the signal according to the absolute value of the signal of the filtering volume V F , and the volume is generated by the filtering volume V F and the lung volume V P . The advantage is to quantify the inflammation level in areas such as the lungs. Another advantage is that the spread of inflammation over time can be quantified.

[0044] According to another embodiment, the method for generating a biomarker is characterized in that:

[0045] ■ Using an MRI system to acquire three-dimensional images configured by the following;

[0046] ○ Proton density or T1-weighted, echo time; and

[0047] ○ Echo time less than a predetermined threshold;

[0048] ■ The automatic calculation of at least one filtering threshold includes;

[0049] ○ Modeling at least two Gaussian functions by adjusting the first function;

[0050] ○ Determining the filtering threshold by calculating the intersection of the first Gaussian function and the second Gaussian function;

[0051] ○ Determining a second threshold corresponding to the minimum value of the first Gaussian function and the minimum value of the signal intensity value of the voxel;

[0052] ■ Make the filtered volume be a first volume corresponding to the voxels quantified by a first Gaussian function between a first threshold and a second threshold, the voxels corresponding to an air medium;

[0053] ■ Normalize the values of the three-dimensional image of the lung volume according to the calculated first threshold and second threshold;

[0054] ■ Generate a first biomarker indicating the ratio of the feature volume to the normalized segmentation volume, the ratio being calculated between the feature volume and the lung volume.

[0055] The advantage of this configuration is to obtain a good segmentation of the lung, while separating the feature volume because a biomarker related to the phenomena associated with emphysema can be identified.

[0056] According to an example of this embodiment, the echo time is less than 1 ms. The advantage of an acquisition configuration with a relatively short TE is that it is insensitive to eddy currents. This configuration enables the signal in the lung parenchyma to be maximized.

[0057] According to an example of this embodiment, the segmentation includes defining a second volume corresponding to the voxels quantified by a second Gaussian function and greater than the first threshold, the voxels corresponding to fat or an intermediate medium. The advantage is an improved segmentation of different volumes in the lung.

[0058] According to an example of this embodiment, the segmentation includes a step of extracting a feature volume, the feature volume including the voxels of the filtered volume, the signal intensity value of the filtered volume being less than a third predetermined threshold, the third predetermined threshold being determined on a normalized scale [0; 1]. The advantage is to obtain a representation of the biomarker of the emphysema phenomenon.

[0059] According to an example of this embodiment, the segmentation includes a step of excluding / deleting disconnected adjacent identically quantified voxels. The advantage is to reduce acquisition artifacts and select a volume that actually corresponds to the feature volume of the pathology.

[0060] According to an example of this embodiment, the modeling of the Gaussian function includes:

[0061] ■ Apply Gaussian smoothing to the acquired image to denoise by reducing the single-shot encoding time and radial acquisition;

[0062] ■ Outline the contour by applying a local filter;

[0063] ■ Represent the distribution frequency of the voxels using a curve adjustment method.

[0064] The advantage is to obtain a realistic and faithful modeling of the distribution of the intensity of the values of the signal in the image. The functionalization from the processing of the Gaussian function enables the generation of the same processing for each patient, which is representative and at the same time adaptive.

[0065] According to an embodiment, the method includes acquiring a signal intensity value of each voxel quantified by the first function quantity, wherein the intensity value corresponds to image contrast data.

[0066] According to an embodiment, the acquisition is performed in a manner synchronized with a respirator. According to an example, the respirator is a navigator or a breathing belt. The advantage is that there is no movement of the lungs during the acquisition of the image.

[0067] According to an embodiment, the method includes performing a step of extracting a volume image based on the image acquired by MRI, and the extracted image is generated at a determined moment of the sequence.

[0068] According to another embodiment, the acquisition of the three-dimensional image is performed by stacking the acquired 2D images, and the thickness of the section is at least equal to the width of the voxel.

[0069] According to another aspect, the present invention relates to a system that at least includes a calculator, a memory, and an interface for receiving the image acquired by an MRI system, and the system is configured to implement the steps of the method of the present invention. Advantageously, the display enables visualization of the acquired MRI image and the processed image. The biomarker can be graphically represented within the generated image. According to an example, the graphical interface enables adjustment of the acquisition parameters and definition of an area of interest or a search area for a pathology.

[0070] According to an embodiment, the first acquisition and the second acquisition of the method according to the present invention are continuously performed within a respiratory cycle, and each acquisition is synchronized with the data characteristics of inhalation and exhalation respectively. The method further includes:

[0071] ■ Generating a first biomarker for each acquisition of the same respiratory cycle;

[0072] ■ Extracting the quantification of the main volume for each acquisition of the same respiratory cycle; and

[0073] ■ Elastically resetting between these two respiratory times to select the voxels representing the corrected difference between the two biomarkers generated in each acquisition.

[0074] According to an embodiment, the method includes normalizing the quantification of the voxels selected between the two acquisitions.

[0075] According to an embodiment, the acquisition is a 4D acquisition, which is configured to acquire the change of the first biomarker per time unit within a predetermined duration.

[0076] According to an embodiment, the 4D acquisition is configured to acquire a voxel volume in time units corresponding to the duration of a respiratory cycle, and the acquisition is synchronized with data characteristics of inhalation and / or exhalation. Other features and advantages of the present invention will become apparent by reading the following detailed description with reference to the accompanying drawings, which illustrate:

[0077] Figure 1 : Steps of an embodiment of the method of the present invention;

[0078] Figure 2 : A first embodiment of the method of the present invention, which is applied to quantify lung air volume to generate a first biomarker;

[0079] Figure 3 : A second embodiment of the method of the present invention, which is applied to quantify the displacement volume of an airway to generate a second biomarker;

[0080] Figure 4 : A 2D slice of a three-dimensional image of the lung, where the slice or the three-dimensional image is acquired by an MRI system;

[0081] Figure 5 : An exemplary use of two functions for determining a first threshold to generate a first biomarker normalized according to the first embodiment;

[0082] Figure 6A : An example of a curve for quantifying different 2D image contrasts obtained using a scanner, and an example of a threshold that enables the generation of an index for characterizing the quantification of emphysema in patients with chronic obstructive pulmonary disease;

[0083] Figure 6B : An example of a curve for quantifying different contrasts of 2D images obtained using a scanner, and an example of a threshold that enables the generation of an index for characterizing the quantification of emphysema in healthy patients;

[0084] Figure 7A : An example of a curve for quantifying different contrasts of 2D images obtained by MRI and processed according to the method of the present invention for automatic threshold calculation, enabling the generation of an index for characterizing the quantification of emphysema in patients with chronic obstructive pulmonary disease;

[0085] Figure 7B : An example of a curve for quantifying different contrasts of 2D images obtained by MRI and processed according to the method of the present invention for automatic threshold calculation, enabling the generation of an index for characterizing the quantification of emphysema in healthy patients;

[0086] Figure 8: Representation of the amount of congestion volume (e.g., secretion of mucus) of the patient's lung volume on a cross-section of an MRI;

[0087] Figure 9 : Representation of a reference threshold that enables filtering of high-intensity voxels to generate a mask.

