EP3025307A2

EP3025307A2 - Method for producing scanography images with table deletion

Info

Publication number: EP3025307A2
Application number: EP14790157.3A
Authority: EP
Inventors: Grégory MALLET; Gad Miller
Original assignee: Global Imaging On Line
Current assignee: Global Imaging On Line
Priority date: 2013-07-25
Filing date: 2014-07-23
Publication date: 2016-06-01
Also published as: WO2015011406A3; FR3009111B1; WO2015011406A2; FR3009111A1

Abstract

The invention relates to a method for producing a digital scanography image of a body in contact with a table, which includes: digitising a real scanography volume VR and determining a first segmentation (14) of the digitised volume (10) VN by selecting only the voxels that correspond to a local density interval INT covering the densities of flesh and/or a muscle and excluding the density of the rigid portions of the table; identifying the connected components of VN (10) in 3D and calculating the relative volumes Vi / VM of said connected components, relative to the connected component having the largest volume VM, and then deleting the connected components having a relative volume of less than a predetermined threshold; creating a 2D convex shell of each elementary 2D segment of VN, and producing a digital image from said 2D convex shells. The invention also relates to a related scanography method and computer program.

Description

Method of producing scan images with elimination of the table

The present invention relates to the technical field of obtaining medical images processed by computer, and more particularly images produced by a scanner.

Computed tomography (CT), computer-calculated axial tomography (CTAT), CT-scan (CT), CAT-scan (CAT), is a technique known as computed tomography (CT). medical imaging using a scanner, which consists of measuring the absorption of X-rays in different areas of a body, then, by computer processing, to digitize and finally reconstruct 2D (two-dimensional) or 3D (three-dimensional) images. dimensions) anatomical structures.

Also known are complementary techniques of medical imaging aiming, from images of a scanner or data corresponding to these images, to perform an operation called "image processing" to eliminate from the image the representation of the image. table on which is lying, or leaning against a patient.

The table can indeed have a density close to that of the bone, and therefore appear on all sections of the body, which can hinder the medical operator (typically the radiologist).

A high-performance method of processing and producing scanner images by means of a computer, for automatic removal of the table during the viewing of the images of the examination, is therefore sought.

Another advantage of an image processing method is to eliminate if possible a number of virtual objects or artifacts that inappropriately add to the image of the part of the body subjected to the scanography. It may be parasitic digital images, similar to a "noise" disturbing the measurements associated with real parts of the body. It may also be representations of inhomogeneous zones, for example zones of interface with a zone of the body (interfaces with the ambient air, or the table, contact zones with clothes, soft parts of the table). etc.).

It is also very desirable to be able to perform such image processing, also sometimes called "automatic removal of the table" in a very short time, for example a few seconds at most, so as to minimize the time of use image processing equipment and the time spent by the medical operator. It is also important not to simultaneously remove portions of the image corresponding to parts of the patient's body.

The methods of the state of the art implement, for example, one or more complex curves for separating the body from the table in a scanner image. As the body rests on the table, the determination of such a curve is complex and therefore requires significant computing power for the computer used for this image processing, and a relatively long calculation time. In addition, the reliability of these techniques, from the point of view of the conservation in the processed image of all parts of the patient's body is difficult to ensure.

The invention aims in particular to improve the aforementioned problems, particularly related to a selective elimination of the table in CT images, without requiring a high calculation time as in conventional methods.

To this end, the subject of the invention is a method of computer-assisted production of a digital scan image of at least a part of a body in contact with a table, in which:

- that at least part of the body being disposed within a real volume V _R, is associated with this actual volume V _R V _N a digital volume consisting of voxels, each voxel V _N corresponding to an elementary real volume V _R ,

each voxel is associated with a numerical value correlated with the X-ray absorption generated by a scanning apparatus in the elementary real volume corresponding to this voxel,

characterized in that the computer performs at least the following steps for the elimination in the digital image of most of the representation of the rigid part of the table:

a first segmentation of VN is determined by selecting the only voxels whose associated numerical value is situated in a predetermined bounded interval INT of values such that:

At least 90%, and preferably 100% of the voxels corresponding to a real grease and / or muscle-containing elementary volume, for a predetermined population sample, have an associated numerical value comprised in the INT interval,

At least 50%, and preferably at least 60% or very preferably at least 90%, voxels corresponding to an elementary real volume containing a rigid part of the table have an associated numerical value which is not included in the INT interval,

determining the related component (s) of the first segmentation, possibly modified, and calculating for each related component of index i, optionally modified, a quantity V, called volume of this connected component, related to the number of voxels of this connected component, the connected component of larger volume V _{M is determined} , possibly after an initial selection of the only related components, possibly modified, having a volume at least equal to a predetermined value,

the set of voxels is gathered and selects the only related components, possibly modified, whose ratio V, / V _M is greater than or equal to a predetermined threshold, to form a second segmentation,

a third segmentation of V _{N is determined} , any part of which in an axial digital elementary slice is a convex set comprising the portion of the second possibly modified segmentation which is included in this same axial digital slice.

and in that at least one digital image is produced from the third segmentation.

