Method for determining a stereotactic brain target

Abstract

The present invention relates to a method for determining a stereotactic brain target (including at least one target point), the method comprising the following steps: - selecting patients in whom the result measured at at least one target point after treatment is greater than or equal to a threshold value, and performing postoperative imaging on each of said patients; - processing the postoperative imaging to determine the coordinates of the at least one target point; - selecting brain marker points; - processing the postoperative imaging to determine the coordinates of the marker points; - creating a learning database that includes the coordinates of the target points and the coordinates of the marker points determined for all selected patients; - determining a prediction function using the learning database and a supervised statistical learning method, the prediction function giving the coordinates of at least one target point based on the coordinates of the marker points; - processing the preoperative imaging of a new patient to be treated to determine the coordinates of the marker points of the new patient; - using the prediction function based on the coordinates of the marker points determined for the new patient to obtain the coordinates of at least one target point of the new patient.

Description

Technical Field

[0001] The present invention belongs to the field of neurosurgery, and more particularly relates to a preparation method before neurosurgical intervention.

[0002] The present invention more particularly relates to a method for precisely determining a stereotactic brain target.

[0003] The present invention may particularly relate to a method for determining a brain target to be stimulated, the determination step being a preparatory step before the implantation of an electrode for deep brain stimulation. Background Art

[0004] Determining a brain target is essential in any neurosurgical treatment protocol, particularly in terms of stereotaxy. Stereotaxy is a neurosurgical technique that uses a method of three-dimensional identification of intracranial structures assisted by medical imaging to precisely determine the volume and location of a certain area of the brain or the volume and location of a lesion in the brain where a neurosurgeon must intervene, based on points located inside the brain.

[0005] One intervention can be deep brain stimulation (DBS). Deep brain stimulation is a very effective surgical technique for alleviating the conditions of patients with neurological pathologies, such as Parkinson's disease, essential tremor, dystonia, obsessive-compulsive disorder, Tourette syndrome, intractable epilepsy, or rather severe treatment-resistant depression, etc.

[0006] Deep brain stimulation involves stimulating hyperactive structures deep in the brain. The stimulation may inhibit or activate neurons, with the goal of adjusting the function of neural networks. It can relieve or even eliminate symptoms and improve the quality of life of patients resistant to conventional treatments.

[0007] Deep brain stimulation is usually carried out by implanting an electrode into a brain structure corresponding to a nucleus, and generally it is part of a nucleus in the gray matter (nerve cell bodies). In the remainder of this specification, the term "nucleus" may refer to a nucleus or a part of a nucleus.

[0008] The electrode receives a low-intensity current and thus delivers it to the target nucleus.

[0009] The brain structures targeted by deep brain stimulation vary according to the pathology to be treated: the subthalamic nucleus (STN) is used to treat Parkinson's disease, the ventral intermediate nucleus of the thalamus (VIM) is used to treat essential tremor, and the internal globus pallidus (GPi) treats dystonia and certain types of Parkinson's disease. Except for the internal globus pallidus, the sizes of these nuclei do not exceed a few millimeters.

[0010] The challenges posed by this technology include identifying (i.e., defining and / or localizing) the nucleus to be stimulated, marking (i.e., targeting) the identified nucleus, and positioning the stimulating electrode.

[0011] The first challenge involves identifying the nucleus to be stimulated. For example, with respect to the VIM, there are multiple nomenclatures for the human thalamus, leading to discrepancies in the identification of the VIM. Compared with the VIM, the problem of the anatomical definition of the STN is not as severe. However, discrepancies still exist, such as which subpart of the STN should be stimulated, or even whether the fibers above the STN ("zona incerta") are actually not the optimal target.

[0012] Even if this identification problem is solved, there are still other challenges: localizing the identified nucleus and thus positioning the stimulating electrode. For example, while magnetic resonance imaging (MRI) can be used to localize the GPi nucleus for the treatment of dystonia, it is much more difficult, or even impossible, to localize the STN nucleus and the VIM nucleus using the MRI that can be used in current stereotactic practice (usually an MRI with a magnetic field of 1.5 or 3 Tesla). Thus, while the localization of some nuclei has indeed been improved due to the progress of MRI - for example, the combination of a series of MRIs enables the visualization of the STN nucleus (3D T1, T2, SWI ("susceptibility-weighted imaging"), FGATIR ("fast gray matter acquisition inversion recovery"), FLAIR ("fluid-attenuated inversion recovery") series) - it has been shown that the STN nucleus visualized radiologically by MRI does not precisely correspond to the STN nucleus where stimulation is effective (which can be measured using the intraoperative electrophysiological recording techniques described later), with an error of more than 5 millimeters visible at the individual level.

[0013] For these less visible or non-localizable regions, it is a known practice to implement so-called "indirect" localization techniques based on stereotactic atlases. The principle of these indirect localization techniques is to have a Cartesian reference system that contains anatomical landmark points that can be easily identified by MRI, and an atlas that gives the positions of different brain regions (e.g., different brain nuclei), but not all regions can be identified by MRI. The principle is to use the atlas to localize the position of any region in any individual brain based on all or part of the landmark points. A function of the ratio between the landmark points and the points in the atlas is used, so that any brain can be calibrated regardless of its size or anatomical characteristics.

[0014] When studying the brain, the use of stereotactic atlases is much broader and far exceeds the application of DBS. A well-known example is the Talairach atlas or coordinates. The center of this reference system (i.e., the point with coordinates x = 0, y = 0, z = 0) is the upper and posterior edges of the anterior commissure in the mid-sagittal plane (i.e., the plane parallel to the inner surface of the brain, also called the interhemispheric plane); this point, called "CA" (also called "AC"), is easily visible in MRI. Another reference point is used: the lower and anterior edges of the posterior commissure (still in the mid-sagittal plane); this point, called "CP" (also called "PC"), is also easily visible in MRI. The three axes of the coordinate system are defined as follows: The Oy axis passes through CA and CP and points towards the front of the skull. The Oz axis is the axis of the sagittal plane, which is perpendicular to Oy and passes through CA. It points towards the top of the skull. Finally, the Ox axis is the axis passing through CA and orthogonal to Oy and Oz; it is oriented from left to right. Another example is the Schaltenbrand atlas, which enables the simplification of the Talairach system by taking the midpoint of the line segment [CACP] (called the mid-commissural point) as the center of the stereotactic reference system. The points CA and CP represent the anatomical centers of the anterior commissure and the posterior commissure, respectively.

[0015] Once the nucleus to be stimulated has been identified according to the pathology to be treated, it can be located in the atlas used using marker points identifiable by MRI. During deep stimulation, the stimulating electrode should be placed in the nucleus indirectly identified in this way.

[0016] The problem with known indirect localization techniques is that they are based on the assumption of the proportion of the brain, like the variability of the distance between two points can explain all the observed anatomical variability, but this is not the case. Therefore, this necessarily leads to inaccurate targeting, which can be extremely disadvantageous for deep brain stimulation.

