WO2018104609A1 - Method for tracking a target in a sequence of medical images, associated device, terminal apparatus and computer programs - Google Patents

Abstract

The invention concerns a method for tracking a target in a sequence of digital medical images, an area of interest having been predetermined in an initial image of sequence, referred to as the reference image (I_t0), the method comprising the following steps, implemented for a current image of the sequence that is distinct from the reference image: - obtaining (E1) an initial position of a model of an area of interest in the current image from a position of a model of an area of interest in a preceding image of the sequence; - obtaining (E2) a confidence measure for each image element for said reference image and for said current image, said measure being representative of a confidence in an intensity of the image element; - transforming (E3) the initial position of the model of the area of interest in the current image, comprising a minimisation of a cost function based at least on a difference between the intensities of the current image and of the reference image, - following the transformation, detecting (E5), in the area of interest of the reference image, at least one image element to be corrected, on the basis of confidence measures obtained for the reference image and the current image, and correcting (E6) the at least one detected element at least on the basis of the intensity of the at least one image element of the area of interest of the current image.

Description

 A method of tracking a target in a sequence of medical images, device, terminal equipment and associated computer programs

1. Field of the invention

 The field of the invention is that of the treatment of medical images.

 More specifically, the invention relates to a method for tracking a clinical target in a sequence of medical digital images.

 The invention finds particular application in the treatment of images obtained by an imaging technique for example ultrasonic or endoscopic type.

 2. State of the art

 Ultrasound or endoscopic imaging techniques are widely used in the medical field to help physicians visualize in real time a clinical target and / or surgical tool during a surgical procedure or invasive examination intended to diagnose a pathology. For example, ultrasound techniques are frequently used during an intervention requiring the insertion of a needle, and especially in interventional radiology.

 It is sometimes difficult for a surgeon to identify by himself certain clinical targets, such as tumors, in an image obtained by an ultrasound imaging technique or endoscopy. In order to help, surgeons have tools to automatically estimate the position of a surgical target in an ultrasound image.

Target tracking methods are known that rely on determining a transformation to align different images over time. To do this, these approaches are usually based on a cost function that measures the similarity between these images. The methods of monitoring can be classified into two categories:

 2.1 Paired Strategies

 A first type of approach consists in determining the position of anatomical structures by successive comparison between different pairs of images. Many methods are based on this model such as the approaches proposed by Shekhar and Zagrodsky and Heyde et al.

 In general, the optimal transformation T is obtained by minimizing the following cost function:

arg min 5. /, ·. T o) + R {T) In this cost function, Ir and It respectively represent the reference image and the current image corresponding to the last image acquired over time. The first term of the cost function, called S, makes it possible to measure the alignment between a reference image (Ir) and a transformed current image (T ° It). A second term, called R, penalizes unrealistic transformations. The choice of the image of the reference image differs according to the methods proposed and can have an impact on the quality of the results of follow-up. Generally, paired strategies can be divided into three subcategories.

 2.1.1 Non-iterative pairing strategies

 At each acquisition, these methods compare the current image with a fixed reference image. Generally, the initial image of the sequence is chosen as the reference image. Thus the cost function described above can be expressed in the following form:

armin S (I, T o I _t ) + R (T)

Where ItO represents the initial image of the sequence. However, the performance of this approach is related to the choice of the reference image. Indeed, inaccurate monitoring results can be obtained if the quality of the reference image is not sufficient.

 2.1.2 Iterative pairing strategies

 To overcome the previous problem, this second type of approach makes it possible to compare the current image with its previous image. The cost function can therefore be expressed in the following form:

arg min jo! .. ·. . '}. 'o / _t ) + R (T)

Where It- 1 represents the image preceding the current image. The optimal transformation Tt-1 represents the optimal transformation calculated in the previous step. However, the disadvantage of this type of approach is related to the drift of the algorithm. Indeed, inaccurate tracking results can be observed because of an accumulation of error over time.