The method according to the invention therefore makes it possible to eliminate at least the largest part, and preferably substantially all of the image of the table, using only digital data processing steps requiring only a power relatively low calculation. Unlike the state of the art, where the first segmentation is often done by thresholding which eliminates the voxels whose associated numerical value is either higher or lower than the threshold, the use of a range of values makes it possible to accelerate and simplify the rest of the process. In fact, at the table correspond important values in Hounsfield due to the density of the matter, whereas in the air and the vacuum correspond low values. The segmentation by an interval allows in a quick and simple calculation step, to eliminate the voxels whose associated values are high, corresponding to the table, and to eliminate at the same time the voxels whose associated values are low, corresponding to the empty and airy. This greatly simplifies the following steps. Thus, it is not necessary to implement complex steps of identification of the related components to know what are the remaining components corresponding to the patient and which are the components corresponding to the table, the latter having been at least largely part eliminated. The following steps are therefore simpler and faster than in the state of the art.

The step of selecting the voxels of the only related components whose ratio V, / V _M is greater than or equal to a predetermined threshold makes it possible in particular to easily eliminate a potentially large number of "false objects", or artifacts:

This predetermined threshold is determined from the usual anatomical proportions of the human being. Indeed, V _M corresponds to a part of the body of largest volume, typically the chest or abdomen. From the anatomical knowledge, it is possible to identify, with respect to V _M the range of relative volumes V, / V _M possible corresponding to relatively low volume parts of the body, for example an arm. This makes it possible to identify possible ratios V, / V _M , and to be able to eliminate (not to select) the objects external to the body, real or virtual, which do not correspond to this interval.

The use of relative volumes is particularly interesting, compared to the use of absolute volumes, for the elimination of artifacts. Indeed, the use of absolute volumes should both make it possible to distinguish, with respect to objects external to the body, real or virtual, any part of the body, both for an adult person, possibly large and of a large body size, only for a child, or a baby. Therefore, the use of absolute volumes can only be used to eliminate virtual or real objects of very small size. On the contrary, the use of relative volumes makes it possible to overcome, at least to a certain extent, the size and body size parameters, because the ratio of the volume of an arm or leg to that of the abdomen or the thorax does not vary significantly according to the above parameters.

The patient is in contact with the table (typically leaning against the table), which determines a main longitudinal axis (the axis of the body, from the feet to the head). The term "axial numerical elementary slice" is conventionally called a two-dimensional elementary subset (2D) of the three-dimensional numerical volume (3D) V _N , corresponding in real volume to an axial elementary slice of V _R , ie that is to say, a slice perpendicular to this longitudinal axis of the body, and whose thickness corresponds to that of an elementary real volume (corresponding in V _N to a voxel). Such an "axial digital elementary slice" corresponds directly, in terms of scanner image, to an axial section of the body at a determined axial level.

An axial digital elementary slice is a two-dimensional numerical space (2D space), formed of voxels. It is therefore a discretized space, as is V _N. Each voxel of an axial digital elementary slice can be defined by its numerical coordinates, corresponding for example to the position of the center of the real elementary volume corresponding to the voxel.

In addition, any axial numerical elementary slice is associated with an axial numerical plane, which is a two-dimensional continuous digital space (non-discretized space, each dimension being defined by a real number).

Each voxel of the associated axial digital elementary slice corresponds, in this continuous space (axial numerical plane), to a point defined by its numerical coordinates, corresponding for example also to the position of the center of the real elemental volume corresponding to the voxel). However, an axial digital plane, because it is a continuous space, makes digital processing operations much more extensive and precise than in a discretized space.

Since such an axial digital plane is the numerical translation of a geometrical plane, planar geometry terms will be used hereinafter to qualify subsets of this axial numerical plane, in particular the terms: "convex set or envelope", "straight lines". These rights may be "parallel" or "perpendicular to each other", "meet", delimit a "plane digital space band", be at a "distance from a point", and so on.

These terms are directly understandable for a person skilled in the art of digital image processing, and correspond for each to the conventional geometric notion in a geometric plane corresponding to the digital plane.

By extension, it will also be possible to use hereinafter geometric terms to qualify an axial digital elementary slice, for example to define a convex set of voxels of this slice, or a convex envelope. In this case also, we will directly understand the term used by considering the associated numerical plane, and finally an associated geometric plane.

The third segmentation, which includes in a convex set the second possibly modified segmentation, aims to be able to produce an unsegmented final image (for example the thorax and the two separate arms). It is typically determined so as not to reintegrate in this segmentation (or only very minor) parts of VN corresponding to rigid parts of the table. The various steps of the method are thus typically adjusted so that the final image produced does not comprise at least 50%, or preferably 60%, or even 80% of the image corresponding to the rigid parts of the table. In particular, there will be a third segmentation encompassing the previous segmentation in a convex set close to the minimum convex set.

Typically, and preferably, any part of the third segmentation included in an axial digital elementary slice is substantially a convex envelope of the portion of the second possibly modified segmentation which is included in this same elementary digital elementary slice. This makes it possible to produce an unfragmented image of substantially minimum dimensions while not excluding any part of the body.

Generally the predetermined threshold, for the ratio Vi A / M is between 0.03 and 0.18 and preferably between 0.05 and 0.12.

The steps described above are not limiting of all the steps of the method according to the invention. In particular, it is possible to add or intercalate between any two successive steps mentioned above one or more additional additional steps, even of minor importance, to further improve the performance of the process. The process steps that are described typically apply to the segmentation obtained in the previously described "possibly modified" step, to indicate that one or more intermediate steps are not excluded.

It is thus advantageously possible to carry out at least one so-called "erosion" step of a first set of voxels, for example of the first segmentation, in which all the voxels of this first set, which are contiguous with one another, are eliminated. voxel not belonging to this first set of voxels. It is thus possible to separate real or virtual objects of small size which appear interconnected by a small number of real or virtual elementary volumes, which is not the case of the parts of a body. These objects of low relative volume can then be eliminated (not selected) easily in a subsequent step. Typically, several successive stages of erosion will be carried out, in general of the first segmentation.

