Patent Application
20180240549
This invention relates generally to a computer-implemented apparatus and method for predicting performance and/or (potential) effects of a surgical strategy and, more particularly but not necessarily exclusively, to such a computer-implemented apparatus and method for assessing performance and/or (potential) effects of a surgical strategy in respect of epilepsy patients.
Epilepsy is the most common serious neurological disease, affecting 1.2% of the population in the UK. Epilepsy is often mistakenly viewed as easily treated, whereas, in fact, 40% of new-onset epilepsy remains uncontrolled after one year of antiepileptic drug (AED) treatment, and over 30% of people with epilepsy never respond to treatment, with consequent morbidity and mortality. Epilepsy is directly responsible for over 100 deaths per year in the UK, and is the fifth most common cause of avoidable years of life lost in men and eighth in women. It is the leading cause of repeated unplanned admissions to NHS hospitals, and is estimated to cost the EU 15.5 billion euros per annum.
Anti-epileptic drugs (AEDs) are usually the first choice of treatment for epilepsy. However, only about 70% of people with the condition are able to control their seizures with AEDs. For some people, brain surgery may be an option. However, this is only the case if removing the area of the brain where epileptic activity starts would not cause damage or disability. If successful, there is a chance epilepsy in the patient will be cured. On the other hand, although surgery is a valuable option for pharmacologically intractable cases of epilepsy, success rates are far from optimal, For example, sustained positive outcome 12 months post-surgery is currently only achieved in around half of patients, and can be as low as 15% in extra-temporal cases. The reasons why surgery can succeed or fail often remain unclear due to a fundamental lack of understanding of the seizure generating (ictogenic) capabilities of the brain, and the inability to assess or predict the consequences of surgical resections. A challenge in identifying the cause of surgery failure lies in the potential involvement of both immediate and longer-term mechanisms. For example, cases with recurrence of seizures in the short-term are most likely to remain intractable, and are presumed to be the result of an inadequate surgical resection, such that the seizure generating (ictogenic) mechanisms that existed before surgery still remain. Long-term failure, however, can occur for cases in which the original ictogenic mechanisms were abolished, but new mechanisms emerged post-surgery. Furthermore, some patients may be seizure-free postoperatively, but seizures can emerge when drugs are withdrawn. It is therefore an object of aspects of the present invention to address the problem in epilepsy management of the optimisation of surgery with a view to achieving maximal benefit to the patient.
A prevalent theoretical concept underpinning the surgical treatment of epilepsy is that of the epileptogenic zone, i.e. “an area of the cortex that is indispensable for the generation of epileptic seizures” (Rosenow, F. and H. Lüders 2001. Presurgical evaluation of epilepsy. Brain, Pp. 1683-1700.). The goal of epilepsy surgery in this paradigm is the removal or disconnection of the epileptogenic zone, since in a “localised” view of epilepsy, such perturbations would render a patient seizure free. Unfortunately, this is a retrospective definition; the epileptogenic zone can only be identified in patients for whom surgery was successful and is not, therefore, a theory that aids the prediction of an optimal surgical strategy. Rather, a combination of observations is acquired during pre-surgical planning regarding aspects of the epileptic brain that are presumed to correlate to the epileptogenic zone. These include, for example, regions of the brain that are first to produce pathologic electrographic activity when seizures occur and regions that display evidence of abnormal structure that perhaps contribute to ictogenicity. Concordance of these features is presumed to indicate a clear epileptogenic zone and in these cases surgery is most likely to succeed. However, non-concordance of these features often occurs and, even in cases for which well-defined focal epileptogenic zones caused by specific lesions are assumed (for example, mesial temporal lobe epilepsy), surgical resection can be unsuccessful.
Thus, it would be highly desirable to provide an apparatus and method that addresses at least some of these issues and provides an improved facility for assessing and/or predicting the effects of a proposed surgical intervention in respect of epilepsy patients.