[0017] To compensate for these inaccuracies in localization to ensure the correct positioning of the electrode in the target, current surgeries are not performed without electrophysiological recording and intraoperative clinical testing. The principle of intraoperative electrophysiological recording is to implant parallel microelectrodes in the area assumed to contain the anatomical target to stimulate the neurons in this area and record the effects of the stimulation during the operation. This technique can optimize the final position of the electrode, thus improving targeting. Although this technique improves the accuracy of targeting, it is extremely time-consuming (the operation time is about 7 to 10 hours), and it brings the risk of infection, and the risk of bleeding increases in direct proportion to the number of recording microelectrodes used. In addition, it is not always easy to precisely achieve the final implantation of the electrode. Finally, the surgical procedure requires the participation of the patient so that the clinical effects of the stimulation can be tested to optimize the position of the electrode. Therefore, it must be performed under local anesthesia, which is extremely uncomfortable for the patient.

[0018] The above technique is based on a method that includes defining the nucleus to be stimulated, then targeting said nucleus, and then positioning the stimulating electrode in the case of DBS. Analysis of the results obtained can lead to a shift in the positioning of the stimulating electrode and / or the anatomical target in the reference frame used (usually the Schaltenbrand or Talairach reference frame).

[0019] Another technique includes an indirect method that includes identifying patients in whom deep brain stimulation has been successful, identifying the position of the electrodes in these patients and correlating them with reference points that are easily recognizable in an MRI, and then using said reference points to apply them to new patients to be treated. This method can avoid mixing the inaccuracies of identifying the nucleus to be stimulated, target localization (targeting), and positioning of the stimulating electrode in the target. In this case, referring to the "clinical target" or "functional target", the stimulation target is determined based on clinical cases that have worked.

[0020] This method is described in Caire's paper "Intraoperative imaging of deep brain stimulation electrodes and proposal of a new method for indirect stereotactic location of the subthalamic target (Proposal of a new method for indirect stereotactic location of the subthalamic target in intraoperative imaging of deep brain stimulation electrodes)". The solution proposed in this paper includes the following steps:

[0021] - Select patients who respond very well to deep thalamic stimulation;

[0022] - Retrieve the postoperative MRI of said patients;

[0023] - For each postoperative MRI, construct a standard CACP stereotactic space: To do this, identify the CA point and the CP point and the interhemispheric plane;

[0024] - Determine the coordinates of the effective stimulation contacts in the postoperative MRI;

[0025] - Determine various anatomical landmark points in postoperative MRI;

[0026] - Calculate the correlation between each coordinate of the effective contact and the coordinates of each corresponding landmark point using a linear regression model;

[0027] - When the correlation is correct, calculate the regression line equation based on the landmark point that provides the best correlation with the x, y, and z coordinates of the effective contact;

[0028] - Use this equation to calculate the coordinates of the theoretical target;

[0029] - Compare the coordinates of this theoretical target with the actual coordinates of the effective contact.

[0030] The drawback of this method is that it only assumes the variability (or homothety) of the brain in three axes while remaining in a proportional system (here the linear regression method). This is better than assuming that the variability of the distance between two points (CA and CP) can explain all the observed anatomical variability, but it is still insufficient. The average error of this method is 2.5 ± 0.6 mm, so this is quite large relative to the size of the structure to be targeted.

[0031] The present invention aims to overcome the above drawbacks of the prior art.

[0032] More specifically, the present invention aims to provide a method for determining a stereotactic brain target, which has better targeting accuracy and is easy to use. The aim is to provide a method for determining a stereotactic brain target, which is used as a preparatory step before neurosurgical treatment, which can improve the effectiveness of this treatment. Summary of the Invention

[0033] One subject of the present invention that allows achieving this objective is a method for determining a stereotactic brain target comprising at least one target point, said method comprising the following steps:

[0034] - Select patients in whom the measurement result at at least one target point after treatment is greater than or equal to a threshold value, and perform postoperative imaging on each of said patients;

[0035] - Process said postoperative imaging to determine all or part of the coordinates of said at least one target point for each selected patient;

[0036] - Select brain landmark points;

[0037] - Process said postoperative imaging to determine all or part of the coordinates of the selected landmark points for each selected patient;

[0038] - Create a learning database that includes the coordinates of the landmark points and the coordinates of the target points determined for all selected patients;

[0039] - Determine a prediction function by using the learning database and a supervised statistical learning method, the prediction function giving the coordinates of at least one target point based on the coordinates of the landmark points;

[0040] - Process the pre-operative imaging of a new patient to determine all or some of the coordinates of the landmark points of the new patient;

[0041] - Based on the coordinates of the landmark points determined for the new patient, utilize the prediction function to obtain the coordinates of at least one target point for the new patient.

[0042] Preferably, the present invention relates to a method for determining a stereotactic brain target comprising at least one target point, the method being implemented before a neurosurgical treatment at the target for a given pathology and comprising the following steps:

[0043] - Select a plurality of patients with clinical cases, the results measured at at least one target point after treatment for the pathology of each of the patients being greater than or equal to a threshold, and perform post-operative imaging on each of the patients;

[0044] - Select a mathematical coordinate system, preferably an orthogonal Cartesian coordinate system;

[0045] - Process the post-operative imaging to determine all or some of the coordinates of the at least one target point in the selected coordinate system for each selected clinical case;

[0046] - Select a plurality of brain landmark points;

[0047] - Process the post-operative imaging to determine all or some of the coordinates of the brain landmark points for each selected clinical case;

[0048] - Create a learning database that includes the determined coordinates of the target points and the determined coordinates of the landmark points for all selected clinical cases;

[0049] - Determine a prediction function by using the learning database and a supervised statistical learning method, the prediction function giving the coordinates of at least one target point based on the landmark points;

[0050] - Process the pre-operative imaging of a new patient to be treated for the pathology to determine all or some of the coordinates of the landmark points of the new patient;

[0051] - Based on the determined coordinates of the landmark points for the new patient, utilize the prediction function to obtain the coordinates of at least one target point for the new patient.

[0052] According to the present invention, the fiducial points are characteristic points of brain anatomical structures that can be located in a given mathematical coordinate system and are anatomical points that are visible, in particular, using conventional imaging.

[0053] According to the present invention, the target points are points where the neurosurgical treatment has been applied to a clinical case or points where the neurosurgical treatment must be applied to a new patient.

[0054] The method according to the present invention is based on data from patients of clinical cases for which the therapeutic efficacy has been determined. Thus, the target points for learning are clinically validated targets. This makes it possible to improve the neurosurgical treatment that can follow the method according to the present invention.

[0055] To this end, the method according to the present invention is based on the available postoperative imaging of these patients, which is processed in order to derive therefrom: the points where the treatment has been applied (target points) and fiducial points selected from among the characteristic points of brain anatomical structures that are visible, in particular, using conventional imaging. The coordinates of these target points and these fiducial points are integrated into a learning database, which makes it possible to determine a prediction function. The prediction function can give at least one target point as a function of the fiducial points. The prediction function is constructed from the learning database using supervised statistical learning methods.