This erosion step is preferably performed after the step of determining the first segmentation, and before the step of determining the related components.

The erosion stage may correspond, in the actual volume, to erosion in 3D (in three dimensions) with elimination of "contiguous 3D voxels" of a voxel not belonging to this first set.

However, it is possible, optionally, to perform an erosion step corresponding, in the actual volume, to erosion in 2D (in two dimensions). During such an erosion step, only the voxels that are "contiguous in 2D" are eliminated from a voxel not belonging to this first set, the term "contiguous in 2D" applying only to the voxels arranged in the same elementary digital elementary slice as this voxel not belonging to this first set.

This then corresponds substantially, in the real volume VR, to an "erosion" performed exclusively in each axial elementary slice. When the numerical cutting of the body is finer in such an axial elementary slice than along the axis of the body, this allows for a finer and more precise erosion in each axial elementary slice of VR. It is also possible to carry out at least one stage, and preferably at least two so-called "expansion" stage (s). a second set of voxels, for example the second segmentation.

In such a dilation step, at least all the voxels of VN that do not belong to this second set but that are contiguous with at least one voxel belonging to this second set are added to this second set.

As for an erosion stage, the contiguous term can be applied in 3D, or, optionally, in 2D. In the latter case, we add only voxels "contiguous in 2D", that is to say the only voxels arranged in the same elementary digital elementary slice that voxel belonging to this second set. This expansion step is preferably performed after the step of determining the related components.

An expansion step is similar to an inverse step of an erosion step. However, it is not systematically reversed, the combination of an erosion step and an expansion step (or vice versa) not necessarily leading to the starting set.

It will be understood that in all cases (2D or 3D erosion, expansion in 2D or 3D), we always start from a three-dimensional set to arrive at another set (subset, or set including starting set) also in three dimensions, the term 2D or 3D applying to the definition of the contiguous term used for the determination of the voxels eliminated or added during the step.

In addition, the term "contiguous" for two voxels can be applied differently depending on the metric used in VN. If we transpose this notion of contiguity to the two elementary volumes of the real volume corresponding to these two voxels, two definitions can be used:

In the first definition, each elementary volume being defined by an elementary parallelepiped, the two elementary volumes (and the two associated voxels) will be considered as contiguous if they have a common edge.

In the second definition, they will be considered contiguous if they have a common edge or a common vertex.

Each of these two definitions can be used for the implementation of the method according to the invention.

Preferably, a greater number of voxels are added by expansion than the number of voxels removed by erosion, for example by using a number of expansion steps greater than the number of erosion stages. Thus, it reduces the risk that areas of the body are not selected, including areas close to the outer skin of the body. It is also possible to perform one or more erosions in 2D, and subsequently an identical or greater number of 3D dilatations, for example a 2D erosion step, then two 3D expansion steps. Advantageously, after, and preferably immediately after calculating the volumes of the related components, a step of eliminating the related components whose volume is less than a predetermined percentage of the volume corresponding to said digital volume VN, for example less than 0.1% of said volume of VN. This makes it possible to directly eliminate related components that are too small to be able to correspond to a part of the body, regardless of the size and body build.

Preferably, to determine the third segmentation, each axial digital elementary slice is associated with a two-dimensional continuous space, called an axial digital plane, comprising a subset, discretized of points each corresponding to a voxel, said trace axial plane of second segmentation possibly modified, associated with the part of the second possibly modified segmentation which is included in the axial numerical elementary slice associated with the axial digital plane,

in each axial numerical plane, a plurality of pairs of digital lines is determined, the digital lines of one and the same pair being parallel to each other and thus delimiting with them in the broad sense an axial digital plane space band, these digital lines being of different directions from a pair of straight lines to another pair of straight lines, each pair of digital straight lines flanking the axial plane trace of a second segmentation possibly modified, so that each of the digital lines of this pair meets said plane trace axial axis of second segmentation possibly modified and that this axial plane trace of second segmentation possibly modified is entirely comprised in the axial digital plane space band delimited by this pair of straight lines,

in each axial numerical plane, a subset of this plane, said convex envelope of an axial trace of second segmentation, possibly modified, is defined as the intersection of the different space bands corresponding to the different pairs of digital lines.

in each axial numerical elementary slice, the subset corresponding to this convex envelop of an axial trace of a second optionally modified segmentation, said convex envelope of an axial slice of second segmentation, possibly modified, is defined,

and defining the third segmentation as the union of all the convex envelopes of axial slice of second possibly modified segmentation corresponding to different axial numerical planes.

To determine a pair of digital lines with a given direction in at least one axial numerical plane, it is advantageous to define in this axial numerical plane a reference axis, called the abscissa axis, parallel to this given direction, and an ordinate axis, perpendicular to this abscissa axis, then for each point of the contour of the axial plane trace of second segmentation possibly modified, the ordinate of this point is calculated,

then we identify two points of the contour of the axial plane trace of second segmentation possibly modified, said end points (vis-à-vis the given direction), whose ordinates are respectively minimum and maximum, then we determine a couple of two straight lines parallel to the given direction passing respectively by one and the other of these two end points.

The contour of the axial plane trace of second possibly modified segmentation is defined by the set of points of the axial plane considered that belong to this plane plane trace of second possibly modified segmentation, and are contiguous of at least one point of the numerical plane axial which does not belong to this plane axial trace of second segmentation possibly modified.

The pairs of lines are typically chosen with variable orientations. Advantageously, any angular sector of space covering an angular interval of 20 °, between 0 ° and 180 °, comprises a direction that is that of at least one pair of straight lines.

