Apparatus for assisting in planning minimally invasive skeletal surgery

Abstract

The invention relates to a data processing device for implementing a method for assisting in selecting at least one optimal trajectory for a minimally invasive bone surgery. Specifically, the device is configured to confirm (102) a target point to be reached in a patient's bone in a three-dimensional anatomical model and to perform at least one iteration, comprising the steps of:-determining (103) a sampling region and a sampling resolution of candidate entry points,-removing (104) candidate trajectories according to at least one predetermined validity criterion,-for each remaining candidate trajectory,-determining (104) the target point to be reached in the patient's bone and performing at least one iteration. Calculating (105) at least one cost function from the plurality of quantized features, displaying (106) candidate trajectories on the anatomical model using a means of quickly visualizing the cost function, selecting (107) an optimal trajectory from the displayed candidate trajectories.

Description

Technical Field

[0001] This patent application relates to the field of minimally invasive medical surgical planning and may involve surgeries assisted by medical robots. Specifically, a data processing device is provided for implementing a method for assisting in selecting an optimal path for a minimally invasive medical surgery in a patient's target anatomical structure. Background Art

[0002] To prepare for a minimally invasive medical surgery aimed at using a medical device to reach a target anatomical region within a patient's target anatomical structure, a doctor typically uses pre-operative or intra-operative medical images to plan the surgery. Specifically, a minimally invasive medical surgery may aim to perform a biopsy or ablation of a tumor in an organ or bone, to perform vertebroplasty or kyphoplasty, or even to stimulate a specific anatomical region. For example, the target anatomical structure can be the lung, kidney, liver, brain, vertebra, tibia, knee, etc. The medical device can be a needle, electrode, probe, etc.

[0003] For example, pre-operative or intra-operative medical images are obtained by computed tomography, magnetic resonance imaging, ultrasound, or positron emission tomography.

[0004] When planning the surgery, the doctor defines a target point in the treatment area of the target anatomical structure. The doctor also defines an entry point for the medical device on the patient's skin. These two points then define the path that the medical device must follow to perform the medical surgery.

[0005] Depending on the type of surgery, certain restrictions must be met. For example, it may be important that the medical device does not pass through high-risk anatomical structures (such as organs, bones, or blood vessels).

[0006] The path is usually defined empirically based on the doctor's knowledge and the information that can be gathered from the pre-operative images. However, this path is not necessarily optimal, i.e., the path that ensures effective treatment of the patient with minimal risk.

[0007] For a given surgery, there are theoretically an infinite number of possible paths. However, the doctor must select a path within a limited and sometimes very short time, especially when performing surgery planning just before the surgery.

[0008] In addition, this task must be completed based on the patient's volume image and is thus in three dimensions, while the traditional method of presenting such images is to provide two-dimensional slices of the volume, which makes it more difficult for the doctor to find the optimal path. Moreover, when the surgery requires defining multiple paths (such as for inserting multiple needles), it becomes even more psychologically complex for the doctor to plan it.

[0009] Ultimately, the quality of path optimization highly depends on the doctor and the time that can be allocated to this task.

[0010] Methods for evaluating the paths proposed by a doctor already exist, for example based on the risk of using a medical device to cross a high-risk area. Such solutions do not enable the doctor to quickly find the optimal or near-optimal path for the medical procedure in question.

[0011] Methods for automatically suggesting entry points, target points, and / or paths to a doctor also exist. However, these solutions generally do not give the user the flexibility to view the optimality of a large number of paths and select a different path that may be sub-optimal but is preferred for various reasons (which may vary from one patient to another and which may not necessarily be considered by the optimality criteria in question). Summary of the Invention

[0013] The aim of the solution provided by this patent application is to overcome all or part of the drawbacks of the prior art, especially those mentioned above.

[0014] To this end, and according to a first aspect, there is provided a data processing device comprising a processor, a computer memory, and a display screen. The computer memory contains program code instructions which, when executed by the processor, configure the processor to implement a method for assisting in the selection of at least one optimal path for a minimally invasive medical procedure for treating a patient's bone. The processor is configured to:

[0015] - Obtain a three-dimensional anatomical model of the patient from pre-acquired medical images of the patient, the anatomical model comprising a representation of the patient's body outer surface and the bone to be treated;

[0016] - Identify at least one target point to be reached within the bone to be treated in the anatomical model;

[0017] - Perform at least one iteration, wherein the processor is configured to:

[0018] o Determine a sampling area and a sampling resolution for sampling candidate entry points on the patient's body outer surface in the anatomical model, each candidate entry point belonging to the sampling area, the sampling resolution representing the minimum distance between two candidate entry points, and each pair formed by a target point and a candidate entry point forms a candidate path;

[0019] o Remove candidate paths according to at least one pre-determined validity criterion;

[0020] o For each remaining candidate path, quantify a plurality of pre-determined features one by one, including the minimum distance between the part of the candidate path located inside the bone to be treated and the cortical bone wall, and calculate at least one cost function from the quantified features;

[0021] o On the display screen, superimposed on the anatomical model, display at least some of the remaining candidate paths, and for each candidate path displayed, there is a means to quickly view the cost function value;

[0022] o Obtain the user's selection of the path to the at least one target point, which path is considered the optimal path among the displayed candidate paths.

[0023] These provisions enable the suggestion of an optimal or near-optimal set of minimally invasive surgical paths to a doctor based on the three-dimensional anatomical model of the patient. The suggested paths take into account various types of constraints (such as practical constraints, performance-related constraints, and / or safety constraints). The doctor can choose a path that may be sub-optimal in terms of a predefined criterion but more preferable in terms of constraints not covered by that criterion. The invention can also quickly and optimally explore all solutions to determine candidate paths. It enables the user to quickly view various candidate paths and their optimality criteria. The proposed solution is flexible because it can adapt to the specific needs of the surgery. In addition, the doctor can also update the optimality criterion (cost function) based on user feedback (evaluation of the selected surgical path).

