Epileptic seizure predicting device, method for analyzing electrocardiogramic index data, seizure predicting computer program, model constructing device, model constructing method, and model constructing computer program
This disclosure relates to an epileptic seizure predicting device ( 1 ) that executes a seizure predicting process ( 12 ). The seizure predicting process ( 12 ) includes: a process of providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of a subject, to an autoencoder (AE) that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data; a process of calculating an error between the input data and the output data of the subject; and a detection process of detecting a sign of an epileptic seizure of the subject on the basis of whether or not the error exceeds a management limit that the error should not exceed in a case of a seizure-free interval of epilepsy.
1 . An epileptic seizure predicting device configured to execute a seizure predicting process,
the seizure predicting process comprising:
a process of providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of a subject, to an autoencoder that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data;
a process of calculating a reconstruction error calculated from an error between the input data and the output data of the subject; and
a detection process of detecting a sign of an epileptic seizure of the subject on the basis of whether the reconstruction error exceeds a management limit, wherein detection that the reconstruction error is equal to or lower than the management limit indicates that the subject is in a seizure-free interval of epilepsy,
wherein the electrocardiographic index data includes a plurality of types of heart rate variability (HRV) indexes comprising meanNN, SDNN, RMSSD, NN50, Total Power, LF/HF, LFnu, and HFnu,
wherein the reconstruction error is a single error calculated as an L1 norm of a vector composed of a reconstruction error of meanNN, a reconstruction error of SDNN, a reconstruction error of RMSSD, a reconstruction error of NN50, a reconstruction error of Total Power, a reconstruction error of LF/HF, a reconstruction error of LFnu, and a reconstruction error of HFnu.
2 . The epileptic seizure predicting device according to claim 1 , wherein
the learning electrocardiographic index data is generated from an electrocardiographic signal in a seizure-free interval of the epilepsy patient.
3 . The epileptic seizure predicting device according to claim 1 , wherein
in the detection process, the sign of the epileptic seizure of the subject is detected when the error continuously exceeds the management limit for a predetermined time.
4 . The epileptic seizure predicting device according to claim 1 , wherein
the electrocardiographic index data is calculated on the basis of an RRI (R-R Interval) generated from the electrocardiographic signal of the subject.
5 . The epileptic seizure predicting device according to claim 1 , wherein
the learning electrocardiographic index data is generated from electrocardiographic signals of a plurality of epilepsy patients, and
the plurality of epilepsy patients include the subject and an epilepsy patient other than the subject.
6 . The epileptic seizure predicting device according to claim 1 , wherein
the epileptic seizure predicting device is configured to further execute an adjusting process for adjusting the management limit stored in advance in a storage device.
7 . A method to be performed by a computer in order to analyze electrocardiographic index data generated from an electrocardiographic signal of a subject, the method comprising the steps, performed by the computer, of:
providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of the subject, to an autoencoder that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data;
calculating a reconstruction error calculated from an error between the input data and the output data of the subject; and
determining whether the reconstruction error exceeds a management limit, wherein detection that the reconstruction error is equal to or lower than the management limit indicates that the subject is in a seizure-free interval of epilepsy,
wherein the electrocardiographic index data includes a plurality of types of heart rate variability (HRV) indexes comprising meanNN, SDNN, RMSSD, NN50, Total Power, LF/HF, LFnu, and HFnu,
wherein the reconstruction error is a single error calculated as an L1 norm of a vector composed of a reconstruction error of meanNN, a reconstruction error of SDNN, a reconstruction error of RMSSD, a reconstruction error of NN50, a reconstruction error of Total Power, a reconstruction error of LF/HF, a reconstruction error of LFnu, and a reconstruction error of HFnu.