Preferably, a number of pairs of lines greater than 5, for example between 10 and 50 pairs of straight lines, of variable orientations, for example 36 pairs of straight lines of increasing orientation of 5 ° of a couple, will be used. from rights to another.

Each so-called "convex envelope of axial trace of second segmentation possibly modified" is thus very close to the corresponding minimal convex envelope in the mathematical sense, which would be obtained with an infinite number of lines. It typically corresponds to a number of voxels typically not exceeding more than 15%, and preferably at most 10%, the minimum convex envelope in the mathematical sense. Its surface typically only exceeds at most 15%, and preferably at most 10%, the area of the minimum convex envelope in the mathematical sense.

The above-mentioned methods of determining straight-line pairs and convex axial trace-segment envelopes are relatively simple and require relatively little computing power or time.

Thus, the method can be realized by allowing a display in almost real time of the image, for example after a computation time of less than 10 seconds per giga-voxels (billion voxels) of the digital space VN, typically less than 5 seconds per giga-voxels of the digital space VN.

Preferably, the predetermined bounded interval INT is chosen so that:

At least 98% of the voxels corresponding to an elementary real volume filled with fat or muscle, for the predetermined population sample, have an associated numerical value comprised in the interval INT,

At least 90% of the voxels corresponding to an elementary real volume filled by a rigid part of the table have an associated numerical value that is not included in the INT interval,

- any voxel corresponding to an elementary real volume filled with air has an associated numerical value which is not included in the INT interval.

This makes it possible to preserve in the final image almost all the areas of flesh and / or muscles of the part of the body subjected to the CT scan, and to eliminate both the peripheral areas of the volume VR that correspond to the air and a very large part of the table.

Typically, a digital image is produced corresponding to the representation of an elementary subset of the third segmentation corresponding to an axial digital elementary slice, that is to say an image of an axial section of the body or of a part from the body.

It is also possible to produce a 3D image (corresponding to a three-dimensional view) of at least a part of the body.

Typically, data is stored in at least one memory, generally a memory of the computer, corresponding to the third segmentation. This data can be used for a display, almost real-time or delayed, of one or more images on one or more screens. They can also be sent for fixation on one or more physical film media (to make x-ray scans), at the request of a medical operator. They can finally be sent via electronic and / or computer means to an external device, for example the computer of a treating physician.

Advantageously, the lower bound of the interval INT is in an interval of values [-300, -100] expressed in Hounsfield, the upper bound being in a range [100, 300].

The subject of the invention is also a computer program characterized in that it comprises program code instructions for executing the steps of the aforementioned method which are carried out by means of a program. computer.

The subject of the invention is also a method of scanning at least a portion of a body in contact with a table, in which a set of numerical values correlated with the absorption of X-rays generated by a scanning apparatus is generated. in elementary real volumes belonging to a real volume VR, characterized in that a digitized image is produced by the aforementioned method, this image being displayed on a screen, preferably automatically, and / or fixed on a physical film medium , at the request of a medical operator.

The subject of the invention is also a film-based physical scan image, characterized in that it has been carried out by the above-mentioned image production method or by the aforementioned scanning method.

The invention also relates to a scanner assembly comprising a scanner and a computer, characterized in that the computer comprises a physical medium on which is recorded a computer program as mentioned above.

The invention also relates to a physical medium of information characterized in that is recorded on this medium a computer program as mentioned above.

The invention also relates to a scanner assembly as mentioned above, wherein the computer is configured to perform the process steps that are performed by computer with a computation time of less than 10 seconds per gigavoxel of the digital volume. VN, and preferably at 5 seconds per gigavoxel of the digital volume VN.

Such a set of scanning can be installed in a single place, for example the scanography service of a hospital. The computer can be separate, or integrated with the scanner. Alternatively, the computer can be remote from the scanner (for example conventionally more than 100 m from this device), this scanner assembly comprising a communication interface between the scanner and the computer.

Finally, a subject of the invention is a method for making digital data available on a telecommunication network for downloading, characterized in that these digital data relate to the third segmentation defined during the execution of the method.

The invention will be better understood on reading the appended figures, which are provided by way of examples and are not limiting in nature. which:

FIG. 1 schematically represents a patient lying on a CT scan table of a CT device according to the invention.

Figure 2 schematically shows a transformed 2D image of a portion of the patient's body, seen in axial section before the removal of the table.

FIG. 3 diagrammatically represents the 2D image of FIG. 2 after a step of the method according to the invention.

FIG. 4 diagrammatically represents the 2D image of FIG. 3 after another step of the method according to the invention.

FIG. 5 diagrammatically represents the 2D image of FIG. 4 after a subsequent step of the method according to the invention.

Figure 6 schematically illustrates an illustration of a method for obtaining the third segmentation, which is a convex envelope in VN of the second segmentation.

FIG. 7 schematically represents a transformed image of the image obtained with the method according to the invention.

Referring now to FIG. 1, which shows the body 2 of an elongate patient, in contact with (against) a CT scan table 4 of a CT device according to the invention, this device also comprising a CT scan 6 and a computer 8, shown in a non-structural and extremely schematic manner. The different elementary volumes of the body can be defined by their coordinates, along the transverse horizontal axis X, the vertical axis Y (oriented from the back to the patient's belly), and the main (longitudinal) axis of the body Z (oriented feet to the patient's head.

FIG. 1 also shows the position of a plane P in an axial section of the body, that is to say in a plane perpendicular to the Z axis of the body, in which it will be possible to follow, in the figures following, the evolution of an image in axial section of the body, transformed according to the different steps of the method according to the invention.