[0024] In a specific embodiment, the device may further comprise one or more of the following features, implemented individually or in any technically possible combination.

[0025] In a specific embodiment, configure the processor to perform at least two iterations. For the nth iteration, where n is an integer greater than one:

[0026] - The sampling region for the nth iteration is defined to be smaller than the sampling region defined for the (n - 1)th iteration and includes the candidate entry points corresponding to the path selected in the (n - 1)th iteration; and

[0027] - The sampling resolution for the nth iteration is defined to be finer than the sampling resolution defined for the (n - 1)th iteration.

[0028] In a specific embodiment, the sampling resolution for the nth iteration is defined based on the variability of the cost function calculated in the (n - 1)th iteration near the path selected in the (n - 1)th iteration.

[0029] In a specific embodiment, the sampling region for the nth iteration is defined based on the variability of the cost function calculated in the (n - 1)th iteration near the path selected in the (n - 1)th iteration.

[0030] In a specific embodiment, to sample the candidate entry points, the processor is configured to determine, in the anatomical model and on the outer surface of the patient's body, a set of points defined by a spherical coordinate system centered on the target point. Each point is defined by a distance r from the target point i,j and two angles α i and β j where the angles α i and β j are defined such that the indices i and j correspond to strictly positive integers:

[0031] [Equation.1]

[0032]

[0033] and

[0034] [Equation.2]

[0035]

[0036] where R represents the minimum distance between two candidate entry points on the outer surface of the patient's body.

[0037] In a specific embodiment, a validity criterion allows for verifying at least one of the following for a given candidate path:

[0038] - The validity of the length of the candidate path for the medical device planned for the procedure,

[0039] - The ability to configure the robotic arm with the medical device so that the medical device will be able to follow the candidate path,

[0040] - The presence of an object equipped on the patient obstructs the candidate path,

[0041] - The candidate path intersects at least one key anatomical structure in the patient's body.

[0042] In a specific embodiment, multiple validity criteria are considered to remove candidate paths, and different validity criteria are evaluated in the order defined for each validity criterion according to the estimated computation time required to evaluate each validity criterion.

[0043] In a specific embodiment, multiple features of the candidate path include at least one of the following features:

[0044] - The minimum distance between the candidate path and a key anatomical structure in the patient's body,

[0045] - The length of the intersection portion between the candidate path and the bone to be treated,

[0046] - The minimum incident angle between the candidate path and the anatomical interface traversed by the candidate path,

[0047] - The minimum distance between the outer surface of the treatment area and the outer surface of the ablation area estimated for the candidate path,

[0048] - The angular difference between the candidate path and the main axis of the treatment area.

[0049] In a particular embodiment, a plurality of cost functions are calculated based on the quantified features, each cost function being calculated using a different set of weights assigned to different quantified features, and the processor is configured to obtain an indication of the cost function to be considered by the user when displaying the candidate path.

[0050] In a particular embodiment, the means for quickly viewing the cost function values of each displayed candidate path includes color coding that associates different colors with different cost function values.

[0051] In a particular embodiment, the processor is configured to display the quantification determined by the user for the features of the selected path.

[0052] In a particular embodiment, the processor is configured to, for at least one of the displayed candidate paths, obtain an evaluation of the candidate path by the user and update the cost function based on the evaluation.

[0053] In a particular embodiment, the processor is configured to:

[0054] - After the surgery, in the anatomical model, determine the actual path followed by the medical device during the surgery based on the medical images of the patient obtained when the medical device was in place;

[0055] - Determine the cost function value of the actual path;

[0056] - Calculate the difference between the cost function value of the candidate path selected for the surgery and the cost function value of the actual path.

[0057] In a particular embodiment, the processor is configured to compare the calculated difference with a pre-determined threshold and display an indication related to the calculated difference value.

[0058] In a particular embodiment, the processor is configured to:

[0059] - Obtain an evaluation of the actual path by the user,

[0060] - Determine a statistical value representing the variability of the cost function based on the calculated difference and the obtained evaluation,

[0061] - Remove the candidate paths that are considered equivalent based on the obtained statistical value. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] The present invention will be more readily understood by reading the following description, which is only a completely non - restrictive example and with reference to Figures 1-7 , which shows:

[0063] Figure 1 A schematic diagram of a data - processing device according to the present invention.

[0064] Figure 2 A schematic diagram of the main steps for assisting in selecting the optimal path for a medical operation.

[0065] Figure 3 A schematic diagram of the definition of the orbital angle and the head - to - tail angle to determine the sampling area.

[0066] Figure 4 A schematic diagram of a set of candidate entry points on the outer surface of a patient's body in an anatomical model.

[0067] Figure 5 A schematic diagram showing a set of candidate paths for a medical operation, the optimality criterion of which is shown by a scale, here shown as a gray - level display.

[0068] Figure 6 A schematic diagram of additional steps considering the user's evaluation of the candidate paths.

[0069] Figure 7 A schematic diagram of additional steps considering the difference between the selected path and the actual path followed by a medical device.

[0070] Figure 8 A schematic diagram of a bone to be treated and two candidate paths to reach the target point within the bone.

[0071] In these figures, the same reference numerals from one figure to another denote the same or similar elements. For the sake of clarity, unless otherwise stated, the elements shown are not necessarily to the same scale.

[0072] DETAILED DESCRIPTION OF AT LEAST ONE EMBODIMENT OF THE INVENTION

[0073] Figure 1 Shows a data - processing device 10 according to the present invention. The data - processing device 10 includes at least one processor 11, at least one computer memory 12, and at least one display screen 13.

