Method for identifying surgically operable target areas in the brains of epilepsy patients

Abstract

The present invention relates to a method for identifying potential surgically operable target areas in the brain of an epilepsy patient. According to the present invention, the method comprises the following steps: providing a computerized platform that models the various areas of the primate brain and the connectivity between the areas; providing a model of the epileptogenic area and a model of the propagation of epileptic discharges from the epileptic area to the propagation area; obtaining a personalized computerized platform for the patient; deriving potential target areas based on modularity analysis; evaluating the effectiveness of the target areas by simulating epileptic seizure propagation in the personalized patient computerized platform; evaluating the safety of the target areas by simulating the spatio-temporal brain activation pattern under defined state conditions and comparing the simulated spatio-temporal brain activation pattern obtained before removing the target areas with the spatio-temporal brain activation pattern obtained after removing the target areas; and identifying the target areas that meet both the effectiveness and safety evaluation criteria as potential surgically operable target areas.

Description

[0001] The present invention relates to a method for identifying a surgically operable target zone in the brain of an epilepsy patient.

[0002] Epilepsy is a chronic neurological disorder defined by the occurrence of repetitive, unanticipated seizures. Seizures, characterized by abnormal synchronization of neural activity, originate in specific brain regions and spread to other regions through the structural interactions between regions that make up an individual's connectome, and produce various ictal symptoms depending on the brain regions recruited.

[0003] For the treatment of epilepsy, pharmacological treatment with antiepileptic drugs is preferably applied, and surgical intervention is often offered as an option for drug-resistant patients (who account for more than 30% of patients). There are two main types of surgical strategies: resection and disconnection. Resection - which removes the brain region generating the seizures - results in seizure-free outcomes in 30 - 70% of postoperative patients, depending on the localization accuracy of the epileptogenic zone (EZ) and the pathology of each patient. Disconnection - which severs the neural pathways that play an important role in seizure spread - can either be curative (i.e., hemispherectomy) or can limit seizure spread (i.e., corpus callosotomy). Although surgical intervention is generally accepted as an effective method for controlling drug-resistant seizures, only about 10% of patients may be considered candidates for surgery because the EZ is typically located in multiple brain regions simultaneously and involves eloquent areas, which are defined as brain regions where lesions would cause neurological complications (such as language, memory, and motor problems). For patients not suitable for conventional surgery, several alternative methods have been attempted, including multiple subpial transection, which can prevent neuronal synchronization in the EZ without altering normal function by severing horizontal intracortical fibers while preserving vertical fibers in the eloquent cortex, but these alternative methods have variable outcomes. Therefore, there is clearly a need to provide more optimized surgical options for these patients. The alternative methods should: 1) be effective in reducing seizures; 2) be able to provide flexible options depending on inoperable EZs or surgically inaccessible areas; and 3) have a minimal impact on normal brain function.

[0004] Research on epilepsy has mainly focused on studying the brain network dynamics of individual patients. By analyzing functional data (such as intracranial electrocorticographic (ECoG) signals and stereotactic electroencephalographic (SEEG) signals), many studies have examined the network properties in each brain state, including interictal, preictal, ictal, and postictal. In particular, graph theory-based network analysis has been able to not only identify the characteristics of the seizure onset zones that will be targeted in resection surgeries, but also observe the changes in network topology in the onset and time course of epileptic seizures. Several studies have shown that a large regular network is formed during epileptic seizures compared to a network composed of several small subnetworks during the interictal period. These results suggest that epileptic seizures can be prevented by disrupting the formation of the large regular network by disconnecting carefully selected subnetworks. In addition, several other studies have demonstrated that epileptic brain networks have more segregated features than healthy brain networks. At the same time, by analyzing MRI-based structural data, many studies have reported structural abnormalities in the epileptic brain that are different from the normal brain, which include not only regional changes but also abnormalities in white matter tracts, i.e., abnormalities in interregional connectivity. From a network perspective, several studies have shown an increase in local network connectivity and a decrease in global network connectivity in the epileptic brain, although the situation is more complex depending on whether the brain regions are involved in seizure generation and propagation. It has also been further reported that healthy brain networks exhibit a wide distribution of hub regions, while epileptic brain networks have hub regions concentrated in specific areas (e.g., in temporal lobe epilepsy, paralimbic / limbic, and temporal association cortices). The results of these studies suggest that the epileptic brain includes a distinct modular structure, and epileptic seizure propagation can be controlled by blocking the interactions between modules (i.e., by cutting connections).

[0005] The translation of any computational modeling method requires the personalization of a brain network model customized for the connectivity and lesions of the patient. Personalized brain network models based on the connectome and clinical information of each patient have been able to simulate individual epileptic seizure propagation patterns.

[0006] Currently, efforts in this field focus on improving the localization of the EZ and developing strategies for the effective removal of the identified areas.

[0007] Accordingly, there is a need for methods that allow the identification of minimally invasive surgical interventions, particularly methods applicable to situations where the EZ is not operable.

[0008] According to a first aspect, the present invention relates to a method for identifying potentially surgically operable target areas in the brain of an epileptic patient, comprising the steps of:

[0009] providing a computerized platform that models various regions of the primate brain and the connectivity between said regions;

[0010] providing a model of the epileptogenic zone and a model of the propagation of epileptic discharges from the epileptogenic zone to the propagation zone, the model of the epileptogenic zone being a mathematical model that describes the occurrence, time course, and offset of epileptic discharges, and loading said model into the computerized platform to obtain a computerized platform for modeling the epileptic primate brain;

[0011] identifying the estimated epileptogenic zone in the patient's brain;

[0012] personalizing the computerized platform for modeling the epileptic primate brain according to the patient's brain structural connectivity and parameterizing the estimated epileptogenic zone as the epileptogenic zone in said computerized platform to obtain a personalized computerized platform for the patient;

[0013] performing a modularity analysis using the patient's brain structural connectivity to derive potential target areas that act as hubs in the interactions between modules, said potential target areas being outside the potential epileptogenic zone and such that if they are surgically manipulated or removed, they would minimize the propagation of epileptic seizures, and evaluating the effectiveness of said potential target areas in minimizing the propagation of epileptic seizures by simulating the propagation characteristics of epileptic seizures and identifying a network simulation of one or more effective target areas in the personalized patient computerized platform, said effective target areas being outside the epileptogenic zone and such that if they are surgically manipulated, they would minimize the propagation of epileptic seizures;

[0014] The safety of maintaining normal brain function in the potential target area is evaluated through network simulation, wherein simulation spatio-temporal brain activation patterns under defined state conditions are obtained from a personalized computerized platform before and after removing the area, and these simulation spatio-temporal brain activation patterns obtained before removing the area are compared with the simulation spatio-temporal brain activation patterns obtained after removing the area. If the spatio-temporal brain activation patterns obtained before removing the area are substantially the same as those obtained after removing the area, the potential target area is identified as a safe target area; and

[0015] Potential target areas that meet both the effectiveness and safety evaluations are identified or recommended as potential target areas that can be surgically operated on.