The table is often arranged horizontally, as shown in Figure 1, and any axial section is in this case a section in a vertical plane. This is however not an obligation, and the orientation of the body (and the table) may be different, for example the body may be substantially vertical, the patient being standing and for example leaning against the table. In such a case, an axial section is a section in a horizontal plane.

An axial section can cover all parts of the body at this level of the Z axis of the body. However, you can also choose to make images only for only part of the body (area of interest), for example by excluding from the cut an area corresponding to one leg, if this leg is located outside the area of interest.

In what follows, we will sometimes assimilate, for the sake of simplicity, digital volumes formed of voxels to real volumes formed of elementary real volumes, and we can for example associate a density with voxels, or even the corresponding pixels. a digital image in an axial section of the body.

Referring now to FIG. 2, which represents a transformed 2D scanography image corresponding to an image of an axial section of the body, associated with an axial digital elementary slice, and an axial numerical plane, in which the image has been transformed. image, initially pixelated conventionally according to very many gray levels, by grouping together in different areas of this image: each homogeneous zone, hatched or not hatched, corresponds to a part of the image for which all the voxels (corresponding to a pixel of the image) have a numerical value within a predetermined interval INT of values. The numerical values used by way of example (in a nonlimiting manner) in this embodiment of the invention are values in Housfield unit (UH) correlated with the absorption of X-rays, and indirectly also with the density of the matter present in the actual elemental volume associated with the voxel considered.

Thus, we must distinguish:

An axial digital elementary slice (discretized space), of identical thickness to that of a voxel along the main axis, a value in UH being associated with each voxel of this slice,

an associated axial digital plane (non-discretized continuous space, comprising a discretized set of points corresponding to the different voxels of the axial digital elementary slice),

An image in axial section (pixellated image of a part of the body, corresponding to the axial digital elementary slice, in which each pixel has a gray level correlated with the UH value of the associated voxel),

A transformed axial section image, such as that of FIG. 2 (pixellated image but in which the pixels have been grouped into homogeneous zones, each corresponding to voxels whose UH value of the associated voxel is within a given interval.

Thus, in the space 10 of this transformed image corresponding to an axial section of the body:

a first homogeneous unhatched zone 12 corresponding to voxels associated with actual elementary volumes whose material has a density less than -200 HU, including air (density around - 1000 HU).

a second homogeneous zone 14, hatched, corresponding to voxels associated with actual elementary volumes whose material has a density belonging to the interval [-200 UH, +200 UH [, especially covering water (density 0 UH)) , fat and muscle,

a third homogeneous area 16, hatched, corresponding to voxels associated with actual elementary volumes whose material has a density belonging to the interval [-200 UH, +600 UH [, and:

- A fourth homogeneous area 18, hatched, corresponding to voxels associated with actual elementary volumes whose material has a density greater than or equal to 600 UH, including bone (about + 1000 UH). The second homogeneous zone 14, hatched, therefore corresponds in particular to voxels associated with actual elementary volumes whose material comprises fat or muscle.

The third homogeneous zone 16 corresponds to real elementary volumes of intermediate density between those of fat and muscle on the one hand, and those of the bone and the rigid part of the table on the other hand.

This may correspond, for example, to body parts close to the skin of the patient, possibly increased density, or to areas of contact of the body with clothes or less rigid parts of the table.

This zone 16 is represented in FIG. 2 artificially more important than on a real image, to make its representation more visible. . The examination table 4 can thus have, typically in a rigid part (which excludes soft parts such as foam), a density comparable to that of the bone, thus corresponding to the fourth homogeneous zone 18.

Figure 2 and the following figures are diagrammatic representations in which, to improve the understanding of these figures, the table 4 is shown relatively far from the body. In reality, it is different, and the areas corresponding to the table and the body are much closer together, and in contact at different points of the image.

FIG. 2 shows a first large area 20 corresponding to a view of the body itself (thorax or abdomen) with the exception of the arms, two zones 22 corresponding to the arms of the patient, and finally an area corresponding to the table 4.

However, because of the presence of clothing, contact areas, non-rigid parts of the table, such as foam, and / or digital artefacts, it appears also around and / or at the table, areas of relatively small and variable size corresponding to densities in the second homogeneous zone 14 or in the third homogeneous zone 16, and which do not correspond to parts of the body.

It is therefore desirable to substantially eliminate from the image both the representation of the table 4 and that of other real or virtual objects that do not correspond to a part of the patient's body. Conversely, it is necessary to preserve the image of the bones, elements of high density, possibly similar to that of the table.

Referring now to FIG. 3, which represents the 2D image of FIG. 2, modified after the step of the method according to the invention corresponding to the selection of only voxels whose associated numerical value is situated in the predetermined interval INT of values corresponding to the second homogeneous zone 14, ie corresponding to voxels associated with real elementary volumes whose material has a density belonging to the interval [-200 UH, +200 UH [. Thus, both high density elements such as the table and low density elements such as air and vacuum have been eliminated. Parts of bone may also have been removed, but since these deleted portions are restricted and located within conserved portions, they will be reintegrated later in the convex hull step.

At this stage, the first segmentation of VN, represented in a manner transformed in FIG. 3, has been obtained because the voxels (and associated pixels) are grouped together and represented in homogeneous zones associated with real elementary volume densities.

At least 90%, and typically substantially 100% of voxels corresponding to a fat and / or muscle-containing elemental real volume, for a predetermined population sample, have an associated numerical value within the INT interval, and are conserved in the image of Figure 3.