[0088] Definition

[0089] In the remainder of this specification, a three-dimensional image includes a plurality of volume pixels (also referred to as voxels). The terms image pixel or voxel will be used interchangeably.

[0090] In the remainder of this specification, "hyposignal" will represent a low-intensity signal that appears as a black or dark gray patch on an MRI image. This is the case in the absence of protons (air), short T2 substances (calcified tissue (cortical bone, enamel, calcification), collagen-rich tissue (tendon, ligament, fascia, etc.), protein-rich fluid, stagnant fluid (urine, LCS, exudate, edema, inflammation, etc.), substances with paramagnetic effects (concentrated gadolinium), high concentrations of iron (hemochromatosis), certain hematomas, etc.).

[0091] In the remainder of this specification, "hypersignal" will represent a high-intensity signal that appears as a white or light gray patch on an MRI image. In T1, it is the signal of substances at short T1: lipids (fat, fatty tumor, fatty marrow, etc.), protein-rich fluid (mucocele, tumor cyst, etc.), substances with paramagnetic effects (gadolinium, etc.), iron (subacute hematoma), free radicals (melanin), posterior pituitary gland, etc. In T2, it is the case of free water (urine, LCS, exudate, synovial fluid, etc.), interstitial water (edema, inflammation, etc.), stagnant blood, low-protein fluid, etc.

[0092] Figure 1 Represents the main steps of an exemplary embodiment of the method of the present invention. The steps include acquiring an image or a sequence of images using an MRI system. For example, the acquisition is performed by acquiring 2D slices, which are then stacked to form a three-dimensional image. According to another example, the acquisition is a direct three-dimensional acquisition, which enables obtaining a 3D image. The present invention is also compatible with so-called "4D" acquisitions, in which the time scale is considered to use image processing algorithms to generate and reconstruct images.

[0093] According to an embodiment, the device for measuring the respiratory rate is used to acquire a signal synchronized with respiration. A respirator or any other device that enables synchronizing the respiratory rate with image acquisition can be used. This feature enables obtaining an acquisition that is less sensitive to movement.

[0094] The method of the present invention includes a step of generating a first distribution function F1 aimed at generating a voxel distribution of an image as a function of signal intensity. According to the acquisition mode of the MRI image, the resolution of the configuration parameters is configured. According to other embodiments, the signal intensity considered is represented by hue, saturation, or brightness level, or a combination of these three factors. According to an example, the signal intensity is a parameter encoded according to an absolute scale based on the acquired image. According to another example, the parameter is encoded according to a relative scale as a function of the magnitude of the variation of the parameter over all the voxels of the acquired image.

[0095] Therefore, the first function F1 makes it possible to take into account the distribution of this signal intensity in the image.

[0096] The method of the present invention includes a step of generating a second function F2 such that a threshold applied to the function F1 can be automatically calculated in order to normalize the obtained distribution values.

[0097] The second function F2 is advantageously a function describing the distribution of the signal intensity of the image. According to different embodiments, this function can be implemented and defined in such a way that the calculation of the threshold is optimized, a threshold is generated to achieve optimized normalization, or a function aimed at identifying patterns in the image or related to the singularities of the distribution of the function F1.

[0098] The method of the present invention includes a step DET_S such that at least one threshold can be automatically determined. The at least one determined threshold is particularly used to filter the distribution of the signal intensity represented by the first function F1 or by the second function F2.

[0099] According to different embodiments, a plurality of thresholds can be determined. The thresholds can be determined using an operation or a combination of operations on at least the second function F2. The operations can include curve intersection calculation, standard deviation calculation, determination of eigenvalues, or alternatively any other type of calculation capable of performing calculations on digital functions.

[0100] The method includes a step SEG which aims to segment the volume of the acquired image filtered with at least one previously determined threshold by selecting the pixels related to the function F1.

[0101] The segmentation includes the segmentation of the lung volume V P and at least one filtered volume V F The filtered volume V F can correspond to the volume obtained after filtering of the function F1 or F2. This segmentation can be performed in several steps. The first step can correspond to the segmentation of the lung volume V P that is, corresponding to the inner membrane of the lung.

[0102] According to an embodiment, the method of the present invention includes a normalization step which particularly makes it possible to normalize in the lung volume VP Upper-segmented filtering volume V F is normalized by the ratio of...

[0103] According to an embodiment, the method of the present invention includes a step aimed at extracting a characteristic volume of a filtering volume representing a physiological sign. It can be a biomarker that enables the quantification or qualification of the presence of a pathology.

[0104] Thus, the method enables the generation of a biomarker B IND such that the characteristic volume V C can be evaluated by the ratio of...

[0105] According to a first embodiment of the present invention, the implementation of the method enables the generation of a first biomarker B IND1 . According to another embodiment, i.e., the second embodiment, the method of the present invention enables the generation of a second biomarker B IND2 .

[0106] The first biomarker

[0107] Figure 2 represents the main steps of the first embodiment of the present invention. In this embodiment, a first configuration CONF1 of the MRI imaging system is implemented. The first configuration CONF1 enables the configuration of weighting of the T1 or proton density type. In both weighting cases, the echo time (temps d’éch, labeled as TE) is defined as less than a predetermined threshold. The repetition time (temps de répétition, labeled as TR) itself is configured within a range of values considered short or long according to the weight used. Those skilled in the art will know how to establish the association between the TE, TR configuration and the selected weight.

[0108] - Acquisition

[0109] This MRI acquisition can be a volume acquisition, that is, the obtained images are directly three-dimensional images. This configuration CONF1 enables the definition of an ultra-short echo time called "UTE", or alternatively the definition of a zero TE sequence (also called ZTE).