[0056] The method according to the present invention can be based on images extracted from a commonly used MRI, such as an MRI of 1.5T. It is not necessary to use a stronger MRI, such as an MRI of 3T or 7T, which is much more expensive (especially in terms of hardware) and requires a higher level of radiological skill than an MRI of 1.5T. These images are processed to determine the coordinates of the fiducial points.

[0057] The method according to the present invention seeks to control the variability of the brain in a multi-dimensional manner rather than on one, two, or three axes. By choosing supervised statistical learning techniques, the method can overcome this variability of the brain, thus achieving better targeting accuracy.

[0058] Furthermore, unlike in other methods where one first seeks to identify the brain structures (STN, VIM, etc.) to be stimulated and then localize them, the target points according to the present invention are determined as clinically validated targets, and then the stimulating electrodes are implanted into the brain structures thus identified and localized. Thus, this makes it possible to avoid mixing inaccurate factors and thus improve the targeting accuracy.

[0059] In this sense, the method according to the invention does not seek to construct an anatomical atlas to visualize brain structures (STN, VIM, etc.) as a function of marker points, but rather seeks target points in order to directly know where to apply the treatment and to apply the treatment by using the target points that are effective for the patient. Indeed, as mentioned above, the inventors have found that even if the center of the anatomical structure (STN, VIM, etc.) is located, it is not guaranteed that this center is the correct target to be stimulated.

[0060] The method according to the invention also makes it possible to determine targets (multiple points) and not just a single point, thus constructing the targets in 3D.

[0061] Once the targets have been determined for a new patient, the position of the targets can be marked in the MRI with a symbol (e.g., a cross). To mark with a cross, the values of five pixels in each of the three directions of the mathematical coordinate system can be changed by assigning the maximum value (corresponding to white) to these pixels. This makes it possible to form a white cross, the center of which allows the professional to visualize the targets.

[0062] Finally, the method according to the invention for determining stereotactic targets can be used as a decision aid, particularly when preparing for any neurosurgical treatment that requires the determination of at least one precise brain target.

[0063] In particular, in the case of deep brain stimulation, the method of the invention can eliminate the electrophysiological techniques normally used to ensure the correct positioning of the electrodes in the brain structure to be stimulated. Thus, the method of the invention is non-invasive for the patient.

[0064] By using appropriate techniques to select and optimize the parameters of the prediction function (also called "meta-model"), the method allows further improvement of the targeting accuracy.

[0065] Thus, according to an advantageous embodiment, the method may further comprise the following steps:

[0066] - Merging the prediction function using a cross-validation method, said merging step resulting in a merged prediction function that gives the coordinates of at least one target point according to the marker points;

[0067] The step of using the prediction function includes: using the merged prediction function.

[0068] According to one embodiment, the image processed to determine the coordinates of the marker points and at least one target point is at least one MRI image, preferably a plurality of MRI images.

[0069] The method may thus include the step of performing a postoperative MRI in a selected patient, and the determination of the landmark points may include the step of processing the images obtained by the MRI, and / or the determination of at least one target point may also include the step of processing the images obtained by the MRI.

[0070] Alternatively or additionally, the method may include the step of performing a postoperative computed tomography (CT) scan in a selected patient, and the determination of at least one target point may include the step of processing the images obtained by the CT, and / or the determination of the landmark points may include the step of processing the images obtained by the CT.

[0071] According to one embodiment, the supervised statistical learning method includes: using a kernel ridge regression method in a reproducing kernel Hilbert space.

[0072] According to another embodiment, the supervised statistical learning method includes using a method of the support vector machine type.

[0073] According to another embodiment, the supervised statistical learning method includes using a method of the neural network type.

[0074] According to one embodiment, the cross-validation method includes using the "leave-one-out cross-validation" method.

[0075] According to another embodiment, the cross-validation method includes using the "leave-k-out cross-validation" method.

[0076] According to a particular embodiment, the selected mathematical coordinate system is an orthogonal Cartesian coordinate system, a straight line passing through the upper and lower edges of the anterior commissure and the lower and upper edges of the posterior commissure forms the Oy axis, the lower and upper edges of the posterior commissure form the center of the coordinate system, and the Oz axis is a straight line perpendicular to the Oy axis in the interhemispheric plane.

[0077] According to a particular embodiment, the landmark points are selected from among the following eighteen points, which are preferably defined with respect to a previously selected mathematical coordinate system:

[0078] - The first landmark point is the mammillothalamic tract on the third axial plane;

[0079] - The second, third, and fourth landmark points are the anterior points of the putamen on each of the first, second, and third axial planes;

[0080] - The fifth and sixth landmark points are the medial points of the putamen on the first and second axial planes;

[0081] - The seventh and eighth fiducial points are the posterior points of the putamen on the first and second axial planes;

[0082] - The ninth fiducial point is the habenular commissure on the second axial plane;

[0083] - The tenth fiducial point is the anterior edge of the thalamus on the second axial plane;

[0084] - The eleventh fiducial point is the posterior edge of the thalamus on the second axial plane;

[0085] - The twelfth fiducial point is the anterior commissure;

[0086] - The thirteenth fiducial point is the medial edge of the third ventricle at the mid-commissure point;

[0087] - The fourteenth fiducial point is the height of the thalamus on the sagittal plane passing through the thirteenth fiducial point;

[0088] - The fifteenth fiducial point is the midpoint of the line segment defined by the thirteenth and fourteenth fiducial points;

[0089] - The sixteenth fiducial point is the anterior edge of the thalamus on a line parallel to the line passing through the upper and lower edges of the anterior commissure and the lower and upper edges of the posterior commissure and passing through the fifteenth fiducial point;

[0090] - The seventeenth fiducial point is the upper edge of the putamen on the coronal plane passing through the fifth fiducial point; and

[0091] - The eighteenth fiducial point is the lateral edge of the putamen on the coronal plane passing through the fifth fiducial point.

[0092] According to one embodiment, all eighteen of these fiducial points are used.

[0093] Other fiducial points may be used.

[0094] Other mathematical coordinate systems may be used, particularly those commonly used in known stereotactic atlases.

[0095] According to one embodiment, the method includes the additional step of adding functional data to the learning database, the functional data being capable of adding at least one confidence metric to the target points and fiducial points of a clinical case.

[0096] According to one embodiment, the method is performed prior to deep brain stimulation. According to another embodiment, the method is performed prior to gamma knife. According to another embodiment, the method is performed prior to focused ultrasound therapy.

[0097] The fiducial points or target points are adjusted according to the planned treatment and thus according to the pathology to be treated.

[0098] A second subject of the present invention is a data processing system including a processor configured to implement all or part of the steps of the method. It can be a computer, a tablet computer, a smart phone, etc.

