Method for automatically planning a trajectory for a medical intervention

Abstract

The invention relates to a method for automatically planning a trajectory to be followed during a medical intervention by a medical instrument (120) targeting an anatomy of interest (130) of a patient (110), said automatic planning method comprising the steps of: - acquiring at least one medical image of the anatomy of interest (130); - determining a target point (145) on the previously acquired image; - generating a set of trajectory planning parameters from the medical image of the anatomy of interest and the previously determined target point, the set of planning parameters comprising coordinates of an entry point on the medical image. The set of parameters is generated using a machine learning method of neural network type. The invention also relates to a guiding device (150) implementing the set of planning parameters obtained.

Description

[0001] Automated trajectory planning method for a medical intervention

[0002] TECHNICAL FIELD OF THE INVENTION

[0003] The field of the invention is that of assistance in planning a medical intervention.

[0004] More specifically, the invention relates to a method for automatically planning the trajectory of a medical instrument to be performed during a medical intervention and an associated guidance device.

[0005] The invention finds applications, in particular, in medical procedures during which a medical instrument is inserted into an anatomical area of ​​interest, for example, to ablate a tumor in an organ, to perform a biopsy, to carry out vertebroplasty or cementoplasty, or even to stimulate an anatomical area. Such a procedure may optionally be assisted by a medical robot and / or an augmented reality device.

[0006] STATE OF THE ART

[0007] It is known from the prior art of techniques for preparing a medical intervention aimed at reaching a target anatomical area in an anatomy of interest of a patient, such as a lung, kidney, liver, brain, tibia, knee, vertebra, etc.

[0008] Traditionally, medical intervention planning is performed manually by an operator from a medical image obtained by a common medical imaging method.

[0009] During the planning phase, the operator defines a target point in the anatomy of interest and an entry point on the patient's skin near the anatomy of interest; these two points define a straight trajectory for a medical instrument used during the procedure. Such an instrument could be, for example, a needle, a probe, or an electrode.

[0010] The operator must pay close attention to the trajectory of the medical instrument, as it must comply with a number of constraints necessary for the proper execution of the medical procedure. For example, it may be important that the medical instrument does not pass through bones or blood vessels, especially those with a diameter greater than three millimeters, or through vital organs.

[0011] To help the operator choose the entry point based on the target point, planning techniques have been developed in which one or more entry points are automatically proposed to an operator based on predefined constraints, associating each corresponding trajectory with a score based on predefined criteria.

[0012] Such a technique is described, for example, in US patent application number 2017 / 0148213 A1, entitled "Planning, navigation and simulation systems and methods for minimally invasive therapy." The method described in this patent application determines trajectories using a classical image processing algorithm in which images are segmented to minimize constraints related to the trajectory. For example, during brain surgery, the trajectory is determined by optimizing several parameters, such as minimizing the number of fibers impacted, the distance between a cortical sulcus boundary and the target, and the volume of white and / or gray matter displaced by the trajectory.

[0013] However, the major drawback of prior art techniques is that they generally rely on minimizing constraints selected by an operator to create a theoretical model, which is often incomplete and imperfect. Furthermore, they require systematic image segmentation to accurately calculate the various possible trajectories. This segmentation proves imprecise and incomplete in some cases, which can lead to errors in the trajectory used by the medical instrument.

[0014] Furthermore, these techniques do not allow for consideration of possible deformation of the medical instrument, such as a needle, during the insertion of its tip into the patient's body.

[0015] Finally, an experienced operator also regularly intervenes to select in the images the regions to avoid, such as blood vessels, and the regions through which the medical instrument must pass, in order to determine the optimal trajectory of the medical instrument.

[0016] Operator interventions prove tedious and demanding, requiring significant attention and experience in the type of intervention. None of the current systems can simultaneously meet all the necessary requirements, namely, to offer an improved technique for automatically planning a medical intervention to reach a target in a patient's anatomy of interest, independent of an operator, while also providing more precise and reliable planning.

[0017] DESCRIPTION OF THE INVENTION

[0018] The present invention aims to remedy all or part of the drawbacks of the prior art mentioned above.

[0019] To this end, the invention relates to a method for automatically planning a trajectory to be followed during a medical intervention by a medical instrument targeting a patient's anatomy of interest, said automatic planning method comprising the following steps:

[0020] - acquisition of at least one medical image of the anatomy of interest;

[0021] - determination of a target point on the previously acquired image;

[0022] - generation of a trajectory planning parameter set from the image of the anatomy of interest and the previously determined target point, the planning parameter set including coordinates of an entry point on the medical image.

