Patent Application
20230405326
The invention is related to a computer program for training a neurological condition detection algorithm e.g. to be used for seizure detection in an implantable neurostimulation device having a target electrode arrangement. This computer program is also referred to as the training computer program. The invention is further related to a method of programming an implantable neurostimulation device using such training computer program and to a computer program in the form of a neurological condition detection algorithm or classifier for detecting neurological conditions from EEG data which has been trained and/or is being trained by such training computer program. This computer program is also referred to as the neurological condition detection computer program. The invention is further related to a neurostimulation device running such a neurological condition detection computer program. The invention is further related to a method for treatment of a neurological condition using a neurostimulation device running such a neurological condition detection computer program.
Despite progress in the development of medication, a significant share of epilepsy patients are resistant to treatment with antiepileptic drugs. As only a minority of these patients are surgical candidates, there is a need for innovative therapeutic approaches. An alternative treatment concept for these patients is the application of electrical stimulation in the early phases of a seizure to interrupt its spread across the brain. This can be accomplished using a closed-loop system, where the electrical brain activity is recorded by a set of electrodes and continuously or intervallic monitored with a seizure detector which triggers electrical stimulation to the seizure onset zone (SOZ) via the same or different electrodes.
Therefore, there is a need for reliable and energy-efficient seizure detectors. It is an object of the invention to provide solutions for this need.
An embodiment of the invention is a computer program for training a neurological condition detection algorithm to be used e.g. for seizure detection in an implantable neurostimulation device having a target electrode arrangement, the computer program comprising the following steps:
The neurological condition which is identified by the neurological activity maybe a seizure, a stroke, neuropathic pain, dementia, Parkinson, tinnitus, aphasia, or any other specific neurological event. In the following, the example of a seizure is mainly used for describing the invention. However, this shall also cover other types of neurological events.
The invention allows for an optimum adaption of the neurological condition detection algorithm to the “real” electrode configuration, namely the pattern of the target electrode arrangement. In this way, the neurological condition detection algorithm can be optimally adapted to the patient's individual neurological condition onset pattern and the individual EEG recording sites of the implanted electrodes where optimum detection of neurological events can be expected. The target electrode arrangement may comprise solely recording electrodes which are used in the neurostimulation device for recording EEG signals from the patient. It is also possible that the target electrode arrangement may comprise solely stimulation electrodes which are used in the neurostimulation device for outputting stimulation signals to the patient. Another possibility is that the target electrode arrangement may comprise a combination of recording electrodes and stimulation electrodes. In some cases, one, more or all of the electrodes of the target electrode arrangement may be used for both purposes, recording EEG signals and outputting stimulation signals, such that they are combined recording and stimulation electrodes.
Typically, the steps of the computer program are executed by a computer. However, the invention is also related to a method comprising the above mentioned steps and/or the following steps. The target electrode arrangement must be defined before the training computer program is run. For example, characteristic data of the target electrode arrangement can be input or programmed in the training computer program. The characteristic data of the target electrode may comprise geometrical data of the location, size and/or configuration of the electrodes of the target electrode arrangement. Generally, the EEG data used for the training comprise a number of measurements over time per recording electrode channel and per patient. The training computer program may need some further manual input, depending on the details which are included in the EEG data set input in the computer program. For example, the EEG data can already be tagged or labelled with information identifying the time and/or location of a neurological condition like seizure activity. Also, information identifying the time and/or location of the neurological condition maybe input manually. In the training step the neurological condition detection algorithm performs detection of neurological activity in the EEG data, and in the course of the training the results of the detection by the neurological condition detection algorithm are compared with the tagging information. Based upon the results of the comparison parameters of the neurological condition detection algorithm are optimized recursively, until a sufficient detection level of the neurological condition detection algorithm is achieved.
According to an embodiment of the invention, such electrode channels are selected out of the available electrode channels which are in closest proximity to the location of the identified neurological activity which corresponds to the neurological condition. This allows for an optimum spatial correlation of the expected neurological activity of the patient and the location of the electrodes of the target electrode arrangement.