[0074] The computer memory 12 contains program - code instructions which, when executed by the processor 11, configure the processor 11 to implement a method for assisting in selecting at least one optimal path that a medical device must follow during a minimally invasive medical operation within a patient's target anatomical structure. ​​​​​​​​

[0075] Specifically, minimally invasive medical procedures can be aimed at achieving tumor biopsies or ablations in organs or bones to treat bone diseases, perform vertebroplasty or kyphoplasty, or even stimulate specific anatomical regions. The target anatomical structure can correspond to an organ or a bone, such as the lung, kidney, liver, brain, vertebra, tibia, femur, hip bone, knee, pelvic bone, pelvis, etc. The medical device can be a needle, electrode, probe, drill, trocar, screw, etc.

[0076] Figure 2 Shown are the main steps of a method 100 for assisting in selecting an optimal path for a minimally invasive medical procedure implemented by a processing device 10.

[0077] As Figure 2 shown, the method 100 includes a step 101 of obtaining a three-dimensional anatomical model of the patient. Specifically, the anatomical model contains a representation of the patient's body outer surface (or a part of the body outer surface) and the target anatomical structure.

[0078] The anatomical model is obtained from pre-acquired medical images of the patient. The medical images used to generate the anatomical model can be obtained, for example, by computed tomography, magnetic resonance imaging, ultrasound, or positron emission tomography (any medical imaging technique that allows reconstruction of a volume in three dimensions can be used). Multiple images are typically required to generate a three-dimensional anatomical model. Alternatively, a three-dimensional anatomical model can also be generated from a single image and a statistical model of the patient.

[0079] The anatomical model can be directly generated by the processing device 10, but optionally it can also be generated by another separate device and transmitted to the processing device 10 by means of communication.

[0080] Traditionally, three-dimensional modeling is performed, for example, using a method of segmenting anatomical structures considered relevant to clinical practice in medical images. The result of the segmentation can take the form of a binary image or a mask. For example, in the context of minimally invasive ablation of a liver tumor, it may be related to segmenting the liver, tumor, lung, gallbladder, bile duct, digestive system organs, bones, certain blood vessels (those with a significant diameter), and the patient's body outer surface. According to another example, in the context of performing minimally invasive surgery on a bone, it may also be related to segmenting the outer surface of the bone to be treated, its cortical part (cortical bone corresponds to the particularly hard peripheral part of the bone; it is the "outer shell" of the bone, i.e., the relatively thick wall of the bone) and / or its cancellous part (cancellous bone corresponds to the bone tissue that forms the porous part of the bone, which is located below the cortical bone). Such three-dimensional anatomical modeling can be represented mathematically as follows:

[0081] [Mathematical formula.3]

[0082]

[0083] where M is a three-dimensional anatomical model, and S i is a segmentation representation considering the generation of the model, and all segmentations are represented within the same anatomical framework.

[0084] Segmentation methods are typically automatic artificial intelligence methods (requiring no user input), semi-automatic artificial intelligence methods (then requiring a single user input, such as points or segmentations), or interactive artificial intelligence methods (by iteratively correcting the segmentation results by the user). Manual segmentation can also be considered, but the time required to perform such segmentation means that this alternative is less relevant to the clinical environment.

[0085] In artificial intelligence methods, convolutional neural networks are currently the most effective. However, other methods may also be advantageous alternatives, such as methods based on, for example, random forests, k-means clustering, watershed segmentation, active contour models, level set methods, etc.

[0086] As Figure 2 shown, the method 100 includes step 102: identifying in the anatomical model target points to be reached within a treatment area (e.g., a tumor) in a patient's target anatomical structure (e.g., the liver). It should be noted that the case of multiple target points that must be considered can also be considered (e.g., when multiple paths must be selected for the insertion of multiple needles). In this case, the set of target points can be determined from a "virtual" target point and the geometric arrangement that the target points must have relative to each other and relative to the virtual target point (e.g., the target points must correspond to the vertices of a specific geometric shape centered on the virtual target point, or actually have a specific spatial distribution around the virtual target point). The virtual target point and the geometric arrangement that the target points must have are input data provided by the user.

[0087] Specifically, the ablation area to be treated can be segmented automatically or semi-automatically. Manual segmentation or correction tools can be used according to the required accuracy. Once a satisfactory segmentation of the treatment area to be treated is obtained, the positions of the target points to be reached can be estimated. It should be noted that the treatment area to be treated can be segmented by the processing device 10, or can also be pre-segmented in the anatomical model by another separate device and then transmitted to the processing device 10 by means of communication.

[0088] For example, segmentation of the treatment area can be used to automatically determine target points (e.g., the center of a circumscribed sphere or ellipsoid determined by calculating the centroid or known parameters). Alternatively, when the surgery involves inserting multiple medical devices, it is conceivable to use segmentation to determine a set of multiple target points, e.g., based on a virtual target point (e.g., corresponding to the center of the treatment area), the geometric arrangement that the target points must have, and / or the desired ablation coverage (specifically, for a given medical device, the ablation area that can be obtained around the target point can be predicted to ensure that it covers the area to be ablated).

[0089] Depending on the type of treatment envisaged (radiofrequency ablation, microwave ablation, cryotherapy or electroporation, biopsy, vertebroplasty, etc.), the target points can be located inside or at the periphery of the treatment area. When the medical device is located at the target point, the position of the target point can optionally be estimated in this way, based on a model of the ablation area that is easily accessible.

[0090] The candidate paths for the surgery will be determined based on one or more target points defined in this way and a set of candidate entry points on the patient's skin. Thus, the set of entry points on the skin forms an initial domain, based on which the optimal path must be confirmed. Specifically, each pair consisting of a target point and a candidate entry point forms a candidate path. Therefore, it is necessary to confirm the candidate entry points on the patient's skin. This can be achieved by a method such as the Otsu method, which is well-suited for voxel classification in the case of a bimodal luminance distribution. However, there are an infinite number of candidate entry points in a continuous space. Even in a discrete domain, for the image resolutions used in clinical practice, there are a large number of candidate entry points. Therefore, sampling is preferably performed to reduce the number of candidate entry points.