In accordance with an aspect of the present invention, there is provided a computer-implemented apparatus for predicting an effect of a proposed surgical strategy for treatment of epilepsy and/or epileptic seizures, the apparatus comprising:
In prior art systems, there is required an a priori assumption or hypothesis of the nature of the brain network and a link between abnormal regions of the brain and the generation of seizures. In contrast, the present invention does not require this a priori assumption or hypothesis of nodes that form a seizure focus to define a seizure onset zone. Instead, the present invention provides a system and method for systematic exploration and quantification of the effect of removing any node on seizure generation, without any a priori assumptions or hypotheses being made as to the pathology of the brain. In an exemplary embodiment of the present invention, the apparatus further comprises a device configured to receive patient brain data, wherein said brain network model is generated from said patient brain data. In an exemplary embodiment, the apparatus comprises a device for selecting, from said multiple proposed surgical strategies, a proposed surgical strategy based on the ΔBNI value calculated in respect thereof. For example, the proposed surgical strategy having the highest ΔBNI value calculated in respect thereof may be selected and, optionally, data representative thereof, for example in the form of which node(s)/edge(s) should be removed, may be output.
Optionally, the synthetic brain activity data comprises, or is configured to simulate, recurrent seizures within said brain network model. In an exemplary embodiment, the apparatus may be configured to place the brain network model close to a transition between brain states. Optionally, one or more parameters of said brain network model may be configured model to place the model close to a saddle-node on invariant circle (SNIC) bifurcation, but the present invention is not necessarily intended to be limited in this regard and other methods of placing the brain network model close to transition between brain states will be apparent to a person skilled in the art.
In an exemplary embodiment, said BNI value is calculated by:
The ΔBNI value, in respect of a brain network having node i removed, may be calculated using the equation:
wherein BNIipre is the BNI value of the brain network before removal of node i and BNIipost is the BNI value of the brain network after removal of the node i.
Apparatus according to an exemplary embodiment of the invention comprises a device configured to identify, from said synthetic brain activity data, nodes having contiguous seizure epochs, calculate the BNI value for each of said identified nodes, and select the highest BNI value thus calculated as a representative BNI for said brain network.
In accordance with another aspect of the present invention, there is provided a computer-implemented method for predicting an effect of a proposed surgical strategy for treatment of epilepsy and/or epileptic seizures, the method comprising:
These and other aspects of the present invention will be apparent from the following specific description, in which embodiments of the present invention are described, by way of examples only, and with reference to the accompanying drawings, in which:
Although surgery is a valuable option for pharmacologically intractable cases of epilepsy, success rates are far from optimal, and the reasons why surgery can succeed or fail have remained unclear, which stems from a fundamental lack of understanding if the seizure generating (ictogenic) capabilities of the brain, and the consequences of surgical resections. The inventors have determined, through extensive innovative effort and novel methods and systems, that seizures are an emergent property of a network within which the existence of pathological zones is neither a necessary nor sufficient condition for seizures to emerge. The novel method and apparatus described hereinafter can be used to demonstrate that typically presumed correlates of the region of tissue to resect, such as the location of pathological nodes or seizure onset zones, do not necessarily predict the best resection strategy. Thus, exemplary embodiments of the present invention provide a framework for assessing, predicting and/or optimising the outcome of surgical strategies in epilepsy patients.
In this exemplary system, the interface 112 of the computing device is connected to a brain data recording machine 114, which may be an EEG machine, MEG machine, MRI scanner, or the like. In the example, the application 110 is configured to receive data 108 from the machine 114 (via any wired or wireless data transfer medium) and process it as described below. In other embodiments, the brain data 108 may be received by the computing device in another manner, e.g. over a communications network or from transportable storage medium, such as a DVD. It will be further understood that in alternative embodiments, at least some of the steps of the application 110 may be performed sequentially, or in parallel on one or more remote processing devices.