[0023] Such a process, used prior to a medical intervention, provides a set of parameters to guide a physician or surgeon during the manipulation of the medical instrument, which may be a needle, probe, electrode, or any other instrument likely to be inserted into the patient's body, using a patient-related reference point. This reference point is generally three-dimensional to guide the medical instrument in space.

[0024] The goal of the medical intervention is to reach a specific anatomical area of ​​the patient's body, for example, to remove a tumor from an organ, perform a biopsy, carry out a vertebroplasty or vertebroplasty, or stimulate an anatomical area. The target anatomical area is located within or on the surface of a part of the patient's anatomy of interest. Examples of such an area of ​​interest include a lung, kidney, liver, tibia, knee, vertebra, or brain.

[0025] The medical image used for planning was obtained, for example, by computed tomography, magnetic resonance imaging, ultrasound, positron emission tomography, or any other medical imaging method.

[0026] According to the invention, the generation of the parameter set is carried out by implementing a neural network type machine learning method, previously trained on a set of so-called training medical images, each training image comprising an anatomy of interest similar to the anatomy of interest of the patient, each training medical image being associated with coordinates of a target point and at least one entry point previously determined.

[0027] Thus, the planning process can be used by any operator who simply has to select a target point on the medical image.

[0028] It should be noted that the planning process is based on machine learning of similar medical images, each associated with an entry point and a target point.

[0029] A similar medical image is defined as an image obtained using an identical or equivalent imaging method and containing the same anatomy of interest as the medical image acquired on a given individual. It is important to note that the type of medical intervention, the type of medical instrument, or the anatomy of interest targeted may differ without affecting the accuracy of the planning parameters obtained. Machine learning allows for the analysis of a new image to determine an optimal trajectory to the target point selected by the operator on the medical image of the patient's anatomy of interest.

[0030] It is important to note that training medical images are generally associated with entry points actually used during the medical procedure undergone by the individuals and target points actually reached by the instrument after its insertion. To supplement the training medical image set, medical images associated with hypothetical entry points, chosen by an operator, can be added to the set.

[0031] Furthermore, the automatic planning process is advantageously based on training non-segmented medical images, meaning images where all or part of the image is characterized according to the type of tissue, organ, or vessel present in that portion. This makes image processing by the planning process faster. The training set of medical images is generally contained within a database or a medical image bank.

[0032] The automated planning process typically provides planning parameters for at least one possible trajectory. When the automated planning process provides planning parameters for multiple possible trajectories, the operator usually manually selects the trajectory they deem best. It is important to note that a trajectory is generally considered best when it meets a number of criteria specific to the medical procedure, such as the angle of incidence relative to a tissue interface (e.g., skin, liver capsule), the proximity of a blood vessel, organ, or bone structure along the trajectory, and so on.

[0033] It should be emphasized that the automatic planning process is implemented before any medical, surgical or therapeutic procedure.

[0034] In particular embodiments of the invention, the machine learning method determines the coordinates of the entry point from the acquired medical image and the target point previously determined in the acquired medical image.

[0035] In particular embodiments of the invention, the machine learning method first generates a probability of being an entry point for each pixel or voxel of the medical image acquired respectively in 2D or 3D, the coordinates of the entry point corresponding to the coordinates of the pixel or voxel having the highest probability.

[0036] In particular embodiments of the invention, the set of similar medical images comprises a plurality of identical images, each identical image being associated with a distinct entry point.

[0037] Thus, learning is improved because the set of medical images includes possible trajectory variations for the medical instrument.

[0038] Advantageously, the set of similar medical images comprises a plurality of identical images, each identical image being associated with a distinct entry point chosen by a distinct operator.

[0039] Thus, the resulting planning parameters are more accurate because they are less sensitive to the choices of a particular operator. It should be noted that the accuracy of the planning parameters depends on the number of operators involved in analyzing the same medical image during the training phase. Preferably, the set of similar medical images comprises at least three identical images, each identical image being associated with a distinct entry point by a distinct operator.

[0040] Thus, at least three operators are involved in generating the database containing the set of medical images used during the learning phase.