[0091] For this purpose, and as Figure 2 shown, the method 100 includes step 103: in the anatomical model, on the outer surface of the patient's body, determine a sampling area and a sampling resolution for sampling candidate entry points.

[0092] The sampling resolution represents the minimum distance between two candidate entry points. A compromise must be found between the fineness of the sampling and the computational time required to evaluate each candidate path associated with each candidate entry point. A very fine "brute force" sampling method for candidate entry points would require very fast calculation of path evaluations, which would lead to strict limitations (limitations of the optimization criteria and / or use of expensive computational means). In addition, the user would need to view a large amount of information.

[0093] Therefore, a sampling resolution can be advantageously considered such that two consecutive paths are sufficiently different from each other. The sampling resolution can be determined by default (for example, it may be required that the minimum distance between two entry points on the body surface of the patient is at least five millimeters), but can also be adjusted by the user as needed (a trade-off between computation time and required accuracy).

[0094] To sample candidate entry points, specifically, it can be envisaged that in the anatomical model, on the outer surface of the patient's body, a set of points defined by a spherical coordinate system centered on the target point is determined (when considering multiple target points, it can be envisaged that each target point is processed sequentially with the aim of determining an optimal path for each target point; alternatively, the set of target points can also be processed together, and in this case, candidate entry points can be sampled relative to a "virtual" target point, which corresponds, for example, to the equidistant point of each target point). Then each point can be defined by the distance r i,j from the target point i and two angles α j and β max and β max . Then the sampling area can correspond to the intersection of the outer surface of the body with a conical angular aperture centered on the target point (described by α 0,0 and β i and β i ). For example, the calculation can be initialized from the first point of the path on the skin corresponding to α0 = β0 = 0; the distance between this point and the origin of the coordinate system (the target point) is denoted as r i and β j . Sampling can be obtained by calculating the angles α

[0095] [Equation 1]

[0096]

[0097] and

[0098] [Equation 2]

[0099]

[0100] in radians, such that the increment between each angle corresponds to the arc length on the outer surface of the patient's body, which corresponds to a predefined resolution parameter. Specifically, for subscripts i and j corresponding to positive integers, the angles α i and β j can be defined as:

[0095] [Equation 1]

[0096]

[0097] and

[0098] [Equation 2]

[0099]

[0100] where R is a predefined resolution parameter representing the minimum distance between two candidate entry points on the outer surface of the patient's body. The angles α and β are continuously increased until they cover the required conical angular aperture.

[0101] As Figure 3As shown, the angle α can correspond to an orbital angle measured with respect to the anteroposterior axis 53 in the transverse plane 51 of the patient 50. The transverse plane 51 is defined by the target point, the anteroposterior axis 53, and the transverse axis 54 of the patient. The angle β can correspond to the cranio-caudal angle formed by the transverse plane 51 and the oblique plane 52. The oblique plane 52 is defined by the target point, the transverse axis 54, and a candidate path 55 formed by the target point and a related entry point.

[0102] Figure 4 Schematically shown is a set of candidate entry points 22 on the outer surface 21 of the patient's body in the anatomical model (the candidate entry points 22 form a "grid" on the outer surface 21 of the patient's body).

[0103] As mentioned above, reducing the number of candidate paths is beneficial for limiting the total computational time of the method for selecting the optimal path. To this end, and as Figure 2 shown, the method 100 includes step 104: removing certain candidate paths according to one or more predetermined validity criteria.

[0104] It is beneficial to consider multiple different validity criteria and evaluate them in a predefined order. This order is defined according to the estimated computational time required to evaluate each validity criterion.

[0105] The problem lies in applying successive simple filters to reduce the number of candidate paths. These filters are computationally simple and can quickly rule out certain impossible paths. These filters are applied in order from the simplest to the most complex in terms of computational time, so that the last filter is applied to as few candidate paths as possible.

[0106] As a non-limiting example, the following are validity criteria that can be considered in sequence.

[0107] In the initial stage, if the medical procedure is assisted by a robot comprising a robotic arm equipped with a medical device, those candidate paths that cannot be configured to move the medical device along the path in question can be removed. Once sampling has been carried out, the candidate paths are filtered based on prior knowledge of the robot's ability to reach the orbital angle and the cranio-caudal angle, as well as the target point in question. This stage depends on the robot model used, the geometry of the tool used to guide the medical device, and the position of the target point. These parameters are known when the candidate paths are determined and can thus significantly reduce the number of candidate paths.

[0108] In the second stage, those candidate paths whose path length is greater than the length of the medical device envisaged for the procedure can be removed.

[0109] In the third stage, those candidate paths that are blocked by objects worn by the patient can be removed. As Figure 4As shown, for example, such an object may be a patient reference 23 for an optical navigation system to guide a robotic arm. According to another example, it may also be a catheter. By segmenting the patient and the objects equipped for the operation, as well as prior knowledge of the three-dimensional geometry of the robotic arm, paths that would cause a collision between the robotic arm and the objects equipped for the operation on the patient can be determined. Such paths can be removed.

[0110] In the fourth stage, candidate paths that intersect at least one critical anatomical structure within the patient can be removed. The "critical anatomical structure" refers, for example, to an organ different from the target anatomical structure (e.g., the lung, spleen, gallbladder, or kidney if the target anatomical structure is the liver) or a high-risk vascular structure (artery, vein, bile duct, digestive tract). Generally speaking, segmentation is not parametrically represented. Therefore, in order to detect the collision between the high-risk anatomical structure and the path, the latter can be finely sampled, and the segmentation value at each sampling point can be determined. A distance equal to half of the image resolution in the acquisition direction (usually on the order of about one millimeter) between points on the path is sufficient to achieve this purpose.

