Evaluation of Painful Disorders by an Expert System

Abstract

A system and method for assessing pain disorders is provided. A stimulus is applied to a subject, and evoked potentials are obtained from at least one electrogram of the subject. A set of features is extracted from the evoked potentials, including features from at least two of a set of features representing connectivity between brain regions, a set of morphological features, a set of features representing time and frequency, a set of features representing signal decomposition, and a set of features representing entropy. From the extracted set of features, a clinical parameter associated with a pain disorder is assigned to the subject using a machine learning model.

Description

Technical Field

[0001] Related Applications This application claims priority to U.S. Provisional Patent Application No. 62 / 828,013, filed on April 2, 2019, with the title "DIAGNOSIS OF PAIN DISORDERS VIA EXPERT SYSTEM", and the entire subject matter thereof is incorporated herein by reference.

[0002] The present invention relates to the diagnosis of pain symptoms and pain disorders by an expert system.

Background Art

[0003] It is difficult to measure the pain experienced by a subject, and currently, most are limited to observing the subject's behavior and self-reporting. However, since both observation and self-reporting are subjective, it is difficult to objectively measure the experienced pain or measure the increase in pain in response to a stimulus. Unfortunately, in many disorders, pain appears as an initial symptom, so if a caregiver cannot objectively measure the subject's pain level, especially the change in pain level in response to a stimulus, the diagnosis of these symptoms and disorders may become complicated.

[0004] For example, neuropathic pain is pain caused by damage or disease that affects the somatosensory nervous system. Neuropathic pain may be accompanied by abnormal sensations (numbness and paresthesia) and / or pain due to stimuli that are not normally painful (allodynia) and / or increased pain due to stimuli that are normally painful (hyperalgesia), and these may occur continuously and / or temporarily. Also, central neuropathic pain may occur as a result of spinal cord injury, multiple sclerosis, or stroke in a part of the central nervous system. Causes of peripheral neuropathy with pain include, in addition to metabolic abnormalities such as diabetes, nutritional deficiencies, toxins, certain viral or bacterial diseases, remote symptoms of malignant tumors, immune-mediated diseases, physical trauma to the nerve trunk, and physical trauma to other tissues such as muscles, joints, bones, and teeth. Fibromyalgia (FM) is characterized by widespread chronic pain, tenderness, mental distress, and fatigue. Fibromyalgia is considered to be a type of disease called "mechanical pain disorder". Other mechanical pain disorders are thought to include temporomandibular joint disorder, chronic fatigue syndrome, fascial pain syndrome, chronic widespread pain, Gulf War syndrome, complex regional pain syndrome, certain post-traumatic stress disorders, certain low back pain, certain vulvovaginitis / vulvodynia, piriformis syndrome, and the like. Although current knowledge about the pathophysiology of FM is limited, neuroimaging studies have revealed the brain's response to experimental pain stimuli in subjects. Thus, it leads to the current assertion that FM, and perhaps other mechanical pain disorders as well, may be due to central sensitization of pain processing. SUMMARY OF THE INVENTION MEANS FOR SOLVING THE PROBLEM

[0005] In one example, the method includes a step of applying a stimulus to a subject and a step of obtaining an evoked potential from at least one electrogram of the subject. From the evoked potential, a set of feature quantities including feature quantities from at least two of a set of feature quantities representing the connectivity between regions of the brain, a set of morphological feature quantities, a set of feature quantities representing time and frequency, a set of feature quantities of signal decomposition, and a set of feature quantities representing entropy is extracted. From the extracted set of feature quantities, a clinical parameter related to a painful disorder is assigned to the subject using a machine learning model.

[0006] In another example, the system includes an electrogram interface that receives an evoked potential recorded from an electrogram of a subject. A feature quantity extractor extracts a set of feature quantities including feature quantities from at least two of a set of feature quantities representing the connectivity between regions of the brain, a set of morphological feature quantities, a set of feature quantities representing time and frequency, a set of feature quantities of signal decomposition, and a set of feature quantities representing entropy from the evoked potential. From the extracted set of feature quantities, a machine learning model assigns a clinical parameter related to a painful disorder to the subject.

[0007] In yet another example, the method includes a step of applying a stimulus to a subject and a step of obtaining an evoked potential from at least one electroencephalogram (EEG) of the subject. Each of a set of feature quantities representing the connectivity between regions of the brain is extracted, and a set of coefficients obtained by one of a set of morphological feature quantities, discrete Fourier transform, autoregressive method, and continuous wavelet transform, a set of feature quantities of signal decomposition, and a set of feature quantities representing entropy is extracted from the evoked potential. Then, the extracted feature quantities are combined to provide a set of composite feature quantities. Using a machine learning model, it is determined whether the subject is likely to benefit from treatment for a painful disorder from the set of composite feature quantities. If it is determined that the treatment will be effective, the treatment is applied to the subject.