[0099] A third subject of the present invention is a computer program including instructions which, when the program is executed by a processor, cause the processor to implement all or part of the steps of the method. Description of the Drawings

[0100] By means of the following description given in a non-limiting illustrative manner with reference to the accompanying drawings, other features and advantages of the present invention will become apparent, in which:

[0101] - Figure 1 is a synthetic image showing the implantation of two stimulating electrodes at two different points and in two different directions;

[0102] - Figure 2 is a synthetic image showing an electrode implanted in a target and including a plurality of active contacts;

[0103] - Figures 3A to 3F is an MRI image showing 18 marker points observed along different cross-sections;

[0104] - Figures 4A to 4C are three MRI images showing the axes of the mathematical coordinate system CA-CP, and Figure 4D show the 3D mathematical coordinate system CA-CP;

[0105] - Figure 5 is a flowchart showing an exemplary method and variants according to the present invention. Detailed Description

[0106] The present invention consists in determining target points for neurosurgical treatment, which are generally located in deep brain structures that are not visible by conventional imaging.

[0107] To this end, the present invention consists in using marker points visible by conventional imaging and a prediction function. The prediction function is constructed using a learning database that includes marker points and target points from patients of clinical cases where a therapeutic efficacy has been observed. As further explained below, the efficacy is determined by at least one measurement result. Supervised statistical learning techniques are used to construct the prediction function (or meta-model) so as to be able to provide the coordinates of at least one target point as a function of the marker points.

[0108] The following description of the embodiments is generally made in the context of implementing the method according to the invention as a preparation for treatment by deep brain stimulation. It should be understood that the invention, and in particular the embodiments described, can be more widely implemented as a preparation for any neurosurgical treatment that requires the determination of at least one precise brain target.

[0109] In a specific application of deep brain stimulation, the target point can also be referred to as the stimulation point.

[0110] Select clinical cases (patients):

[0111] Patients are selected in whom electrode stimulation has been effective at at least one stimulation point when treating the same pathology. The efficiency is determined by at least one measurement result.

[0112] The pathology can be Parkinson's disease, essential tremor, dystonia, obsessive-compulsive disorder, Tourette syndrome, intractable epilepsy, or moderately severe treatment-resistant depression.

[0113] The target structures to be stimulated can be: the subthalamic nucleus (STN), which is used to treat Parkinson's disease; the ventral intermediate nucleus of the thalamus (VIM), which is used to treat essential tremor; the internal globus pallidus (GPi), which is used to treat dystonia and certain forms of Parkinson's disease.

[0114] The results measured after neurosurgical treatment can be postoperative clinical evaluation, evaluation by anatomical or radiological localization, or electrophysiological evaluation.

[0115] Postoperative clinical evaluation is performed using a specific scale for the treated pathology and / or a quality of life scale (e.g., three months after treatment). This makes it possible to select a first criterion and thus determine a first threshold for the clinical case (patient).

[0116] Evaluation by anatomical or radiological localization is defined by preoperative anatomical or radiological targeting and the distance between the preoperative targeting and the postoperative treatment point (e.g., the position of the electrode in the case of DBS). This makes it possible to select a second criterion and thus determine a second threshold for the clinical case (patient).

[0117] Electrophysiological evaluation is defined by intraoperative target identification and the distance between the intraoperative identification and the postoperative treatment point (e.g., the position of the electrode in the case of DBS). This makes it possible to select a third criterion and thus determine a third threshold for the clinical case (patient).

[0118] The various evaluation criteria can be combined to define a combined threshold.

[0119] The selected patients will make it possible to establish a statistical learning database that includes marker point / target point pairs for at least one pathology.

[0120] In addition, the method may include the step of adding functional data to the patient's learning database. Such data may include parameters for evaluating the improvement degree of preoperative and postoperative pathological symptoms. For example, the UPDRS scale (Unified Parkinson's Disease Rating Scale) for Parkinson's disease or the Fahn-Tolosa-Marin scale for essential tremor may be used.

[0121] In addition, in the case of deep brain stimulation, such data may include the electrical parameters of the stimulation (amplitude, frequency, and duration).

[0122] Therefore, in addition to the target points and marker points, all such functional data can be input into the learning database and the meta-model. All or part of such functional data can be used as at least one confidence index in the patient data in the cost function. In this case, such data is mathematically transformed into a weight function in the meta-model, which will weight the cost function to be minimized.

[0123] Marker points are characteristic points of the brain anatomy, which can be located in the mathematical coordinate system used and are anatomical points visible using conventional imaging, especially MRI.

[0124] Therefore, marker points can be obtained by using the imaging from existing postoperative MRI.

[0125] In the specific embodiment described, the target point (or stimulation point) is the implantation point of the electrode. The stimulation point is characterized by its coordinates in the mathematical coordinate system used. As Figure 1 and Figure 2 shown, it can also be characterized by the directions of the electrodes (E1, E2) implanted at the points (PC1, PC2).

[0126] For example, each stimulation point is determined by performing and using postoperative imaging such as by MRI or computed tomography (“CT”) and by reconstructing the electrodes in the reference frame based on distal artefacts and artefacts generated by various contacts.

[0127] In this specification, the target point or group of target points constitutes the “target” or “stimulation target” in a specific application of deep brain stimulation. The target is characterized by the coordinates of each target point. In a specific application of deep brain stimulation, it can also be characterized by the direction of the electrode (coordinates of the direction vector of the electrode) at the stimulation point.

[0128] Therefore, these marker points and target points determined for the selected patient constitute a statistical learning database, which can be enriched with other data.

[0129] The method according to the invention comprises using a regression method based on supervised statistical learning, using said learning database so as to establish a prediction function between the landmark points and one or more target landmark points.

[0130] Specific reference Figures 3A to 3F and Figures 4A to 4D The following description given with reference to

[0131] Steps for constructing and / or initializing a learning database

[0132] Mathematical coordinate system used

[0133] "CA" represents the upper and lower edges of the anterior commissure, and "CP" represents the lower and upper edges of the posterior commissure.

[0134] In Figures 4A to 4D is shown the mathematical coordinate system used. Figure 4A is a coronal section, Figure 4B is a median sagittal section, and Figure 4C is an axial section. Figure 4D is a 3D view showing the three axes.

[0135] The illustrated coordinate system is an orthogonal Cartesian coordinate system with the point CP as the origin, whose Oy axis is the line passing through the points CA and CP (labeled "CACP"). The Oz axis is the line perpendicular to the line (CACP) in the interhemispheric plane, and the Ox axis is derived. This coordinate system is labeled CA-CP in this specification.