[0110] The UTE sequence can be used to acquire isotropic 3D images. The 3D images can be obtained, for example, by means of a cylindrical acquisition with stacked radial planes or a spherical acquisition using radial projections in three directions. One possible option is to take the end portions of the rays on the surface of the sphere distributed on a helix. Such an acquisition pattern is described, for example, in Joseph Yazbek's "IRM pulmonaire 3D à temps d’écho ultracourt par acquisition spiralée ou sphérique de l'espace-k". The filling of the volume is done by the readout gradient. This ends in a discretized surface in a substantially uniform manner in spherical coordinates.

[0111] The ZTE sequence is also a 3D acquisition technique. This technique consists of applying a steady amplitude gradient and sending RF pulses with a very short duration and a small flip angle. The acquisition starts after a given duration corresponding to the flip between transmission and reception. Between two TRs, the amplitude of the readout gradient gradually increases. Due to the fact that there is no constraint on the rate of change of the intensity of the gradient during the acquisition time, this configuration has the advantage of being insensitive to Foucault currents.

[0112] The advantage of the UTE or ZTE sequence is to maximize the available signal in the lung parenchyma.

[0113] The acquired images can be processed by image processing algorithms in order to reduce noise, eliminate artifacts, or adapt the contrast scale.

[0114] According to an embodiment, when the acquisition is performed in 2D instead of 3D, the steps of reconstructing a 3D image can be performed for each stack of 2D layers. Then, the obtained and optionally processed three-dimensional image includes a plurality of voxels.

[0115] - Image processing

[0116] Figure 4 Represents a cross-section of a three-dimensional image of each lung having a lung volume V P The volume of the thorax is represented, for example, here by the volume V T In this image, the inner membranes of each lung are illustrated, in particular due to the voxel signal intensity discontinuities delimiting the volume of each lung with the volume of the thorax.

[0117] According to an embodiment, the MRI imaging system is configured with a PETRA sequence. According to an exemplary embodiment, a contour enhancement filter can be applied to the acquired images. Of interest is obtaining a better definition of the contrast of the images. The contours of the contours can be obtained from the application of local filters.

[0118] According to an embodiment, the method of the present invention includes a step of applying Gaussian smoothing to the acquired image, which reduces the one-time encoding time and performs denoising using radial acquisition. The purpose of this processing is to reduce the noise obtained during the acquisition of the assumed regularity of the curve obtained by this processing method.

[0119] - Extraction of lung volume

[0120] The method of the present invention enables the generation of a curve C1 representing the diffusion or distribution of voxels according to signal intensity values or chromaticity parameters. The distribution can be limited to the lung volume V by first performing the extraction of the 3D inner membrane of the lung. P This extraction can be automatically performed using a contour processing algorithm based on the analysis of signal intensity values or the analysis of chromaticity parameters of the voxel and its neighborhood.

[0121] The signal intensity values or chromaticity parameters enabling the realization of the distribution of voxels are advantageously contrast data. Any other type of parameter can be used as a function of the acquired image. For example, brightness, hue, or saturation data can be used in combination or used in place of contrast data.

[0122] Thus, the first function F1 represents the voxel distribution curve in the lung volume V according to a given resolution. The given resolution corresponds to the resolution of image acquisition. Therefore, the resolution depends on the image acquisition configuration. P The given resolution corresponds to the resolution of image acquisition. Therefore, the resolution depends on the image acquisition configuration.

[0123] The function F1 can be a continuous function or a discrete function. In the latter case, the voxels can be clustered together according to a scale defining contrast units. This scale can be predefined or automatically calculated as a function of, for example, the minimum and maximum values and the lung volume and / or calculated as a function of the number of voxels considered. The curve C1, which is a graphical representation of the function F1, is shown in Figure 5 .

[0124] According to an embodiment, the extraction of a volume (such as the filtered volume V P or the characteristic volume V F or V C ) within the lung volume V includes image processing that enables the exclusion or even elimination of voxels having the same contrast range that are disconnected from the voxels in their neighborhood. Of interest is the elimination of false positives, acquisition errors, or voxels outside the region of interest representing physiological or pathological signs.

[0125] The method of the present invention includes a step of automatically calculating at least one threshold S1 such that the voxels of the distribution represented by the function F1 or the voxels of the function modeled by the function F1 can be filtered.

[0126] In this first embodiment, two thresholds S1 and S2 are calculated. The threshold S1 defines the maximum value, and the second threshold defines the minimum value of the signal intensity value or chromaticity parameter to be considered for generating the distribution of voxels.

[0127] - Definition of the second function

[0128] According to this first embodiment, in order to determine the optimal threshold, the method includes a step of aiming to model at least two Gaussian functions Fg1, Fg2 by adjusting the first function F1.

[0129] These two Gaussian functions Fg1, Fg2 are obtained such that they each model the distribution of the signal of the lung volume V P (that is, the distribution of the voxels within the lung volume V P ).

[0130] The first Gaussian function Fg1 basically models the signal of the lung volume V corresponding to the air medium P . The second Gaussian function Fg2 basically models the signal of the lung volume V corresponding to the intermediate medium (such as the heart, muscle, blood vessels or bone marrow) P .

[0131] By adjusting the modeling of the curve C1, two functions Fg1 and Fg2 representing two Gaussian distributions can be obtained. Figure 5 Represents the curves associated with the functions Fg1, Fg2. The third curve Fg3 also represented in Figure 5 can be obtained by adjusting the curve C1. The function Fg3 basically models the signal of the lung volume V corresponding to the fat medium P .

[0132] - Generation of the thresholds

[0133] The first threshold S1 is automatically generated by calculating the intersection of the curves Fg1 and Fg2. The first threshold S1 defines the maximum value of the quantization of the signal in the lung volume V P . In the case of Figure 5 , it is the maximum value of the contrast parameter. The maximum value S1 enables filtering of the voxels of the distribution of C1 with values below the first threshold S1.

[0134] The advantage of defining the first threshold S1 is to obtain an optimal segmentation between the values of the lung parenchymal tissue and the soft tissue.

[0135] The second threshold S2 is automatically generated by considering the minimum contrast value that defines the first voxel with the first contrast level. Figure 5 Represents the second threshold S2.

[0136] The advantage is that there is no variability in the absolute value of the image signal, which may depend from one patient to another or from one given configuration to another given configuration. The method of the invention makes it possible to consider the minimum value of the signal of the image as a value normalized to 0.

[0137] Then a step of segmenting SEG of the lung volume is performed, the aim of which is to extract the filtered volume V delimiting the voxels of the curve C1 F , the signal intensity value of which or the chromaticity parameter of which is included between a second threshold S2 (minimum value) and a first threshold S1 (maximum value).