[0111] The validity criteria listed by way of examples above may apply to both minimally invasive surgery on soft organs and minimally invasive surgery on bones.

[0112] As Figure 2 shown, method 100 includes step 105: calculating at least one cost function for each remaining candidate path (i.e., each candidate path not removed in removal step 104) based on multiple features related to, for example, the safety or performance of the operation. These features are quantifiable, that is, a measurement and standardized value is assigned to each feature under discussion (in other words, each quantified feature corresponds to an index representing a safety or performance criterion).

[0113] As a non-limiting example, all or some of the following features can be considered when calculating the cost function. Unless otherwise specified, each of these features can be considered for minimally invasive surgery on soft organs or minimally invasive surgery on bones.

[0114] According to the first example, the minimum distance between the candidate path and the critical anatomical structure within the patient can be considered. For this purpose, the anatomical model can include the segmentation of the internal critical anatomical structure. Then, the distance transform (also known as the distance map) of the relevant segmentation can be used to calculate the distance between each point on the path and the given anatomical structure. When this transform is applied to a binary image, the result is a new image where each voxel value corresponds to the shortest distance between the voxel and the image edge. The minimum distance between the candidate path and the internal critical anatomical structure corresponds to a safety margin that must take into account factors affecting the imprecision of the medical robot and / or biomechanical variables that cannot be controlled during the insertion of the medical device.

[0115] According to another example, it is conceivable to consider the length of the intersection portion of the candidate path with the target anatomical structure. This feature is related to the safety criteria of a medical procedure. Specifically, it may require a minimum distance for a medical device to pass through the target anatomical structure to prevent the spread of tumor cells outside the target anatomical structure during its removal. This feature is also related to performance criteria because the longer the intersection portion between the candidate path and the target anatomical structure, the better the stability of the medical device within the target anatomical structure during the procedure. Therefore, a compromise needs to be found based on the clinical goal being targeted; this compromise may vary for each procedure. In the case where the goal of minimally invasive surgery is to insert a screw into a bone, the minimum depth of screw insertion into the bone is also related to the stability of the screw within the bone.

[0116] According to another example, the minimum incident angle between the candidate path and the anatomical interface through which the candidate path passes can be considered. When inserting a medical device (such as a needle) in minimally invasive surgery, the device may pass through certain interfaces such as the skin or the wall of certain organs (such as the liver capsule). The incident angle of the device with these interfaces may affect the effectiveness of device placement (potential device deflection or organ movement). This parameter also has an impact on safety because an incident angle that is too tangential to the anatomical interface may promote the formation of a hematoma. Calculating the incident angle requires knowing the surface normal at the path entry point. Methods based on surface meshing exist, but they can be expensive, limiting their use in a clinical setting. To reduce the calculation time, the normal can be calculated as the gradient of the distance variation of the segmentation evaluated at the entry point. It is recommended to calculate the gradient by convolution with the derivative of a Gaussian function in order to obtain a more robust normal estimate against noise (sensitivity to noise is a drawback of the finite difference method). In bone surgery, too small an incident angle may increase the risk of the medical device sliding on the bone.

[0117] According to another example, an estimated value representing the stability of a medical device positioned along the candidate path to a target point under the influence of its own weight can be considered. Specifically, this value can be calculated based on the insertion length of the path in question: the shorter the insertion length, the easier it will be for the medical device to move under the action of its own weight or an external force. This value can also be calculated using a biomechanical model that provides, for example, viscoelastic information of the organs passed through. A path that preferably results in the minimum risk of the medical device moving under its own weight is selected. This feature is more relevant to soft organs compared to bones.

[0118] According to another example, when the minimally invasive procedure corresponds to ablation, the minimum distance between the outer surface of the region to be treated and the outer surface of the ablation region estimated for the candidate path can be considered. This minimum distance corresponds to the estimated ablation boundary. For a given medical device, models of the ablation region are available. These models allow the estimation of the theoretical coverage area obtained for each path. For example, it is advantageous to optimize the coverage of the cancer region by ablation (maximizing the boundary between the ablation region and the cancer region), while minimizing damage to healthy parenchymal tissue and nearby organs. Generally, the models of the ablation region are elliptical and eccentric with respect to the distal end of the device. The shape and orientation of the tumor are important aspects to consider (the tumor is not necessarily spherical). Therefore, in order to obtain better coverage of the tumor, certain paths may be preferred depending on their orientation and the position of the distal end of the device at the end of the path.

[0119] According to another example, the angular difference between the candidate path and the major axis of the region to be treated can also be considered. Specifically, in practice, to ablate a tumor, it is recommended to attack the tumor along its major axis because treatment is more effective along the axis of the medical device.

[0120] According to another example, when the minimally invasive procedure corresponds to ablation, the coverage value of the region to be treated can also be considered. This coverage value can be estimated based on the segmentation of the region to be treated and the estimation of the ablation region obtainable using the candidate path under discussion. Specifically, for example, in the case of tumor ablation, a minimum ablation boundary may be required to minimize the recurrence rate. In the case of bone tumors, another important factor is that the ablation region remains within the bone (therefore, the limitation of the coverage range is determined by the anatomy and location of the lesion).

[0121] According to another example, when the target anatomy is bone, for partial paths within the bone, the minimum distance between the partial candidate path and the cortical layer of the bone can also be considered. Specifically, it is advantageous to maximize the minimum distance between the partial candidate path within the bone and the cortical layer of the bone to minimize the risk of re - contacting the cortical layer after the medical device is inserted into the bone. Specifically, due to the imprecision of the placement of the medical device or biomechanical effects (deflection of the medical device, movement of the target anatomy due to patient breathing or due to effects associated with the insertion of the device, etc.), the actual path followed by the medical device may be different from the selected theoretical path. Maximizing the minimum distance between the partial candidate path within the bone and the cortical layer of the bone is equivalent to maximizing the safety margin of the candidate path.