[0136] Available marked points

[0137] In an exemplary embodiment, the number of available landmark points is 18 (m = 18) corresponding to the points particularly shown in Figures 3A to 3F The eighteen available landmark points are:

[0138] - The first landmark point (FMT): the mammillothalamic tract on the third axial plane;

[0139] - The second, third and fourth landmark points (PA1, PA2, PA3): the anterior points of the putamen on the first, second and third axial planes;

[0140] - The fifth and sixth landmark points (PM1, PM2): the middle points of the putamen on the first and second axial planes;

[0141] - The seventh and eighth landmark points (PP1, PP2): the posterior points of the putamen on the first and second axial planes;

[0142] - The ninth landmark point (CH): the habenular commissure on the second axial plane;

[0143] - Tenth landmark point (BAT): The anterior edge of the thalamus on the second axial plane;

[0144] - Eleventh landmark point (BPT): The posterior edge of the thalamus on the second axial plane;

[0145] - Twelfth landmark point (CA): The anterior commissure;

[0146] - Thirteenth landmark point (A): The middle edge of the third ventricle at the mid-commissure point, called the "MCP" point;

[0147] - Fourteenth landmark point (B): The height of the thalamus on the sagittal plane passing through the thirteenth landmark point (A);

[0148] - Fifteenth landmark point (C): The midpoint of the line segment [AB] defined by the thirteenth and fourteenth landmark points;

[0149] - Sixteenth landmark point (D): The anterior edge of the thalamus on the line parallel to (CACP) and passing through the fifteenth landmark point (C);

[0150] - Seventeenth landmark point (P culm ): The upper edge of the putamen on the coronal plane passing through the fifth landmark point (PM1); and - Eighteenth landmark point (P lat ): The outer edge of the putamen on the coronal plane passing through the fifth landmark point (PM1).

[0151] The axial reference plane (or transverse plane) is the plane passing through the center of the CA-CP coordinate system and the Ox and Oy axes. The first, second, and third axial planes are defined relative to the axial reference plane.

[0152] The first axial plane corresponds to the plane parallel to the axial reference plane passing through the z(C)-5mm coordinate (a point 5mm below the z coordinate of the fifteenth landmark point and with its x and y coordinates equal to zero).

[0153] The second axial plane corresponds to the plane parallel to the axial reference plane passing through the z(C) coordinate (the z coordinate of the fifteenth landmark point and with its x and y coordinates equal to zero).

[0154] The third axial plane corresponds to the plane parallel to the axial reference plane passing through the z(C)-10mm coordinate (10mm below the z coordinate of the fifteenth landmark point and with its x and y coordinates equal to zero).

[0155] The sagittal plane should be understood as any plane that coincides with or is parallel to the plane formed by the Oy and Oz axes. In addition, the sagittal plane selected in the embodiment passes through the x coordinate of the thirteenth landmark point (A).

[0156] The coronal plane should be understood as any plane that coincides with or is parallel to the plane formed by the Ox axis and the Oz axis. Additionally, the coronal plane selected in the embodiment passes through the y coordinate of the fifth landmark point (PM1).

[0157] Figures 3A to 3F MRI images observed along respective sections are shown. Figure 3A is an axial section (axial reference plane). Figure 3B is a sagittal section along the sagittal plane defined above. Figure 3C is an axial section (first axial plane). Figure 3D is an axial section (second axial plane). Figure 3E is an axial section (third axial plane). Figure 3F is a coronal section along the coronal plane defined above.

[0158] Other landmark points can be used according to the planned neurosurgical intervention. Thus, since the target points depend on the pathology to be treated, landmark points that can be related to the target points and are easily recognizable in conventional imaging such as MRI can generally be selected.

[0159] According to a specific embodiment, the landmark point PR can be represented by a matrix X, where m represents the number of landmark points, and m×3 represents all the coordinates of all these landmark points.

[0160] Each row of the matrix X represents a given landmark point. Each column represents one of the three coordinates.

[0161] By concatenating the coordinates of the landmark points, the matrix X can be reshaped into a vector In the remainder of this specification, a vector X will be considered, which includes multiple (3×m) scalar elements.

[0162] Each of the 18 landmark points is characterized by its three coordinates x PR , y PR , z PR in the mathematical coordinate system used. This results in a total of 54 coordinates.

[0163] Generally, the determination is made hemisphere by hemisphere: the right hemisphere (denoted as "HD" in this specification) and the left hemisphere (denoted as "HG" in this specification). Then there are 18 points or 54 coordinates in HD, and 18 points or 54 coordinates in HG. In this case, m = 36, and a total of 108 coordinates are obtained.

[0164] Therefore, one prediction function F G that can determine the left target can be constructed for the left hemisphere, and another prediction function F DAlternatively, a single function F can be constructed that is capable of determining two targets (a right target and a left target).

[0165] Target marked points (or stimulation points)

[0166] A target (grouping one or more target points PC together) is represented by a vector Y, where the vector Y includes multiple (p) scalar elements.

[0167] The number of target points in a target is variable. The value of p depends on the number of target points multiplied by the number of coordinates per point. The coordinates can be the Cartesian coordinates x PC , y PC , z PC .

[0168] In the case of deep brain stimulation, the coordinates can also include coordinates characterizing the direction of the stimulating electrode at that point. By knowing the stimulation point and the direction of the electrode at that point, other stimulation points on the same electrode can be found. The stimulation point is also called the "stimulation area". Each electrode E1, E2 can have four areas, for example, as Figure 2 shown. Point 0 can represent target points PC1, PC2. Thus, the patient can be stimulated via one or two regions of each hemisphere. In the Figure 2 example, regions 0 and 1 are entirely within the target, region 2 is partially within the target, and region 3 is outside the target. For example, each region is 1.5 mm long and the two regions are spaced 0.5 mm apart.

[0169] If only the coordinates of the target points are sought, then p = 3 corresponds to the three coordinates x PC , yPC, zPC of the target points in the mathematical coordinate system used.

[0170] In the case of deep brain stimulation, if the direction of the electrode of the stimulation point is also sought, then p = 6 corresponds to the target x PC , y PC , z PC plus the coordinates v xPC , v yPC , v zPC of the direction vector of the electrode along the three axes of the mathematical coordinate system used.

[0171] Other coordinates can be added according to the planned neurosurgical intervention. For example, for gamma knife or focused ultrasound treatment, the coordinates of the target points can be adjusted.

[0172] If it is desired to establish the coordinates of the target points for the right hemisphere on the one hand and for the left hemisphere on the other hand, then p = 6 (only Cartesian coordinates) or p = 12 (including vector coordinates).

[0173] Since there may be one or more target points (or regions) in the target, the value p is equal to the number of points multiplied by the coordinates used for each point.

[0174] Learning database

[0175] For patient i (i = 1...n), there is a vector X of landmark points i (x of 3×m values i ) and a vector Y of target points i (y of p values i ). The vectors (X i , Y i ) i＝1...n extracted from n patients define the entire learning database.

[0176] Learning steps (constructing the function Y = F(X))

[0177] Using a supervised statistical learning method, the method according to the present invention is used to construct a prediction function F such that Y≈F(X).

[0178] In this specification, the prediction function may be abbreviated as "function".

[0179] The function F takes X or a part of X as an argument to generate Y.

[0180] For a new patient for whom the target must be determined, the vector X' of landmark points must be determined from an MRI (e.g., a preoperative MRI performed before surgery).

[0181] Then the position Y' of the target is determined using the function F as shown below:

[0182]

[0183] is the value obtained for the position Y' of the target according to the method of the present invention.

[0184] According to a specific embodiment, the supervised statistical learning method includes: using the kernel ridge regression method in a reproducing kernel Hilbert space.

[0185] The following description describes this specific embodiment.