[0138] If not already done previously, the volume V of the lung forming the inner membrane of each lung is also extracted in this step P .

[0139] - Normalization

[0140] The method includes a step of normalization NORM which makes it possible to normalize the values of the signal included between two thresholds S2 and S1. The minimum value corresponding to the threshold S2 is defined as the value 0 and the maximum value corresponding to the threshold S1 is defined as the value 1. The value at 0 is close to the value of the voxels corresponding to the air medium in the lung. The value at 1 is close to the value of the voxels corresponding to the soft tissues of the chest.

[0141] Thus, the method of the invention makes it possible to select a first adaptive threshold S1 which makes it possible to quantify the amount of signal the value of which is less than this first threshold S1.

[0142] Advantageously, the method makes it possible to take into account the possible variations in the threshold levels according to the patient.

[0143] - 4D acquisition for monitoring biomarkers

[0144] The method of the invention can be applied from static three-dimensional MRI images or from dynamic three-dimensional MRI images with an additional time component. This is then called 4D acquisition. The advantage of 4D acquisition is to generate dynamic biomarkers which make it possible to evaluate the temporal variations of a first biomarker BIO IND1 over a given duration. This acquisition is particularly interesting when measuring the variations in an indicator between the inspiration and the expiration of the patient. Then, the first biomarker BIO IND1 makes it possible to quantify the functional impairment of the small airways.

[0145] In fact, the overall measurement of the changes in lung signals over a period of time cannot systematically distinguish between fixed destructive lesions (termed emphysema) and potential reversible damage to the small airways or alveolar trapping. Accordingly, the object of the present invention is to define dynamic biomarkers such that quantification of voxels can be obtained, which allows the distinction between emphysema and reversible damage to the small airways.

[0146] The 4D acquisition method of the present invention is particularly suitable for processing low signals of the lung and effectively processes the low signals of the lung.

[0147] 4D acquisition of the lung can be achieved according to two embodiments.

[0148] In a first embodiment, two static acquisitions are performed. The first acquisition is carried out during inspiration, and the second acquisition is carried out during expiration or when the patient holds their breath. Each acquisition can be triggered jointly in synchronization with the target expected respiration. For example, synchronization can be performed from a device intended to measure the air exhaled or inhaled by the patient.

[0149] In a second mode, acquisition is carried out during free breathing, and retrospective respiratory synchronization enables the decomposition of the entire respiratory cycle.

[0150] According to the first embodiment, the signal representative of the target of emphysema and the entire signal of the lung parenchyma are extracted such that elastic resetting can be performed between these two respiratory times, and thus the voxels identified as emphysema are subtracted from the voxels associated with the entire change in the lung parenchyma in order to obtain an accurate and objective measurement of alveolar trapping. This discrimination is not easily achieved by a simple overall measurement of the changes in lung signals. Therefore, when quantifying this discrimination, this acquisition mode is particularly advantageous for generating dynamic biomarkers representing the voxels of interest.

[0151] According to an embodiment, when the first acquisition is carried out during inspiration and the second acquisition is carried out during expiration, the acquisition is configured to be triggered at the cycle instants of each respiratory cycle. In this embodiment, two acquisitions are performed during the respiratory cycle.

[0152] According to an alternative embodiment, acquisition is carried out during each inspiration, and the differential volume is measured between two inspirations. According to another alternative embodiment, acquisition is carried out during each expiration, and the differential volume is measured between two expirations. In this case, a single acquisition is performed during the respiratory cycle.

[0153] According to the second embodiment, during inspiration or expiration, the change in the volume of emphysema over time enables the volume change of this emphysema per unit time to be obtained. Accordingly, this dynamic biomarker is a reflection of lung compliance.

[0154] According to one or another of two embodiments, the threshold setting of the voxel intensity can be configured during acquisition in the inspiration phase and / or during the expiration phase. The method of the present invention enables measurement of changes in the threshold processing for each configuration. These changes can be measured between two static acquisitions or during 4D acquisition.

[0155] The advantage is to obtain a dynamic marker with discriminative ability during the respiratory dynamics. It is interesting to more effectively distinguish the volumes of voxels corresponding to different pathologies.

[0156] According to one usage pattern, the dynamic biomarker can be used as a supplement to the static biomarker of the present invention to confirm or weaken the first characterization of the acquired images and the pathological condition.

[0157] - Define a third threshold for generating the biomarker

[0158] Figure 6A and Figure 6B represent a signal distribution curve of the segmented volume from the images obtained using a scanner. In this example, a predefined threshold S3 at a given scale enables the distinction of the lung including the Figure 6A characteristic volume V represented in C which is larger than the characteristic volume V obtained in the case represented in C The volume V Figure 6B represented in C can correspond to an intermediate medium that confirms a given physiological signal. C The curves obtained in Figure 6A and Figure 6B can correspond to the curves of two patients. Figure 6A is characteristic of the presence of emphysema. Figure 6B is characteristic of a healthy subject.

[0159] Figure 7A and Figure 7B represent the distribution curve of the signal included between two thresholds S1 and S2 of the characteristic segmented volume V C obtained by implementing the method of the present invention. For example, Figure 7A corresponds to the same patient as Figure 6A and Figure 6B corresponds to the same patient as Figure 7B The curves in Figure 7A and Figure 7B are obtained from an MRI imaging system according to the method of the present invention.

[0160] Figure 7A represents the characteristic volume V obtained for voxels with values less than a predetermined threshold on a scale normalized from 0 to 1 C . In Figure 7A and​​​​​​Figure 7B In this case, the predetermined threshold corresponds to a value of 0.2. When the low signal is significantly lower than the third threshold S3, this threshold makes it possible to obtain a good representation of the patient's pulmonary emphysema.

[0161] When the threshold S3 included between the threshold S2 and the threshold S1 is low, that is to say, close to 0, for example between 0 and 0.30, the first biomarker BIO IND1 can be used for the detection of specific signs of the pulmonary emphysema phenomenon.

[0162] When the threshold S3 included between the thresholds S2 and S1 is a threshold with an intermediate value, that is to say, a value range defined around the median or the average value of the scale [0; 1], for example between 0.35 and 0.65, the biomarker BIO IND1 can be used as a capture marker for the airways.

[0163] Thus, this first threshold level makes it possible to generate the biomarker BIO IND1 , so as to make it possible to classify a specific patient based on physiological data.