[0041] In particular embodiments of the invention, information relating to the anatomy of interest is associated with each medical image in the set of medical images, the information including a type of anatomy of interest or tumor present in the anatomy of interest, the machine learning method being trained on a portion of the set of medical images restricted to images associated with the same type of anatomy or tumor.

[0042] In particular embodiments of the invention, the automatic planning method also includes a step of assigning a score to a defined trajectory between the entry point of the planning parameter set and the target point previously determined on the acquired image.

[0043] Thus, the operator is assisted in choosing a trajectory from among the possible trajectories provided by the automatic planning process. The score is generally assigned according to criteria specific to the medical intervention.

[0044] The trajectory defined between the input point of the planning parameter set and the predetermined target point on the acquired image is generally rectilinear. However, it is possible for the trajectory to be curvilinear, for example, approximately along a circular arc with a maximum radius of curvature to account for the rigidity of the medical instrument. Generally, a curvilinear trajectory is either concave or convex. In other words, the derivative of a curvilinear trajectory is usually of constant sign, negative or positive, between the input point and the target point.

[0045] Preferably, the trajectory score is assigned based on at least one of the following criteria:

[0046] - proximity to a blood vessel;

[0047] - the proximity of an organ;

[0048] - the proximity of a bony structure;

[0049] - the angle of incidence relative to a tissue interface; - the length of the trajectory;

[0050] - the fragility of a tissue traversed by the trajectory.

[0051] In particular embodiments of the invention, the assignment of the score for the trajectory takes into account a probability that the medical instrument will deform upon contact with a tissue interface.

[0052] This deformation generally occurs when the medical instrument has a flexible part, that is to say, one that is likely to deform upon contact with a tissue interface, such as when inserting the medical device through the patient's skin.

[0053] In particular embodiments of the invention, the assignment of the score for the trajectory takes into account a relapse rate or a recovery time associated with a trajectory similar to the planned trajectory.

[0054] Thus, the score assigned to the trajectory is negatively impacted if the planned trajectory results in a recurrence rate or a recovery time that is too long for the patient.

[0055] In particular embodiments of the invention, the automatic planning method also includes a step of comparing the score assigned to the trajectory with a threshold score, the trajectory being validated when the trajectory score is greater than or equal to the threshold score.

[0056] In particular embodiments of the invention, the automatic planning process also includes a step of modifying the entry point when the score assigned to the trajectory is less than the threshold score.

[0057] In particular embodiments of the invention, the acquired medical image is two-dimensional or three-dimensional.

[0058] In particular embodiments of the invention, the medical image is acquired by magnetic resonance, ultrasound, computed tomography or positron emission tomography.

[0059] The invention also relates to a medical instrument guidance device, comprising means for guiding a medical instrument according to the set of planning parameters obtained by the automatic planning process according to any of the preceding embodiments.

[0060] The guidance system can be robotic, a navigation system (with or without a robotic device), an augmented reality device, a patient-specific guide, or a three-dimensional model of the patient's anatomy. It is important to emphasize that the medical instrument's guidance system assists a practitioner performing the medical procedure.

[0061] BRIEF DESCRIPTION OF THE FIGURES

[0062] Other advantages, purposes and particular features of the present invention will become apparent from the following non-limiting description of at least one particular embodiment of the devices and methods of the present invention, with reference to the accompanying drawings, in which:

[0063] - Figure 1 is a schematic view of a medical intervention during which a medical instrument is guided according to a set of parameters established by an automatic planning process according to the invention;

[0064] - Figure 2 is a synoptic diagram of an automatic planning process according to a particular method of implementing the invention;

[0065] - Figure 3 is an example of a medical image acquired during the first step of the planning process in Figure 2;

[0066] - Figure 4 is an example of a medical image used during the training of the neural network implemented by the process of Figure 2;

[0067] - Figure 5 is a schematic view of a training phase of the neural network implemented by the process of Figure 2;

[0068] - Figure 6 is a schematic view of the unfolding of the neural network implemented by the process of Figure 2, and trained according to the training phase of Figure 5;

[0069] - Figure 7 is a schematic view of the unfolding of the neural network implemented by the process of Figure 2, and trained according to an alternative training phase;

[0070] Figure 8 shows two medical images of the same patient, one with a medical instrument inserted and the other corresponding to the same view without the instrument, used during the training of a neural network configured to define a curvilinear trajectory of a medical instrument. DETAILED DESCRIPTION OF THE INVENTION

[0071] The present description is given by way of non-limiting attribution, each feature of an embodiment being able to be advantageously combined with any other feature of any other embodiment.