[0122] Figure 8 Schematically shows the bone 80 to be treated having a cortical layer 81 and a cancellous bone portion 82. Reference 83 represents the target point that the medical device must reach to treat the bone 80. Figure 8Also shown are two candidate paths 31-1 and 31-2 that reach a target point 83 from two candidate entry points 22-1 and 22-2 on the outer body surface 21 of a patient, respectively. In Figure 8 In the illustrated example, candidate path 31-1 may be more preferable than candidate path 31-2 because the minimum distance d1 between the partial candidate path 31-1 within the bone 80 and the bone cortical layer 81 is greater than the minimum distance d2 between the partial candidate path 31-2 within the bone 80 and the cortical layer 81.

[0123] According to another example, when the target anatomical structure is bone, the incident angle between the candidate path and the bone wall can also be considered. Specifically, it is advantageous for the candidate path to have an incident angle as close as possible to a right angle (90° angle) to prevent the medical device from sliding on the bone wall during its insertion (which may damage the bone and / or cause the medical device to deviate from the selected path during the insertion of the medical device). The "incident angle" refers to the angle formed between the candidate path and the bone at the interface where the path penetrates the interior of the bone.

[0124] In Figure 8 the illustrated example, candidate path 31-1 may be more preferable than candidate path 31-2 because the incident angle i1 formed between candidate path 31-1 and the wall of the bone 80 is closer to a right angle than the incident angle i2 formed between candidate path 31-2 and the wall of the bone 80.

[0125] According to another example, when the target anatomical structure is bone, the estimated risk of surgically induced fracture for the candidate path under discussion can also be considered. For example, when the surgery aims to insert a screw into the bone to reinforce it, a biomechanical model that can model the bone structure (with solid regions and more fragile regions of the bone) and the mechanical stress borne by the bone (e.g., depending on patient measurements) can be used to estimate the fracture risk due to the screw inserted along the candidate path under discussion. Optionally, this risk can be compared with the fracture risk without surgery (i.e., without the reinforcing screw) to estimate to what extent the surgery reduces the fracture risk. The risk of fracture can also be determined from a database of past surgeries that includes paths that prevented fractures and paths that failed to prevent fractures.

[0126] The above-described path characteristics can be quantified and used to construct a cost function, thereby allowing the paths to be classified according to objective criteria. Multiple mathematical functions can be considered to combine these characteristics in order to effectively represent the clinical objectives. The simplest function is a linear combination, in which case the cost function C(T) for a path T takes the form of a weighted sum of the above-described characteristics:

[0127] [Mathematical formula.4]

[0128]

[0129] where the weight wi Allows the control of feature y i The "balance" between

[0130] Various sets of weights w with predefined values i Various cost functions can be formed, and if they are proven to match well with the doctor's clinical practice, they can be used as they are. The doctor can also, based on the goals of the surgery under discussion, create their own cost function by modifying the weights w i values to assign more or less importance to feature y i This can be done manually, for example through a user interface, or even automatically based on previous surgeries. In fact, for each previous surgery, an evaluation (score) can be assigned to the paths used and the features calculated. Based on this information, a system of equations (with the number of equations N significantly higher than the number of features K under discussion) can be created and solved using regression methods. Neural networks can also be used to generate cost functions.

[0131] It may be advantageous to use various cost functions to represent various optimization categories (such as "performance" or "safety"). This allows predefined criteria or user-customized criteria to be considered. Then, the user can provide an indication of the cost function to be considered when displaying candidate paths on the display screen 13 through the user interface.

[0132] As Figure 2 shown, method 100 includes step 106: on the display screen 13, superimposed on the anatomical model, display at least some of the remaining candidate paths, and for each displayed candidate path, have a means for quickly viewing the cost function value.

[0133] It should be noted that the display screen 13 can take various forms: for example, it can be a flat LCD computer screen (LCD monitor), but it can also be a head-mounted screen for augmented reality, in which case the displayed information can be directly projected onto the patient.

[0134] It is conceivable not to display candidate paths whose cost function is less than a threshold.

[0135] In a specific embodiment, the means for quickly viewing the cost function value of the displayed candidate paths corresponds to color coding: various colors are associated with various values of the cost function. This display allows for the quick localization of the optimal path according to established clinical criteria and selection using the user interface. When the user selects a path, the quantification determined for the features of that path can be displayed, with the aim of providing the user with a detailed view of each feature used to calculate the path cost function. Of course, the system provides the ability to select any candidate path, which allows for the management of optimality criteria not captured by the cost function.

[0136] Figure 5 Schematically shows the display of a set of candidate paths 31 on the display screen 13. Each candidate path 31 is represented by a color related to the cost function value calculated for the candidate path 31 under discussion. Also shown is the color coding 30 used. In Figure 5 the example shown, the candidate paths 31 are shown on the outer surface 21 of the patient's body. When the user selects one of these candidate paths 31, various views of the selected path can be displayed in the anatomical model (e.g., in different cross-sections containing the selected path).

[0137] As Figure 2 shown, the method 100 includes step 107: the user selects, from the displayed paths, the path considered to be optimal for the target point under discussion.

[0138] For example, the selected optimal path can be used to configure the movement of a robotic arm suitable for holding or guiding a medical device intended for performing a medical procedure.

[0139] All of the following steps: determining 103 regions and sampling resolutions, removing 104 certain candidate paths, calculating 105 at least one cost function, displaying 106 candidate paths, and selecting 107 paths, can be iterated multiple times to optimize the selection of candidate paths to be used for performing a medical procedure. To this end, as Figure 2 shown in step 108, for example, the user can be asked to indicate via a user interface whether they wish to optimize the selected candidate path.