[0164] Figure 7A Relative to Figure 6A the advantage is that the characteristic volume V obtained by using the method of the present invention is more important than the part obtained by using the method of the prior art. In addition, C the difference between the healthy subjects shown and the subjects suffering from pulmonary emphysema is more significant in Figure 7B and Figure 7A and Figure 7B than in Figure 6A and Figure 6B .

[0165] On the normalized scale, the threshold S3 defined as 0.2 enables the doctor to obtain satisfactory results, so as to make it possible to distinguish the characteristic volume of the presence of an intermediate medium in the lungs.

[0166] The second biomarker

[0167] Figure 3 represents the main steps of the second embodiment of the present invention. In this embodiment, a second configuration CONF2 of the MRI imaging system is implemented. The second configuration CONF2 makes it possible to configure the T2-type weighting.

[0168] The purpose of this embodiment is to generate a biomarker representing the volume of mucus load and / or the volume of edema within the lung volume. The advantage of the T2-type weighting is to represent the image signal specific to stagnant fluids such as water, mucus, non-circulating blood. For this purpose, the method of the present invention includes the step of filtering a certain volume to separate the volume representing the mucus load.

[0169] The second configuration CONF2 of the acquisition ACQ2 can advantageously be completed by means of a complementary acquisition ACQ1 that uses T1-weighted or proton density-weighted images. This complementary acquisition ACQ1 significantly makes it possible to improve the segmentation of the lung volume V P which is preferably configured as Figure 2 the acquisition of the first embodiment represented in

[0170] - Acquisition

[0171] According to a second embodiment, the T2 weighting is advantageously achieved with a long echo time TE and a repetition time TR that are considered too long. Recall that the terms long TE, short TE, long TR, short TR are known to those skilled in the art who know how to determine the appropriate configuration of an MRI system with T2 weighting. Possible configurations correspond to an echo time: TE > 80 ms, and a repetition time: TR > 2000 ms.

[0172] For example, this acquisition ACQ2 is performed in 2D or 3D. In the first case, the reconstruction step of the 3D image can be carried out by stacking 2D slices. The obtained and optionally processed three-dimensional image then includes a plurality of voxels. This step is represented by step GEN 3D in Figure 3 which aims to generate a three-dimensional image from a 2D acquisition.

[0173] - RTSE acquisition

[0174] According to an embodiment, the acquisition using T2 weighting is a radial acquisition sequence called "turbo spin echo". It is more commonly labeled T2-RTSE.

[0175] This T2-RTSE sequence is a 2D spin echo-based sequence that has a radial trajectory in k-space. During the acquisition of radial data, the center of k-space is obtained by naturally using the averaging effect produced by the acquisition configuration. This property makes it possible to improve artifacts related to motion. The immunity to radial movement is enhanced by using a so-called PACE (Prospective Acquisition and CorrEction) acquisition technique.

[0176] This acquisition technique is compatible with various sequences, in particular with the RTSE sequences among various sequences. It particularly makes it possible to eliminate some kinetic movements and increase the signal-to-noise ratio.

[0177] T2-RTSE acquisition includes sampling density of rings according to increasing radius, and then using linear interpolation to obtain the radial projection of adjacent echo times TE. This process is repeated using an algorithm with a pseudo-golden angle ratio, such as that described in "Multi-Echo Pseudo-Golden Angle Stack of Stars Thermometry with High Spatial and Temporal Resolution Using k-Space Weighted Image Contrast", until the maximum radius is reached.

[0178] Several echoes are acquired and reconstructed using a k-space stratification method that preserves the contrast of the images acquired at each TE. This technique then includes: generating a synthetic image constructed from all the images acquired at multiple consecutive TEs, and using the T2-weighted reconstructed images from multiple echo acquisitions.

[0179] According to an example, the consecutive echo times TE succeed each other as the duration increases. According to an example, the signal is acquired at 5 TEs, for example, the signal has the following corresponding values: 20 ms, 50 ms, 85 ms, 100 ms, and 150 ms.

[0180] Using a long echo time TE up to 150 ms allows for improved cancellation of the signal from the blood volume circulating in the vessels. This function of canceling the signal from the circulating blood volume can be improved by using acquisitions with spatial saturation and / or double inversion recovery and / or diffusion modules. Double inversion recovery is an MRI sequence labeled as DIR and known to those skilled in the art. In contrast, non-circulating fluids with short to intermediate duration TEs can be erased at long TEs. Therefore, the availability of short TEs in the synthetic image reconstructed over the entire TE is natural, allowing the information of short TEs not to be lost. Thus, multi-TE acquisition achieves an optimal compromise between effective visualization of non-circulating fluids and cancellation of soft parts or circulating vessels.

[0181] Advantageously, the second configuration CONF2 of the MRI imaging system includes parameterization of the TR time and the TE time. Appropriate parameterization allows for acquisitions aimed at eliminating the circulating volume (such as the vascular volume). In fact, the parameter TR allows for control of the proton inflow phenomenon, while the TE allows for control of the proton outflow phenomenon. In the spin echo acquisition configuration, when the circulation speed increases, the acquired signal rapidly decreases. In fact, the circulating protons do not rephase.

[0182] This phenomenon is even more pronounced when the long TE time is parameterized because, when the signal is collected, almost all the excited protons have flowed out. Since a spin echo sequence is used, the signal intensity of the blood vessels becomes weaker. It is worth noting that using a long TE of up to 150 ms and spatial saturation upstream of each echo train makes it possible to further eliminate the signal intensity of the blood vessels in the circulation.

[0183] The second configuration CONF2 advantageously also includes presaturation or saturation. This technique is based, for example, on the choice of the emission frequency, which can be achieved by emitting radiofrequency pulses. This pulse makes it possible to eliminate the longitudinal magnetization of the protons. Before each TE period, the pulse is advantageously emitted in the form of a prepulse. The radiofrequency pulse disperses the spin phases without being followed by a rephasing gradient, but with a spoiler that eliminates the remaining magnetization. For example, a presaturation strip placed upstream of a stack of slices makes it possible to disrupt the magnetization of the protons entering the slice. The signal of these protons will thus be reduced, or even zero, whereas otherwise they would be liable to have a detectable signal in the absence of presaturation. This configuration makes it possible to reduce or eliminate the circulating flow. Thus, the method of the invention includes steps aimed at eliminating the signal of moving protons and retaining the signal of non-moving protons. This operation can be carried out due to the saturation configuration of the signal.

[0184] According to an exemplary embodiment, the following parameters can be defined to establish the second configuration CONF2: reconstructed TE = 20 ms, 50 ms, 85 ms, 100 ms, and 150 ms, TR = 2350 ms, flip angle = 145°, pixel size = 1.6 × 1.6 mm2, section thickness = 1.6 mm, no interleaving, average acquisition time = 12 minutes.