Brief Description of the Drawings

[0008]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

DETAILED DESCRIPTION OF THE INVENTION

[0009] As used herein, "painful disorder" refers to a condition having pain as an initial symptom, including, but not limited to, somatic symptom disorder, neuropathy, and chronic pain of unknown origin. The painful disorders evaluated by the systems and methods described herein include pain caused by peripheral mechanisms (e.g., inflammatory, nociceptive, neuropathic, traumatic, or metabolic) or pain caused by central mechanisms (e.g., without an identifiable organic cause in the body), as well as pain that presents acutely, subacutely, and / or chronically (i.e., as persistent pain lasting longer than 6 months) and pain that presents as neuropathic pain.

[0010] As used herein, "subject" refers to a human who is receiving medical evaluation or care by a healthcare provider or who is a participant in a research project.

[0011] Objectively measuring the pain experienced by others is difficult, especially when there is no obvious physiological cause for the pain. Therefore, medical practitioners often have to rely on subjective reports when diagnosing or monitoring patients or subjects in clinical trials under their care. Even worse, subjects may not be able to clearly convey the type of pain they are feeling or various pain phenomena they may experience, making diagnosis and treatment complicated. In the systems and methods described in this specification, a combination of feature quantities extracted from the potentials induced in response to stimuli is used to evaluate one or more of the presence, nature, cause, and severity of the pain experienced by a patient. By combining the feature quantities obtained through multiple different analyses of the evoked potential, each of the presence, nature, cause, and severity of the pain can be accurately determined, and the pain experienced by the patient can be grasped.

[0012] FIG. 1 shows an example of a system 100 for evaluating a subject according to one aspect of the present invention. In one embodiment, this system is implemented as computer-readable instructions executed by a processor and stored in a non-transitory computer-readable medium. The system 100 includes an electrogram interface 102 that receives electrogram data about a subject from a set of electrodes (not shown) and formats this data into a form suitable for use in the system 100. In practice, the electrogram data is recorded after applying a stimulus to the subject to provide an evoked potential. As used in this specification, "evoked potential" is intended to refer to either the result in one electrogram after the introduction of a stimulus or the independent event-related potential obtained from multiple electrograms.

[0013] The feature extractor 104 extracts parameters from the evoked potential. Specifically, the feature extractor extracts parameters regarding the discrimination between a subject with a painful disorder and a subject without a painful disorder, which are referred to as "features" in this specification. Examples of features include various parameters from each of five categories, namely, a first category including parameters representing the form of a selected waveform of electrogram data, a second category using coefficients obtained by signal decomposition of electrogram data, a third category including one or more entropy measurement values for electrogram data, a fourth category including parameters representing the time and frequency characteristics of the signal, and a fifth category including parameters representing the connectivity between various regions of the brain. In the illustrated example, the feature extractor 104 extracts features from at least two of these categories, but it will be understood that depending on the embodiment, features from three, four, or all of the categories can be used.

[0014] The machine learning model 106 assigns clinical parameters to a subject using a plurality of extracted feature quantities. In this machine learning model 106, it is possible to use one or more classification algorithms or regression algorithms. Each of these algorithms analyzes the feature quantities extracted to assign continuous parameters or categorical parameters related to painful disorders, or analyzes a subset of the extracted feature quantities, and provides this information to the user by an appropriate output device such as a display. Clinical parameters can represent any of 1.) the presence of some painful disorder, 2.) the presence of a specific painful disorder (e.g., fibromyalgia), 3.) the predicted or actual response to treatment of a painful disorder, 4.) the presence or absence of a predetermined tender point, 5.) the type of pain, 6.) the subtype of a painful disorder, 7.) the severity of pain, 8.) the severity of a painful disorder, 9.) the change over time in the severity of a painful disorder, 10.) the presence or absence of a specific pain phenomenon (central sensitization, hyperalgesia, allodynia, hypoalgesia, etc.), 11.) the intensity of a specific pain phenomenon, 12.) the likelihood that a painful disorder will occur in a subject in general, 13.) the likelihood that a specific painful disorder will occur in a subject, 14.) the symptoms of a painful disorder, or 15.) the findings in a physical examination. In the illustrated embodiment, the machine learning model 106 includes one or both of a support vector machine and a random forest classifier.