[0016] Preferably, - clinically estimate the epileptogenic area in the patient's brain; - the target area is a node or edge involved in the propagation of epileptic seizures, and the node and edge correspond to a brain region and a fiber bundle between brain regions, respectively; - reconstruct the structural brain connectivity from the image data of the patient's brain obtained using magnetic resonance imaging, diffusion-weighted magnetic resonance imaging, nuclear magnetic resonance imaging, and / or magnetic resonance tomography; - personalize a computerized platform for modeling the epileptic primate brain based on the patient-specific brain connectivity and the patient's functional data; - the functional data is obtained through electroencephalogram (EEG) or stereotactic EEG (SEEG) techniques; - perform modularity analysis to derive the potential target area; - to achieve modularity analysis, add constraint conditions to prevent inoperable nodes from being derived as target areas; - perform system simulation in the patient's personalized computerized platform, wherein if the identified target area does not meet the evaluation criteria, new target areas are derived by feeding the simulation results back to the analysis again; - the defined state condition is the resting state condition; - use multiple resting state conditions for simulation; and - modularity analysis provides a non-overlapping modular structure that minimizes the edges between modules and maximizes the edges within modules.

[0017] Accordingly, focusing on the fact that epileptic brain networks have distinct segregation properties, the present invention employs modularity analysis utilizing the structural brain connectivity from each patient in order to derive brain regions and fiber tracts as the target zones (TZ) that should be removed for resection and disconnection surgeries, respectively. Assuming a worst-case scenario where the EZ is inoperable, such that the proposed in silico surgical method induces seizure remission by suppressing the spread of seizures to other brain regions, although it cannot prevent seizure generation in the EZ. Reducing the involvement of the propagation network is a major factor in reducing the impact of seizures, particularly loss of consciousness. The obtained TZ is evaluated in terms of the effectiveness in controlling seizure spread and the safety in maintaining normal brain function through personalized brain network simulations, and then optimized based on the results.

[0018] Other features and aspects of the present invention will become apparent from the following description and the accompanying drawings, in which:

[0019] Figures 1A to 1G Illustrates a brain simulation performed using a virtual brain according to the method of the present invention. Figures 1A to 1D Shows a network simulation for effectiveness evaluation. The brain network model evaluates effectiveness by identifying seizure propagation characteristics. These figures show the simulation signals at each brain node before ( Figure 1B ) and after ( Figure 1C and 1D ) removing node 23 or node 21 when nodes 3, 22, and 27 act as the epileptogenic zones, respectively. Figures 1E to 1G Shows a network simulation for safety evaluation. The brain network model evaluates safety by studying the integrity of the instantaneous spatio-temporal trajectories after electrical stimulation at certain nodes. When the stimulation is applied to node 3, the stimulation induces different response signals (solid lines, Figure 1F and 1G ) at each node. When all connections from node 23 are removed, the response signals (dashed lines, Figure 1F ) at each node are changed compared to before removal. On the other hand, when node 21 is eliminated, the response signals (dashed lines in Figure 1G ) are not significantly different from the response signals before removal. The color bar represents the similarity coefficient between the response signals at each node before and after elimination.

[0020] Figures 2A to 2D Illustrates the target zones derived from modularity analysis for a specific patient according to the method of the present invention. Figure 2AShows the modular structure when the epileptogenic zone is set as the inoperable zone with a resolution parameter of 1.25. The brain network is divided into seven modules, and the EZ sub-module (the upper module in this figure) is subdivided into four sub-modules such that each EZ (nodes 61, 64, large circles) and its adjacent nodes belong to the same sub-module. Based on this modular structure, three nodes (black triangles) and eight edges (gray dashed lines) are derived as the target nodes and target edges respectively. For visualization, only the edges with a connection weight greater than 0.08 are drawn. Figure 2B Shows the list and anatomical locations of the obtained target areas. The middle gray nodes represent the epileptogenic zone, and the light gray nodes and gray edges indicate the target nodes and target edges. Figure 2C Shows the modular structure when key nodes (gray triangles) are added to the inoperable zone in modularity analysis with a resolution parameter of 1.25. The brain network is divided into 8 modules, and the epileptogenic zone sub-module (the upper module) is subdivided into 2 sub-modules such that each inoperable zone and its adjacent nodes belong to the same sub-module. Based on this modular structure, three nodes (black triangles) and five edges (gray dashed lines) are derived as the new target nodes and target edges respectively. Figure 2D Shows the list and anatomical locations of the newly obtained target areas. The middle gray nodes represent the epileptogenic zone, and the light gray nodes and green edges indicate the target nodes and target edges.

[0021] Figure 3A and 3B Illustrates the safety evaluation of the target area according to the method of the present invention. Figure 3A Is the network simulation result of the safety assessment. When a stimulus is applied to the brain nodes that can reproduce each resting state network, the difference between the response signals before and after removing the target area is presented as a similarity coefficient. The similarity coefficient is calculated independently for all brain nodes, and the gray shading in this figure indicates the value of the similarity coefficient. Figure 3B Illustrates the identification of key nodes. When each node belonging to the initially obtained target nodes is eliminated, the degree of change in the response network corresponding to the memory network M is expressed as a similarity coefficient.