[0185] According to this example, the 3D image can then be constructed by stacking 2D images. A volume can then be obtained that consists of isotropic voxels, for example, of size {1.6 mm} 3 each.

[0186] - Using complementary acquisitions weighted by T1 or proton density

[0187] When a second MRI acquisition is carried out using T1 weighting or proton weighting in order to segment the intima of the lung volume V P the configuration using T1 weighting can include, for example: echo time: TE = 10 ms to 20 ms, repetition time: TR = 400 ms to 600 ms. This second MRI acquisition can be a volume acquisition, that is to say, the images obtained are directly three-dimensional images. This configuration then substantially approaches the configuration CONF1 of the first embodiment. This final configuration can be achieved by defining an ultra-short echo time UTE (or alternatively, a zero TE sequence).

[0188] Then, using the images acquired by T2-weighted acquisition, reset the T1-weighted or proton-weighted acquisition of the images. The lung volume V extracted from the T1-weighted acquisition P The contour of can be reset on the images acquired by T2-weighted acquisition to segment the lung volume V of the images acquired by T2-weighted acquisition P .

[0189] The advantage of this resetting is that it enables the recovery of the contour of the lung volume, the contrast of which has been reduced due to the treatment of the circulating blood volume.

[0190] According to another example, a sequence with TR = 4.1 ms, TE = 0.05 ms, and flip angle = 5° can be defined.

[0191] Similarly, the images acquired by this second T1-weighted MRI acquisition can be processed by an image processing algorithm to reduce noise, eliminate artifacts, or adjust the contrast.

[0192] - Extraction of lung volume

[0193] The three-dimensional images obtained from the first T2-weighted acquisition and from the second T1-weighted acquisition can be superimposed in the form of voxels of the images inside the filtered lung volume V P Finally, this T1-weighted acquisition also enables the outlining of the volume of the heart and its subtraction from the volume of interest.

[0194] The purpose of the T1-weighted or proton density-weighted acquisition is to achieve a mask of the lung inner membrane. To achieve this 3D mask, the 3D UTE acquisition is interpolated to correspond to the spatial resolution of the T2-RTSE acquisition. According to the example, images have been recorded using an intensity-based optimized multimodal algorithm. An iterative algorithm for selecting the optimal intensity threshold can be implemented. This algorithm calculates the average values of the two classes that separate the background and foreground images. In addition, it calculates the optimal separation threshold iteratively and stops whenever this calculation does not change. This algorithm can divide the histogram into two images based on two voxel intensity classes in order to create a lung mask. Next, for example, the lung mask is applied to the recorded T2-RTSE images in order to isolate the voxels included in the lung volume V P Other techniques enable the acquisition of a mask of the lung volume. For example, the mask can be obtained from another imaging system or from images acquired previously and recorded in memory.

[0195] - First function F1

[0196] The method of the present invention includes a step of generating a first distribution function F1 of the voxel distribution of the images acquired by T2-weighted acquisition as a function of the signal intensity. To obtain the function F1 representing the mucus load inside the lung volume, for example, the lung volume V is outlined from the second T1-weighted acquisition PThe steps of the contour are as described above. In addition, as described above, based on the T2-weighted image acquisition mode, the step of filtering the voxels corresponding to the blood fluid circulating in the blood vessel is also performed.

[0197] Therefore, the first function F1 enables the distribution of chromaticity parameters (such as signal intensity) in the image representing the volume of the fluid (such as water) to be considered. Thus, the function F1 enables the volume of mucus substantially formed by 95% water or the volume of stagnant edema substantially formed by 90% water in the lung volume to be distinguished. The signal intensity is preferably a contrast parameter.

[0198] Once the function F1 is generated, the curve C1 representing the voxel distribution in the lung volume is obtained. For the absolute scale of the voxels, the distribution of the voxels is obtained, and its resolution depends on the image acquisition configuration.

[0199] - The second function F2

[0200] The function F2 that can define the automatic threshold enables the filtered volume V F and the characteristic volume V C to be extracted. Recall that the filtered volume is the volume extracted from the segmented image corresponding to the region of interest. The characteristic volume V C is the volume obtained from the filtered volume in a manner that generates biomarkers.

[0201] For this purpose, the reference volume V A is identified in the acquired image. A The reference volume V A is preferably selected at the level of the air region, that is, the volume corresponding to the volume of air. The reference volume V A corresponds to the region of the smallest size, so that the region corresponding to the air volume rather than the artifact of the image or a single region can be fully represented. In other words, a large enough region is selected to obtain the value representing the air contrast level. This reference volume V P can be automatically selected at a given position within the pulmonary intima V P or at a given position outside the pulmonary intima V A For example, the position at a predetermined distance from the boundary of the surface of the chest wall enables a high probability of the air region to be obtained. This distance can include, for example, between 1 cm and 4 cm, preferably between 1.5 cm and 3 cm. A distance of 2 cm enables compatibility with a large number of subjects. The reference volume V A enables the distribution of chromaticity parameters to be selected, that is, for example, the distribution of the signal intensity values of the reference volume V A , such as the distribution value of the contrast level. The function representing the distribution of the voxels of this reference volume VA 。

[0202] The method of the present invention comprises steps aimed at measuring the standard deviation σ A of a second function F A . The standard deviation σ A is calculated over a reference volume V A . The method of the present invention then comprises calculating a reference threshold S A , which reference threshold S A is calculated as a function of the calculated standard deviation σ A . According to an embodiment, the reference threshold S A is a multiple of the standard deviation σ A . According to an example, the value of the reference threshold S A is included between 8 times and 16 times the value of the standard deviation σ A , plus the value of the main distribution pattern of the voxels of the lung parenchyma. To this end, a base volume V B of voxels is automatically selected in the lung region. This base volume V B can be predefined, for example, at a minimum distance from the inner membrane of the lung, or determined around the region of interest. The region of interest can be defined as being close to or adjacent to respiratory bronchioles, alveolar ducts, and alveoli. According to an exemplary case, the entire volume of each lung is considered. According to another example, different base volumes V B are calculated at different positions and then an average value is generated. According to another example, the method is repeated for different base volumes V B . Then, the main pattern of the base volume V B is labeled M B .