 2.1.3 Hybrid Pair Strategies

To compensate for this drift, a third type of approach, proposed by Matthews et al., 2004, combines the two previous strategies. To do this, the optimal transformation T is initialized by minimizing a cost function between the current image and the previous image. Then, this transformation is refined using a cost function between the current image and the initial image. However, this type of approach remains sensitive to the quality of the initial image. 2.2 Group Strategies

Unlike the methods presented above, group policies compare all images in the sequence to the current image. With each new current image acquired, these approaches take into account all the information in the sequence in order to obtain a more robust tracking result. For applications based on the ultrasound image, various examples have been proposed by Metz et al. and Vijayan et al. These methods make it possible to follow an anatomical target by comparing the current image with a synthetic image obtained from the preceding images. The estimation of target displacements is obtained from the minimization of the following cost function arg min 5 (/ _m? T o I _t ) → ^■ 1 {{ ^' ] ^' }

Where Im represents the synthetic image generated from the previous images. To obtain it, Metz et al. proposes to use the following equation:

Ni represents the number of images in the sequence. P (tk) represents the vector of the positions of the voxels belonging to the target at the time Tk. These positions depend directly on the transformation T. Itk (Ptk) represents the intensity of the pixels / voxels of the image at time tk. It can be seen that the synthetic reference image is obtained from the average of all the images preceding the current image. Thus, this method has the advantage of avoiding selecting a reference image. In addition, it eliminates problems related to drift. On the other hand, the main limitation of these strategies is that the intensity of the structure must remain constant over time. However, the intensity of a structure is often affected by various undesirable effects such as the presence of noise, artifacts, or shadows.

3. Disadvantages of prior art

A disadvantage of these methods is that their performance is generally disturbed by artifacts present in the images of the sequence and whose type differs according to the modality used. For example, ultrasound images are affected by shadows and mirrors effects. In addition, changes in gain and probe orientation may also impact these images. Endoscopic imaging can also be affected by different artifacts such as specular reflections, occultations, and illumination changes. Also known from Royer et al entitled "Real-time target tracking of soft tissue in 3D ultrasound images based on robust Visua information and mechanical simulation", published in the journal "Medical Image Analysis", in September 2016, a method a pairwise strategy-based target tracking that implements a local confidence measure of the intensities of the reference image and the next image and exploits it to minimize the impact of these artifacts on monitoring target.

 A disadvantage of this method is that its performance depends on the quality of the reference image.

 4. Objectives of the invention

 The invention improves the situation.

 The invention particularly aims to overcome these disadvantages of the prior art.

 More specifically, an objective of the invention is to propose a solution that guarantees efficient monitoring, when the reference image includes artifacts in the area of interest.

 5. Presentation of the invention

 These objectives, as well as others that will appear later, are achieved by using a target tracking method in a sequence of medical digital images, an area of interest that encompasses the target having been predetermined in an initial image of the sequence, called reference image, said method comprising the following steps, implemented for a current image of the sequence, distinct from the reference image:

 Obtaining an initial position of a model of a zone of interest in the current image from a position of a model of a zone of interest in a previous image of the sequence; - Obtaining a confidence measure per image element for said reference image and for said current image, said measurement being representative of a confidence in an intensity of the image element;

 Transforming the initial position of the model of the zone of interest into the current image comprising minimizing a cost function based at least on a difference between the intensities of the current image and the reference image,

According to the invention, the method further comprises, following the transformation, a step (E4) of mapping of the image elements of the area of interest of the reference image and the image elements of the image. zone of interest of the current image, a step of detecting in the area of interest of the reference image of at least one image element to be corrected, comprising a comparing (E51) the confidence measure of a picture element of the area of interest of the reference picture with that of at least one corresponding picture element in the area of interest of the picture current, an image element to be corrected being detected (E52) when its confidence measurement is lower than that of the at least one corresponding image element and, when at least one image element has been detected, a step of correction of the at least one detected element at least from the intensity of at least one picture element of the area of interest of the current picture.

 According to the invention, the area of interest of the reference image is enriched by reliable information contained in the current image.

 Thus, the invention is based on an entirely new and inventive approach, which consists in reconstructing an improved reference image, based on the following images of the sequence, by exploiting a local confidence measurement associated with the intensities of the image elements. images of the sequence.

 Unlike the prior art which uses either the original preceding images or a synthetic image derived from an averaging of the original preceding images, the invention proposes to dynamically form an improved reference image as the sequence is processed. . The reconstructed reference image for the current image becomes the reference image for the next image.

 One advantage is to make the following images of the sequence of the contributions of the previous images benefit. The quality of the reference image can thus improve over time and any acquisition artifacts are gradually erased by the successive corrections.

 The proposed method has the advantage of being faster than the group methods since it relies only on the comparison of two images. It also makes it possible to avoid the accumulation of the error since it locally corrects the structures of the reference image only when the current image offers information likely to improve the quality of the reference image.

Another advantage is that only those image elements of the reference image that are identified as less reliable than their corresponding in the current image are corrected.

The invention thus makes it possible to solve the technical problem of the management of the artifacts present in a sequence of medical images and, in particular, in the image chosen as a reference for target tracking. According to another aspect of the invention, the correction of an image element detected in the reference image comprises a replacement of its intensity by the intensity of the at least one corresponding image element in the current image.