At least 50%, and preferably at least 90%, voxels corresponding to an elementary real volume containing a rigid portion of the table have an associated numerical value that is not in the INT range. Thus, at least the major part of the representation of the table has been eliminated from FIG.

On the other hand, the images of a certain number of small objects, real or virtual, which are not part of the body have not been eliminated, but nevertheless have associated densities, corresponding to the second homogeneous zone 14.

These parasitic images appear at the bottom of the overall image.

Referring now to FIG. 4, which represents the 2D image of FIG. modified after additional steps of the method according to the invention corresponding to:

1) determination of the related components remaining in the image at this stage of the process. This calculation is typically done in 3D.

A connected component groups together all the voxels that can be linked together (for any pair of voxels belonging to that connected component) by a path of contiguous voxels that also belong to that related component. Two voxels are considered contiguous, in this embodiment of the invention, if the actual elementary volumes corresponding to these two voxels, typically two parallelepipeds, have at least one vertex or a common edge.

Then we calculate the volume of each connected component. The volume of a related component may be the number of voxels of that related component, or a correlated value (for example proportional) to that number of voxels.

2) the elimination of related components whose volume is less than a predetermined percentage of the volume corresponding to the digital volume VN, for example less than 0.1% of said volume of VN.

This last step, optional, eliminates images of very small objects that can not match parts of the body, even when the patient is a toddler or a baby.

Reference is now made to FIG. 5, which represents the 2D image of FIG. 4 modified after additional steps of the method according to the invention corresponding to:

determination of the connected component of larger volume VM.

the calculation, for each connected component of volume Vi, the ratio Vi / VM.

the selection of related components whose Vi / VM ratio is greater than or equal to a predetermined threshold.

This eliminates the image of parasitic objects that may have a medium volume, taking advantage of anatomical knowledge.

In fact, the ratio of the volumes between the main part of the body (thorax or abdomen) and an arm does not vary considerably according to the patient, who can be a large adult, a child or even a baby, this patient can be of low or large build. Therefore, in evaluating the relative volume of a particular connected component relative to that of the larger volume connected component, it is possible to determine in many cases that a particular connected component does not have sufficient volume to match to one arm, and that it is therefore necessary to eliminate his image. Conversely, with this method, we do not eliminate related components sought substantial relative volumes, even though their absolute volume is very low (this may for example correspond to the case of an arm of a baby).

After this step, the voxels of the selected connected components are combined to obtain the second segmentation, corresponding to all the voxels of the only related components that may each correspond to a part of the body (head, arm, thorax, abdomen, leg, foot), ie the thorax (or abdomen) and both arms, in the case of Figure 5.

Reference is now made to FIGS. 6 and 7, which schematically show an illustration of a method for obtaining the third segmentation.

To define this third segmentation, we must place ourselves in each elementary digital elementary slice, and define there the component of this third segmentation in this slice. For this, we realize several steps:

a) consider for each slice, the associated axial digital plane. This axial digital plane comprises an axial plane trace of the second segmentation

(optionally modified) formed by a set of points each corresponding to the center of a voxel of this second segmentation.

b) A set of pairs of parallel numerical straight lines is then determined, each of these pairs of lines flanking the axial plane trace of the second segmentation (possibly modified), each pair of lines delimiting an axial digital plane space band.

c) The intersection of the different bands of axial digital plane space is determined, which substantially forms a convex envelope of the axial plane trace of the second segmentation (possibly modified). Thus, a convex envelope of points corresponding to the second (possibly modified) segmentation is substantially produced, but exclusively in a 2D space (each axial digital plane), and not a convex envelope in 3D. FIGS. 6 and 7 illustrate the operations carried out in an axial digital plane, starting from a transformed axial sectional image (pixellated image of the digital elementary slice corresponding to the axial digital plane, but in which the pixels have been grouped into homogeneous zones , as previously stated).

A large number of pairs of lines are used, the lines of the same pair of lines being parallel to each other, to frame the image in a large number of directions and to eliminate the parts of the image which are not all located at the same time. inside the different bands of space delimited by the pairs of straight lines.

FIG. 6 shows only two pairs of straight lines, D1 and D'1 of a part, and D2 and D'2 on the other hand, delimiting respectively the space bands E1 and E2, which correspond to conserved image areas, for these two pairs of lines.

In practice, a significantly larger number of pairs of straight lines, for example several tens, with different orientations covering a very wide range of directions, will preferably be used.

Referring now to Figure 7, which schematically shows substantially a convex envelope of the hatched homogeneous area 14 of Figure 6. In addition to the hatched area 14, the convex envelope comprises another hatched area 23 which completes the area 14 to form the convex hull.

Thanks to the use of a large number of pairs of straight lines, for example orientations varying from 5 ° between two pairs of straight lines of neighboring orientation, the whole formed by the two arms and the thorax (or abdomen). Preferably, straight lines approaching outer tangents 24 to an arm and to the thorax (or to the abdomen) are accurately applied.

To determine a pair of parallel straight lines surrounding the axial trace of a second segmentation that may be modified, it is advantageous to define in this axial numerical plane a reference axis, called the abscissa axis, parallel to this given direction, and an ordinate axis, perpendicular to this abscissa axis, then for each point of the contour of the axial plane trace of second segmentation possibly modified, the ordinate of this point is calculated, then two points of the contour of the axial plane trace of second segmentation are identified. modified, said end points (vis-à-vis the given direction), whose ordinates are respectively minimum and maximum, then a pair of two parallel digital straight lines is determined to the given direction passing respectively by the one and the other of these two endpoints.