[0015] An example of a system 200 for evaluating a subject according to one aspect of the present invention is shown in FIG. 2. An electroencephalogram (EEG) device 202 can be used to record electroencephalogram data obtained from the subject. This data is provided to a pain evaluation system 210. The pain evaluation system 210 is implemented as computer-readable instructions executed by a processor 214 and stored on a non-transitory computer-readable medium 230. The pain evaluation system 210 includes a feature extractor 232 that extracts a vector of feature values from the electroencephalogram data. Examples of features include various parameters representing the form of selected waveforms of electroencephalogram data, signal decomposition of electroencephalogram data, one or more entropy measurements for electroencephalogram data, parameters representing the time and frequency characteristics of the signal, and parameters representing the connectivity between various regions of the brain. In the illustrated example, the feature extractor 232 extracts a significant number of features from these categories and uses the composite features generated via a feature reduction process 234 to generate a set of composite features for consideration.

[0016] In one embodiment, the feature extractor 232 can extract features from individual EEGs obtained in response to a stimulus applied to the subject. The stimulus can be, for example, any of heat, cold, mechanical pressure, electrical stimulation, or laser stimulation applied to a selected location on the subject's body. In one example, mechanical force can be applied to one of the identified tender points of the subject, or to one or more of the standard tender points associated with fibromyalgia. This mechanical force can be applied by hand or using a mechanical device for standardizing the applied force. In one embodiment, a rubber stopper with a surface area of 1 square centimeter was used to apply a stimulus at a speed of 200 millimeters per second with a force equivalent to 2 kilograms. In another embodiment, it is possible to acquire multiple EEGs after each stimulus, and extract features from the average value of the EEG signals to remove background noise and isolate event-related potentials. As an example, 30 stimuli are applied at intervals of 10 to 14 seconds to the selected site, and the event-related potential is generated as Woody Filter Mean. In another embodiment, the intensity of the stimulus can be increased step by step until the subject reports feeling pain, and an evoked potential can be obtained only when the stimulus reaches an intensity sufficient to induce a pain response.

[0017] Parameters representing the form of the selected waveform of EEG data can include anything as long as it is a parameter that describes the shape of the detected waveform. The extracted feature quantities can include the amplitude or width of the N1 peak, the amplitude and width of the P1 trough, the voltage between peaks, and the duration of the event-related potential. Parameters representing the entropy that can be calculated from the waveform or event-related potential include the Petrosian fractal dimension (PFD), the fractal dimension by the Higuchi method (HFD), the Hjorth parameter, the spectral entropy parameter, the spectral entropy (SE) that calculates the amount of entropy in the spectral region as a scalar feature quantity using the relative intensity ratio (RIR) of the power spectral density (PSD) of the signal, and the singular value decomposition (SVD) entropy which is a dictionary-based analysis where the signal is decomposed based on a dictionary and the signal can be expressed as a linear sum of dictionary components, etc.

[0018] As an example of the feature quantity of connectivity, there is the oscillatory synchronization of neurons. The oscillatory synchronization of neurons means that there is some definite relationship in the oscillatory modulation of neural activities based on different neurons or neuron groups. The phase synchronization of these oscillatory neurons or neuron groups, especially the temporal and spatial frequency abnormalities of phase synchronization, can be quantified using the frequency band phase synchronization approach for multi-channel EEG signals. Specifically, the phase difference ΔΦ xy (t) between two signals from two electrodes x and y is defined as Φ x (t) - Φ y (t). The frequency band phase synchronization feature quantity BS xy for two electrodes can be calculated as follows as the sum of the magnitudes of the phase differences between defined epochs for a predetermined frequency band (for example, alpha, beta, gamma, delta, or theta).

Equation

[0019] Here, N is the number of samples of the defined epoch.

[0020] For each of the possible electrode pairs for each frequency band, the phase synchronization feature amount can be calculated, but it can be understood that the potential number of feature amounts is large. In a system using 32 electrodes such as the standard 10-20 method, approximately 500 pairs are available for each frequency band. Therefore, in one embodiment, the set of feature amounts can be reduced to the number of pairs most relevant to classification.

[0021] As feature amounts of signal decomposition, it can include signals derived by principal component analysis, empirical mode decomposition, discrete wavelet transform, or similar processes. For time and frequency feature amounts, they can be determined as coefficients obtained by discrete Fourier transform, autoregressive method, or continuous wavelet transform. In one embodiment using wavelet decomposition, the feature amounts of signal decomposition include, for each electrode, a scale-dependent feature amount and a scale-invariant feature amount. For the time series x of the voltage value from the i-th electrode i the wavelet coefficients W a (n) generated by wavelet decomposition can be defined as follows.

Equation

[0022] Here, Ψ is the wavelet function, M is the length of the time series, and a and n define the coefficient calculation positions.

[0023] The scale-dependent feature amount SD can be determined as follows.

Equation

[0024] Here, N = M / a is the number of coefficients at a predetermined scale a, and Wa is the mother wavelet function.