 An advantage of this embodiment is its simplicity. The additional information extracted from the current image is injected into the reference image in order to enrich it.

According to yet another aspect of the invention, the correction of an image element detected in the reference image comprises a replacement of its intensity by an intensity calculated from a correction function which takes into account the intensities. image elements of the area of interest of the reference image and those of the corresponding elements in the area of interest of the current image and their associated confidence measurements.

 An advantage of this embodiment is that the calculation of the new intensity takes into account the intensities of the other image elements of the area of interest which are sufficiently trustworthy, and therefore any differences in intensity level. in the rest of the area of interest between the current image and the reference image. This makes it possible to compensate in particular for a variation of the gain level between the two images or, for a sequence of ultrasound images, a change of orientation of the probe and thus to obtain a corrected image presenting a zone of interest at more intensities. homogeneous.

In yet another aspect of the invention, the intensity correction function of a detected image element calculates an intensity corrected by a ratio between a joint probability density that defines the probability that a pixel in the area of interest takes a first intensity value in the current image and a second value in the reference image and a probability that this image element takes the second value in the reference image. An advantage of this embodiment is that it is simple to calculate and adapted to a real-time application.

According to yet another aspect of the invention, the method further comprises a step of correcting the confidence measurements associated with the intensities corrected in the reference image at least from the confidence measurements of the corresponding picture elements of the picture. current image. One advantage is that the target tracking in the next image will not only benefit from a corrected reference image, but also from updated confidence measurements following the correction. Of this way, it is guaranteed to stop the dynamic correction of the reference image as soon as it has reached a sufficient level of confidence.

 According to another aspect of the invention, the confidence measure of a corrected picture element is replaced by that of the corresponding picture element of the current picture.

 The various embodiments or features mentioned below may be added independently or in combination with each other, to the characteristics of the authentication method defined above.

The invention also relates to a tracking device adapted to implement the tracking method according to one of the particular embodiments defined above. This device may of course include the various features relating to the method according to the invention. Thus, the features and advantages of this device are the same as those of the method, and are not detailed further.

Correlatively, the invention also relates to a terminal equipment comprising an image acquisition module, an acquired image processing module, a processed image display module, a user interaction module and a tracking device. target according to the invention.

The invention also relates to a computer program comprising instructions for implementing the steps of a method as described above, when this program is executed by a processor.

This program can use any programming language. It can be downloaded from a communication network and / or saved on a computer-readable medium.

The invention finally relates to a recording medium, readable by a processor, integrated or not to the target device according to the invention, possibly removable, respectively storing a computer program implementing a monitoring method, such as previously described.

6. List of figures

Other advantages and characteristics of the invention will emerge more clearly on reading the following description of a particular embodiment of the invention, given as a simple illustrative and nonlimiting example, and the appended drawings, among which: : FIG. 1 schematically illustrates in two images of a medical image sequence, an area of interest delimited around a clinical target;

 FIG. 2 is a block diagram of the steps of a clinical target tracking method according to an embodiment of the invention;

 Fig. 3 is a view of a segmented contour of a target in a section of an image of a sequence;

 FIG. 4 illustrates a confidence measurement image for a zone of interest of a section of an image of the sequence;

 FIG. 5 schematically illustrates a mapping function between a picture element of the area of interest of the reference picture and picture elements of the area of interest of the current picture;

 FIG. 6 shows sections of images of an ultrasound sequence processed by the target tracking method according to the invention; and

 Figure 7 schematically shows an example of a hardware structure of a clinical target tracking device according to the invention.

7. Description of a particular embodiment of the invention

 The general principle of the invention is based on the detection of image elements to be corrected in an area of interest of a reference image from local confidence measurements obtained for the reference image and the current image. and on the correction of the intensities of the detected picture elements using the intensities of the corresponding picture elements in the area of interest of the current picture.

 We now detail an example of implementation of the invention. In this particular embodiment of the invention, the image sequence is obtained by ultrasound imaging. It is a sequence of three-dimensional images, the elements of which are voxels.

 Of course, the invention is not limited to this example and could relate to a sequence from another imaging modality, such as for example an endoscopic sequence and apply to a sequence of two-dimensional images.

In relation to FIG. 1, consider a sequence of images comprising T images with T integer greater than or equal to 3 images, among which a reference image I _t o, a previous image I _t -i and a current image I _t . For simplicity, 2D images are shown. In each images appears a target Cb whose contour is represented schematically by a closed curve.