[0022] Figures 4A to 4D Illustrates the network simulation results for effectiveness verification according to the method of the present invention, as well as the local field potentials in all brain nodes. They further illustrate the propagation characteristics of the seizures originating from the EZ (nodes 61 and 64). Figure 4A The results shown in are the results obtained before removing the target area, Figure 4B The results shown in are the results obtained after removing three target nodes, Figure 4C The results shown in are the results obtained after removing five target edges, andFigure 4D The results shown were obtained after removing 3 random nodes.

[0023] Figure 5A and 5B Illustrates the safety evaluation results for the new target area obtained through feedback and the initial target area according to the method of the present invention. The histogram illustrates the average of the similarity coefficients between the response activation patterns due to stimulation in all brain regions before and after removing the target nodes and target edges, respectively. Although the elimination of the initial TZ has a value below the threshold (0.75) when the stimulation is applied to node 10 to reproduce the memory network (M), the removal of the new TZ has a value above this threshold at all stimulation sites.

[0024] Figures 6A to 6C Illustrates that the target area depends on the location of the epileptogenic zone. Figure 6A Shows the target nodes according to the location of the epileptogenic zone and their cumulative results. The gray horizontal bars in each column indicate the target nodes when the epileptogenic zone is located in each different node. Figure 6B Shows the target edges according to the location of the epileptogenic zone and their cumulative results. The connecting lines between the nodes in each slice indicate the target edges when the epileptogenic zone is located in each different node. The cumulative results identify several nodes and edges that are frequently used as the target area. Here, the resolution parameter for modularity analysis is set to 1.0. Figure 6C Shows the anatomical locations of the nodes and edges that are frequently obtained as the target area. The gray shading for the nodes and the thickness of the edges indicate the frequency of being used as the target area.

[0025] The present invention relates to a method for identifying surgically operable target areas in the epileptic brain of a primate epilepsy patient. The patients are particularly human patients who are drug-resistant epilepsy patients. The method according to the present invention constitutes a bioinformatics surgical method that is based on graph-theoretic analysis (specifically modularity analysis) using the patient-specific connectome of the brain, as well as personalized brain network simulation, and that suggests effective and safe intervention options by minimizing the impact on the signal transmission ability of the brain.

[0026] The method according to the present invention includes the following steps, according to which a computerized platform, i.e., a brain network, is provided that models the various regions of the primate brain and the connectivity between said regions. Such a computerized platform constitutes a virtual brain. In " The Virtual Brain:a simulator of primate brain network dynamicsExamples of virtual brains are disclosed in the publication document of "", which is incorporated herein by reference. In this document, the virtual brain is disclosed as a neuroinformatics platform for whole-brain network simulation using biologically realistic connectivity. This simulation environment enables model-based inference of neurophysiological mechanisms across different brain scales, which are the basis for generating macroscopic neuroimaging signals including functional magnetic resonance imaging (fMRI), EEG, and magnetoencephalogram (MEG). It allows the reproduction and evaluation of the personalized configuration of the brain by using individual subject data.

[0027] According to a further step of the present invention, a model of the epileptogenic zone (EZ) and a model of the propagation of epileptic discharges from the epileptic zone to the propagation zone (PZ) are provided. These models are then loaded into the computerized platform to obtain a computerized platform for modeling the epileptic primate brain.

[0028] The model of EZ is a mathematical model that describes the occurrence, time course, and offset of epileptic discharges. Such a model is disclosed in a publication document entitled "", Jirsa et al., Brain 2014, 137, 2210-2230, which is incorporated herein by reference. This model is named Epileptor. On the nature of seizure dynamics

[0029] The model of PZ is the same as the model of EZ, however, its excitability parameter is lower than the critical value x 0C = −2.05. All other brain regions can be modeled by Epileptor with excitability values far from this threshold, or equivalently by a standard neural population model, as disclosed by Paula Sanz Leon et al. on June 11, 2013, which is incorporated herein by reference. The coupling between brain regions follows a mathematical model disclosed in a publication document entitled "", Timothée Proix et al., The Journal of Neuroscience, November 5, 2014, 34(45): 15009-15021, which is incorporated herein by reference. Permittivity Coupling across Brain Regions Determines Seizure Recruitment in Partial Epilepsy

[0030] According to a further step of the present invention, the structural and functional data of the brain of an epileptic patient are obtained. The brain connectome is reconstructed based on the structural data, and the epileptogenic zone is estimated based on the functional data.

[0031] The structural data is, for example, image data of a patient's brain obtained using magnetic resonance imaging (MRI), diffusion-weighted magnetic resonance imaging (DW-MRI), nuclear magnetic resonance imaging (NMRI), or magnetic resonance tomography (MRT). The functional data is, for example, EEG or SEEG signals. The estimation of the epileptogenic zone can be clinical or can be provided using non-clinical methods, such as the method disclosed in the international application published under number W02018 / 015779.

[0032] According to a further step of the present invention, a computerized platform for modeling an epileptic primate brain is personalized according to the brain structural connectivity of the patient. The estimated epileptogenic zone is further parameterized as an epileptogenic zone in the computerized platform to obtain a personalized computerized platform for the patient, i.e., a personalized brain network.

[0033] According to a further step of the present invention, potential target zones are derived based on modularity analysis, and the effectiveness of the target zones is evaluated through network simulation. The propagation characteristics of epileptic seizures are simulated in the personalized patient computerized platform, and one or more effective target zones are identified. The effective target zones are outside the epileptogenic zone. In addition, they are such that if they are manipulated or removed surgically, they minimize the propagation of epileptic seizures.

[0034] According to a further step of the present invention, the safety of the potential target zones is evaluated through network simulation. The simulated spatio-temporal brain activation patterns under the defined state conditions are obtained from the personalized computerized platform before and after removing the zones. The defined state condition is, for example, the resting state condition. However, this can be other state conditions, for example, a state condition where the brain is in a memory mode, a state where the brain is in a mode of perceiving things, or a state where the brain is in an attention mode. According to a further step of the present invention, the simulated spatio-temporal brain activation pattern obtained before removing the target zone is compared with the spatio-temporal brain activation pattern obtained after removing the zone. If the spatio-temporal brain activation pattern obtained before removing the zone is substantially the same as the spatio-temporal brain activation pattern obtained after removing the zone, the potential target zone is identified as a safe target zone.