8 . A non-transitory computer-readable storage medium with a seizure predicting computer program stored thereon, wherein the program is configured to cause a computer to execute a seizure predicting process,
the seizure predicting process comprising:
a process of providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of a subject, to an autoencoder that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data;
a process of calculating a reconstruction error calculated from an error between the input data and the output data of the subject; and
a detection process of detecting a sign of an epileptic seizure of the subject on the basis of whether the reconstruction error exceeds a management limit, wherein detection that the reconstruction error is equal to or lower than the management limit indicates that the subject is in a seizure-free interval of epilepsy,
wherein the electrocardiographic index data includes a plurality of types of heart rate variability (HRV) indexes comprising meanNN, SDNN, RMSSD, NN50, Total Power, LF/HF, LFnu, and HFnu,
wherein the reconstruction error is a single error calculated as an L1 norm of a vector composed of a reconstruction error of meanNN, a reconstruction error of SDNN, a reconstruction error of RMSSD, a reconstruction error of NN50, a reconstruction error of Total Power, a reconstruction error of LF/HF, a reconstruction error of LFnu, and a reconstruction error of HFnu.
9 . A model constructing device configured to execute a model constructing process for epileptic seizure prediction,
the model constructing process comprising:
a process of providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of a subject, to an autoencoder that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data;
a calculation process of calculating a reconstruction error calculated from an error between the input data and the output data; and
a setting process of setting, on the basis of the reconstruction error, a management limit, wherein detection that the reconstruction error is equal to or lower than the management limit indicates that the subject is in a seizure-free interval of epilepsy,
wherein the electrocardiographic index data includes a plurality of types of heart rate variability (HRV) indexes comprising meanNN, SDNN, RMSSD, NN50, Total Power, LF/HF, LFnu, and HFnu,
wherein the reconstruction error is a single error calculated as an L1 norm of a vector composed of a reconstruction error of meanNN, a reconstruction error of SDNN, a reconstruction error of RMSSD, a reconstruction error of NN50, a reconstruction error of Total Power, a reconstruction error of LF/HF, a reconstruction error of LFnu, and a reconstruction error of HFnu.
10 . The model constructing device according to claim 9 , wherein
in the setting process, the management limit is set such that: a predetermined proportion of errors that accounts for a majority of a plurality of the errors calculated in the calculation process does not exceed the management limit; and a remainder of the plurality of the errors exceeds the management limit.
11 . The model constructing device according to claim 10 , wherein
the predetermined proportion that accounts for the majority is a proportion in a range of not less than 90% and less than 100%.
12 . A method for constructing a model for epileptic seizure prediction, the method comprising:
providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of a subject, to an autoencoder that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data;
calculating a reconstruction error calculated from an error between the input data and the output data; and
setting, on the basis of the reconstruction error, a management limit, wherein detection that the reconstruction error is equal to or lower than the management limit indicates that the subject is in a seizure-free interval of epilepsy,
wherein the electrocardiographic index data includes a plurality of types of heart rate variability (HRV) indexes comprising meanNN, SDNN, RMSSD, NN50, Total Power, LF/HF, LFnu, and HFnu,
wherein the reconstruction error is a single error calculated as an L1 norm of a vector composed of a reconstruction error of meanNN, a reconstruction error of SDNN, a reconstruction error of RMSSD, a reconstruction error of NN50, a reconstruction error of Total Power, a reconstruction error of LF/HF, a reconstruction error of LFnu, and a reconstruction error of HFnu.
13 . A non-transitory computer-readable storage medium with a model constructing computer program stored thereon, wherein the program configured to cause a computer to execute a model constructing process,
the model constructing process comprising:
a process of providing, as input data, electrocardiographic index data generated from an electrocardiographic signal of a subject, to an autoencoder that has been provided with and has learned with, as learning input data, learning electrocardiographic index data generated from an electrocardiographic signal of an epilepsy patient, and obtaining output data which is reconstruction data of the input data;
a calculation process of calculating a reconstruction error calculated from an error between the input data and the output data; and
a setting process of setting, on the basis of the reconstruction error, a management limit, wherein detection that the reconstruction error is equal to or lower than the management limit indicates that the subject is in a seizure-free interval of epilepsy,
wherein the electrocardiographic index data includes a plurality of types of heart rate variability (HRV) indexes comprising meanNN, SDNN, RMSSD, NN50, Total Power, LF/HF, LFnu, and HFnu,
wherein the reconstruction error is a single error calculated as an L1 norm of a vector composed of a reconstruction error of meanNN, a reconstruction error of SDNN, a reconstruction error of RMSSD, a reconstruction error of NN50, a reconstruction error of Total Power, a reconstruction error of LF/HF, a reconstruction error of LFnu, and a reconstruction error of HFnu.