According to an embodiment of the invention, such electrode channels are selected out of the available electrode channels which have the closest geometrical match with the electrodes of the target electrode arrangement. In this way, neurological condition detection using the electrodes of the target electrode arrangement can be optimized to a level which performs comparable to neurological condition detection using a relatively complex 10-20 or 10-10 EEG electrode system, without needing such high number of electrodes in the target electrode arrangement.
According to an embodiment of the invention, the electrodes of the target electrode arrangement are arranged in a pseudo-Laplacian pattern. This allows for highly localized EEG recording. Through the pseudo-Laplacian configuration, external signal noise and/or artifacts from muscle activity or movement will be reduced significantly, improving neurological condition detection.
According to an embodiment of the invention, such electrode channels are selected out of the available electrode channels which are arranged in a pseudo-Laplacian pattern. This allows for a close match of the selected electrode channels with the pattern of the target electrode arrangement in pseudo-Laplacian pattern. In a pseudo-Laplacian pattern, the selected electrode channels comprise a center electrode and at least two circumferential electrodes which surround the center electrode.
According to an embodiment of the invention, step d) comprises the steps:
In this way, further virtual electrode channels can be generated by such calculation steps. This allows for a refinement of the spatial signal resolution of the electrode pattern, without needing further hardware electrode channels.
According to an embodiment of the invention, exactly 5 electrode channels are selected out of the available electrode channels. This allows for an optimum match of the selected electrode channels with the target electrode arrangement having also five electrodes.
According to an embodiment of the invention, the computer program comprises at least two training cycles of the neurological condition detection algorithm:
In this way the training of neurological condition detection algorithm can be further optimized and accelerated.
According to an embodiment of the invention, the computer program is arranged for evaluating data tags which are assigned to the EEG data which are input in the computer which executes the computer program, wherein the data tags are used for selecting the subset of electrode channels out of the available electrode channels. The data tags may comprise information about the location of an electrode channel on the patients head during clinical measurements of the EEG signals.
Another embodiment of the invention is a method of programming an implantable neurostimulation device, comprising the following steps:
In this way, the same advantages as mentioned before can be achieved. In addition, for training purposes a different computer can be used than the computer of the implantable neurostimulation device. This allows for using a more powerful computer for running the training computer program, which saves significant time.
According to an embodiment of the invention, the implantable neurostimulation device is a closed-loop neurostimulator which is arranged for recording EEG signals, for calculating stimulation signals based upon the recorded EEG signals and for outputting the stimulation signals. In this way, the neurostimulation therapy can be optimized and specifically adapted to the needs of the individual patient. The stimulation signals are sent out via the electrodes of the target electrode arrangement. For recording the EEG signals the same electrodes of the target electrode arrangement can be used. It is also possible that different electrodes are used for recording the EEG signals and outputting the stimulation signals, for example electrodes which are specifically optimized for EEG signal recording and electrodes which are specifically optimized for signal outputting.
According to an embodiment of the invention, the electrodes of the target electrode arrangement are arranged in a pseudo-Laplacian configuration. This allows for optimum stimulation success when the electrodes are used for neurostimulation. Through the pseudo-Laplacian pattern deep stimulation can be provided to the patient even with an electrode pattern which is flat. For example, the electrodes can comprise a center electrode which is surrounded by at least two stimulation electrodes. It is also possible that there is a higher number of stimulation electrodes surrounding the center electrode, for example four stimulation electrodes.
According to an embodiment of the invention, in step g) the computer program is run on an external computer which is not part of the implantable neurostimulation device.
Another embodiment of the invention is a computer program in the form of a neurological condition detection algorithm or classifier for detecting neurological conditions from EEG data which has been trained and/or is being trained by a training computer program of the aforementioned kind. The computer program can be configured for implementation on a microcontroller or on a digital signal processor (DSP) or on a field programmable gate array (FPGA) or on an application specific integrated circuit (ASIC). According to an embodiment of the invention, the computer program is optimized for lowest power consumption. According to another embodiment of the invention, the computer program is optimized for highest performance. According to another embodiment of the invention, the computer program is optimized for lowest power consumption under defined performance constraints. According to another embodiment of the invention, the computer program is optimized for highest performance under defined power constraints.