[0186] Kernel ridge regression is a relatively simple method. It generally uses the entire learning database and is applicable to "small" learning databases (number of patients (n) less than 20) or "medium" databases (number of patients (n) between 20 and 100).

[0187] In the Hilbert space the least squares method and a regularization term are used Seek the function F, where λ is the regularization coefficient. Regularization ensures the norm and thus the function F does not degenerate.

[0188] Therefore, the selected Hilbert space is a reproducing kernel Hilbert space based on a Gaussian kernel:

[0189]

[0190] According to the property given by the following equation (1), it can be said that the kernel "reproduces" in the space :

[0191]

[0192] where is the scalar product in

[0193] Consider the following regularized least squares problem in the Hilbert space : Given the learning database where n is the number of elements in the database, the goal is to solve the following equation (2):

[0194]

[0195] where λ > 0 is the given regularization coefficient.

[0196] The Riesz representation theorem states that the solution of equation (2) can be written as:

[0197]

[0198] where is a new unknown.

[0199] By defining the Gram matrix K = (K ij ) i,j＝1…n and K ij = K(x i , x j ), the equation (2) to be solved is equivalent to solving the following equation (3):

[0200]

[0201] whose solution is simply given by equation (4):

[0202] α = (K + λnI) -1 y (4)

[0203] I is the identity matrix in

[0204] Since the target (grouping one or more target points together) is not defined by scalar characteristics alone, but by a Y vector of size p, which contains the coordinates of each target point and / or the orientation of the stimulating electrodes to be positioned at each point (characterized by vector coordinates on three axes) and / or other coordinates depending on the planned neurosurgical intervention, the problem is to solve the following equation (5):

[0205] A = (K + λnI) -1 Y (5)

[0206] I is the identity matrix in

[0207] And

[0208] Each column of A corresponds to the coefficient α for a given coordinate of the target (between 1 and p). Each column of Y corresponds to the value of a given coordinate of the target (between 1 and p).

[0209] As is known, as described above, K is the Gaussian kernel defined by the Gram matrix:

[0210]

[0211] Reconstruction steps (for new patients)

[0212] Once the learning step is completed, i.e., the function F has been constructed, the target can be reconstructed, i.e., a vector of the coordinates of one or more target points can be determined based on, for example, a set of landmark points determined from the patient's MRI to determine using the following equation (7):

[0213]

[0214] where

[0215]

[0216] In the case of adding functional data to the database , the following notation is introduced.

[0217] At least one vector X i takes into account all preoperative anatomical data, such as 18 landmark points, but also functional data, such as the scoring of pathological symptoms. Thus, the vector X i will be of size 3×m + M, where M is the number of functional data, such as the parameters resulting from the patient's preoperative symptom evaluation (or "preoperative score").

[0218] For each patient i in the database, a vector postOpVector is definedi For example, the vector can include parameters generated by postoperative symptom evaluation (or "postoperative score"), and / or electrical parameters of the stimulation (frequency, amplitude, and duration).

[0219] The function to be minimized when considering the functional data is:

[0220]

[0221] where W is a weight function and measures the confidence of the patient data. This weight function can be chosen to be complex or simple. An example of the weight function is the reciprocal of the stimulation amplitude or the reciprocal of the current injected per second, which corresponds to the reciprocal of the product of the stimulation parameters.

[0222] In the latter case:

[0223]

[0224] Another example of the weight function is the reciprocal of the sum of the patient's postoperative scores, which reflects the fact that the lower the score, the better the electrode placement.

[0225] The weight function W can also take another expression that combines the postoperative score and the electrical parameters.

[0226] Selection steps

[0227] The function obtained by solving equation (5) does not include any choice of the landmark points or the coordinates of these points. In this sense, the 54 coordinates of the 18 landmark points extracted from the MRI (or 108 coordinates of 36 points if both cerebral hemispheres are considered) are used in the learning and reconstruction steps. However, some coordinates are irrelevant to the target. This has an impact on the accuracy when determining the target.

[0228] Preferably, the method includes a selection step to use only the coordinates of the landmark points relevant to the target.

[0229] The selection step may include using a projection operator where is the set of all projection operators in

[0230] By solving the following equation (10), the selection of the coordinates of the optimal landmark points can be combined with the construction of the prediction function F:

[0231]

[0232] This selection corresponds to the optimization of the projection operator π. In fact, this is done by calculating the sensitivity of the function to be minimized with respect to each feature (coordinates of the landmark points) considered.

[0233] This can improve the accuracy of target determination.

[0234] Parameter optimization steps

[0235] The method includes constructing a function F that depends on a relatively small number of parameters (which must be optimized to obtain the most accurate possible function and thus the most accurate (and therefore most effective) target): the regularization coefficient λ, the width σ of the Gaussian kernel, and the projection operator π.

[0236] The optimization of these parameters is solved iteratively.

[0237] The parameters λ, σ, and π are optimized using a "leave-one-out" type of "cross-validation" procedure.

[0238] The principle of "leave-one-out" type cross-validation is to use all but one (patient) of the examples (here patients) in the learning database to construct a prediction function and evaluate the performance of the constructed function on the example removed from the learning database. This process is repeated by repeating the process for each patient in the entire learning database.

[0239] According to one embodiment, the "leave-one-out cross-validation error" (also referred to as "LOOE") is used for the set values of σ, λ, and the operator π, which is obtained by the following formula (11):

[0240]

[0241] where S = K(K + λI) -1 and

[0242] The matrix K incorporates the projection operator π because:

[0243] K = (K ij ) i,j＝1...n and K i,j = k(π(x i ), π(x j ))

[0244] According to a first variant, two parameters σ and λ and the operator π are directly optimized using an algorithm for global optimization by evolutionary strategies. This algorithm is based on the CMA-ES (abbreviation for "Covariance Matrix Adaptation - Evolutionary Strategy") method; the following equation (12) can be solved:

[0245]

[0246] According to this first variant, the method includes the following steps corresponding to the direct optimization:

[0247] - Step 1: Construct / Initialize the Learning Database

[0248] ○ Load data

[0249] ○ Define the input data (landmark points PR)

[0250] ○ Define the output data (target points PC)

[0251] ○ Construct the initial function F, F P

[0252] - Step 2: Optimize the parameters

[0253] ○ Solve equation (12)

[0254] - Step 3: Construct the combined function F using the optimal parameters λ, σ, and π (π integrates the optimal coordinates of the landmark points PR) C

[0255] - Step 4: Reconstruct for a new patient:

[0256] ○ Perform a preoperative MRI (e.g., a 1.5T MRI)

[0257] ○ Identify 18 landmark points PR' visible in the MRI

[0258] ○ Use the combined function F C and the landmark points PR' to determine the target points PC' (the targets).

[0259] According to this first variant, the selection step and the optimization step are performed simultaneously, with the selection made by optimizing the projection operator π.

[0260] The problem with this first variant embodiment is the size of the projection operator π, which is 18 × 3 = 54, or 18 × 3 × 2 = 108 in the case of combining two hemispheres to reconstruct the target. In this case, the number of parameters to be optimized is too large because, in addition to the two parameters λ and σ, the number of possible operators π is 2 54 or 2 108 , depending on the size of the operator.