[0035] According to a further step of the present invention, potential target zones that meet both the efficacy and safety evaluation criteria are proposed as target zones that can be surgically operated on. To evaluate the efficacy and safety of the identified TZs, brain network simulations are employed. Based on a patient-specific network model constructed from the structural brain connectivity and the estimation of the EZ for each patient, the efficacy of the TZ is evaluated by simulating the seizure propagation characteristics before and after removing the TZ. Reducing the participation of the propagation network is a major factor in reducing the impact of seizures, particularly loss of consciousness. Loss of consciousness is one of the main signs and is significantly associated with synchronization in the propagation network, particularly with synchronization in the fronto-parietal network during temporal lobe seizures. It is recognized that good outcomes after epilepsy surgery can include patients having residual subjective symptoms (aura), but no more objective signs (automatisms, loss of consciousness).

[0036] According to the method of the present invention, modularity analysis is generally used to study the synchronization characteristics between brain regions. The brain network of each patient has a distinct modular structure. Based on the patient-specific modular structure, the nodes and edges that connect the EZ sub-module to other sub-modules or modules are extracted as the surgical option TZ, thereby suppressing seizure propagation in a patient-specific manner. By adding constraints to the existing modularity analysis, a flexible TZ that excludes inoperable zones is derived, which can provide an alternative surgical method that can produce seizure remission for patients who are considered unsuitable for conventional surgery - due to the fact that EZ resection may lead to severe neurological complications. In addition, parameter sweeps in modularity analysis obtain different modular structures, ultimately generating multiple TZ scenarios. This multiplicity is crucial because, considering the number of interventions and the degree of seizure suppression, clinicians can select the surgical target within multiple options. Clinicians can also consider not only specific regions that should be excluded from surgery based on their clinical experience, but also technically challenging regions.

[0037] Based on the patient-specific modular structure obtained from the structural brain connectivity and the estimation of the EZ for the patient (particularly clinical estimation), the brain regions and fiber bundles that act as hubs (i.e., connect different modules) in the interaction between modules are identified as the TZs for surgical intervention. The obtained TZs are evaluated for their efficacy and safety through personalized brain network simulations.

[0038] In fact, considering the location of the EZ and the individual brain connectome, the results include several TZ variants suitable for each patient's situation. The final TZ variant is an effective surgical target for preventing seizure propagation while maintaining normal brain function.

[0039] System simulations allow for the identification of different TZs based on the location of the EZ. These results can be used not only to identify the main nodes and edges involved in seizure propagation, but also, if there are several clinical hypotheses regarding the location of the EZ, as a reference for deriving reasonable surgical targets.

[0040] Since the impact on seizure reduction is highly dependent on the location or contribution of each node in the network, a systematic method is performed according to the present invention to identify target nodes. In addition, the network impact is studied at the whole-brain scale. In particular, the derived TZs are based on clinical estimates and brain connectivity analysis for each patient, and the impact of TZ removal in the seizure propagation network is examined through personalized brain network simulations based on the individual connectome.

[0041] Critical for surgical intervention outside the EZ is the safety study of the process according to the present invention. Assuming that the signal transmission properties of the brain network are directly related to brain function, the concept of preservation of these signal transmission properties operationalizes safety. Advantageously, at least implicitly, brain functional capacity is quantified through functional connectivity in the resting state (RS). This attempts to quantify the nature of the attractor state at rest through construction. The TZ is evaluated by distinctly quantifying the changes in network characteristics in the resting state in pre- and post-surgical conditions. Perturbation of the attractor state allows for sampling of additional properties of the brain network, such as attractor stability, convergence and divergence of the flow, and thus significantly enhances the characterization of its dynamic properties. Stimulation is a reliable way to induce perturbations of various states, which generates spatio-temporal response patterns according to the stimulation location and brain connectivity. Stimulation is used to reproduce each RS network, and the changes in network characteristics before and after eliminating the TZ are explicitly quantified. To best estimate the instantaneous spatio-temporal trajectories due to stimulation applied to individual brain regions, the spatial and temporal properties before and after eliminating the TZ are compared. Thus, the instantaneous trajectories are highly constrained by the structural properties of the network and show unexpectedly low-dimensional behavior after the initial local stimulation artifact. These instantaneous trajectory properties are used to quantify the differences in the response network due to stimulation, and it is assumed that changes in the response pattern after TZ removal indicate a negative impact on brain functionality, i.e., if the differences in the response pattern before and after TZ removal are large, the removal of the TZ is interpreted as unsafe.

[0042] Figure 1 illustrates an example of target zone evaluation. As Figures 1A to 1D shown, the effectiveness of controlling seizure propagation is evaluated by the degree of seizure propagation inhibition. Figure 1BThe figure shows the simulated signals at several brain nodes identified in Figure 1A when nodes 3 (ctx-lh-caudal middle frontal), 22 (ctx-lh-posterior cingulate), and 27 (ctx-lh-superior frontal) are EZs. After a seizure is generated from the EZ, these nodes seem to be mobilized by the seizure with some delay depending on the connectivity between the nodes. On the other hand, after removing a specific node in the seizure propagation path (i.e., node 23 (ctx-lh-precentral) or node 21 (ctx-lh-postcentral)), the simulated brain signals show that the propagation beyond each node can be prevented even if a seizure still occurs from the EZ. This is shown in Figure 1C and 1D . In this way, the characteristics of seizure propagation are observed after eliminating the target node or target edge for the effectiveness evaluation of the TZ. Figure 1A After a seizure is generated from the EZ, these nodes seem to be mobilized by the seizure with some delay depending on the connectivity between the nodes. On the other hand, after removing a specific node in the seizure propagation path (i.e., node 23 (ctx-lh-precentral) or node 21 (ctx-lh-postcentral)), the simulated brain signals show that the propagation beyond each node can be prevented even if a seizure still occurs from the EZ. This is shown in Figure 1C and 1D . In this way, the characteristics of seizure propagation are observed after eliminating the target node or target edge for the effectiveness evaluation of the TZ. Figure 1C and 1D In this way, the characteristics of seizure propagation are observed after eliminating the target node or target edge for the effectiveness evaluation of the TZ.

