US 12,140,652 B2
Systems and methods for provisioning training data to enable neural networks to analyze signals in NMR measurements
Silvère Lux, Schirrhein (FR)
Assigned to Bruker BioSpin GmbH & Co. KG, Ettlingen (DE)
Filed by Bruker BioSpin GmbH Co. KG, Ettlingen (DE)
Filed on Dec. 15, 2023, as Appl. No. 18/542,259.
Application 18/542,259 is a continuation of application No. PCT/EP2022/056581, filed on Mar. 14, 2022.
Claims priority of application No. 21179968 (EP), filed on Jun. 17, 2021.
Prior Publication US 2024/0151793 A1, May 9, 2024
Int. Cl. G01R 33/46 (2006.01); G01N 24/08 (2006.01); G01N 35/00 (2006.01); G06N 3/0455 (2023.01); G06N 3/0464 (2023.01); G06N 3/09 (2023.01)
CPC G01R 33/4625 (2013.01) [G01N 24/087 (2013.01); G01N 35/00693 (2013.01); G06N 3/0455 (2023.01); G06N 3/0464 (2023.01); G06N 3/09 (2023.01); G01N 2035/00702 (2013.01)] 16 Claims
OG exemplary drawing
 
1. A computer-implemented method for generating a data record of a training dataset set configured to train a neural network for determination of a concentration of a target molecule in an NMR sample, comprising:
obtaining an NMR spectrum associated with a known concentration of the target molecule, with the obtained NMR spectrum having a region of interest in which the NMR spectrum exceeds a predefined noise threshold value;
adjusting the obtained NMR spectrum by applying a random shift in a range from −0.2 ppm to +0.2 ppm of the region of interest to generate an adjusted NMR spectrum;
randomly determining a number N of multiplets representing a background of a resulting NMR spectrum wherein N is determined by multiplying a multiplet density with a width of the region of interest, with the multiplet density being randomly chosen from a predefined multiplet density range;
repeating, for N iterations:
generating a mathematical model of a multiplet of the number of multiplets;
adjusting the generated mathematical model of the multiplet by applying a random shift in the range of the region of interest to obtain an adjusted mathematical model of the multiplet;
adding the adjusted mathematical model to the region of interest of the adjusted NMR spectrum; and
storing, after iteration N of the N iterations, information about the concentration of the target molecule together with the adjusted NMR spectrum as the resulting NMR spectrum as the data record of the training dataset.