After performing the steps a), b) and c) above, the following steps are carried out:

d) One returns in the axial numerical elementary slice, to determine the corresponding convex envelope, called convex envelope of axial slice of second segmentation (possibly modified).

e) The third segmentation is defined as the union of all the convex envelopes of axial slice of second possibly modified segmentation corresponding to different axial numerical planes.

It would not be departing from the scope of the invention if the transition from a zone of an axial digital plane to a portion of the axial digital elementary slice This was done not in step d) but between steps b) and c). In such a step, to determine whether a voxel belongs to the desired (voxel) space, it is examined whether the point corresponding to the center of this voxel belongs to the axial digital plane space zone considered.

It will be understood that the different segmentations carried out by the steps of the method according to the invention (and which are not necessarily limited to the segmentations described) correspond to selections of voxels in the digital volume VN, and / or to selections of points in a axial numerical plane, while Figures 2 to 7 represent only illustrations (images) of different segmentations, viewed in 2D and corresponding to axial sections of the body, these images being furthermore transformed by the fact that each voxel (or pixel corresponding) is not represented in relation to its associated numerical value, but with a numerical value common to all the voxels (pixels) of the same homogeneous zone corresponding to a determined density interval.

The image production or scanning method according to the invention also advantageously comprises one or more erosion and / or expansion steps, and may comprise other intermediate data or image processing steps.

By way of example, the following steps can advantageously be implemented:

1) Determination of a digital volume VN cut into voxels, corresponding to a real volume VR.

2) Scan of the actual volume and assignment of numerical values to the different voxels, in relation to the density of the corresponding elementary real volumes.

3) Determining a first segmentation, corresponding to the selection of only voxels whose numerical value is within a predetermined range of values, covering the values corresponding to flesh or muscle.

4) Realization of one or two erosion steps, performed in 2D (for each elementary slice of VN corresponding to an axial section of the body).

5) Calculation of the different convex components (in 3D).

6) Calculation of the volume Vi of each related component.

7) Elimination of the convex components of volume lower than a first threshold.

8) Determination of the connected component of maximum volume VM.

9) Determination of the ratios (relative volumes) Vi / VM.

10) Elimination of related components whose ratio (relative volume) Vi / VM is less than a second threshold. 11) Determination of the second fusion segmentation of the remaining connected components.

12) Realization of two, three, or four expansion stages, performed in 2D (for each elementary slice of VN corresponding to an axial section of the body) and / or in 3D.

12) Determination (in 2D, for each axial digital plane of VN, corresponding to an axial section of the body) of a convex envelope of the corresponding part of the second modified segmentation (after expansion), using pairs of straight lines.

13) Projection of one or more 2D and / or 3D images on a screen and / or on a film support.

At the end of each of the steps, the set of data corresponding to this step is typically stored in a memory of the computer.

The invention is not limited to the embodiments presented and other embodiments will become apparent to those skilled in the art. In particular, by choosing an interval INT such that the lower bound of the interval is in an interval of values [-300, -100] expressed in Hounsfield and such that the upper bound is in an interval [100, 300], a first efficient segmentation is obtained which makes it possible to eliminate in a single operation, at the same time a large part of the air and the vacuum and a large part of the table.

In particular, it is possible to carry out other intermediate data processing or image processing steps, in particular steps already known to those skilled in the art. The scanner may be designed according to different technologies, and may include an integrated or separate computer, or separate separate and / or integrated computers, without departing from the scope of the invention. Similarly, it is possible to use different screen technologies, and / or different known physical media, such as hard disks, CDs, DVDs, etc. for the computer program, and / or various film supports, without departing from the scope. of the invention.

Claims

1. A method of producing a digital scan image of at least a portion of a body (2) in contact with a table (4), wherein:

this at least part of the body (2) being disposed inside a real volume

VR (10), this real volume VR is associated with a digital volume VN formed of voxels, each voxel of VN corresponding to an elementary real volume of VR,

- we associate with each voxel a numerical value correlated with the absorption of rays

X generated by a CT scanner in the elementary real volume corresponding to this voxel,

characterized in that at least the following steps are performed by means of a computer for the elimination in the digital image of at least the largest part of the representation of the rigid part of the table:

^A first segmentation (14) of VN is determined by selecting the only voxels whose associated numerical value is located in a predetermined bounded interval INT of values such that:

At least 90%, and preferably 100% of the voxels corresponding to an elementary real volume containing fat and / or muscle, for a predetermined population sample, have an associated numerical value comprised in the interval INT,

At least 50%, and preferably at least 90%, of the voxels corresponding to an elementary real volume containing a rigid part of the table have an associated numerical value which is not included in the INT interval,

the related component (s) of the first segmentation, possibly modified, is determined and for each connected component of index i, which may be modified, is calculated a variable Vi, referred to as the volume of this connected component, related to the number of voxels of this component connected, the connected component of larger volume VM is determined, possibly after an initial selection of only the related components, possibly modified, having a volume at least equal to a predetermined value,

^ we gather and select the set of voxels of the only related components, possibly modified, whose ratio Vi / VM is greater than or equal to a threshold predetermined, to form a second segmentation,

a third segmentation of VN is determined, of which any part comprised in an axial digital elementary slice, is a convex set comprising the part of the second segmentation or the second modified segmentation which is included in this axial digital elementary slice,

and in that at least one digital image is produced from the third segmentation.

2. The method of claim 1, wherein any portion of the third segmentation included in an axial digital elementary slice is substantially a convex envelope of the portion of the second segmentation possibly modified in the same axial digital elementary slice.