[0025] The scale-independent feature quantity SI can be calculated as the sum of the q-th powers of the maximum values of the wavelet functions in Equation 3, and represents the different fractal characteristics of the time series of voltage values at different scaling components τ(q).

Number

[0026] It will be understood that the specific wavelet function, scale a, and values of q can vary depending on the application. In one example, the Morlet function can be used.

[0027] The autoregressive feature quantity is derived from a spectral analysis model in which the voltage value x(n) of each electrode is modeled as the output of a linear system characterized by a rational structure. A set of parameters is estimated from a given data sequence x(n) (0 ≤ n ≤ N - 1), and the power spectral density (PSD) is calculated therefrom. The PSD can be calculated by solving a series of linear equations, whereby the data is modeled as the output of a causal all-pole discrete filter with white noise as the input. The autoregressive model of each electrode is expressed as follows with degree p.

Number

[0028] Here, a(k) are p autoregressive coefficients, and w(n) is a white noise signal whose variance is equal to the variance of the signal x(n).

[0029] In one embodiment, the autocorrelation function is determined by the Burg method using an appropriate order p for the autocorrelation selected according to the Akaike Information Criterion (AIC). Instead of using the entire spectrum as a feature quantity, six types of feature quantities can be extracted from each of the five major frequency ranges used in EEG analysis, namely alpha, beta, gamma, delta, and theta. Examples of the feature quantities include: 1) the average power of each frequency range; 2) the variance power of each frequency range; 3) the average frequency of the average power observed in each frequency range; 4) the variance frequency of the average power observed in each frequency range and the ratio between them; 5) the average power of each frequency range; and 6) the variance of each frequency range. However, it will be understood that other descriptive statistics representing the frequency components of the signal may also be used.

[0030] In another embodiment, changes in two or more of the above-described feature quantities between the first EEG acquired at the first time and the second EEG acquired at the second time may be used as feature quantities for classification. Thereby, for example, it becomes possible to monitor the temporal change in the state of a subject who responds to treatment. In another example, in order to track the rate of change of features over time, one or more time derivatives of the feature quantities can be included in the feature quantities for classification. In some embodiments, higher-order time derivatives can also be used as feature quantities.

[0031] The feature reduction process 234 generates composite features from the extracted multiple features and reduces the dimensionality of the feature space for classification. Appropriate composite features can be generated by feature reduction algorithms such as Lasso, ElasticNet, Randomized PCA, ISO MAP, SPECTRAL Embedding, Random Projections, Tree Based Methods, Recursive Feature Selection, Multivariate feature reduction, etc. In one embodiment, univariate methods such as mutual information measurement are used to identify features that are substantially independent for use in a classifier. These features can be linearly combined with weighted terms to form composite features, which can be used for classification. Instead of or in addition to the above, it is also possible to employ principal component analysis to generate a set of composite features for classification from the extracted features. Thereby, the complexity of the classifier is reduced, and the likelihood that the classifier is overfitted to known training samples is reduced.

[0032] The pattern recognition classifier 238 uses the generated composite features to classify the subject into one of a plurality of classes based on the composite features. These classes each represent one of the presence of pain disorders in general or a specific pain disorder (e.g., fibromyalgia), the type of pain, findings or symptoms obtained in a physical examination related to the pain disorder, the subtype of the pain disorder, the severity of the pain, the severity of the pain disorder, the presence or absence of pain phenomena, the response to treatment for the pain disorder, the change in the patient's condition, the presence or absence of predetermined tender points, the likelihood that the subject will develop a pain disorder in general or a specific pain disorder. In one embodiment, the classification is binary between a "pain disorder" class and a "non-pain disorder" class, but it will be understood that additional classes may be included, such as a class representing the range of likelihood that the subject will develop a pain disorder. In another embodiment, the classes may represent the change in the subject's condition regardless of the presence or absence of treatment, such as classes of "improvement", "no change", "deterioration", or a class representing the degree of change.

[0033] In addition to the composite feature quantity, it will be understood that it is also possible to use the original extracted feature quantity and feature quantities from other clinical data representing the patent. Clinical parameters include the subject's medical history including demographic parameters (e.g., age, gender), physiological parameters (blood pressure, blood glucose level, heart rate, oxygen saturation, etc.), categorical variables representing the presence or absence of various conditions, the drugs the subject is taking, and the dosages of these drugs, but are not limited thereto.