[0072] It should be noted from the outset that the figures are not to scale.

[0073] Example of a particular implementation method

[0074] Figure 1 is a schematic view of a medical procedure in which a patient 110 lying on a table 115 is treated using a medical instrument 120. In this non-limiting example of the invention, the medical procedure consists of the ablation of a tumor in an anatomical structure of interest 130, which is in this case the liver of patient 110, using the medical instrument 120, which is a semi-rigid needle. The procedure is a percutaneous one, meaning that the body of patient 110 is not opened. Furthermore, the procedure can be performed according to various treatment parameters. Such treatment parameters include, for example, the duration and power of the ablation treatment, the voltage applied in the case of electroporation treatment, or the frequency applied in the case of radiofrequency ablation.It should be emphasized that the present example is given for illustrative purposes only and that a person skilled in the art can implement the invention described below for any type of medical intervention using any medical instrument aimed at an anatomy of interest to the patient.

[0075] In this example, the medical instrument 120 is advantageously guided by a device 150 along a straight trajectory thanks to the prior development of a set of planning parameters comprising the coordinates of an entry point 140 on the patient's skin 110, and even an angle to follow in a three-dimensional coordinate system linked to the patient 110 to aim at a previously determined target point 145. The set of planning parameters is established by means of an automatic planning method 200 according to the invention, as illustrated in Figure 2 in the form of a block diagram.

[0076] The method 200 for automatically planning the trajectory to be followed by the medical instrument 120 during the medical intervention includes a first step 210 of acquiring at least one medical image of the anatomy of interest 130 of the patient 110.

[0077] Medical imaging is usually taken prior to the medical intervention with equipment dedicated to medical imaging, such as a magnetic resonance imaging (MRI) machine, a computed tomography (CT) scanner, a spectral scanner or an ultrasound scanner.

[0078] An example of a medical image 300 obtained by computed tomography showing a model, commonly referred to by the English term "phantom", corresponding to the anatomy of interest 130 of patient 110 is presented in figure 3. The medical image 300 corresponds to a cross-sectional view of patient 110 along a plane substantially perpendicular to the axis of the spine of patient 110. In addition to the anatomy of interest 130, a vertebra 310 of the spine and six ribs 320 are notably visible in the medical image 300.

[0079] In the previously acquired medical image 300, the target point 145 is determined during a second step 220 of the automatic planning process 200 either manually by an operator or automatically by image analysis.

[0080] Target point 145 is associated with coordinates in medical image 300. These coordinates are two-dimensional or three-dimensional depending on the type of medical image acquired. In the case of a two-dimensional medical image 300, target point 145 corresponds approximately to a pixel of the image. In the case of a three-dimensional medical image 300, target point 145 corresponds approximately to a voxel of the image.

[0081] In order to determine the coordinates of an entry point of a set of trajectory planning parameters to be followed by the medical instrument 120 from the medical image 300 and the target point 145, a machine learning algorithm, here of the neural network type, is loaded during a third step 230 of the automatic planning process 200.

[0082] The neural network was previously trained in a 290-phase learning process on a set of medical training images, each comprising an anatomy of interest similar to the anatomy of interest 130. The medical training images were typically acquired from a cohort of individuals, with each medical training image associated with coordinates of a target point and an entry point previously determined, usually by at least one operator.

[0083] Advantageously, the training medical image set includes multiple copies of the same medical image but associated with distinct entry points determined generally by at least three operators.

[0084] Figure 4 shows an example of the same medical image 400, each time including the same target point 420. This medical image 400, included nine times in the training set of medical images, was processed by three separate operators, 01, 02, and 03, each of whom provided three input points, respectively 41001, 41002, and 41003.

[0085] The training of the neural network can be advantageously restricted to images associated with a given piece of information, such as the type of anatomy of interest or tumor present in the anatomy of interest, in order to increase consistency by decreasing the variability of the planning parameter sets that the neural network can obtain.

[0086] It is important to note that there may be hardware limitations to training a neural network, particularly when the training set of medical images includes three-dimensional images of the anatomy of interest. To overcome these hardware limitations, it is possible to reduce the resolution of each medical image, but this risks reducing the accuracy of the parameters obtained by the neural network. It is also possible to restrict training to trajectories parallel to a predetermined plane, such as a plane perpendicular to the axis of the patient's spine. Another solution to overcome hardware limitations is to use chips commonly known as "tensor processor units," which are dedicated to machine learning.