At step 202, the application 110 receives patient brain data from a source as described above. At step 204, the application generates a network model based on the patient brain data. Network analysis has been proposed, generally, in the study of brain function and, more particularly, in the study of epilepsy, and offers a framework to characterise the organisation of brain networks. Network analysis reduces complex systems such as the brain to a collection of “nodes” and “edges”. Nodes represent functional or structural elements of the network, in this case the brain regions of the patient brain data, and edges are any type of relation between brain regions, representing either a structural or functional connection between brain regions, as recorded in the source patient brain data. Together, these two building blocks enable characterisation of the organisation of brain networks.
The structure of the above-mentioned network model may be inferred from the patient brain data using any one of a number of methods, as set out, for example, in Eric van Diessen, Sander J. H. Diederen, Kees P. J. Braun, Floor E. Jansen and Cornelis J. Stam, Functional and structural brain networks in epilepsy: What have we learned?. Epilepsia, 54(11): 1855-1865, 2013 (the contents of which are hereby incorporated by reference), Benjamin, O., Fitzgerald, T. H. B., Ashwin, P. et al, A phenomenological model of seizure initiation suggests network structure may explain seizure frequency in idiopathic generalised epilepsy. J Math Neurosci 2012, Vol. 2, No. 1, pp. 1-41 (the contents of which are hereby incorporated by reference) or alternatively using another measure of nonlinear correlation. Alternatively, a method for inferring brain networks from data, that involves learning parameters of a specific model from such data, may be used, as set out in, for example, Freestone, D. R., Aram, P., Dewar, M., Scerri, K., Grayden, D. B., & Kadirkamanathan, V. (2011). A data-driven framework for neural field modeling. NeuroImage, 56(3), 1043-1058; and Freestone, D. R., Karoly, P. J., Nešić, D., Aram, P., Cook, M. J., & Grayden, D. B. (2014). Estimation of effective connectivity via data-driven neural modeling. Frontiers in Neuroscience, 8 (November), 383.
However, in general and as stated above, the model comprises a set of nodes that correspond to brain regions of the patient brain data and connections between the nodes of the network data structure correspond to connections between the brain regions, as recorded in the source brain data. In the case of EEG data, for instance, the brain regions will correspond to regions of the brain close to the electrodes of the EEG machine.
At step 206, the apparatus generates synthetic brain activity data based on at least some of the nodes of the network model. This can be a dynamic computational model based on features of human seizure data, e.g. one that phenomenologically or physiologically models transitions from interictal to ictal states in brain regions over time. In alternative embodiments, the synthetic brain activity data can be generated using a probabilistic model, for example, representation by a Markov process. Mathematically, seizures in the above-described model can arise due to several different mechanisms, and apparatus according to an exemplary embodiment of the present invention may be configured to place the brain network model close to a transition between brain states. In an exemplary, non-limiting embodiment, parameterisation of the model may, for example, be designed to place the model close to a saddle-node on invariant circle (SNIC) bifurcation, as discussed in Blenkinsop, A. Valentin, A., Richardson, M. P., and Terry, J. R. The dynamic evolution of focal-onset epilepsies-combining theoretical and clinical observations. The European Journal of neuroscience, 25(3):756-770. In this regime, a model node can generate spiking due to noise and this can propagate through networks. This model therefore has the property that it can generate recurrent seizure events, and therefore is a model for both seizure generation as well as epilepsy, and it can be shown that the proximity of the model to the above-mentioned bifurcation determines its degree of “excitability” in the sense of how likely it is to generate spikes. However, other suitable methods for placing the network model close to a transition between brain states will be apparent to a person skilled in the art.