[0203] The main pattern represents the value of the maximum distribution of chromaticity parameters or signal intensity values, which is labeled M B . This main pattern makes it possible to generate an adaptive threshold depending on the patient and his own physiological data. This constitutes a real advantage of the present invention in order to measure and quantify the mucus volume relative to the individual. In order to obtain a filtered volume for discriminating the mucus-loaded volume, the value of the reference threshold S A is the main pattern M B plus 10 times to 14 times the value of the standard deviation σ A . According to a preferred example, the value of the reference threshold S A is defined between 11 times and 13 times the value of the main pattern M B plus the value of the standard deviation σ A . In particular, the value of the reference threshold is S A = 12·σ A + M B .

[0204] Figure 9Indicates the threshold S calculated from the main mode A as an example, which main mode is obtained from the most representative distribution of the signal of the lung volume V P The signal intensity acquired with T2 is shown on the Y-axis CI, and the distance metric specific to the image represented in the section is shown on the X-axis. The distance "distance (VP)" is expressed in millimeters and corresponds to the distance from the inner membrane of the lung to the other end of the lung.

[0205] According to the first example, the distance function (VP) is defined by means of a straight-line segment defined by a first point located on the wall of the lung volume and perpendicular to the wall, and a second point defined at the intersection opposite the first point on the lung wall.

[0206] According to another example, the distance function (VP) is defined by a straight-line segment starting from a first point located on the wall of the lung volume and passing through the center of the lung volume. The center of the lung volume is a fictional center that can be defined according to a geometric function, such as a function for determining the centroid and applied to the inner membrane of the lung volume V P of the inner membrane.

[0207] According to another example, the distance function (VP) is defined by considering the segmentation between two points considered on the wall of the lung volume.

[0208] The main mode M B can also be calculated within a partial volume of the lung volume V P of the partial volume.

[0209] The signal S N corresponds to the signal intensity in the sectional image (2D). Therefore, it is measured from the distance of the inner membrane of the lung. The threshold S A is obtained by calculating the main mode M B which, for the purpose of representation, the main mode M B is represented as the distribution of the signal intensity of the section. To obtain the threshold S A , a multiple (usually between 12 times and 16 times) of the standard deviation of the signal measured on the reference air volume inhaled outside the lung volume has been added to the main mode.

[0210] - Segmentation

[0211] The method includes the step of segmenting the volume of the voxels greater than the threshold S A . The volume obtained is the filtered volume V F . The filtered volume makes it possible to preserve the high-intensity voxels and makes it possible to eliminate a part of the signal corresponding to the noise.

[0212] According to an embodiment, the function F A (i.e., the second function F2) defines the filter of the function F1. The filter is applied in order to extract by the function FA Filtered voxels (i.e., those voxels with a contrast level greater than the threshold S A ). In this case, the function F A is defined as a mask applied to the function F1.

[0213] Filtered volume V F is defined by the volume composed of the voxels of the distribution of the function F1 filtered by the reference threshold S A . Thus, the voxels of the filtered volume have a contrast intensity greater than the reference threshold S A . Of interest is obtaining the T2-weighted intensity values of the mucus regions present in the lung volume. The function F2 is used as a volume mask for filtering the acquired high-signal-intensity volume.

[0214] Figure 8 An example of a cross-section representing two lung lobes P A , P B is shown, where region 10 corresponding to the volume portion of the region with an intensity greater than the threshold S A is represented in 2D. Thus, Figure 8 an example of the volume filtered by the function F2 is shown.

[0215] The method includes the step of normalizing the voxel values of the filtered volume V F . According to a first alternative embodiment, the contrast values are normalized on a scale, for example, between 0 and 1.

[0216] According to another alternative embodiment (which can be combined or implemented as an alternative to the first alternative), the normalization includes calculating the volume intensity product, labeled VIP. The VIP is obtained by the product of the volume of the signal and the intensity T2. It enables quantification of the volume intensity of the hyper-signal and thus quantification of the level of inflammation. Additionally, it can quantify the evolution of the region through the correlation between the increase or decrease in the volume of the inflamed area and the increase or decrease in the signal intensity.

[0217] Thus, due to the threshold S A , the normalization includes the chromaticity parameter of the voxels or the absolute value of the signal intensity (here represented by the quantization of the signal T2, for example, in ms), and the normalization of the filtering threshold corresponding to the mask of the applied filtered volume.

[0218] According to an example, the VIP can be expressed in milliseconds and is obtained from the following equation:

[0219]

[0220] ■i corresponds to the value of T2 for which the distribution of the values of the T2 signal is achieved;

[0221] ■ The maximum value of the T2 signal corresponding to max;

[0222] ■ T2i represents the signal of the horizontal i of the three-dimensional image;

[0223] ■ HSV i corresponding to the volume associated with the T2 value level of the distribution, the volume being included in the mask of the three-dimensional image corresponding to the high-intensity filtering volume V F of the three-dimensional image;

[0224] ■ V P corresponding to the lung volume.

[0225] The advantage is to obtain a normalized value using an index that reflects the intensity per unit volume. Given the adaptive threshold obtained due to the threshold S A The method of the present invention makes it possible to generate exploitable biomarkers while limiting the variability in the exploitation and interpretation of images from one individual to another. In addition, the biomarker makes it possible to obtain a time reference point for measuring the evolution of mucus development over time.

[0226] According to an embodiment, a step of performing a merger including UTE images and RTSE images is executed. In this case, a merger of a high-resolution morphological image obtained using a sequence with an ultra-short echo time UTE and a mask of the volume of the segmented T2 hyper-signal is performed in order to combine two items of information. A person skilled in the art can apply this technique to PET-TDM.

Claims

1. A method for generating a biomarker, comprising: Acquire images using an MRI system; Process the MRI images to generate three-dimensional images of the lungs; Generate a first function corresponding to the distribution of different signal intensity values of each voxel of a portion of the acquired three-dimensional image; Automatically calculate at least one filtering threshold of the first function according to a second function of at least one different signal intensity value distribution of each voxel of a portion of the acquired three-dimensional image; Segment a volume including the following volumes: A main volume corresponding to the lung volume; A filtered volume of the volume of voxels quantified by the first function and filtered at least by means of the calculated filtering threshold; Normalize the values of the three-dimensional image of the lung volume according to the absolute value of the signal intensity value of the voxels of the image and at least the calculated filtering threshold; Generate a biomarker indicating the normalized segmentation volume ratio; The three-dimensional image acquired by the MRI system is configured by the following: T2 weighted; And An echo time TE greater than a predetermined threshold; The automatic calculation of at least one filtering threshold includes: Obtain a reference volume; Generate a reference function corresponding to the distribution of different signal intensity values of each voxel of the reference volume, where The reference function is a second distribution function; Calculate the standard deviation of the reference function; Determine a filtering threshold according to the calculated standard deviation of the reference function, where the filtering threshold is designated as a reference threshold.

