	US 12,408,864 B2
	Artificial neural network-based nuclear magnetic resonance and magnetic resonance spectroscopy data processing method and apparatus thereof
Hyeonjin Kim, Seoul (KR); and Hyeong Hun Lee, Gyeonggi-do (KR)
Assigned to SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION, Seoul (KR)
Appl. No. 17/430,461
Filed by SEOUL NATIONAL UNIVERSITY R&DB FOUNDATION, Seoul (KR)
PCT Filed Apr. 7, 2020, PCT No. PCT/KR2020/004655 § 371(c)(1), (2) Date Aug. 12, 2021, PCT Pub. No. WO2020/209566, PCT Pub. Date Oct. 15, 2020.
Claims priority of application No. 10-2019-0041059 (KR), filed on Apr. 8, 2019; and application No. 10-2020-0041497 (KR), filed on Apr. 6, 2020.
Prior Publication US 2022/0117552 A1, Apr. 21, 2022
Int. Cl. A61B 5/00 (2006.01); A61B 5/055 (2006.01)

CPC A61B 5/4866 (2013.01) [A61B 5/055 (2013.01); A61B 5/7264 (2013.01)]

8 Claims

1. An apparatus for processing nuclear magnetic resonance and magnetic resonance spectroscopy data, the apparatus comprising:

a data storage unit configured to store training data, the training data including a plurality of pairs of incomplete data including at least one truncated section and ground truth data having no truncated section;

a data input unit configured to receive the training data from the data storage unit, receive input data from a magnetic resonator, and classify the input data into the ground truth data and the incomplete data;

a data recovery unit comprising a first artificial neural network training unit and a first artificial neural network prediction unit, wherein:

the first artificial neural network training unit is configured to receive the training data from the data input unit and train a first neural network to recover the incomplete data of the training data into the ground truth data of the training data, and

the first artificial neural network prediction unit is configured to receive the incomplete data of the input data from the data input unit, generate recovered data having no truncated section by recovering the truncated section from the incomplete data of the input data by using the first artificial neural network, and determine the input data as bad data when the incomplete data of the input data cannot be recovered;

a disease diagnosis unit comprising a second artificial neural network training unit and a second artificial neural network prediction unit, wherein:

the second artificial neural network training unit is configured to train a second artificial neural network based on at least one of the ground truth data received from the data input unit, the recovered data received from the data recovery unit, a signal obtained by fast Fourier transform (FFT) of the ground truth data, and a signal obtained by FFT of the recovered data to generate metabolite quantification data, and

the second artificial neural network prediction unit is configured to determine a target metabolite that can distinguish between normal groups and patient groups for a specific disease using results obtained by categorizing and classifying the metabolite quantification data into sections according to a concentration range of a specific metabolite by using the second artificial neural network, and diagnose diseases based on concentration of the target metabolite; and

a result output unit configured to output an error rate based on a true value for concentration of the target metabolite received from the second artificial neural network training unit and a predicted value for concentration of the target metabolite received from the second artificial neural network prediction unit,

wherein the second artificial neural network prediction unit is configured to use a Moore-Penrose pseudoinverse matrix to calculate a multiplication coefficient value of each metabolite signal from the metabolite quantification data predicted by the second artificial neural network, the calculated multiplication coefficient value referring to an intrinsic concentration of each metabolite.

7. A method of processing nuclear magnetic resonance and magnetic resonance spectroscopy data by using a computing apparatus, the method comprising:

storing training data, the training data including a plurality of pairs of incomplete data including at least one truncated section and ground truth data having no truncated section;

receiving the training data and input data from a magnetic resonator;

classifying the input data into the ground truth data and the incomplete data;

training a first artificial neural network based on the training data to recover the incomplete data of the training data into the ground truth data of the training data;

generating recovered data having no truncated section by recovering the truncated section from the incomplete data of the input data by using the first artificial neural network;

determining the input data as bad data when the incomplete data of the input data cannot be recovered;

training a second artificial neural network based on at least one of the ground truth data, the recovered data, and a signal obtained by FFT of the ground truth data, and a signal obtained by FFT of the recovered data to generate metabolite quantification data;

determining a target metabolite that can distinguish between normal groups and patent groups for a specific disease using results obtained by categorizing and classifying the metabolite quantification data into sections according to a concentration range of a specific metabolite by using the second artificial neural network;

using a Moore-Penrose pseudoinverse matrix to calculate a multiplication coefficient value of each metabolite signal from the metabolite quantification data predicted by the second artificial neural network, the calculated multiplication coefficient value referring to an intrinsic concentration of each metabolite;

diagnosing diseases based on concentration of the target metabolite; and

outputting an error rate based on a true value for concentration of the target metabolite and a predicted value for concentration of the target metabolite.