Next, at step 208, the apparatus computes the probability of a brain network to generate seizures, hereinafter referred to as Brain Network Ictogenicity (BNI) and introduced in Chowdbury et al, 2014. A practical approach to quantifying this is by summing the total time spent in seizure compared to a reference time period (seizure days per year, for example):
Thus, using the above-described model, seizure epochs are defined by the presence of spiking activity in the model, which is identifiable by the amplitude of the model output in comparison to periods of time where non-spiking dynamics occur, as illustrated in
It will be appreciated that, whilst understanding seizure generation as an emergent property of a network naturally introduces contributions from both nodes (brain regions) and edges (connections between them), from the perspective of surgery, it would not necessarily be intended to just alter edges. Rather, and possibly more typically, the aim would be to resect nodes. Thus, the present invention seeks to assess and/or predict the (potential) effect of surgery based on the contribution of each node to the ictogenicity of the network. More specifically, aspects of the present invention seek to quantify the effect of surgery by measuring the magnitude of the change in BNI that would be observed following the removal of a specific node and/or edge or, more likely, sets of nodes and/or edges. It is this change in BNI, termed herein ΔBNI, that quantifies the contribution to the overall ictogenicity made by a node. Specifically, referring back to
Given the choice of intrinsic parameters (chosen to place the model close to the above-mentioned SNIC bifurcation) and connectivity structure, free parameters that remain are the global scaling of connectivity strength (α) and the variance of white noise added to each node (σ). We fix σ=650, which is considered sufficient to generate recurrent seizures. Having fixed the noise amplitude,
In practice, ΔBNI is equivalent to a quantification of whether resection of a specific brain region (i.e. node(s)/edge(s) i) will sufficiently control seizures and, by estimating this quantity using patient brain data, optimal resections can be planned. Thus, at step 212, the apparatus outputs quantitative values corresponding to the predicted effect (i.e. change) on brain network ictogenicity of resection of one or more respective brain regions, thereby enabling a quantitative evaluation to be made of a number of different surgical strategies and optimal strategy selected. Networks containing nodes with ΔBNI values close to 1 would indicate good prognoses, whereas networks with a lower ΔBNI would be less likely to be cured by a localised resection, and more elaborate surgical strategies might, potentially, be considered. In most cases, the surgical strategy having the highest ΔBNI should be selected. In an exemplary embodiment, at step 214, the apparatus selects a surgical strategy having a highest ΔBNI and outputs data representative thereof, possibly in the form of data indicative of the node(s)/edge(s) that should be removed in accordance with the selected surgical strategy.
Thus, having established a model derived from patient brain data and placed in a state of recurrent seizures, the inventors have developed an apparatus and method for assessing the relative contribution of each node to the overall ictogenicity of a network. The classical approach to surgery assumes that an epileptogenic zone exists, and thus removal of this zone would render a patient seizure free. However, the inventors have determined, through extensive innovative effort and novel methods and systems, that recurrent seizures can arise in networks that do not possess an intrinsically pathological node (i.e. the underlying properties of each region is the same). In this case, the distribution of ΔBNI across the nodes of a network with identical nodes can be heterogeneous. This means that the removal of some nodes can cause a greater reduction in BNI than others, despite all of the nodes having the same intrinsic properties.
The inventors have also determined that the existence of pathological nodes within the network can alter the BNI profile, depending on where in the network the pathology resides. The framework proposed by exemplary embodiments of the present invention can be used to show this using the concept of “seizure onset zones”. For each seizure of a simulation, the “seizure onset zone” is defined as the first node to display a spike at the start of the seizure, as indicated in
Thus, the present invention moves beyond the traditional qualitative approach and instead quantifies the contribution of each node to ictogenicity by calculating its ΔBNI, i.e. the relative change in ictogenicity brought about by removing one or more nodes/edges from the network, thus enabling one or more potential surgical strategies to be assessed, and the outcome thereof to be predicted, so as to optimise the probability of success of any surgical resection strategy undertaken.
It will be appreciated by a person skilled in the art, from the foregoing description, that modifications and variations can be made to the described embodiments without departing from the scope of the invention as defined by the appended claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 1514740.8 | Aug 2015 | GB | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/GB2016/052556 | 8/18/2016 | WO | 00 |