[0034] The pattern recognition classifier 238 can utilize one or more pattern recognition algorithms. Each of these pattern recognition algorithms analyzes the extracted feature quantity or a subset of the extracted feature quantities to classify the subject into one of a plurality of classes and provides this information to the display 240. When using multiple classification models or regression models, it is possible to use an arbitration element to provide consistent results from the multiple models. The training process of a given classifier varies from implementation to implementation, but in training, it is generally necessary to statistically aggregate the training data into one or more parameters associated with the output class. The training process can be achieved by a remote system and / or local devices, wearables, and applications. In a rule-based model such as a decision tree, when selecting rules for classifying a user using the extracted feature quantities, domain knowledge such as that provided by one or more experts can be used instead of or to complement the training data. Various techniques can be used for classification algorithms, including support vector machines (SVMs), regression models, self-organizing maps, fuzzy logic systems, data fusion processes, boosting and bagging methods, rule-based systems, or artificial neural networks (ANNs).

[0035] For example, an SVM classifier can conceptually divide a boundary in an N-dimensional feature space by using a plurality of functions called hyperplanes. Here, each of the N dimensions represents one of the feature quantities associated with the feature vector. The boundary defines the range of the feature quantity values associated with each class. Therefore, depending on the position in the feature space with respect to the boundary, the output class and the associated confidence value for a given input feature vector can be determined. In one embodiment, the SVM can be implemented by a kernel method using a linear kernel or a non-linear kernel.

[0036] An ANN classifier includes a plurality of nodes having a plurality of interconnections. The values obtained from the feature vector are provided to a plurality of input nodes. Each input node provides these input values to a layer consisting of one or more intermediate nodes. A given intermediate node receives one or more output values from the previous node. The received values are weighted according to a series of weights established during the learning of the classifier. The intermediate node converts the received values into one output according to the transfer function of that node. For example, the intermediate node can sum the received values and apply the sum value to a binary step function. The final layer consisting of a plurality of nodes provides the confidence value for the output class of the ANN. Each of these nodes has an associated value representing the confidence level for one of the associated output classes of the classifier.

[0037] Many ANN classifiers are fully connected feedforward. However, convolutional neural networks include convolutional layers where nodes from the previous layer are connected only to a subset of the nodes in the convolutional layer. Recurrent neural networks are a type of neural network in which the connections between nodes form a directed graph along a temporal order. Different from feedforward networks, recurrent neural networks can incorporate feedback from states resulting from previous inputs, so the output of a recurrent neural network for a given input can be a function not only of that input but also of one or more previous inputs. As an example, there is an improved version of recurrent neural networks, the long short-term memory (LSTM) network, which makes it easier to store past data in memory.

[0038] Rule-based classifiers apply a set of logical rules to the extracted features to select an output class. Generally, the rules are applied in order, and the logical results at each step affect the analysis in subsequent steps. The specific rules and their order can be determined from any or all of training data, analogy from past cases, and existing domain knowledge. As an example of a rule-based classifier, there is the decision tree algorithm. In the decision tree algorithm, the values of the features included in the feature set are compared with the corresponding thresholds of a hierarchical tree structure to select the class of the feature vector. The random forest classifier is an improvement of the decision tree algorithm in the bootstrap aggregation, i.e., "bagging" method. In this method, multiple decision trees are learned for random samples of the training set, and the average result (average value, median, mode, etc.) of the multiple decision trees is returned. In the case of a classification task, since the results from each tree will be classificatory, the mode result can be used.

[0039] In one example, an unsupervised machine learning model can be applied using the extracted feature quantities. In this example, the identities and numbers of the categories to which the subjects can be assigned are not known in advance. For example, in the study of new types and subtypes of painful disorders, an unsupervised learning process can be utilized to identify sets of subjects with similar characteristics or sets of subjects deviating from the expected parameters. Examples of unsupervised learning algorithms that can be used for this purpose include clustering algorithms such as the k-means method and hierarchical clustering, self-organizing maps, and anomaly detection systems.

[0040] Considering the structural and functional feature quantities described above, an exemplary method will be better understood with reference to FIGS. 3 and 4. For the sake of simplicity, the exemplary methods of FIGS. 3 and 4 are illustrated and described as being executed continuously, but it should be understood that these examples are not limited to the illustrated order. This is because in other examples, some actions may occur in a different order, multiple times, and / or simultaneously than those illustrated and described herein. Furthermore, it is not necessary to perform all the actions described to carry out a certain method.

[0041] FIG. 3 shows an example of a method 300 for diagnosing a painful disorder. At 302, a subject is stimulated. At 304, at least one electrogram of the subject is acquired, providing an evoked potential. It will be understood that when acquiring multiple electrograms, it is possible to apply the stimulation at 302 prior to each electrogram. The electrogram includes a plurality of electrogram signals, each of which is acquired from an electrode placed on the subject's scalp. In one example, a plurality of electrograms can be acquired in response to the applied stimulation, providing a plurality of event-related potentials. These are then averaged to isolate the event-related potential.