[0261] According to a second variant, particularly as described in reference Figure 5 , the two parameters σ and λ and the operator π are optimized indirectly. An iterative method is used that can optimize the parameters λ and σ and the projection operator π.

[0262] First, solve the following equation (13):

[0263]

[0264] Next, test the sensitivity of the LOOE function to one of the coordinates (k) of the removed marked points. When the k-th coordinate is removed, two possible results are obtained and processed:

[0265] - If the LOOE increases, it means that the target point and the k-th coordinate are relevant, and this coordinate must be retained;

[0266] - If the LOOE remains unchanged or decreases, it means that the target point and the k-th coordinate are not relevant, and the k-th coordinate may be removed.

[0267] Among the coordinates that may be removed, the coordinate with the least relevance to the target is removed.

[0268] Each time a coordinate is removed, the projection operator π (with one less coordinate) is updated and the parameters σ and λ are optimized using the LOOE function.

[0269] Repeat this process until no coordinates need to be removed. In this case, all the remaining coordinates are necessary to retain the best solution within the meaning of the LOOE.

[0270] According to this second variant, the method includes the following steps, corresponding to indirect optimization:

[0271] - Step 1: Construct / initialize the learning database

[0272] ○ Load data

[0273] ○ Define the input data (marked points PR)

[0274] ○ Define the output data (target points PC)

[0275] ○ Initialize the projection operator π

[0276] ○ Assume π is equal to the identity function and construct the initial function F, F P

[0277] - Step 2: Optimize the parameters

[0278] ○ Perform the first optimization of σ and λ using Equation (13)

[0279] ○ Perform a loop operation: when the coordinates existing in the operator π are not optimal (within the meaning of the LOOE):

[0280] · For each coordinate k of the marked points (k = 1 to 3×m):

[0281] · Perform the calculation of the LOOE without the coordinate k;

[0282] · If the LOOE remains unchanged or decreases, note that the coordinate k of the learning database may be removable; otherwise, retain the coordinate k and consider the next coordinate k+1;

[0283] · Remove the coordinate k from among the possibly removable coordinates min ; the coordinate k min is the one with the least relevance to the target within the meaning of the LOOE. In other words, if it is taken into account, it degrades the LOOE the most;

[0284] · Update the operator π in the absence of the coordinate k min and re-optimize σ and λ using equation (13).

[0285] ○ End this while loop.

[0286] - Step 3: Construct the merging function F using the optimal parameters λ, σ, and the operator π (which incorporates the optimal coordinates of the fiducial points PR) C

[0287] - Step 4: Reconstruct for the new patient

[0288] ○ Perform a preoperative MRI (e.g., a 1.5T MRI)

[0289] ○ Identify 18 fiducial points PR' visible in the MRI

[0290] ○ Use the merging function F C and the fiducial points PR' to determine the target point PC' (the target).

[0291] As Figure 5 shown, the method may also add other steps:

[0292] - Step 5: Evaluate the efficacy of the stimulation (more generally, the neurosurgical treatment) performed at the target determined according to the method: This step is carried out in the months following the treatment. For example, the UPDRS-3 scale for Parkinson's disease (STN and GPi targets) and the Fahn-Tolosa-Marin scale for essential tremor (VIM target) can be used;

[0293] - Step 0: Input to the learning database: Specifically, new case patients with the best clinical outcomes as determined by postoperative evaluation (whose targets are determined by the method of the present invention or otherwise) can be used to input the database.

[0294] Step 5 and / or Step 0 can enrich the learning database with data from new patients whose treatments have been evaluated as effective.

[0295] Other supervised statistical learning methods can be used, such as methods of the support vector machine (SVM) type, which use only a subset (support vectors) instead of the entire learning database, which is different from kernel ridge regression. This method can be used when the learning database is "large", i.e., when the number of patients (n) in the database is greater than or equal to 100.

[0296] When the learning database is very large, methods of the neural network type can also be used.

[0297] Advantageously, when the learning database is enriched with data from new patients until it reaches a critical size of 100 patients, methods of the support vector machine or neural network type can be used as supervised statistical learning methods to construct the function F(X)=Y.

[0298] Advantageously, a step of statistical classification (whether supervised or unsupervised) can also be introduced in order to define multiple classes in the learning database, each class containing a subset of patient data that is different from the data of another class (e.g., according to pathology, age, other diseases, etc.). Next, for each class, a meta-model is constructed, i.e., a prediction function F, for example as described in the second variant of reference Figure 5 as described.

[0299] Statistical classification (whether supervised or unsupervised) can also be performed according to the coordinates of the marked points of the individual patients in the database. In the latter case, these classes may group together patients with similar coordinates, and thus their targets may also be similar.

[0300] To construct a target for a given new patient, all that needs to be done is to identify which class the new patient belongs to and use the function F corresponding to that class to predict the target. This can improve the targeting accuracy.

[0301] Statistical classification can be performed using one of the following methods:

[0302] - Logistic regression

[0303] - SVM

[0304] - Hierarchical classification method

[0305] - Neural network

[0306] - Random forest

[0307] Any other suitable statistical classification method can be used.

[0308] The various embodiments presented can be combined with each other.

[0309] All or part of the following steps: steps of constructing and / or initializing a learning database, a learning step, a selection step, a parameter optimization step, a reconstruction step, defining a mathematical coordinate system and / or the landmark points described in the embodiments, can be used for any application of a method other than deep surgical stimulation, typically in preparation for any neurosurgical treatment that requires determination of at least one precise brain target.

[0310] Furthermore, the present invention is not limited to the above-described embodiments, but extends to any embodiment falling within the scope of the claims.

[0311] The method, system and computer program according to the present invention can find applications other than in preparation for deep brain stimulation. The present invention can be used in the preparation steps of a treatment of the "gamma knife" type to determine the area to which an ionizing beam is applied. It can also be used in the preparation steps of a high-intensity focused ultrasound (HIFU) treatment to determine the area to which an ultrasonic beam is applied.

[0312] Generally speaking, the present invention can find application in the preparation for any neurosurgical treatment that requires determination of at least one brain target.

Claims

1. A method for determining a stereotactic brain target, the target including at least one target point PC', the method being implemented on a processor before a neurosurgical treatment is performed at the target for a given pathology and including the following steps: - Selecting a plurality of clinical case patients, the results measured at at least one target point PC after treatment for the pathology being greater than or equal to a threshold for each of the patients, and performing postoperative imaging on each of the patients; - Selecting a mathematical coordinate system, which is an orthogonal Cartesian coordinate system; - Process the postoperative imaging to determine, for each selected clinical case, all or part of the coordinates (X PC , Y PC , Z PC , V xPC , V yPC , V zPC ) of the at least one target punctuation point PC in the selected coordinate system, where the coordinates are p in number and the value of p depends on the number of target punctuation points multiplied by the number of coordinates of each target punctuation point; - Selecting a plurality of brain landmark points PR, the plurality being m; - Process the postoperative imaging to determine the coordinates (X PR , Y PR , Z PR ) of the brain landmark points PR for each selected clinical case, where the coordinates are 3×m; - Creating a learning database that includes the determined coordinates of the target points PC of all selected clinical cases and the determined coordinates of the landmark points PR; - Determining a prediction function F by using the learning database and a supervised statistical learning method, the prediction function F giving the coordinates of at least one target point PC based on the landmark points PR; - Process preoperative imaging of a new patient to be treated for the pathology to determine the coordinates (X PR' , Y PR' , Z PR' ) of the fiducial point PR' of the new patient; - Using the prediction function F to obtain the coordinates of at least one target point PC' of the new patient based on the coordinates of the landmark points PR' determined for the new patient.