[0140] Thus, in certain embodiments, at least two iterations are performed. Advantageously, for a given iteration (different from the first iteration), the new sampling region contains an entry point corresponding to the path selected in the previous iteration, the new sampling region is defined as smaller than the sampling region defined for the previous iteration, and the sampling resolution is defined as finer than the sampling resolution defined for the iteration (which means that the minimum distance between two candidate entry points is smaller).

[0141] In certain embodiments, for a given iteration (different from the first iteration), the new sampling resolution and / or the new sampling region are defined based on the variability of the cost function calculated for the previous iteration in the vicinity of the path selected in the previous iteration. For example, the variability of the cost function can be measured as a function of angles α and β. The higher the variability in the previous iteration, the more a fine resolution is recommended for the new iteration. The lower the variability, the more a broad sampling region is recommended for the new iteration.

[0142] As Figure 6 shown, the method 100 can also include step 110: the user evaluates at least one of the displayed candidate paths, and step 111: updating the cost function based on the evaluation obtained.

[0143] For each displayed path, the user can evaluate the path by assigning it a score. This score can be used to trigger an optional update of the cost function parameters (e.g., if the cost function is a linear combination of multiple quantified features, the weights of the linear combination can be modified to adapt to the score provided by the user, e.g., by a regression algorithm or a neural network).

[0144] The path actually followed by the medical device during a medical procedure may deviate from the selected candidate path, especially due to imprecise needle placement or biomechanical effects (needle deflection, movement of the target anatomy due to breathing or due to effects related to the insertion). Thus, it is related to a retrospective cost function that determines the actual path followed by the medical device, and to the difference ΔC(T) between the cost function value C(T) calculated for the candidate path selected for the procedure and the cost function value C'(T) of the actual path.

[0145] Thus, and as Figure 7 shown, after the medical procedure, method 100 may further comprise the following steps:

[0146] - Determine 120 in the anatomical model the actual path actually followed by the medical device during the procedure (specifically, this actual path can be determined based on medical images of the patient acquired when the medical device is in place, i.e., when it reaches the target point and is in the correct position for performing the treatment);

[0147] - Determine 121 the cost function value of the actual path;

[0148] - Calculate 122 the difference between the cost function value of the candidate path selected for the procedure and the cost function value of the actual path.

[0149] Calculation of the cost function value C'(T) of the actual path requires estimating the position of the medical device after insertion from control images. This can be done manually (by clicking on the entry point and the end of the device), or automatically by using image processing methods to segment the device and then extract the required information.

[0150] Specifically, the retrospective determination of the cost function of the actual path can be used as a means to alert the user of an incorrect execution of the procedure (the placement quality of the device is lower than planned). This alert may be triggered when the change ΔC(T) in the cost function is significant. The significance threshold can be determined by retrospective analysis of past paths to which evaluation criteria (scores) have been assigned.

[0151] Thus, and as Figure 7 shown, method 100 may include step 123: displaying on the display screen 13 an indication related to the difference calculated in step 122.

[0152] Statistics of the change ΔC(T) in the cost function of the path followed according to the plan can be retrospectively analyzed to determine the typical value of ΔC(T) for determining path equivalence classes. Specifically, for the set of paths judged to follow the plan, statistics of ΔC(T) can be derived to determine what are typical changes in the cost function. This value can then be used in method 100 to select the optimal path to avoid suggesting paths that are too similar to the user (i.e., paths with not enough difference in their cost functions).

[0153] Thus, and as Figure 7 shown, method 100 may include the following steps:

[0154] - Obtain 124 the user's evaluation of the actual path;

[0155] - Determine 125 a statistical value representing the variability of the cost function based on the calculated differences and the obtained evaluation;

[0156] - Remove 126 candidate paths considered equivalent based on the statistical value thus obtained.

[0157] In method 100 for assisting in selecting the best path, the steps 101 - 108, 110, and 111 described in reference Figure 2 and Figure 6 are all performed before the medical operation; the steps 120 - 126 described in reference Figure 7 are performed after the medical operation. Thus, these steps do not involve performing a surgical operation. Steps 108, 110, 111, and 120 - 126 are optional.

[0158] The above data processing device 10 enables a set of optimal or near - optimal minimally invasive surgical paths to be suggested to a doctor based on a three - dimensional anatomical model of a patient. The suggested paths take into account various types of constraints (e.g., practical constraints, performance - related constraints, and / or safety constraints). The doctor is able to select a path that may be sub - optimal in terms of a predefined criterion but is preferred in terms of constraints not covered by that criterion.

[0159] The proposed solution allows for a fast and optimized exploration of the solution set to determine candidate paths (a "multi - resolution" approach to sampling candidate entry points). It allows the user to quickly view various candidate paths and their optimality criteria. The proposed solution is flexible as it is able to adapt to the specific needs of the surgery (the need to improve placement accuracy, the need to increase the safety margin due to potential weaknesses of the patient's condition, etc.). The doctor is also able to update the optimality criterion (cost function) based on user feedback (evaluation of the path selected for the surgery).

[0160] The above description mainly targets the case of a single target point (e.g., "single-needle" surgery), or actually targets the case of continuously considering multiple target points to separately determine the optimal path for each target point (e.g., "multi-needle" surgery). However, in the case where multiple medical devices must be inserted during surgery (e.g., "multi-needle" surgery), it is also possible to consider this set of target points simultaneously (and thus consider the set of candidate paths associated with various target points). Specifically, this can be achieved by taking into account the characteristics that depend on the mutual relationship between candidate paths when calculating the cost function. For example, this can consider the maximum relative angle between two of the various paths when calculating the cost function (it may be beneficial to minimize this relative angle). According to another example, this can consider the difference between the distance between two of the various paths and a target value (specifically, the target value can be derived from the recommendations of the manufacturer of the medical device used).