 With reference to FIG. 2, the steps of an exemplary method of monitoring a clinical target in a sequence of images according to the invention are schematically illustrated in block diagram form.

 During a first step E1, an initial position of a model of an area of interest of the current image is obtained from a zone of interest Z1 (It-i) of the preceding image . For example, a simple transposition of the model of the zone of interest of the preceding image into the current image is carried out, placing it at the same position as in the preceding image. As illustrated in FIG. 1, the initial area of interest is therefore identical in terms of position, size and shape to that of the previous image.

According to a first embodiment of the invention, the area of interest of the reference image I _t o has itself been obtained by simple input of the user who delimited by hand the contours of this area of interest in the image, so as to completely encompass the target Cb he wishes to follow throughout the sequence. This area of interest usually takes a simple geometric shape such as a rectangle, as shown in Figure 1.

According to a second embodiment, the zone of interest ZI (I _t o) of the reference image was obtained by delimiting a contour of the target, by a segmentation method known per se, which can be manual or automatic. The contour of the segmented target was then smoothed to remove the sharp edges and discontinuities of shape that appeared on its outline.

A zone of interest ZI (I _t o) delimiting the segmented contour of the target in the reference image

Ir is then obtained by constructing a representation of the interior of the target contour, for example by generating a tetrahedral mesh.

An example of a mesh of the zone ZI (I _t o) is illustrated in FIG. 3. This figure corresponds to an echographic image section comprising a target partly situated in a shaded area, hatched in white. In this example, the mesh of the zone ZI has N _c vertices delimiting tetrahedral cells. Zone Z includes Nu voxels in total.

During a step E2, Uto and Ut confidence measurements are obtained from the image elements of the zones of interest Zl (Ito) and ZI (It).

In the following, a measure of confidence of a picture element is a value indicative of a probability or likelihood that the intensity of the pixel / voxel with which it is associated represents the target and is not affected by different disturbances such as shadows, specular reflections, or occultations generated by the presence of other objects. In other words, this local confidence measure provides information on a quality of the intensity of the picture element.

 Methods for estimating a confidence measure per pixel are known, for example that described by Karamalis et al. in a paper titled "Ultrasonic confidence map using random walks" in the journal Medical Image Analysis, published by Elsevier in 2012, pp.1101-1112. In this paper we measure the confidence of a pixel / voxel of the ultrasound image as the probability that a random walk in this pixel / voxel reaches the transducers of the ultrasound probe. The path is constrained by the model of propagation of an ultrasonic wave in the soft tissues. The value of the confidence measure that is assigned to each voxel is for example between 0 and 255.

 With this method, low confidence measurement values, for example less than 20, are attributed to the intensity of each voxel located in a shaded portion PO of zone ZI, such as that shown hatched in FIG.

 It will be understood that this method of measuring a confidence value of the intensities of the elements of the image gives an indication of the location of any outliers in the area of the target.

 An example of an image of a confidence map U1 obtained for the zone ZI is illustrated in FIG. 4. In this figure, the higher the confidence value of a pixel, the greater the confidence value of a pixel. high intensity and therefore appears clearer.

A map U _t of local confidence measurements of the current image It can therefore be obtained in E2, on the basis of this method. A map U _t o local confidence measurements of the reference image Ito was obtained previously, for example with this method.

 Alternatively, the step E2 of obtaining confidence measurements for the area of interest of the current image It comprises a sub-step of detecting shaded parts in the area of interest.

For example, this detection of the shaded parts of the zone ZI (It) implements a technique known per se, for example described in the document by Pierre Hellier et al, entitled "An automatic geometric and statistical method to detect acoustic shadows in intraoperative ultrasound brain images "in the journal Medical Image Analysis, published by Elsevier, in 2010, vol. 14 (2), pp.195-204. This method involves analyzing ultrasound lines to determine positions corresponding to noise and intensity levels below predetermined thresholds. For the detection of clear parts, of the halo or specular type, more suitable for endoscopic type images, reference will be made, for example, to the detection technique described in the document by Morgand et al, entitled "Generic detection and real-time specularities". ", Published in the Proceedings of the Francophone Days of Young Researchers in Computer Vision, held in Amiens, June 2015. Specularities are detected by dynamically thresholding the image in the HSV space for Hue-Saturation- Value (or HSV for "Hue Saturation Value"). The value of the thresholds used is automatically estimated according to the overall brightness of the image.

In a second step, a confidence measurement is calculated for each image element taking into account the part of the zone where it is located. For example, a bit mask is applied to the intensities of the area of interest of the target. Voxels belonging to an outlier will get zero confidence and voxels outside an outlier will get a confidence measure of 1.