2. The method according to claim 1, further including the following steps: - Combine the prediction function F using a cross-validation method, and the combining step results in a combined prediction function F C , the prediction function F C Give the coordinates of at least one target point PC according to the marked point PR; The step of using the prediction function F includes: using the combined prediction function F C .

3. The method according to claim 1, the imaging processed to determine the coordinates of the landmark points PR, PR' and / or the at least one target point PC is at least one MRI image.

4. The method according to claim 3, the imaging processed to determine the coordinates of the landmark points PR, PR' and / or the at least one target point PC is a plurality of MRI images.

5. The method according to claim 2, the imaging processed to determine the coordinates of the landmark points PR, PR' and / or the at least one target point PC is at least one MRI image.

6. The method according to claim 5, the imaging processed to determine the coordinates of the landmark points PR, PR' and / or the at least one target point PC is a plurality of MRI images.

7. The method according to any one of claims 1-6, wherein the supervised statistical learning method comprises: Using a kernel ridge regression method in a reproducing kernel Hilbert space.

8. The method according to any one of claims 1-6, wherein the supervised statistical learning method comprises: Using a method of the support vector machine type.

9. The method according to any one of claims 1-6, wherein the supervised statistical learning method comprises: Using a method of the neural network type.

10. The method according to one of claims 2 and 5-6, wherein the cross-validation method comprises: Using a leave-one-out cross-validation method.

11. The method according to one of claims 2 and 5-6, wherein the cross-validation method comprises: Using a leave-k-out cross-validation method.

12. The method according to any one of claims 1-6, the mathematical coordinate system being an orthogonal Cartesian coordinate system, a line CACP passing through the upper and lower edges of the anterior commissure CA and the lower and upper edges of the posterior commissure CP forming the Oy axis, the lower and upper edges of the posterior commissure CP forming the center of the coordinate system, and the Oz axis being a line perpendicular to the line CACP in the interhemispheric plane.

13. The method according to claim 7, the mathematical coordinate system being an orthogonal Cartesian coordinate system, a line CACP passing through the upper and lower edges of the anterior commissure CA and the lower and upper edges of the posterior commissure CP forming the Oy axis, the lower and upper edges of the posterior commissure CP forming the center of the coordinate system, and the Oz axis being a line perpendicular to the line CACP in the interhemispheric plane.

14. The method according to claim 8, wherein the mathematical coordinate system is an orthogonal Cartesian coordinate system, a straight line CACP passing through the upper and lower edges of the anterior commissure CA and the lower and upper edges of the posterior commissure CP forms the Oy axis, the lower and upper edges of the posterior commissure CP form the center of the coordinate system, and the Oz axis is a straight line perpendicular to the straight line CACP in the interhemispheric plane.

15. The method according to claim 9, wherein the mathematical coordinate system is an orthogonal Cartesian coordinate system, a straight line CACP passing through the upper and lower edges of the anterior commissure CA and the lower and upper edges of the posterior commissure CP forms the Oy axis, the lower and upper edges of the posterior commissure CP form the center of the coordinate system, and the Oz axis is a straight line perpendicular to the straight line CACP in the interhemispheric plane.

16. The method according to claim 10, wherein the mathematical coordinate system is an orthogonal Cartesian coordinate system, a straight line CACP passing through the upper and lower edges of the anterior commissure CA and the lower and upper edges of the posterior commissure CP forms the Oy axis, the lower and upper edges of the posterior commissure CP form the center of the coordinate system, and the Oz axis is a straight line perpendicular to the straight line CACP in the interhemispheric plane.

17. The method according to claim 11, wherein the mathematical coordinate system is an orthogonal Cartesian coordinate system, a straight line CACP passing through the upper and lower edges of the anterior commissure CA and the lower and upper edges of the posterior commissure CP forms the Oy axis, the lower and upper edges of the posterior commissure CP form the center of the coordinate system, and the Oz axis is a straight line perpendicular to the straight line CACP in the interhemispheric plane.

18. The method according to any one of claims 1-6 and 13-17, wherein the marked point PR is selected from among the following eighteen points: - The first marked point FMT is the mammillothalamic tract on the third axial plane; - The second, third, and fourth marked points PA1, PA2, PA3 are the anterior points of the putamen on each of the first, second, and third axial planes; - The fifth and sixth marked points PM1, PM2 are the middle points of the putamen on the first and second axial planes; - The seventh and eighth marked points PP1, PP2 are the posterior points of the putamen on the first and second axial planes; - The ninth marked point CH is the habenular commissure on the second axial plane; - The tenth marked point BAT is the anterior edge of the thalamus on the second axial plane; - The eleventh marked point BPT is the posterior edge of the thalamus on the second axial plane; - The twelfth marked point CA is the anterior commissure; - The thirteenth marked point A is the middle edge of the third ventricle at the mid-commissure point; - The fourteenth marked point B is the height of the thalamus on the sagittal plane passing through the thirteenth marked point A; - The fifteenth marked point C is the midpoint of the line segment AB defined by the thirteenth and fourteenth marked points; - The sixteenth marked point D is the anterior edge of the thalamus on a straight line parallel to the straight line CACP passing through the upper and lower edges of the anterior commissure and the lower and upper edges of the posterior commissure and passing through the fifteenth marked point C; - The seventeenth landmark point P culm is the upper edge of the putamen on the coronal plane passing through the fifth landmark point PM1; and - The eighteenth landmark point P lat is the lateral margin of the putamen on the coronal plane passing through the fifth landmark point PM1.

19. The method according to any one of claims 1-6 and 13-17, further comprising an additional step of adding functional data to the learning database, the functional data being capable of adding at least one confidence index to the marked point PR and the target point PC of the clinical case.

20. The method according to claim 18, further comprising the additional step of adding functional data to the learning database, the functional data being capable of adding at least one confidence metric to the fiducial point PR and the target point PC of the clinical case.

21. The method according to any one of claims 1-6, 13-17 and 20, which is implemented on a processor prior to deep brain stimulation, gamma knife or focused ultrasound treatment.

22. A data processing system comprising a processor, which is configured to implement the steps of the method according to any one of claims 1-21.

23. A computer program product comprising instructions, which when executed by a processor, cause the processor to implement the steps of the method according to any one of claims 1-21.

Images