Claims

1. A data processing device (10) comprising a processor (11), a computer memory (12) and a display screen (13), wherein the computer memory (12) comprises program code instructions which, when executed by the processor (11), configure the processor to implement a method (100) for assisting in selecting at least one optimal path for a minimally invasive medical procedure on a patient's bone, the processor (11) being configured to: - obtain (101) a three-dimensional anatomical model of the patient from pre-acquired medical images of the patient, the anatomical model comprising the patient's body outer surface (21) and a representation of the bone to be treated; - identify (102) at least one target point to be reached within the bone to be treated in the anatomical model; - perform at least one iteration, wherein the processor is configured to: o determine (103) a sampling region and a sampling resolution for sampling candidate entry points (22) on the patient's body outer surface (21) in the anatomical model, each candidate entry point (22) belonging to the sampling region, the sampling resolution representing the minimum distance between two candidate entry points (22), and each pair formed by the target point and a candidate entry point (22) forming a candidate path; o remove (104) candidate paths according to at least one pre-determined validity criterion; o for each remaining candidate path, quantify a plurality of pre-determined features one by one, including the minimum distance between the part of the candidate path located inside the bone to be treated and the cortical bone wall, and calculate (105) at least one cost function from the quantified features; o on the display screen (13), superimposed on the anatomical model, display (106) at least some of the remaining candidate paths, and for each displayed candidate path (31), have a means for quickly viewing the cost function value; o obtain (107) a user's selection of a path to the at least one target point, which path is considered to be the optimal path among the displayed candidate paths (31).

2. The device (10) according to claim 1, wherein the processor (11) is configured to perform at least two iterations; and wherein for the nth iteration, n being an integer strictly greater than one: - the sampling region of the nth iteration is defined as being smaller than the sampling region defined in the (n - 1)th iteration and includes the candidate entry points (22) corresponding to the path selected in the (n - 1)th iteration; and - the sampling resolution of the nth iteration is defined as being finer than the sampling resolution defined in the (n - 1)th iteration.

3. The device (10) according to claim 2, wherein the sampling resolution of the nth iteration is defined according to the variability of the cost function calculated in the (n - 1)th iteration in the vicinity of the path selected in the (n - 1)th iteration.

4. The device (10) according to any one of claims 2-3, wherein the sampling region of the n-th iteration is defined based on the variability near the path selected in the (n-1)-th iteration according to the cost function calculated in the (n-1)-th iteration.

5. The device (10) according to any one of claims 1 - 4, wherein, in order to sample the candidate entry points (22), the processor (11) is configured to determine, in the anatomical model, on the outer surface (21) of the patient's body, a set of points defined by a spherical coordinate system centered on the target point, each point being defined by a distance r from the target point i,j and two angles α i and β j where the angle α i and β j are defined such that the subscripts i and j correspond to strictly positive integers: And wherein R represents the minimum distance between two candidate entry points (22) on the outer surface (21) of the patient's body.

6. The device (10) according to any one of claims 1-5, wherein a validity criterion allows for verifying at least one of the following for a given candidate path: - the validity of the length of the candidate path for the medical device planned for the surgery, - the ability of the robotic arm configured with the medical device so that the medical device will be able to follow the candidate path, - the presence of an object equipped on the patient obstructs the candidate path, - the candidate path intersects at least one key anatomical structure in the patient's body.

7. The device (10) according to any one of claims 1-6, wherein multiple validity criteria are considered to remove candidate paths, and different validity criteria are evaluated in the order defined for each validity criterion according to the estimated calculation time required to evaluate each said validity criterion.

8. The device (10) according to any one of claims 1-7, wherein the multiple features of the candidate path include at least one of the following features: - the minimum distance between the candidate path and a key anatomical structure in the patient's body, - the length of the intersection portion between the candidate path and the bone to be treated, - the minimum incident angle between the candidate path and the anatomical interface traversed by the candidate path, - the minimum distance between the outer surface of the treatment area and the outer surface of the ablation area estimated for the candidate path, - the angular difference between the candidate path and the main axis of the treatment area.

9. The device (10) according to any one of claims 1-8, wherein multiple cost functions are calculated based on the quantified features, each cost function being calculated using a different set of weights assigned to different quantified features, and the processor (11) is configured to obtain an indication of the cost function to be considered by the user when displaying the candidate path.

10. The device (10) according to any one of claims 1-9, wherein the means for quickly viewing the cost function values of each displayed candidate path (31) includes color coding associating different colors with different cost function values.

11. The device (10) according to any one of claims 1-10, wherein the processor (11) is configured to display the quantification determined by the user for the features of the selected path.

12. The device (10) according to any one of claims 1-11, wherein the processor (11) is configured to obtain (110) the user's evaluation of at least one displayed candidate path (31) and update (111) the cost function based on the evaluation.

13. The device (10) according to any one of claims 1-12, wherein the processor (11) is configured to: - After the surgery, in the anatomical model, determine (120) the actual path followed by the medical device during the surgery based on the medical images of the patient obtained when the medical device was in place. - Determine (121) the cost function value of the actual path. - Calculate (122) the difference between the cost function value of the candidate path selected for the surgery and the cost function value of the actual path.

14. The apparatus (10) according to claim 13, wherein the processor (11) is configured to compare the calculated difference with a predetermined threshold and display (123) an indication related to the calculated difference value.

15. The apparatus (10) according to claim 13, wherein the processor (11) is configured to: - Obtain (124) the user's evaluation of the actual path, - Determine (125) a statistical value representing the variability of the cost function based on the calculated difference and the obtained evaluation, - Remove (126) the candidate paths that are considered equivalent based on the obtained statistical value.

Images