 It is understood that a picture element will get a lower confidence measure if it is in a part detected as aberrant.

As regards the reference image I _t o, these confidence measurements have already been estimated during the processing of the reference image and are available in memory, for example in the form of a confidence card. Step E2 thus comprises a reading of the Uto confidence card in this memory.

At the end of this step E2, there is available for the current image I _t a map or image of confidence Ut and for the reference image I _t o of a map or image of confidence Uto-

During a step E3, transforming the initial position of the model corresponding to the zone of interest ZI (I _t -i) in the current image so as to coincide with the new position of the target in the current image It. To do this, the similarities between the reference image and the current image are exploited. The objective of this step is to determine a transformation of the model positions to optimize a cost function that defines the similarity between the image elements of the area of interest of the current image and the image elements. the area of interest of the reference image. For this, we define a relation allowing to link the positions of the model "q" and the positions of the elements of images "p".

(3.N_C), définissant un ensemble de coefficients bilinéaires. Grâce à la minimisation de la fonction de coût C, nous obtenons les nouvelles positions des sommets q(t) ainsi que l'ensemble des nouvelles positions des pixels p(t) de la zone d'intérêt dans l'image courante. According to the first embodiment of the invention, the model representing the initial area of interest is of simple geometrical shape, for example rectangular for a 2D or parallelepipedic image for a 3D image. It includes the summits of the form geometric and defines from these vertices the tetrahedral cells of a mesh of the zone of interest delimited by this geometric form. This type of model further defines a function for linking the positions of the image elements of the area of interest to the position of the vertices of the rectangle. In order to compute the displacements of the model, an intensity-based approach is implemented which consists in minimizing a cost function C based on the intensity differences between the pixels / voxels of the zone of interest in the reference image and of the area of interest in the current image. More precisely, this approach calculates the new positions of the vertices of the model constituting the target by using the function connecting the positions of the N _u voxels with the positions of the N _c vertices of the rectangle. This function is expressed in the form p = Mq, where p is a vector, of dimension (3. Nu) xl, representing the positions of the N _u voxels of the zone of interest, q is a vector, of dimension (3 N _c ) xl, representing the positions of the N _c vertices of the rectangle, and M is a matrix with constant coefficients, of dimension

(3.N _C ), defining a set of bilinear coefficients. Thanks to the minimization of the cost function C, we obtain the new positions of the vertices q (t) as well as all the new positions of the pixels p (t) of the zone of interest in the current image.

According to the second embodiment of the invention, the initial zone of interest of the current image is delimited by a segmented contour. For example, consider a model comprising vertices of this contour and defining a mesh of the zone of interest delimited by this contour.

 In a manner analogous to that of the first embodiment, an intensity-based approach is used to calculate the so-called external displacements of the target contour, which consists in minimizing a cost function C based on the intensity differences between the vertices. Contours of the target in the reference image and in the current image. This approach, described in the document Royer et al already cited above, is based on a mesh of the area of interest.

More precisely, this approach consists in calculating a deformation of tetrahedral cells constituting the target and applies a piecewise affine function connecting the positions of the u voxels of the target with the positions of the N _c vertices of the mesh, expressing in the form p = mq, where p is a vector of dimension (3. Nu) l, representing the positions of _Ν υ voxels of the target, q is a vector of dimension (3.N _c) xl, representing N _c vertices positions of the mesh, and M is a matrix with constant coefficients, of dimension (3.Nu) x (3.N _C ), defining a set of barycentric coordinates. To obtain the optimal position of the segmented contour, this estimate of the external displacements undergone by the contour of the target is combined with that of internal displacements resulting from the simulation of the deformation of a mechanical mass-spring-damper system applied to the target.

 An optimal displacement of the vertices of the contour of the target is estimated iteratively by taking into account the two types of internal and external displacements.

 In the solution described, the confidence measurements associated with the image elements, respectively at the vertices of the tetrahedral cells, are advantageously used to weight the external displacement, so as to minimize the external displacement of image elements associated with a measurement of low confidence.

At the end of this step E3, we have a transformed zone of interest ZI _T (I _t ) whose shape, size and position in the current image are optimized with respect to the actual position of the target to follow in this image. More precisely, this transformed area of interest is represented by a vector p (t) whose picture elements px (t) are the transforms of picture elements px (tl) of the vector p (tl) of the area d interest of the previous image.

To determine the correspondence between two non-successive images of the image sequence, we concatenate the different local transformations applied to the successive images.