[0042] In 306, a set of feature quantities including at least two of a set of feature quantities representing the connectivity between regions of the brain, a set of morphological feature quantities, a set of feature quantities representing time and frequency as feature quantities of signal decomposition, and a set of feature quantities representing entropy is extracted from the evoked potential. In one embodiment, feature quantities of all five types are extracted. The set of morphological feature quantities can include, for example, any or all of the amplitude of the N1 peak in the evoked potential, the depth of the N1 peak in the evoked potential, the amplitude of the P1 trough in the evoked potential, the depth of the P1 trough in the evoked potential, the voltage between peaks, and the duration of the event-related potential. The set of feature quantities representing entropy can include any or all of the Petrosian fractal dimension, the fractal dimension by the Higuchi method, the Hjorth parameter, the spectral entropy parameter, and the singular value decomposition entropy.

[0043] The set of feature quantities of signal decomposition can include values derived by one of principal component analysis, empirical mode decomposition, and discrete wavelet transform. The set of feature quantities representing time and frequency can include coefficients obtained by one of discrete Fourier transform, autoregressive method, and continuous wavelet transform. The set of feature quantities representing the connectivity between regions of the brain can include, as described above, a measurement of the oscillation synchronization of neurons. In one embodiment, according to a feature quantity reduction algorithm applied, for example, during the training of a machine learning model, the extracted sets of feature quantities can be combined to provide a set of composite feature quantities as part or all of the extracted sets of feature quantities.

[0044] At 308, a machine learning model assigns clinical parameters related to a painful disorder to a subject from the set of extracted features. "Related to a painful disorder" means that the clinical parameter represents any one of the presence of a painful disorder in general or a specific painful disorder, a predicted or actual change in the subject's condition, a painful disorder, the presence or absence of a predetermined pressure point, the type of pain, a subtype of a painful disorder, the severity of pain, the severity of a painful disorder, the presence or absence of a specific pain phenomenon, the intensity of a specific pain phenomenon, the likelihood that a painful disorder in general will occur in the subject or the likelihood that a specific painful disorder will occur in the subject. In one example, a subject is classified into one of a plurality of classes by, for example, a support vector machine or a random forest classifier trained with known training samples obtained from subjects whose associated classes are known, according to the extracted features. If it is determined that the subject is likely to benefit from treatment, treatments such as behavioral biofeedback, psychotherapy focused on sleep disorders (e.g., cognitive behavioral therapy), training in relaxation techniques, and pharmaceutical interventions can be performed.

[0045] FIG. 4 shows an example of a method 400 for determining whether a subject is likely to respond to treatment, according to one aspect of the present invention. At 402, a stimulus is applied to the subject. At 404, an evoked potential is obtained from at least one electroencephalogram (EEG) of the subject. In one embodiment, a plurality of event-related potentials are obtained from the subject in response to each stimulus, and the evoked potential is generated as the Woody filter average of the plurality of event-related potentials.

[0046] At 406, a set of feature quantities representing the connectivity between brain regions is extracted from the evoked potential. At 408, a set of morphological feature quantities is extracted from the evoked potential. At 410, a set of coefficients obtained by one of discrete Fourier transform, autoregressive method, and continuous wavelet transform is extracted from the evoked potential. At 412, a set of feature quantities of signal decomposition is extracted from the evoked potential. At 414, a set of feature quantities representing entropy is extracted from the evoked potential. At 416, each of the set of feature quantities representing the connectivity between brain regions, the set of morphological feature quantities, the set of coefficients obtained by one of discrete Fourier transform, autoregressive method, and continuous wavelet transform, the set of feature quantities of signal decomposition, and the set of feature quantities representing entropy is combined to provide a set of composite feature quantities. It will be understood that an appropriate feature reduction algorithm can be applied during the training of the machine learning model to determine an appropriate set of composite features for analysis.

[0047] At 418, it is determined from the set of composite feature quantities in the machine learning model whether the subject is likely to benefit from treatment for a painful disorder. It will be understood that the machine learning model can provide an output indicating whether the subject has a painful disorder or is likely to have a painful disorder, determine whether a certain treatment is likely to be successful, or determine whether the ongoing treatment is effective for the treatment of the painful disorder. If it is determined that the likelihood that the treatment will benefit the subject is low (N), the method ends. Otherwise (Y), at 420, treatment is administered to the subject. In the illustrated example, this treatment can include behavioral biofeedback, psychotherapy, training in relaxation techniques, and / or pharmaceutical intervention.

[0048] FIG. 5 is a schematic block diagram showing an exemplary system 500 of hardware components that can implement an example of the systems and methods disclosed herein. System 500 can include various systems and subsystems. System 500 can be a personal computer, laptop computer, workstation, computer system, appliance, application specific integrated circuit (ASIC), server, server BladeCenter, server farm, or the like.