[0087] The neural network training phase 290, as illustrated in more detail in Figure 5, generally comprises two main steps 510 and 520, which can be repeated, and requires a database 501 containing a set of medical images where each image is associated with an entry point and a target point. Optionally, information about the properties of the instrument used to perform the intervention, such as the instrument length or the instrument stiffness coefficient, is also associated with each medical image in the database 501. After the training phase 290, an optional testing phase 550 can be implemented.

[0088] The 501 medical image database is partitioned into three databases, 502, 503, and 504, each containing separate medical images. These three databases are referred to as the training database, validation database, and test database, respectively.

[0089] In this non-limiting example of the invention, 60 to 98% of the medical images in database 501 are grouped in training database 502, 1 to 20% in validation database 503 and 1 to 20% in test database 504. The percentages, which are generally functions of the number of images in database 501, are given here for illustrative purposes.

[0090] During the first step 510 of the training phase, medical images 515 from the training base 502 are used to determine a weight W and a bias b for each neuron of the neural network 530 used to obtain the coordinates of the entry point of the trajectory planning parameter set.

[0091] To determine the weight W and bias b of each neuron, each medical image 515 from the training set 502 is presented to the neural network 530 in two variants: a first variant 515i containing only the target point Ce, and a second variant 5152 containing both the target point Ce and the predetermined entry point p. From the first variant of the medical image 515i, the neural network 530 then makes a prediction 535 on the position of the entry point p'. The coordinates of the predicted entry point p' are compared to the coordinates of the position of the predetermined entry point p, associated with the second variant of the medical image 5152. The error between the coordinates of the predicted entry point p' and the predetermined entry point p is then used to adjust the parameters W and b of each neuron in the neural network 530. A model 518 is obtained at the end of the first step 510 of the training phase.

[0092] During the second step 520 of the training phase, the medical images 525 from the validation base 503, advantageously distinct from the medical images 515, are used to validate the weight W and the bias b of each neuron in the neural network 530.

[0093] During this second step 520 of the training phase 290, a variant 525i of each medical image, containing only the position of a target point Cv, is presented to the neural network 530. The neural network 530 then makes a prediction 536 on the position of the input point d'. The coordinates of the predicted input point d' are compared to the coordinates of the position of the predetermined input point d, associated with the medical image 525 used for validation. The error between the coordinates of the predicted input point d' and the predetermined input point d is then used to verify the parameters W and b of each neuron in the neural network 530, determined during the first step 510.

[0094] In the event that the prediction error of the neural network is too large at the end of this second step 520, the neural network 530 is re-trained according to the two steps 510 and 520 of the training phase 290 described previously, reusing the same medical training images 515 and validation images 525.

[0095] Alternatively, during the retraining of the neural network 530, the first step 510 uses all or part of the validation images 525. The second step 520 of retraining the neural network uses as many training images 515 as validation images 525 used for the first step 510 of retraining.

[0096] It should be noted that the 530 neural network can be retrained as many times as necessary to reduce prediction error.

[0097] When the two steps 510 and 520 of the training phase 290 are implemented at least once, the final performance of the neural network can be tested during a possible test phase 550 using the medical images 555 from the test set 504. These medical images 555, advantageously distinct from images 515 and 525, allow verification that the neural network 530, as configured with the parameters W and b for each neuron, can predict with good accuracy the coordinates of an entry point in all situations that the neural network 530 is likely to encounter. A comparison is thus performed between the coordinates of the entry point f as predicted by the neural network 530 and the predetermined entry point f in the so-called test medical image 555. This comparison is identical to that performed during the second step 520 of the training phase.However, unlike step 520, this testing phase 550 does not result in a new training cycle for the neural network 530. If the performance of neural network 530 is poor at the end of step 550, training phase 290 is then restarted with a new, untrained neural network. It should be noted that the images 555 used during testing phase 550 are generally carefully selected to cover different positions of the target point and within the anatomy of interest in order to best test the predictive capabilities of the training network 530.

[0098] In an alternative training phase, the neural network can be trained to provide, for each pixel or voxel in a medical image, a probability that it actually corresponds to the input point. The set of medical images used for this alternative training can be the same as the set of medical images used previously. However, it may be preferable for this alternative training to use medical images with multiple input points on the same image. Advantageously, the input points displayed on the same image are determined by at least three distinct operators. The alternative training of the neural network proceeds in three stages similar to the training phase described previously.