It will be understood that the local transforms successively applied to the vector p (tl) thus define a global transform that makes it possible to pass from the vector p (to) of the reference image to the vector p (t) of the current image. In other words, a picture element pxtodu p (to) has the corresponding pixel px _t in the current image.

It should be noted that the coordinates of the pixel px _t may not be integer because of the successive transformations. In this case, it is associated with the image elements having the neighboring integer positions, as illustrated by Figure 5. It is therefore possible that a pxto image element has several correspondents in the current image.

During a step E4, the image elements of the vector p (to) of the reference image are scanned, in a predetermined order, for example that of the N components of the vector, with N non-zero integer and we associate to a current pixel, px _t o of vector p (t), its corresponding picture element pxt in the vector p (t). In E5i, the current picture element px _t o of the vector p (to) is compared with the corresponding picture element px _t of the vector p (t).

The comparison more precisely relates to the confidence measurements of the Uto and Ut images associated with the corresponding image elements in the Ito and It images. We compare the confidence measures Ut (pxt) and Uto (pxto) and we determine in E52 if Ut (pxt) is greater than U (pxto).

If this is the case, the picture element px _t o of the reference picture is detected as correct.

Note that if several pxt image elements correspond to the pxto image element, the result of each of their comparisons contributes to the decision on the pixel px _t o. For example, the average of their confidence measures is compared with that of px _t o or alternatively, the highest confidence measure.

According to the invention, a dynamic correction of the pixel px _t o (i) detected in the area of interest of the reference image Ito is carried out at E6.

In a first embodiment of the invention, it is implemented as follows:

I _d represents the corrected or dynamic reference image. px _t o is an image element detected as to be corrected.

It is understood that the intensity of the reference image is corrected dynamically in the areas where the local confidence of the elements of the reference image is less than that of the corresponding elements of the reference image. In this mode, the intensity of the picture element to be corrected is simply replaced by that of the picture element Pt which corresponds to it in the reference picture.

If, as illustrated in FIG. 5, several image elements of the current image correspond to the current pixel px _t o, a correction value is calculated, for example by averaging the intensities of the elements pxto, pxto2 , pxto3, pxto4 corresponding or by interpolation. According to a second embodiment of the invention, the correction of a picture element of the reference picture also takes into account the intensities of the other picture elements in the reference picture and in the picture. and their associated confidence measures.

Advantageously, a corrected intensity Id (pxto) is calculated for the pixel px as follows:

Le terme pr,_tko(i,j) représente une fonction de densité de probabilité jointe qui définit la probabilité qu'un élément d'image dans la zone d'intérêt prenne la valeur d'intensité i dans l'image courante et la valeur j dans l'image de référence. Le terme pr_h(i) représente la probabilité qu'un élément d'image prenne la valeur i dans l'image courante It.

The term pr, _tko (i, j) represents a joint probability density function that defines the probability that a picture element in the area of interest will take the intensity value i in the current picture and the value j in the reference image. The term pr _h (i) represents the probability that an image element takes the value i in the current image It.

This calculation takes into account a ratio between an attached probability density that an image element of the reference image takes a value j and an image element of the current image takes the value i and the probability that the pixel px _t o takes the value j in the reference image.

This makes it possible to compensate for a possible difference in gain between the two images and to avoid introducing into the reference image a corrected intensity that is not homogeneous with the local structure. Advantageously, the joint probability density is calculated in the following way: pr _Vm (.i, j) (Pkto) H _t Pkt) Siiit (Pkt)) SjOto Pkto))

L-l

Vr _hho i ⁱ , i) = _j P _hho (i, i)

H *., And H * represent weighting matrices associated with the corrected reference image and the current image. These matrices make it possible to weight the influence of the pixels / voxels as a function of their associated confidence measures. They are obtained from the confidence maps U _t and representing the local confidence measurement associated with the image elements of the current image and the initial image. The function is 1 when x = k, and 0 elsewhere.

In this way, it is ensured that the corrected intensity in the reference image Id corresponds to the intensities of the other image elements of the reference image Ito. This makes it possible to take into account a possible global variation in intensity between the reference image and the current image, which could make the correction of the incoherent reference image possible. In the case of ultrasound, such variation may be caused by a change in gain or probe orientation.

During a step E7, the Uto confidence card associated with the reference image is corrected in turn, by replacing the Uto confidence measurement (pxto) of an image element whose intensity has been corrected. by a corrected confidence measurement, a function of the confidence measurement or measurements of the corresponding picture element (s) in the current Ito image.

A dynamic confidence image U _d is obtained which represents the quality of the dynamic reference image I _d .