[0043] The safety evaluation of this intervention relies on maximizing the signal transmission properties of the brain network. The latter is evaluated by stimulating the relevant brain regions and quantifying the subsequent transient trajectories of brain network activation. More specifically, the safety is evaluated by assessing the similarity of the spatio-temporal brain activation patterns after electrical stimulation before and after removing the TZ. To study the changes in the resting state (RS) network, the brain regions where the stimulation is applied can reproduce a response network similar to each of the 8 well-known RS networks.

[0044] Figures 1E to 1G The figure shows an example of the response network when the stimulation is applied to a specific node - node 3. The stimulation first locally activates the stimulated node, and then propagates and sequentially mobilizes through the connectome, thereby generating a unique spatio-temporal response pattern specific to the stimulation site. The solid and dashed lines show the situation after eliminating a specific node - node 23 in Figure 1F and node... in Figure 1G Figure 1F in Figure 1F and is node 23 and in Figure 1GIn the middle is Node 21 - the simulated signals obtained from several brain nodes before and after. Compared with the response pattern before removal, the response signal is changed after removing Node 23, while the response signal seems unaffected after removing Node 21. The bar quantitatively represents the degree of these differences, that is, as a similarity coefficient. These results illustrate the sensitivity of spatio-temporal seizure organization to network changes. In this way, by systematically stimulating to reproduce specific nodes of each RS network and comparing the response patterns before and after eliminating the target node or target edge, the safety of the TZ is evaluated. If the TZ is judged to be insufficient based on the network simulation results, another TZ is derived by applying the results to modularity analysis again. Through this feedback method, an optimized TZ can be obtained, which effectively prevents seizure propagation while minimally affecting normal brain function.

[0045] Finally, the method according to the present invention proposes a personalized bioinformatics surgical method that can recommend effective and safe surgical options for each epilepsy patient. It particularly focuses on deriving effective alternative methods for those cases where the EZ is inoperable due to problems related to neurological complications. Preferably based on modularity analysis using the structural brain connectivity from each patient, the TZ considered to be the surgical site is obtained. The obtained TZ is evaluated in terms of effectiveness and safety through personalized brain network simulation. Through a feedback method that combines modularity analysis and brain network simulation, an optimized TZ option that minimizes seizure propagation without affecting normal brain function is obtained. It demonstrates the possibility of the field of computational neuroscience to construct a personalized medicine paradigm by deriving innovative surgical options suitable for each patient and predicting surgical outcomes.

[0046] Example 1: Materials and Methods

[0047] The method according to the present invention is advantageously based on graph theory analysis and brain network simulation. Preferably, based on modularity analysis considering the inoperable area, the brain regions and fiber bundles that act as hubs in the interaction between modules are derived as the TZ. Then, using the Virtual Brain (TVB) - a platform for simulating brain network dynamics - the obtained TZ is evaluated in terms of effectiveness and safety through personalized brain network simulation. If the TZ does not meet the evaluation criteria, a new TZ is derived by feeding back the simulation results to modularity analysis again. Through this feedback method, an optimized TZ option is obtained, which minimizes seizure propagation without affecting normal brain function.

[0048] Structural brain network reconstruction

[0049] Neuroimaging data were obtained from seven patients with drug-resistant epilepsy. These patients had EZs at different locations and underwent a comprehensive preoperative evaluation. Using the SCRIPTS TM pipeline, the structural brain network of each patient was reconstructed based on diffusion MRI scans and T1-weighted images (Siemens Magnetom Verio TM 3T MR scanner). Each patient's brain was divided into 84 regions, including 68 cortical regions based on the Desikan-Killiani atlas and 16 subcortical regions. The connection strength between brain regions was defined based on the number of streamlines as fiber bundles, and the bundle length for determining the signal transmission delay between regions was also derived.

[0050] Target region derivation based on patient-specific modular architecture

[0051] To analyze the modular structure of the brain network, a Matlab toolbox was used. The modularity analysis based on the Newman spectral algorithm provided a non-overlapping modular structure that minimized the edges between modules and maximized the edges within modules. However, modularity analysis can be performed using other toolboxes that may be based on the Newman spectral algorithm or other algorithms. For example, another such toolbox, which is a Matlab toolbox, is disclosed in the document "Complex network measures of brain connectivity, uses and interpretations" by Mikail Rubinov et al. in NeuroImage, Vol. 52, Issue 3, Sept. 2010, p. 1059-1069. The modularity analysis performed using the Matlab toolbox based on the Newman algorithm allowed the calculation of the leading eigenvector of the modularity matrix of the following equation B and the network nodes were partitioned into two modules according to the signs of the elements in this eigenvector.

[0052]

[0053] In this equation, A ij represents the weight value between node i and node j , k i and k j indicate the degree of each node, mRepresents the total number of edges in the network. α Is the resolution parameter for this analysis, and the classical value is 1. The partition is finely adjusted by the node movement method to obtain the maximum modularity coefficient Q. The modularity coefficient has values ranging from 0 to 1, and a value of 0.3 or higher usually indicates a good partition. s i and s j Represents the group membership variable, whose value is +1 or -1, depending on the group to which each node belongs. Each module partitioned based on the eigenvector algorithm is further partitioned into 2 modules until there is no effective partition that produces a positive modularity coefficient. Constraint conditions are added to the existing toolbox to prevent inoperable nodes from being exported as TZs. First, identify the group membership variable values of the nodes classified by the eigenvector algorithm. Then, if an inoperable node and its adjacent nodes (i.e., adjacent nodes based on the weight matrix) do not have the same value, it sets their values to the value that the majority of them have. In other words, this constraint condition restricts the inoperable node and its neighborhood nodes to belong to the same module, so that the inoperable node does not act as a hub connecting these modules. At the same time, the resolution parameter α Is swept from 0.5 to 1.5 at intervals of 0.25 to obtain multiple modular structures. The resolution parameter determines the size of each module when partitioning the network nodes into modules, that is, the number of modules. A high parameter value leads to a modular structure composed of small modules (i.e., a large number of modules), and a low parameter value obtains a structure composed of large modules (i.e., a small number of modules).

