Apparatus and method for distinguishing neural waveforms
A neural waveform distinguishment apparatus includes: a neural waveform obtainment unit that obtains multiple neural waveforms in a pre-designated manner from neural signals sensed by way of at least one electrode; a preprocessing unit that obtains multiple gradient waveforms by calculating pointwise slopes in each of the neural waveforms; a feature extraction unit comprising an encoder ensemble composed of multiple encoders, which have a pattern estimation method learned beforehand and include different numbers of hidden layers, where the feature extraction unit obtains multiple codes as multiple features extracted by the encoders respectively from the gradient waveforms and concatenates the codes extracted by the encoders respectively to extract a feature ensemble for each of the gradient waveforms; and a clustering unit that distinguishes the neural waveforms corresponding respectively to the gradient waveforms by clustering the feature ensembles extracted respectively in correspondence to the gradient waveforms according to a pre-designated clustering technique.
1. An electronic device for representing neural waveforms by providing a multi-dimensional grouping of clustered feature ensembles comprising:
a processor including a central processing unit capable of executing a computer program,
a display, and
a memory connected to the processor,
wherein the memory stores program instructions for providing a multi-dimensional grouping of clustered feature ensembles that, when executed, cause the processor to
obtain a plurality of neural waveforms from neural signals sensed by at least one electrode;
calculate pointwise slopes in each of the plurality of neural waveforms to obtain a plurality of gradient waveforms;
extract a plurality of features from the plurality of gradient waveforms using an encoder ensemble including a plurality of encoders executing a pre-learned pattern estimation method, each of the plurality of encoders using a different number of hidden layers to extract the plurality of features from the same plurality of gradient waveforms;
extract a plurality of codes for each of the plurality of gradient waveforms by concatenating the plurality of extracted features, respectively;
extract a plurality of feature ensembles for each of the plurality of gradient waveforms by concatenating the plurality of extracted codes, respectively; and
cluster the plurality of feature ensembles to distinguish the plurality of neural waveforms corresponding each of the plurality of gradient waveforms,
wherein the encoder ensemble includes a plurality of decoders that correspond respectively to the plurality of encoders, each of the decoders configured to receive a code extracted from the gradient waveforms by a corresponding encoder and to recover the gradient waveforms inputted to the corresponding encoder according to a learned pattern recovery method,
wherein learning by the plurality of encoders is performed as an error, calculated from differences between the gradient waveforms and recovered waveforms recovered by the decoders, that is back-propagated by way of the decoders, and
wherein the clustered feature ensembles distinguishing the plurality of neural waveforms are displayed on the display,
wherein the different number of hidden layers is set for each of the plurality of encoders so that each of the plurality of encoders outputs different features for the same plurality of gradient waveforms,
wherein the same plurality of gradient waveforms is input to each of the plurality of encoders.
2. The electronic device of claim 1 , wherein the learning by each of the plurality of encoders is performed in the same manner using the same gradient waveforms.
3. The electronic device of claim 1 , wherein the plurality of feature ensembles are clustered into at least one clustered feature ensemble according to a density-based spatial clustering of applications applying a noise (DBSCAN) technique.
4. The electronic device of claim 1 ,
wherein the neural signals are obtained by sampling raw-level neural signals in analog form sensed by the at least one electrode and converting the raw-level neural signals into digital neural signals, extracting the digital neural signals with intensities at or above a pre-designated threshold intensity, and aligning the digital neural signals with intensities at or above a pre-designated threshold intensity in a pre-designated manner.
5. A method for distinguishing neural waveforms, the method comprising:
providing a processor including a central processing unit capable of executing a computer program, a display, and a memory connected to the processor, wherein the memory stores program instructions for providing a multi-dimensional grouping of clustered feature ensembles, including instructions for
obtaining a plurality of neural waveforms in a pre-designated manner from neural signals sensed by way of at least one electrode;
obtaining a plurality of gradient waveforms by calculating pointwise slopes in each of the plurality of neural waveforms;
obtaining a plurality of codes using an encoder ensemble, the encoder ensemble composed of a plurality of encoders having a previously learned pattern estimation method and each having different numbers of hidden layers, the plurality of codes obtained as a plurality of features extracted by the plurality of encoders from the same plurality of gradient waveforms;
extracting a feature ensemble for each of the plurality of gradient waveforms by concatenating the plurality of codes extracted by the plurality of encoders, respectively;
distinguishing the plurality of neural waveforms corresponding respectively to the plurality of gradient waveforms by clustering a plurality of feature ensembles extracted respectively in correspondence to the plurality of gradient waveforms according to a pre-designated clustering technique; and
displaying the clustered plurality of feature ensembles that distinguish the plurality of neural waveforms on a display,
wherein the encoder ensemble includes a plurality of decoders that correspond respectively to the plurality of encoders, each of the decoders configured to receive a code extracted from the gradient waveforms by a corresponding encoder and to recover the gradient waveforms inputted to the corresponding encoder according to a learned pattern recovery method, and
wherein learning by the plurality of encoders is performed as an error, calculated from differences between the gradient waveforms and recovered waveforms recovered by the decoders, that is back-propagated by way of the decoders,
wherein the different number of hidden layers is set for each of the plurality of encoders so that each of the plurality of encoders outputs different features for the same plurality of gradient waveforms,
wherein the same plurality of gradient waveforms is input of each of the plurality of encoders.
6. The method of claim 5 , wherein the distinguishing the plurality of neural waveforms comprises clustering the plurality of feature ensembles into at least one cluster according to a density-based spatial clustering of applications with a noise (DBSCAN) technique.
7. The method of claim 5 , wherein the obtaining of the plurality of neural waveforms comprises sampling raw-level neural signals of an analog form sensed by way of the at least one electrode and converting the raw-level neural signals into a digital form; and
obtaining the plurality of neural waveforms by extracting neural signals having intensities of a pre-designated threshold intensity or greater from among the neural signals and aligning the extracted neural signals in a pre-designated manner.