[0099] The previously trained neural network makes it possible to determine during the fourth step 240 of the automatic planning process 200 at least one set of planning parameters for the trajectory to be followed by the medical instrument 120 from the analysis.

[0100] In the case where the neural network is trained according to the 290 training phase, the 530 neural network will provide, from the medical image I and the coordinates of the target point T, three-dimensional coordinates (x,y,z) of the entry point in the acquired medical image, as illustrated in figure 6.

[0101] In the case where the neural network is trained according to the alternative training phase, the 530 neural network will provide, from the medical image I and the coordinates of the target point T, a probability for each pixel or voxel of the medical image to be the entry point, as illustrated in Figure 7. The pixel or voxel with the highest probability is then selected as the entry point.

[0102] The automatic planning process illustrated in Figure 2 includes a fifth step, which is implemented once a trajectory is determined using a set of planning parameters generated by the neural network. During this fifth step, a score is assigned to the trajectory defined by the line connecting the entry point and the target point.

[0103] For example, the score assigned to the trajectory ranges from 0 to 100, with a score of 100 corresponding to an ideal trajectory. In variations of this particular embodiment of the invention, the trajectory is curvilinear, obtained, for example, by calculating the most probable trajectory on the acquired, previously segmented medical image, or by a neural network that has previously learned the trajectories followed during previous medical procedures by a similar or identical medical instrument, particularly in terms of rigidity and length. The parameter set then includes additional parameters for defining the predicted trajectory between the entry point and the target point.

[0104] To illustrate these variant embodiments of the invention, Figure 8 shows two medical images 810, 820 of a patient 830, one including and one not including a medical instrument 840. By comparing the two medical images 810 and 820, it is possible to determine the actual trajectory taken by the medical instrument 840. This trajectory can also be determined by recognizing the medical instrument 840 in the medical image 810, for example, by detecting strong variations in intensity or contrast at the pixel / voxel level of the medical image 810 in order to isolate the medical instrument 840 from the medical image 810.

[0105] The trajectory score is generally determined based on criteria that can be ranked in order of importance. It should be noted that the examples of criteria described below are not exhaustive and that other criteria specific to a given medical intervention may be used to determine the trajectory score.

[0106] The trajectory score can, for example, be calculated based on its proximity to a blood vessel. Indeed, when the trajectory of a medical instrument is likely to cross a blood vessel, there is a risk of bleeding. Consequently, the greater the number of blood vessels along the trajectory, the lower the score assigned to the trajectory.

[0107] It is important to note that the size of a blood vessel can be factored into the scoring. For example, if a blood vessel with a diameter of 3 mm or greater is located on or near the path calculated by the neural network, points are automatically deducted from the score—for example, 50 points on a scale of 0 to 100—because these blood vessels can be vital to the patient. When a blood vessel encountered is a vena cava, a portal vein, or the aorta, the score is automatically set to 0, which might be the case, for instance, during the removal of a liver tumor.

[0108] The trajectory score can also be calculated based on the proximity of the trajectory to an organ and / or a bone structure.

[0109] Indeed, for certain procedures, for example on soft tissue, no bony structure should be in the path of the procedure. If this is the case, the score assigned to the path is zero.

[0110] For other procedures, such as those involving bony structures like the knee or shoulder, passing through a bony structure does not negatively impact the assigned score. More specifically, if the trajectory passes through a predetermined bony structure, the assigned score may be increased.

[0111] With regard to organs, the trajectory score is generally reduced when a high-risk organ, such as a lung, intestine, or muscle, is located at least in close proximity to the trajectory. This is also the case when a nerve, bile duct, ligament, tendon, or neighboring organ of interest is located at least in close proximity to the trajectory.

[0112] The trajectory score can also be calculated based on the angle of incidence of the trajectory with a tissue interface at the point of entry.

[0113] For example, when inserting a semi-rigid needle along a trajectory tangential to a tissue interface, such as the skin or the liver capsule, there is a risk that the needle will bend and not follow the planned trajectory. The smaller the angle between the trajectory and the tissue interface, the lower the trajectory score. This criterion can be interpreted as the optimal trajectory corresponding to an angle between the tissue interface and the trajectory greater than 20°.