[0054] To derive the target zones from modularity analysis, it is preferable to first set the EZ and the inoperable zones. The EZ is fixed according to the clinical evaluation of each patient, and the inoperable zones are arbitrarily set to all EZs, i.e., assuming the worst-case scenario where all EZs cannot be removed surgically. In fact, it is desired to obtain the following TZs: These TZs exclude all EZs for resection surgery and all fiber bundles connected to the EZs for disconnection surgery. The strategy for suppressing seizure propagation is to partition the brain network of each patient into multiple modules and then remove the connections from the module containing the EZ module to other modules, i.e., nodes or edges. However, in modularity analysis, when using low-resolution parameters, a relatively large number of nodes may belong to the same module as the EZ, and even if the TZ is eliminated, there may still be a considerable number of nodes mobilized by seizures. To control this problem, i.e., to prevent a large number of nodes from becoming seizure-active nodes, it is chosen to partition the EZ module into sub-modules again, and the TZ is defined as the following nodes / edges: The nodes / edges connect the sub-module containing the EZ sub-module to other sub-modules or modules. The nodes and edges obtained for resection and disconnection surgery are named target nodes and target edges, respectively. Since the resolution parameter is controlled in the modularity analysis of the two partitioning processes, multiple modular structures can be obtained for the same patient, and thus it can provide multiple intervention options for the target nodes and target edges. All the processes described above are automatically performed by the developed Matlab model. According to the positions of the EZ and the inoperable zones, the model can generate multiple TZ options.

[0055] Brain network simulation using The Virtual Brain

[0056] A patient-specific network model is constructed using the Virtual Brain to verify the effectiveness of the derived TZs. Specifically, a six-dimensional model named Epileptor is adopted to describe the network nodes, and the reconstructed structural connectivity is used to connect these nodes. Epileptor is a phenomenological neural population model that reproduces seizure characteristics and consists of 5 state variables and 6 parameters. Each Epileptor is coupled to other Epileptors via permittivity coupling of a slow time-scale variable z that replicates the extracellular effect. In the following equation, K ij represents the node i and the node j between the connection weights, and τ ij represents the time delay determined by the track length between these two nodes.

[0057]

[0058]

[0059] wherein

[0060]

[0061] Clinically, the degree of epileptogenicity can be plotted against the excitability parameter x 0, where the epileptogenic zone (EZ) that generates spontaneous seizure activity, the propagation zone (PZ) that is mobilized by seizure propagation from the EZ, and other zones that are not mobilized during propagation are distinguished. In this example, the excitability parameter x 0 is set to -1.6 for the EZ, and depending on the structural connectivity of each patient, the excitability parameter x 0 is set to a value between -2.150 and -2.095 corresponding to the PZ of all other nodes, in order to simulate the worst-case scenario in which seizure activity originating from the EZ propagates to most other brain nodes. For the other parameters in the equation, l 1 = 3.1, l 2 = 0.45, γ = 0.01, τ 0 = 6667 and τ 2 = 10 are used. In addition, zero-mean white Gaussian noise with a standard deviation of 0.0003 is linearly added to the variables x 2 and y 2 in each Epileptor for stochastic simulation. These noise environments excite each Epileptor and thereby generate interictal spikes as baseline activity.

[0062] Using a patient-specific network model, the seizure propagation characteristics before and after eliminating the target node or target edge are simulated. In particular, the seizure propagation inhibition ratio as shown in the following equation is quantified and used to compare the removal effects of each TZ. The x 1 + x 2 waveforms of each Epileptor are observed to reproduce the local field potential at each node.

[0063] SR, seizure propagation inhibition ratio =

[0064] N bef is the number of nodes mobilized by seizures before removing the TZ, N af is the number of nodes mobilized by seizures after removing the TZ.

[0065] To evaluate normal brain function, a stimulation paradigm was adapted in which, after a transient stimulus, the information transmission ability of the network was quantified by the spatio-temporal nature of its trajectory in the resting state. Eight specific well-known RS networks were tested, including the default mode, visual, auditory - phonological, somato - motor, memory, ventral stream, dorsal attention, and working memory. Simulating specific brain regions can reproduce a dynamic response network similar to the brain activation pattern in the RS network.

[0066] Table 1 below shows the stimulation sites that can reproduce the response pattern that best matches the brain activation pattern in each RS network. The numbers in parentheses indicate the node indices.

[0067] 。

[0068] It was chosen to apply an electrical pulse of 2.5 s to specific cortical regions and observe the response signals in all brain regions. The stimulation sites used to test each RS network are shown in Table 1. In this simulation, a previously published patient - specific network model was used, which utilized a neural mass model of a general two - dimensional oscillator in Equation (4) below instead of Epileptor, in order to replicate the damped oscillations due to this stimulation. For these parameters, τ = 1, a = - 0.5, b = - 15.0, c = 0.0, d = 0.02, e = 3.0, f = 1.0 and g = 0.0 were used. Each oscillator was coupled to other oscillators via differential coupling based on the individual structural brain connectivity. Here, each oscillator or brain node was operated at a stable focus near the instability point - the supercritical Andronov - Hopf bifurcation - but never reached the critical point. Each node did not show any activity without stimulation, but when stimulated (or receiving input from other nodes via the connectome), it generated damped oscillations by operating near the critical point. Since the working distance from the critical point is determined by the connectivity of each node (connection weights and time delays), each node generated different damped oscillations with different amplitudes and decay times, thus producing a specific energy dissipation pattern (response activation pattern) according to the stimulation location and brain connectivity.

[0069]

[0070] Then, compare the response spatio-temporal activation patterns before and after removing the target node or target edge. To this end, the subspace in which the trajectories evolve after the stimulus is quantified by adopting pattern-level cognitive subtraction (MLCS) analysis. A reference coordinate system is derived based on the principal component analysis (PCA) performed using the response signals in all brain nodes after bioinformatics surgery, i.e., the eigenvectors of the covariance matrix of the response signals are calculated. . Then, select three principal components (PCs), and project the response signals in these two cases (before and after removing TZ) under the PCs q b , q a ), and the reconstructed response signals are obtained at each brain node q r,b , q r,a :

[0071] (5).