[0049] System 500 can include a system bus 502, a processing device 504, a system memory 506, memory devices 508 and 510, a communication interface 512 (e.g., a network interface), a communication link 514, a display 516 (e.g., a video screen), and an input device 518 (e.g., a keyboard, touch screen, and / or mouse). System bus 502 can communicate with processing device 504 and system memory 506. Also, additional memory devices 508 and 510, such as hard disk drives, servers, stand-alone databases, or other non-volatile memories, can communicate with system bus 502. System bus 502 interconnects processing device 504, memory devices 506 - 510, communication interface 512, and display 516 with each other. In some examples, system bus 502 also interconnects with additional ports (not shown), such as universal serial bus (USB) ports.

[0050] Processing device 504 can be a computing device and can include an application specific integrated circuit (ASIC). Processing device 504 executes a series of instructions for implementing the operations of the examples disclosed herein. The processing device can include a processing core.

[0051] The additional memory devices 506, 508, and 510 can store data, programs, instructions, databases queries in text or compiled form, and any other information that may be required to operate the computer. Memories 506, 508, 510 can be implemented as computer-readable media (integral or removable), such as memory cards, disk drives, compact discs (CDs), or servers accessible via a network. In one example, memories 506, 508, 510 can include text, images, video, and / or audio, some of which may be available in a format understandable by humans.

[0052] In addition to or instead of the above, system 500 can access an external data source or query source via a communication interface 512 communicable with system bus 502 and communication link 514.

[0053] In operation, system 500 can be used to implement one or more portions of the pain assessment system according to the present invention, particularly the feature extractor 104 and the machine learning model 106. The computer-executable logic for implementing the pain assessment system is in one or more of the system memory 506, memory devices 508, and 510 in one example. The processing device 504 executes one or more computer-executable instructions derived from the system memory 506, memory devices 508, and 510. As used herein, the term "computer-readable media" refers to the media involved in providing instructions to the processing device 504 for execution. This media may be distributed among a plurality of discrete assemblies all operably coupled to a common processor or related set of processors.

[0054] In the above description, details have been shown specifically so as to enable a complete understanding of the embodiments. However, it is understood that these embodiments can be implemented even without such specific details. For example, physical components can be shown in block diagrams so that the embodiments are not obscured by unnecessary details. In other examples, well-known circuits, processes, algorithms, structures, and techniques can be shown without unnecessary details in order to avoid obscuring the embodiments.

[0055] The above-described technologies, blocks, processes, and means can be implemented in various ways. For example, these technologies, blocks, processes, and means can be implemented in hardware, software, or a combination thereof. In the case of hardware implementation, the processing device can be implemented in one or more of the following: application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), processors, controllers, microcontrollers, microprocessors, and other electronic units designed to perform the above-described functions and / or combinations thereof.

[0056] It should also be noted that multiple embodiments can be described as processes illustrated as flowcharts, flow diagrams, data flow diagrams, structure diagrams, or block diagrams. In a flowchart, operations can be described as sequential processes, but many of these operations can be executed in parallel or simultaneously. Also, the order of operations can be changed. When the operations of a process are completed, the process ends, but there may be additional steps not shown. One process can be made to correspond to a method, function, procedure, subroutine, subprogram, etc. When a process corresponds to a function, its end corresponds to returning the function to the calling function or the main function.

[0057] Furthermore, embodiments can be implemented by hardware, software, script languages, firmware, middleware, microcode, hardware description languages, and / or any combination thereof. When implemented in software, firmware, middleware, script languages, and / or microcode, program code or code segments for performing the necessary tasks can be stored in a machine-readable medium such as a storage medium. Code segments or machine-executable instructions can represent procedures, functions, subprograms, programs, routines, subroutines, modules, software packages, scripts, classes, or any combination of instructions, data structures, and / or program statements. A code segment can be combined with other code segments or hardware circuits by passing information, data, arguments, parameters, and / or memory contents. Information, arguments, parameters, data, etc. can be passed, transferred, or transmitted via any suitable means including memory sharing, message passing, ticket passing, network transmission, etc.

[0058] In a firmware and / or software implementation, the methodology can be implemented with modules (e.g., procedures, functions, etc.) that perform the functions described herein. When implementing the methodology described herein, it is possible to use any machine-readable medium that explicitly embodies the instructions. For example, software code can be stored in memory. Memory can be implemented either within the processor or outside the processor. As used herein, the term "memory" refers to long-term, short-term, volatile, non-volatile, or any other type of storage medium, and is not limited to a particular type of memory or multiple memories, or the type of medium in which the memory is stored.