[0114] The trajectory score can also be calculated based on the angle of incidence of the trajectory with a bony structure.

[0115] For example, in the case of an intervention on a bony structure, there is a risk that the medical instrument will slip on the bone when inserted tangentially to it. The criterion then translates to the fact that the greater the angle between the trajectory and the bony structure, the lower the trajectory score.

[0116] The trajectory score can also be calculated based on the trajectory length to minimize the length of the path and the inherent risk of causing damage to the patient's body. The trajectory score can also be calculated based on the fragility of the tissue traversed.

[0117] For example, in the specific case of an intervention on a patient's brain, the trajectory score may be reduced if the planned trajectory passes through fragile tissues.

[0118] In the case of a semi-rigid needle insertion, the score can also be calculated based on the probability of needle deformation during insertion. This probability is calculated using information about the type of needle used, such as its length, stiffness coefficient, or bevel shape, combined with previously determined information, namely the type of tissue traversed, the angle of incidence, and / or the length of the trajectory.

[0119] It is important to note that to calculate the trajectory score, the acquired medical image may have been previously segmented to identify the different types of elements present in the image, such as tissue, blood vessels, bone structures, etc., located on or near the defined trajectory between the predicted entry point and the predetermined target point. This segmentation of the acquired image is used only when the medical instrument's trajectory is generated, and not during trajectory generation by the neural network.

[0120] The trajectory score, obtained from criteria specific to the medical intervention, can be weighted according to a recurrence rate and / or with a recovery time.

[0121] Regarding the relapse rate, the score obtained is reduced when the trajectory planned by the neural network is similar to a trajectory used with the same treatment parameters during previous medical interventions implementing the same medical instrument, and for which the relapse rate of individuals who underwent these medical interventions is notable.

[0122] Similarly, regarding recovery time, the score obtained is reduced when the recovery time previously observed for individuals who underwent a medical intervention with a trajectory similar to the planned trajectory is high, for example more than three days.

[0123] The score assigned to the planned trajectory is then compared to a threshold score during a sixth step 260 of the automatic planning process 200. For example, on a scale of 0 to 100, the planned trajectory can only be validated if the assigned score is greater than or equal to 50. Preferably, the planned trajectory is validated if its score is greater than or equal to 70.

[0124] If the trajectory score is lower than the threshold score, the operator can manually modify the entry point during a possible seventh step 270 of the automatic planning process 200. The modification is made, for example, via a graphical interface until the score of the modified trajectory is higher than the threshold score.

[0125] Alternatively, the trajectory can be modified automatically using a gradient algorithm, a graph algorithm, or any other optimization algorithm (Momentum, Nesterov Momentum, AdaGrad, RMSProp, Adam, etc.).

[0126] Finally, when the score of the trajectory provided by the neural network, possibly modified, is greater than or equal to the threshold score, the trajectory is validated during an eighth step 280 of the automatic planning process 200.

[0127] The validated trajectory can then be used during the medical intervention to guide the insertion of the medical instrument 120 into the anatomy of interest 130 of the patient 110 with very good accuracy and with the maximum chance that the medical intervention will go as smoothly as possible.

[0128] It should be noted that the reference frame used for guidance generally corresponds to table 115 on which patient 110 is lying. The coordinates of the target point are advantageously transferred to the guidance frame, in which characteristic points of patient 110 have been previously calibrated. This transfer and calibration operation of the guidance frame is common practice.

[0129] The guidance device 150 can then be used to guide the medical instrument 120 by following the validated trajectory planning parameter set.

[0130] The guidance device 150 can be robotic, a navigation system associated or not with a robotic device, an augmented reality device, a patient-specific guide 110 or a three-dimensional model of the patient's anatomy 110.

[0131] The augmented reality device could be, for example, a pair of glasses in which the planned trajectory is projected onto at least one lens. The augmented reality device could also be a screen placed near the patient, displaying the planned trajectory. The augmented reality device could also include a projector projecting the planned trajectory onto the patient's body or be a holographic device. The guidance system could include optical navigation means, electromagnetic navigation means, or an inertial measurement unit (IMU) with acceleration and rotation sensors.

[0132] The validated trajectory planning parameter set can be used to construct a patient-specific guide. This guide is typically used during open surgery of a bony structure. It is important to note that the patient-specific guide is a personalized, single-use medical device, usually 3D printed. The patient-specific guide helps prevent inaccuracies during surgery, ensuring the procedure is performed as planned. The guide generally conforms to the shape of the bony structure corresponding to the anatomy of interest and guides the insertion of a medical instrument according to the planned trajectory orientation.