[0072] To compare the reconstructed response patterns, for each brain node, calculate the amount of overlap between the powers of the reconstructed response signals before and after eliminating TZ. The values obtained in each brain node are normalized by this overlap value using only the signal power before removing TZ, and then this value is defined as the similarity coefficient (if the value > 1, it is defined as the deviation from 1; thus, the similarity coefficient has a value between 0 and 1). Here, if the average value of the similarity coefficients in all brain regions is lower than 0.75, the derived TZ is considered to have a high risk. In other words, it indicates that the elimination of TZ may affect the corresponding RS network. The TZ with a high risk is called the non-operable area. If the TZ contains more than one node, find the key node whose response activation pattern has been severely changed due to the stimulus, and then designate this node as the non-operable area. This key node is defined as the node that produces the lowest similarity coefficient when the same simulation is repeated after removing each node belonging to the TZ. Apply the updated non-operable area (added with this key node) to the modularity analysis again, which produces a new TZ. Evaluate the effectiveness and safety of the newly obtained TZ again through network simulation. Iterate these feedback processes until a TZ that meets the safety criteria is obtained.

[0073] Example 2: Target area derivation

[0074] In this example, several surgical intervention options outside the EZ are presented for a specific patient. The patient has two EZs, namely, ctx-rh-lingual (node 61) and ctx-rh-parahippocampal (node 64), which are designated as inoperable regions.

[0075] Using modularity analysis, considering Figure 2A the inoperable regions identified in, a patient-specific modular structure is constructed. The brain network nodes are partitioned into 7 modules with a modularity coefficient of 0.3912, and the green module including the EZ is further subdivided into 4 sub-modules. Based on this modular structure, 3 target nodes (black triangles) and 8 target edges (gray dashed lines) are identified, which connect the EZ sub-module to other sub-modules or modules. The anatomical location of the initial TZ is shown in Figure 2B .

[0076] In the network simulation for evaluating the effectiveness of the TZ, before removing the TZ, most brain nodes were mobilized after seizure activity was generated in the EZ. However, when the 3 target nodes were removed, seizure activity was almost isolated in the EZ, where the seizure spread inhibition ratio SR was 95.65%. When the 8 target edges were disconnected, the nodes mobilized by the seizure were significantly reduced, where SR was 91.30%, although seizure activity was still observed in several adjacent nodes of the EZ. These results demonstrate that eliminating the derived TZ can prevent seizure spread. At the same time, in the network simulation for evaluating the safety of the TZ, by stimulating specific brain regions to test several RS networks, the similarity coefficient between the response activation patterns before and after removing the TZ is calculated, as shown in Figure 3A . A low similarity coefficient indicates that the response pattern due to stimulation has been severely changed after removing the TZ. In this case, the result means that eliminating the obtained TZ may lead to greater network disorganization and then a higher risk of negative cognitive effects (especially on memory function). If the removal of the TZ distorts the response pattern by more than 25% of the original pattern, that is, if the average value of the similarity coefficient in all brain regions is lower than 0.75, the TZ is considered unsafe.

[0077] Since the obtained TZ may have a negative impact on the memory network, the next step is to identify the key nodes that cause the most significant changes. Figure 3BShows the impact on the memory network when each of the initially derived target nodes is removed. Compared with before removal, eliminating the left-cerebellum-cortex (node 35) produced the lowest average similarity value (this similarity value was 0.58 when the stimulus was applied to node 10), making this node be defined as a key node and thus being designated as an inoperable area.

[0078] By feeding back this updated inoperable area to the modularity analysis, a new modular structure was obtained. Figure 2C Shows the modular structure when this key node (gray triangle, node 35) and two EZs (nodes 61 and 64) are set as inoperable areas. The brain network nodes are divided into 8 modules with a modularity coefficient of 0.3995, and the green module containing the EZ is subdivided into 2 sub-modules, such that each inoperable area and its adjacent nodes belong to the same sub-module. Based on this modular structure, new target nodes (black triangles) and target edges (gray dashed lines) were obtained. Figure 2D Shows the anatomical location of the new TZ.

[0079] Figure 4 shows the network simulation results for the effectiveness evaluation of the newly obtained TZ. The results present the time series data in all brain nodes, i.e., the local field potential. Before removing the TZ, the seizure activity originating from the EZ propagated to other nodes after some delay, i.e., most nodes were mobilized by the seizure, as Figure 4A shown. In the case where the new TZ has been eliminated, compared with the simulation before removal, a significant reduction in the area mobilized by the seizure was identified, although they have some more areas mobilized by the seizure than when the initial TZ was removed. As Figure 4B and 4C shown, the SR after removing 3 new target nodes is 89.86%, and the SR after removing 5 new target edges is 85.51%. Figure 4D Shows the simulation results when the same number of random nodes (excluding the EZ) as the derived target nodes are removed. Comparing the degree of reduction of the nodes mobilized by the seizure, it proves that eliminating the TZ obtained from the proposed method can effectively inhibit seizure propagation. In this example, the SR after removing 3 random nodes is 31.88%. At the same time, the simulation results show that after removing the TZ, even though the seizure activity is inhibited in each brain node, persistent spikes will appear. These interictal spikes are caused by the noise environment applied for the random simulation. Gaussian noise was applied to all brain nodes (Epileptor) to account for the background internal activity, such that each node generates random spike events as baseline activity. The occurrence of these spikes is regulated according to the state of each segment (such as pre-seizure, seizure, and post-seizure).

[0080] Figure 5 shows the difference between the security evaluation results of the initial TZ and the new TZ. The histogram shows the average of the similarity coefficients of the response patterns due to stimulation in all brain regions before and after removing the TZ. Comparing the values between the two groups, it indicates that: eliminating the new TZ can keep all RS networks at a level similar to that before removal (average similarity coefficient > 0.75), while eliminating the initial TZ may disrupt the memory network. In other words, this means that the newly derived TZ has less impact on the transmission properties of the brain networks that maintain normal brain function. The results also show that: compared with excising the brain region corresponding to the target node, disconnecting the fiber bundle corresponding to the target edge has less impact on normal brain function. In this example, the new TZ obtained from a single feedback meets the security criteria. However, if the newly derived TZ does not meet this criterion, the iterative feedback process (finding the key nodes among the new TZs; setting them as inoperable areas; and obtaining a new modular structure) continues until a TZ that meets this criterion is derived.