For example, this correction is made from the following equation:

The correction here consists of a simple replacement of the confidence measure associated with the pixel px _t o by that of the corresponding pixel P from which it has been corrected.

In the case where the picture element px _t o corresponds to several picture elements px _t in the current picture, the corrected confidence measure Ud (pxto) is obtained from the image element confidence measurements. pxt, for example by averaging or interpolating their values.

The operations which have just been described are repeated for all the pixels of the area of interest of the reference image Ito. In the embodiment that has just been presented, the image elements of the area of interest of the reference image are reviewed and, as soon as an image element detected as to be corrected in E5, it is directly corrected in E6, and its confidence measurement updated in E7, before moving on to the next element. Of course, the invention is not limited to this embodiment. One could for example provide an implementation of the invention according to which each step is performed all the elements of the vector ptO of the reference image, before going to the next step.

In connection with FIG. 6, examples of sections of a processed image sequence according to the invention are presented. The first line represents sections of the images of the sequence over time. The second line represents an evolution over time of the dynamically corrected reference image according to the invention. The third line shows an enlargement of the area of interest in the corrected reference image and its evolution over time. As can be seen in the images a), e) and i), the initial image Ito is disturbed by the presence of a shadow. The images e) to h), respectively i) to I) in their enlarged versions, highlight the improvement of the quality of the area of interest of the reference image over time, as and when successive corrections by the trustworthy intensities of the current image It.

The proposed method has the advantage of being faster than the group methods of the prior art because it relies only on the comparison of two images. In addition, the invention makes it possible to avoid the accumulation of the error since it locally replaces the structures of the reference image until its quality is sufficient. It has improved performance compared to a tracking method based on a fixed reference image, potentially of poor quality.

Figure 7 shows the simplified structure of a target tracking device adapted to implement the target tracking method of the invention. The device 100 is adapted, for a current image of the sequence, è:

 obtaining an initial position of a model of a zone of interest in the current image from a position of a model of area of interest of a previous image;

obtaining a confidence measure per image element in said zone for said reference image and for said current image, said measurement being representative of a confidence associated with the intensity of the pixel; transforming the initial position of the area of interest in the current image by minimizing a cost function based at least on a difference between the intensities of the current image and the reference image,

 following the transformation, detecting in the area of interest of the reference image of at least one image element to be corrected, based on the obtained confidence measurements and,

when at least one picture element has been detected, correcting it at least from the intensity of at least one picture element of the area of interest of the current picture. The device 100 is further configured to implement the various embodiments of the invention which have just been described in relation to FIG. 2.

 According to a particular embodiment of the invention, the steps of the monitoring method are implemented by computer program instructions. For this purpose, the device 100 has the conventional architecture of a computer and notably comprises a memory MEM1, a processing unit UT1, equipped for example with a microprocessor μΐ, and driven by the computer program Pgl stored in memory MEM1 . The computer program Pgl includes instructions for implementing the steps of the tracking method as described above, when the program is executed by the processor μΐ.

At initialization, the code instructions of the computer program Pgl are for example loaded into a RAM memory before being executed by the processor μΐ, which implements, in particular, the steps of the tracking method described above, according to the instructions of the computer program Pgl.

According to another particular embodiment of the invention, the method is implemented by modules or functional units (the) s. For this, the tracking device 100 further comprises the following modules: GET ZI (I _t -i) obtaining an initial position of the area of interest in the current image;

obtain GET U _t , U _t o a confidence measure per image element in said zone of interest of said reference image and of said current image;

transforming TRANS Zl (It-i) the initial position of the area of interest in the current image by minimizing a cost function based at least on a difference between the intensities of the current image and the image of reference, following the transformation, detecting DET EI (I _t o) in the area of interest of the reference image of at least one image element to be corrected, based on the confidence measurements obtained and,

 when at least one picture element has been detected, the CORR correction El (Ito) at least from the intensity of at least one picture element of the area of interest of the current picture.

 Advantageously, the device further comprises a CORR Uto module for correcting the confidence measurement associated with the corrected element.

The processing unit UT1 cooperates with the various functional modules described above and the memory MEM1 in order to implement the steps of the monitoring method.

According to one embodiment of the invention, the device 100 furthermore comprises a unit Mi for storing the position of the area of interest of the reference image, the images of confidence Ut, UtO, of the image of Ud corrected reference, of the Udto corrected confidence image.

These units are driven by the μι processor of the processing unit 110.