2. The method for generating a biomarker according to claim 1, wherein, The acquisition is parameterized and includes: The acquisition of the image is recombined from multiple images acquired over one or more cycles of at least 4 TEs, where the echo time is configured according to increasing durations; A spin echo sequence; and Parameterization aimed at emitting signals to pre-saturate or saturate the acquired signals.

3. The method for generating a biomarker according to any one of claims 1 to 2, wherein, Perform a second acquisition of the image using a first configuration to outline the lung volume, the first configuration defining a parameterization of T1 or a proton density weighted acquisition, and perform the steps of image processing to combine the image acquired using T2 weighted acquisition with the image acquired by the second acquisition.

4. The method for generating a biomarker according to claim 3, wherein, Perform a merging operation between at least one image acquired with T1 weighted with an ultra-short echo time and at least one image acquired with T2 weighted to generate an image in which the data from each acquired image has been combined to segment the lung volume.

5. The method for generating a biomarker according to claim 1, wherein, The reference threshold is established according to a combination of the reference distribution value of the reference function and a value between 10 times and 20 times the value of the standard deviation of the reference function.

6. The method for generating a biomarker according to claim 5, wherein, The reference distribution value of the reference function is the main mode of the distribution of the signal intensity values of the image acquired with T2 weighted within the lung volume.

7. The method for generating a biomarker according to claim 1, wherein, The method includes a normalization step, and the normalization step includes calculating the volume intensity product of the signal according to the absolute value of the signal of the filtered volume, and the volume is generated by the filtered volume and the lung volume.

8. The method for generating a biomarker according to claim 1, wherein, The method includes acquiring the signal intensity value of each voxel quantified by the first function, where the intensity value corresponds to the image contrast data.

9. The method for generating a biomarker according to claim 1, wherein, The acquisition is performed in a manner synchronized with a respirator, and the respirator is a navigator or a respiratory belt.

10. The method for generating a biomarker according to claim 1, wherein, According to the image acquired by MRI, perform the step of extracting a volume image, and the extracted image is generated at a determined moment of the sequence.

11. The method for generating a biomarker according to claim 1, wherein, The acquisition of the three-dimensional image is performed by stacking the acquired 2D images, and the thickness of the cross-section is at least equal to the width of the voxel.

12. A method for generating a biomarker, comprising: Images are acquired using an MRI system; The MRI images are processed to generate a three-dimensional image of the lung; A first function corresponding to the distribution of different signal intensity values of each voxel of a part of the acquired three-dimensional image is generated; At least one filtering threshold of the first function is automatically calculated according to a second function of at least one different signal intensity value distribution of each voxel of a part of the acquired three-dimensional image; Segment a volume including the following volumes: A main volume corresponding to the lung volume; A filtered volume of the volume of voxels quantified by the first function and filtered at least by means of the calculated filtering threshold; Wherein: The three-dimensional image acquired using the MRI system is configured by the following; Proton density or T1 weighting with a long repetition time TR or a long TE; An echo time less than a predetermined threshold; The automatic calculation of at least one filtering threshold includes; Model at least two Gaussian functions by adjusting the first function; Determine a filtering threshold called the first threshold by calculating the intersection of the first Gaussian function and the second Gaussian function; Determine a second threshold corresponding to the minimum value of the first Gaussian function and the minimum value of the signal intensity value of the voxel; make the filtered volume be a first volume corresponding to the voxels quantified by the first Gaussian function and included between the first threshold and the second threshold, and the voxels correspond to the air medium; Normalize the value of the three-dimensional image of the lung volume according to the calculated first threshold and second threshold by the absolute value of the signal intensity value of the voxels of the image; Generate a first biomarker indicating the ratio of the characteristic volume to the normalized segmented volume, and the ratio is calculated between the characteristic volume and the lung volume.

13. The method for generating a biomarker according to claim 12, wherein, The echo time is less than 1 ms.

14. The method for generating a biomarker according to claim 12, wherein, The segmentation includes defining a second volume corresponding to the voxels quantified by the second Gaussian function and greater than the first threshold, and the voxels correspond to fat or intermediate media.

15. The method for generating a biomarker according to claim 12, wherein, The segmentation includes a step of extracting a characteristic volume, and the characteristic volume includes the voxels of the filtered volume, and the signal intensity value of the filtered volume is less than a third predetermined threshold, and the third predetermined threshold is determined on a normalized scale [0; 1].

16. The method for generating a biomarker according to claim 14, wherein, The segmentation includes a step of excluding / deleting disconnected adjacent identically quantified voxels.

17. The method for generating a biomarker according to claim 12, wherein, The modeling of the Gaussian function includes: Apply Gaussian smoothing to the acquired images to denoise by reducing the single-shot encoding time and radial acquisition; Outline the contour by applying a local filter; Use a curve adjustment method to represent the distribution frequency of voxels.

18. The method for generating a biomarker according to claim 12, wherein, The method includes acquiring the signal intensity value of each voxel quantified by the first function, wherein the intensity value corresponds to the image contrast data.

19. The method for generating a biomarker according to claim 12, wherein, The acquisition is performed in a manner synchronized with a respirator, and the respirator is a navigator or a respiratory belt.

20. The method for generating a biomarker according to claim 12, wherein, According to the images acquired by MRI, perform a step of extracting a volume image, and the extracted image is generated at a determined moment of the sequence.

21. The method for generating a biomarker according to claim 12, wherein, The acquisition of the three-dimensional image is performed by stacking the acquired 2D images, and the thickness of the cross-section is at least equal to the width of the voxel.

22. The method for generating a biomarker according to claim 12, wherein, The acquisition is a 4D acquisition configured to acquire the variation of the first biomarker per unit of time over a predetermined duration.

23. The method for generating a biomarker according to claim 22, wherein, The 4D acquisition is configured to obtain voxel volumes at time units corresponding to the duration of a respiratory cycle, and the acquisition is synchronized with data characteristics of inhalation and / or exhalation.

24. A system comprising at least a calculator, a memory, and an interface for receiving images acquired by an MRI system, the system being configured to implement the steps of the method according to any one of claims 1 to 11.

25. A system comprising at least a calculator, a memory, and an interface for receiving images acquired by an MRI system, the system being configured to implement the steps of the method according to any one of claims 12 to 23.

Images