According to an embodiment of the invention, the neurological condition detection algorithm or classifier for detecting neurological conditions is an artificial intelligence algorithm, e.g. Random Forest, Support Vector Machine, Multi-layer Perceptron, Convolutional Neural Network, Recurrent Neural Network, e.g. a Long Short-Term Memory Network, or Attention based Networks. This allows for an easy and optimum adaption of the neurological condition detection algorithm to typical neurological condition onset pattern occurring on real patients.
The computer can be located in the cloud (server type computer) or be any commercially available computer, like a PC, Laptop, Notebook, Tablet or Smartphone, or a microprocessor, microcontroller, DSP or FPGA, or a combination of such elements. The computer program can be stored on a non-transitory computer-readable medium.
As far as a closed-loop control is mentioned, a closed-loop control differs from a open-loop control in that a closed-loop control has a feedback or feedback of measured or internal values, with which the generated output values of the closed-loop control are influenced in turn in the sense of a closed-loop control circuit. In a closed-loop control system, only a variable is controlled without such feedback or feedback.
Further exemplary embodiments of the invention are described using the following figures, which show in:
The control processor 6 can be a microcontroller unit (MCU) or any other unit, which can perform control steps via processing of computer programs, e.g. in the form of hardware, firmware or software programs.
The signal generation circuit 3 is able to create and deliver stimulation pulses to the stimulation electrodes 21, 22, 23, 24 upon command from the control processor 6. The signal generation circuit 3 may comprise amplifier components.
The control processor 6 can detect neuro-signals and/or brain activities through the sensors 7, 8. The detected neuro-signals and/or brain activities can be processed and used for event driven delivery of stimulation pulses to any of the stimulation electrodes 21, 22, 23, 24.
The battery pack 9 supplies the aforementioned elements of the power unit 1 with electrical energy. The battery pack 9 may comprise rechargeable batteries.
The control processor 6 is arranged for executing the trained neurological condition detection algorithm. The training of the neurological condition detection algorithm is done on the computer 13, as it will be described in the following.
The neurological condition detection algorithm for such a system must be trained on recordings configured similar to the subgaleal electrodes. Since there is a high similarity between EEG data from subcutaneous and proximate scalp electrodes in patients with neocortical epilepsy, we use surface EEG recordings obtained from electrode configurations representing the implanted subgaleal electrode for training and evaluation of the neurological condition detection algorithm. This approach could be used to pre-train an individualized detector using scalp-based neurological condition recordings prior to a device implantation.
Electrode selection can be performed to represent the target electrode arrangement 2 during long-term video-EEG monitoring of neurological conditions. Expert epileptologists may define the seizure onset zone (SOZ) by visual exploration of the seizure onset electrodes and considering inter-electrode distances. Our proposed methods automate the electrode selection by considering the geometrical dimensions of the system shown in
In our system, after development and implementation of two automatic electrode selection methods, we designed and implemented four energy-efficient neurological condition detection algorithms using Random Forest (RF) classifier, Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Convolutional Neural Network (CNN) that can perform reliably with a reduced electrode set for detection of focal-onset seizures. Finally, we compared their detection performance to evaluate their suitability for implantable devices.
The EEG dataset used for training included surface EEG recordings of 50 patients with a total number of 358 seizures. Patients were selected, having the SOZ covered by electrodes positioned according to the 10-10 system in addition to the traditional 10-20 electrode layout. EEG data were recorded at a 250 Hz sampling rate on a 256-channels DC amplifier with a resolution of 24 bit. Electrodes were re-referenced to the centrally positioned electrode. For antialiasing, a low-pass filter with a cut-off frequency of 100 Hz was applied. EEG data from ten patients were used for training the hybrid model of the CNN. Therefore, to keep the test dataset consistent over all the classifiers, the remaining 40 patients with a total number of 286 seizures were used for evaluating the classifiers.
To improve the data quality and removing data contaminated with artifacts, first, data were filtered by a Chebyshev II band-pass (order:10, band stop=40 dB) with low and high cut-off frequencies of 0.1 Hz and 48 Hz. Next, very high amplitudes representing artifacts were removed from the analysis.