Advantageously, such a device 100 may be integrated with a user terminal equipment ET, such as a medical image acquisition system, for example an ultrasound system or an endoscope. It includes an acquisition module such as an ultrasound probe, a data processing module acquired, a data display module such as a screen and a user interaction module such as a keyboard , a mouse etc.

The various functional modules described above can be in hardware and / or software form. In a software form, such a functional module may include a processor, a memory, and program code instructions for implementing the function corresponding to the module when the code instructions are executed by the processor. In a material form, such a functional module can be implemented by any type of suitable encoding circuits, such as, for example and without limitation microprocessors, signal processing processors (DSP for Digital Signal Processor). , application-specific integrated circuits (ASICs for Application Specifies Integrated Circuit in English), FPGAs for Field Programmable Gate Arrays in English, logic unit wiring. It goes without saying that the embodiments which have been described above have been given for purely indicative and non-limiting purposes, and that many modifications can easily be made by those skilled in the art without departing from the scope. of the invention.

Claims

A method of tracking a target in a sequence of medical digital images, an area of interest that encompasses the target having been predetermined in an initial image of the sequence, said reference image, said method comprising the following steps, implemented for a current image of the sequence, distinct from the reference image:  Obtaining (El) an initial position of a model of a zone of interest in the current image from a position of a model of a zone of interest in a previous image of the sequence;  Obtaining (E2) a confidence measure per image element for said reference image and for said current image, said measurement being indicative of a confidence in an intensity of the image element;  Transforming (E3) the initial position of the model of the area of interest into the current image comprising minimizing a cost function based at least on a difference between the intensities of the current image and the image of the reference, characterized in that it further comprises, following the transformation, a step (E4) of mapping of the image elements of the area of interest of the reference image and the image elements of the area of interest of the current image, a detection step (E5) in the area of interest of the reference image of at least one image element to be corrected, comprising a comparison (E51) of the measurement of a picture element of the area of interest of the reference picture to that of at least one corresponding picture element in the area of interest of the current picture, a picture element to be corrected being detected (E52) when its confidence measure is lower than that of the at least one corresponding picture element and, when at least one picture element has been detected, a step of correcting (E6) the at least one detected element at least from the intensity of at least one picture element the area of interest of the current image. A target tracking method according to claim 1, characterized in that the correction of an image element detected in the reference image comprises a replacement of its intensity by the intensity of the at least one corresponding image element in the current image. A target tracking method according to claim 1, characterized in that the correction of an image element detected in the reference image comprises a replacement of its intensity by an intensity calculated from a correction function which takes the intensities of the image elements of the area of interest of the reference image and those of the corresponding elements in the area of interest of the current image and their associated confidence measurements. A target tracking method according to claim 3, characterized in that the correction function of the intensity of a detected image element calculates an intensity corrected by a ratio between a joint probability density which defines the probability that a image element in the area of interest takes a first intensity value in the current image and a second value in the reference image and a probability that this image element takes the second value in the image of reference. A target tracking method according to one of claims 1 to 4, characterized in that it further comprises a step of correcting the confidence measures associated with the intensities corrected in the reference image at least from the confidence measurements corresponding picture elements of the current picture.  A target tracking method according to claim 5, characterized in that the confidence measure of a corrected picture element is replaced by that of the corresponding picture element of the current picture. A target tracking device (100) in a medical digital image sequence from a current image, an area of interest (Zl (Ito)) which encompasses the target having been predetermined in a reference image of the sequence , comprising a machine dedicated to or configured to implement, for a current image of the sequence:  Obtaining an initial position of a model of a zone of interest in the current image from a position of a model of a zone of interest in a previous image of the sequence; - Obtaining a confidence measure per image element (Uto) for said reference image and (Ut) for said current image, said measurement being representative of a confidence in an intensity of the pixel; Transforming the area of interest into the current image comprising minimizing a cost function based at least on a difference between the intensities of the current image and the reference image,  characterized in that, following the transformation, it is able to:  detecting, in the area of interest of the reference image, at least one image element to be corrected, based on the obtained confidence measurements and  when at least one picture element has been detected, correcting it at least from the intensity of at least one picture element of the area of interest of the current picture. 8. Terminal equipment (ET) comprising an acquisition module (ACQ) of images, a processing module (PROC) acquired images, a module (DISP) display processed images, an interaction module (INT ) with a user, characterized in that it comprises a target tracking device (100) according to claim 7. 9. Computer program (Pgl) comprising instructions for implementing the target tracking method according to any one of claims 1 to 6, when said program is executed by a processor. A computer-readable recording medium, on which is recorded a computer program comprising program code instructions for performing the steps of the method according to one of claims 1 to 6.