3. The method of claim 1 or 2, wherein the predetermined threshold is between 0.03 and 0.18 and preferably between 0.05 and 0.12.

4. Method according to one of claims 1 to 3, wherein is carried out at least one so-called "erosion" step of a first set of voxels, for example of the first segmentation, in which at least every voxels of this first set, which are contiguous to a voxel not belonging to this first set of voxels.

5. The method according to claim 4, wherein during the erosion step, only the voxels which are contiguous to a voxel not belonging to this first set are eliminated, the contiguous term applying only to the voxels arranged. in the same axial digital elementary slice as the voxel not belonging to this first set, this erosion step being preferably performed after the step of determining the first segmentation, and before the step of determining the related components.

6. Method according to any one of claims 1 to 5, wherein there is carried out at least one step, and preferably at least two step (s) said (s) "expansion" of a second set of voxels, by example of the second segmentation, in which one adds to this second set at least all the voxels of VN not belonging to this second set, but which are contiguous to at least one voxel belonging to this second set, the term contiguous to optionally applying only to the voxels arranged in the same axial digital elementary slice as this voxel belonging to this second set, this expansion step being preferably performed after the step of determining the related components.

7. Process according to claims 4 and 6 taken together, in which a greater number of voxels are added by expansion than the number of voxels removed by erosion, for example by using a number of expansion steps greater than a number of stage (s) of erosion.

8. A method according to any one of claims 1 to 7, wherein is also carried out, after, and preferably immediately after the step of calculating the volumes of related components, a step of eliminating related components whose volume is less than a predetermined percentage of the volume corresponding to said digital volume VN, for example less than 0.1% of said volume of VN.

The method of claim 2, wherein:

each axial elementary digital slice is associated with a two-dimensional continuous space, called an axial digital plane, comprising a discretized subset of points each corresponding to a voxel, said axial plane trace of a second possibly modified segmentation, associated with the part of the second possibly modified segmentation which is included in the axial digital elementary slice associated with the axial digital plane,

in each axial numerical plane, a plurality of pairs of digital lines is determined, the digital lines of one and the same pair being parallel to each other and thus delimiting, in the broad sense, an axial digital plane space band, these digital lines being of different directions from a a pair of straight lines to another pair of straight lines, each pair of digital lines flanking the axial planar trace of a second segmentation, possibly modified, so that each of the digital lines of this pair meets said axial plane trace of second segmentation, possibly modified, and this axial plane trace of second segmentation possibly modified is entirely comprised in the axial digital plane space band delimited by this pair of straight lines,

in each axial numerical plane, a subset of this plane, said convex envelope of axial trace of second segmentation, possibly modified, is defined as the intersection of the different space bands corresponding to the different pairs of digital lines,

10. The method of claim 9, wherein, to determine a pair of digital lines having a given direction in at least one axial digital plane, defines in this axial digital plane a reference axis, said abscissa axis, parallel to this given direction, and an ordinate axis, perpendicular to this abscissa axis, and for each point of the contour of the axial plane trace of second segmentation possibly modified, the ordinate of this point is calculated,

then we identify two points of the contour of the axial plane trace of second segmentation possibly modified, said end points vis-à-vis the given direction, whose ordinates are respectively minimum and maximum, then we determine a couple of two straight digital parallel to the given direction passing through each of these two end points respectively.

11. A method as claimed in any one of claims 1 to 10, wherein the predetermined bounded interval INT of values is such that: At least 98% of the voxels corresponding to an elementary real volume filled with fat or muscle, for the predetermined population sample, have an associated numerical value comprised in the interval INT,

12. Method according to any one of claims 1 to 11, wherein produces a digital image corresponding to the representation of an elementary subset of the third segmentation corresponding to an axial axial digital elementary slice.

13. A method according to any one of claims 1 to 12, wherein is stored in at least one memory of the computer, data corresponding to the third segmentation.

The method of any one of claims 1 to 13, wherein the lower bound of the INT interval is within a range of values [-300, -

100] expressed in Hounsfield, the upper bound being in a range [100, 300].

15. Computer program characterized in that it comprises program code instructions for performing the steps of the method according to any one of claims 1 to 14 which are performed by means of a computer.

16. A method of scanning at least a portion of a body in contact with a table, in which a set of numerical values correlated with the X-ray absorption generated by a CT apparatus is generated in elementary real volumes belonging to to a real volume VR, characterized in that a digitized image is produced by the method according to any one of the claims 1 to 14, this image being displayed on a screen, preferably automatically, and / or fixed on a physical film medium, at the request of a medical operator.

17. Scanning image on physical film support, characterized in that it has been carried out by the image production method according to any one of claims 1 to 14, or according to the scanning method according to claim 16.

18. Physical information medium characterized in that is recorded on this medium a computer program according to claim 15.

19. A scanner assembly comprising a scanner and a computer, characterized in that the computer comprises a physical medium on which is recorded a computer program according to claim 15.

The scanner assembly of claim 19, wherein the computer is configured to perform the method steps of claim 1 which are computer-implemented with a computation time of less than 10 seconds per giga-voxels of the digital volume VN, and preferably at 5 seconds per giga-voxels of the digital volume VN.

21. The scanner assembly of claim 19 or 20, wherein the computer is remote from the scanner apparatus, the scanner assembly including a communication interface between the scanner and the computer.

22. A method of providing digital data over a telecommunication network for downloading, characterized in that these digital data relate to the third segmentation defined during the execution of the method according to any one of claims 1. at 14.