[0081] To easily describe the process of deriving the TZ, only the results when the resolution parameter in modularity analysis is fixed at 1.25 are presented in this example in Figures 2 to 5. However, since the proposed method involves a parameter sweep of the resolution parameter from 0.5 to 1.5 at intervals of 0.25, multiple modular structures are obtained according to the parameter values (the resolution parameter determines the size of each module, i.e., the number of modules), thus generating multiple TZ options. For the specific patient in this example, initially 5 variants for the target node and 7 variants for the target edge are obtained. After applying the feedback, finally 7 variants for the target node and 9 variants for the target edge are derived.

[0082] Example 3: System Analysis According to the Location of the Epileptogenic Zone

[0083] To demonstrate the robustness of the proposed method, additional simulation results are presented, which show how the TZ changes according to the location of the EZ. Figure 6A and Figure 6BThe target nodes and target edges in another specific patient are shown, which are obtained by performing a system simulation in which an EZ is placed among all possible brain nodes, and the EZ is assumed to be an inoperable area. The cumulative results of the TZ identify the nodes and edges that are frequently used as the TZ. The frequently obtained nodes and edges play an important role in the spread of seizure activity from a localized area to the whole brain and can be effectively controlled by being removed. In this patient, the most frequently derived node is ctx-lh-postcentral (node 70), and the most frequently derived edge is the connection between ctx-lh-supramarginal (node 30) and ctx-lh-postcentral (node 21). Figure 6C The anatomical locations of the cumulative results are presented. At the same time, when deriving the TZ, the frequency of the initially obtained target nodes is positively correlated with the node strength (the sum of the weights of the chains connecting to other nodes), that is, the nodes with high strength are frequently derived as the TZ (correlation coefficient: 0.7842). However, the final target nodes obtained from the feedback process tend to be more concentrated at a few nodes, and thus the frequency of the finally obtained target nodes is not significantly correlated with the node strength, with a correlation coefficient of 0.3059.

[0084] The key nodes used for the feedback strategy to consider the safety of normal brain function do not differ significantly among all patients. In particular, the superior frontal cortex (nodes 27 and 76) often appears as key nodes, which means that these nodes are effective in controlling the spread of seizure activity, but removing them may cause problems for normal brain function. The network simulation results identify that the elimination of these nodes severely distorts the RS network corresponding to vision, working memory, the ventral stream, and the default mode. In fact, the superior frontal cortex has been studied as a node that is frequently used as the shortest path connecting two different brain nodes and has been shown to play an important role in the interhemispheric spread of seizures.

Claims

1. A method for identifying potentially surgically manipulable target areas in the brain of an epilepsy patient, comprising the following steps: Providing a computerized platform that models various regions of the primate brain and the connectivity between said regions; Providing a model of the epileptogenic zone and a model of the propagation of epileptic discharges from the epileptogenic zone to the propagation zone, the model of the epileptogenic zone being a mathematical model that describes the occurrence, time course, and offset of epileptic discharges, and loading said model into the computerized platform to obtain a computerized platform for modeling the epileptic primate brain; Identifying the estimated epileptogenic zone in the patient's brain; Personalizing the computerized platform for modeling the epileptic primate brain according to the patient's brain structural connectivity, and parameterizing the estimated epileptogenic zone as the epileptogenic zone in the computerized platform to obtain a personalized computerized platform for the patient; Performing modularity analysis using the patient's brain structural connectivity to derive potential target areas that act as hubs in the interactions between modules, the potential target areas being outside the potential epileptogenic zone and, if they are surgically manipulated or removed, would minimize the propagation of epileptic seizures, and evaluating the effectiveness of the potential target areas in minimizing the propagation of epileptic seizures by simulating the propagation characteristics of epileptic seizures in the personalized patient computerized platform and identifying a network simulation of one or more effective target areas, the effective target areas being outside the epileptogenic zone and, if they are surgically manipulated, would minimize the propagation of epileptic seizures; Evaluating the safety of the potential target areas in maintaining normal brain function by network simulation, wherein simulation spatio-temporal brain activation patterns under defined state conditions are obtained from the personalized computerized platform before and after removing the areas, and comparing these simulation spatio-temporal brain activation patterns obtained before removing the areas with the simulation spatio-temporal brain activation patterns obtained after removing the areas, and if the spatio-temporal brain activation patterns obtained before removing the areas are substantially the same as the spatio-temporal brain activation patterns obtained after removing the areas, then identifying the potential target area as a safe target area; and Identifying potential target areas that meet both the effectiveness and safety evaluation criteria as potentially surgically manipulable target areas.

2. The method according to claim 1, wherein the estimated epileptogenic zone in the patient's brain is clinically estimated.

3. The method according to claim 1 or 2, wherein the target area is a node or an edge involved in the propagation of epileptic seizures, the node and the edge corresponding to a brain region and a fiber bundle between brain regions, respectively.

4. The method according to claim 1 or 2, wherein the structural connectivity is reconstructed from image data of the patient's brain obtained using magnetic resonance imaging.

5. The method according to claim 4, wherein the magnetic resonance imaging is diffusion-weighted magnetic resonance imaging.

6. The method according to claim 1 or 2, wherein the computerized platform for modeling the epileptic primate brain is personalized according to the patient-specific brain connectivity and the patient's functional data.

7. The method according to claim 6, wherein the functional data is obtained by electroencephalogram (EEG) technology.

8. The method according to claim 7, wherein the EEG technology is stereotactic EEG (SEEG) technology.

9. The method according to claim 1 or 2, wherein, for the implementation of modularity analysis, constraint conditions are added to prevent inoperable nodes from being exported as target areas.

10. The method according to claim 1 or 2, wherein system simulation is performed in a personalized computerized platform of a patient, and wherein if the target area does not meet the evaluation criteria, a new target area is derived by feeding back the simulation result to the analysis again.

11. The method according to claim 1 or 2, wherein the defined state condition is a resting state condition, or a state condition in which the brain is in a memory mode, or a state in which the brain is in a mode of perceiving things, or a state in which the brain is in an attention mode.

12. The method according to claim 11, wherein multiple defined state conditions are used for simulation.

13. The method according to claim 1 or 2, wherein modularity analysis provides a non-overlapping modular structure that minimizes the edges between modules and maximizes the edges within modules.

Images