[0059] Furthermore, as disclosed herein, the term "memory medium" can represent one or more memories for storing data, including read-only memory (ROM), random access memory (RAM), magnetic RAM, core memory, magnetic disk storage media, optical storage media, flash memory devices, and / or other machine-readable media for storing data. The term "machine-readable medium" includes, but is not limited to, portable or fixed storage devices, optical storage devices, wireless channels, and / or various other storage media that can contain or hold instructions and / or data.

[0060] The above description is illustrative. Of course, it is impossible to describe all possible combinations of components and methodologies, but those skilled in the art will recognize that many more combinations and permutations are possible. Accordingly, the present disclosure is intended to cover any such changes, modifications, and variations that are within the scope of this application, including the appended claims. As used herein, the term "comprising" means including but not limited to. The expression "based on" means at least partially based on. Further, when the present disclosure or the claims recite "a," "an," "first," "another," element or their equivalents, it should be construed to include one or more such elements without excluding or requiring two or more of such elements.

Claims

1. A method of operating a pain evaluation system for a painful disorder, comprising the following steps performed by the pain evaluation system: (1) A step of applying a stimulus to a subject; (2) A step of obtaining an evoked potential from at least one electrogram of the subject; (3) A step of extracting a set of feature quantities from the evoked potential; (4) A step of classifying the subject into one of a first class representing the presence of the painful disorder and a second class representing the absence of the painful disorder, using the set of the extracted feature quantities by a machine learning model; The set of the extracted feature quantities includes feature quantities from at least two of a set of feature quantities representing the connectivity between regions of the brain, a set of morphological feature quantities, a set of feature quantities representing time and frequency, a set of feature quantities of signal decomposition, and a set of feature quantities representing entropy, The machine learning model is trained with known training samples obtained from subjects whose associated classes are known. A method of operating a pain evaluation system.

2. The set of feature quantities representing entropy includes Petrosian fractal dimension, fractal dimension by the Higuchi method, Hjorth parameter, spectral entropy parameter, and singular value decomposition entropy. The operating method according to claim 1.

3. The set of feature quantities representing the connectivity between regions of the brain includes a measurement of the oscillatory synchronization of neurons. The operating method according to claim 1.

4. Further comprising a step of providing a set of composite feature quantities by combining the set of the extracted feature quantities, and the step of classifying the subject into one of the first class or the second class by the machine learning model according to the set of the extracted feature quantities includes the step of classifying the subject into one of the first class or the second class by the machine learning model according to the set of the composite feature quantities. The operating method according to claim 1.

5. The step of applying the stimulus to the subject and obtaining the evoked potential from the at least one electrogram of the subject comprises: applying a first stimulus to the subject; after applying the first stimulus, obtaining a first event-related potential from the at least one electrogram; applying a second stimulus to the subject; after applying the second stimulus, obtaining a second event-related potential from the at least one electrogram; obtaining the evoked potential by averaging at least the first event-related potential and the second event-related potential, the operating method according to claim 1.

6. The subject is one of a plurality of participants in a research project, and the step of assigning clinical parameters comprises applying the machine learning model to a set of extracted features for each person in a subset of the plurality of participants, and assigning each person in the subset of the plurality of participants to a group of participants having similar values for the set of extracted features, the operating method according to claim 1.

7. The stimulus is a first stimulus, the evoked potential is a first evoked potential, the electrogram is a first electrogram acquired at a first time, the method further comprises the step of applying a second stimulus to the subject at a second time and obtaining a second evoked potential from a second electrogram of the subject, the set of features represents a change in features from at least two of the set of features representing connectivity between regions of the brain, the set of morphological features, the set of features representing time and frequency, the set of features of signal decomposition, and the set of features representing entropy, and the clinical parameter represents a change in the state of the subject between the first time and the second time, the operating method according to claim 1.

8. Applying a stimulus to the subject includes applying mechanical pressure to a selected location on the subject's body, and the painful disorder is a clinical parameter representing one of the existences of mechanical painful disorders. The operation method according to claim 1.

9. An electrogram interface that receives the evoked potential recorded from the electrogram of the subject, A set of feature quantities including feature quantities from at least two of a set of feature quantities representing the connectivity between regions of the brain, a set of morphological feature quantities, a set of feature quantities representing time and frequency, a set of feature quantities of signal decomposition, and a set of feature quantities representing entropy. A feature quantity extractor that extracts from the evoked potential, A machine learning model that classifies the subject into one of a first class representing the presence of the painful disorder and a second class representing the absence of the painful disorder using the set of the extracted feature quantities. The machine learning model is trained with known training samples obtained from subjects whose associated classes are known. A system for evaluating a painful disorder.

10. The set of feature quantities includes feature quantities from each of the set of feature quantities representing the connectivity between regions of the brain, the set of morphological feature quantities, the set of feature quantities representing time and frequency, the set of feature quantities of signal decomposition, and the set of feature quantities representing entropy. The system according to claim 9.

Images