[0133] In certain medical procedures, a three-dimensional model of the patient's anatomy, commonly referred to as a "phantom," is constructed for training purposes prior to the procedure. This three-dimensional model can then advantageously include an indication of the entry point of the medical instrument, as defined in the trajectory planning parameters obtained through the automatic planning process.

[0134] The results obtained by the automated planning process can also be used to present the medical intervention to peers, such as physicians, surgeons, or radiologists, or even to the patient who is to undergo the procedure. These results can also be used to train peers on the medical intervention.

Claims

Demands 1. A method (200) for automatically planning a trajectory to be followed during a medical intervention by a medical instrument (120) targeting an anatomy of interest (130) of a patient (110), said automatic planning method comprising the steps of: - acquisition (210) of at least one medical image (300) of the anatomy of interest; - determination (220) of a target point (145) on the previously acquired image (300); - generation (240) of a trajectory planning parameter set from the medical image of the anatomy of interest and the previously determined target point, the planning parameter set comprising coordinates of an input point (140) on the medical image (300); characterized in that the generation of the parameter set is carried out by implementing a neural network type machine learning method (530), previously trained on a set of so-called training medical images, each training medical image comprising anatomy of interest similar to the anatomy of interest (130) of the patient (110), each training medical image being associated with coordinates of a previously determined target point and at least one input point.

2. Automatic planning method according to claim 1, wherein the machine learning method determines the coordinates of the entry point from the acquired medical image and the target point previously determined in the acquired medical image.

3. Automatic planning method according to any one of claims 1 to 2, wherein the machine learning method first generates a probability of being an entry point for each pixel or voxel of the medical image acquired respectively in 2D or 3D, the coordinates of the entry point corresponding to the coordinates of the pixel or voxel having the highest probability.

4. An automated planning method according to any one of claims 1 to 3, wherein the set of similar medical images comprises a plurality of identical images, each identical image being associated with a distinct entry point.

5. An automatic planning method according to any one of claims 1 to 4, wherein the set of similar medical images comprises a plurality of identical images, each identical image being associated with a distinct entry point chosen by a distinct operator.

6. Automatic planning method according to any one of claims 1 to 5, wherein information relating to the anatomy of interest is associated with each medical image in the set of medical images, the information including a type of anatomy of interest or tumor present in the anatomy of interest, the machine learning method being trained on a portion of the set of medical images restricted to images associated with the same type of anatomy or tumor.

7. Automatic planning method according to any one of claims 1 to 6, also comprising a step of assigning a score to a defined trajectory between the entry point of the planning parameter set and the target point previously determined on the acquired image.

8. An automatic planning method according to claim 7, wherein the acquired image is mapped, and the trajectory score assignment is a function of at least one of the following criteria: - proximity to a blood vessel; - the proximity of an organ; - the proximity of a bony structure; - the angle of incidence relative to a tissue interface; - the length of the trajectory; - the fragility of a tissue traversed by the trajectory.

9. Automatic planning method according to any one of claims 7 to 8, wherein the assignment of the score for the trajectory takes into account a probability that the medical instrument will deform upon contact with a tissue interface.

10. Automatic planning method according to any one of claims 7 to 9, wherein the assignment of the score for the trajectory takes into account a recidivism rate associated with a trajectory similar to the planned trajectory.

11. An automatic planning method according to any one of claims 7 to 10, wherein the scoring for the trajectory takes into account a recovery time associated with a trajectory similar to the planned trajectory.

12. Automatic planning method according to any one of claims 7 to 11, also comprising a step of comparing the score assigned to the trajectory with a threshold score, the trajectory being validated when the trajectory score is greater than or equal to the threshold score.

13. Automatic planning method according to any one of claims 7 to 12, also comprising a step of modifying the entry point when the score assigned to the trajectory is less than the threshold score.

14. A medical instrument guidance device, comprising means for guiding a medical instrument according to the planning parameter set obtained by the automatic planning method according to any one of claims 1 to 13.

15. A guidance device according to claim 14, being either a robotic guidance device, a navigation system associated or not with a robotic device, an augmented reality device, a patient-specific guide, or a three-dimensional model of the patient's anatomy.

Images