Electrode selection was performed to obtain an electrode set with the maximum number of electrodes covering the seizure onset area at inter-electrode distances below a threshold mapping to the design of the implantable system shown in
In this method, first, the mean of the five selected electrode coordinates was calculated and the nearest scalp electrode to this position was found. This electrode was considered as the center electrode. Any of the selected five electrodes that had a distance from the center electrode greater than the threshold distance was excluded from the list. As depicted in
In this method, in each step, one electrode was selected as the central electrode and similar to method 1, the number of electrodes from the initial SOZ electrodes whose distance from the central electrode was smaller than the threshold distance was counted. This process was repeated for all EEG electrode positions over the scalp, and a list of the selected initial SOZ electrodes enclosed with each electrode was generated. The electrode that contained the maximum number of initial SOZ electrodes was chosen as the optimal central electrode. If required, electrodes with a minimum distance from the central electrode were added to the list to yield exactly five electrodes for seizure detection.
Several features in the time and frequency domain were selected for seizure detection. Time-domain features were mean, maximum, mean absolute deviation, variance, skewness, kurtosis, line length, autocorrelation, and entropy. Frequency domain features included mean, maximum, and variance of the power spectrum, power in the theta (4-8 Hz), beta (13-30 Hz), and gamma band (30-45 Hz), spectral entropy, and epileptogenicity index. These features were used for classification by Random Forest, SVM, and MLP classifiers. For SVM and MLP, as the range of calculated features affect their weight as well as the subsequent decision boundaries, features were scaled for classification.
The number of binary decision trees was set to 100. We selected entropy for impurity measurement as the branching index. Four features were randomly selected at each node. To keep the size of the trees limited, the maximum depth of the trees was limited to 10. Bootstrap samples were used during building decision trees. The sample weights for each class were adjusted inversely proportional to class frequencies in the input data. The “leave one out” method was used for crossvalidation.
We selected the Gaussian Radial Basis Function (RBF) as the kernel function for handling non-linearity between the features and class labels. Two hyperparameters needed to be set for this purpose: the kernel coefficient of the Gaussian function was set to 0.01, and the penalty parameter of the error term, which behaves as a regularization parameter for the SVM, was set to 0.1. The sample weights for each class were adjusted so that they were inversely proportional to class frequencies in the input data. The “leave-one-out” method was used for cross-validation.
The MLP network consists of at least three layers of nodes: an input layer, one or more hidden layers, and an output layer. We implemented an MLP classifier consisting of one hidden layer with 20 neurons. We selected “Adam” as the solver for weight optimization. The logistic sigmoid function was selected as the activation function and an adaptive learning rate was selected to schedule for weight updates. The L2 penalty (regularization term) parameter was set to 10−4. The “leave one out” method was used for cross-validation.
A CNN consists of an input layer, multiple hidden layers, and an output layer. The hidden layers consist of convolutional layers, pooling layers, and fully connected layers. The architecture of our proposed CNN is shown in Table I. In the first layer, to be able to learn spatio-temporal patterns efficiently, we used a kernel size that extends over all the channels and detection time window (2 seconds=500 data points). In all hidden layers, we used batch normalization after the convolutions and Rectified Linear Units (ReLu) were used as the activation function. Dropout regularization was applied during the training to reduce overfitting. In the two last layers, we used two fully connected layers.
The following table shows a preferred architecture of our proposed CNN:
For training, because the available data for each patient was limited, we used the transfer learning method that consists of freezing the bottom layers in a model and only training the top layers. Therefore, we pre-trained the network over data from 10 patients, and subsequently, for each of the remaining 40 patients, the last convolution layer and the two fully connected layers were fine-tuned using the patient-specific data. Because the classes are imbalanced, the class indices were weighted to balance the weighting of the loss function during the training. Each model was trained with a batch size of 512 for 500 epochs. For weight optimization, we used Adam solver with a learning rate of 10−3. We used binary cross-entropy as the loss function. For evaluation, we used 3-fold cross-validation.
| Number | Date | Country | Kind |
|---|---|---|---|
| 20209813.3 | Nov 2020 | EP | regional |
| 21183521.0 | Jul 2021 | EP | regional |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/EP2021/082334 | 11/19/2021 | WO |
