RESUMEN
The detection of spontaneous magnetic signals can be used for the non-invasive electrophysiological evaluation of induced pluripotent stem cell-derived cardiomyocytes (iPS-CMs). We report that deep learning with a dataset that combines magnetic signals estimated using numerical simulation and actual noise data is effective in the detection of weak biomagnetic signals. To verify the feasibility of this method, we measured artificially generated magnetic signals that mimic cellular magnetic fields using a superconducting quantum interference device and attempted peak detection using a long short-term memory network. We correctly detected 80.0% of the peaks and the method achieved superior detection performance compared with conventional methods. Next, we attempted peak detection for magnetic signals measured from mouse iPS-CMs. The number of detected peaks was consistent with the spontaneous beats counted using microscopic observation and the average peak waveform achieved good similarity with the prediction. We also observed the synchronization of peak positions between simultaneously measured field potentials and magnetic signals. Furthermore, the magnetic measurements of cell samples treated with isoproterenol showed potential for the detection of chronotropic effects. These results suggest that the proposed method is effective and has potential application in the safety assessment of regenerative medicine and drug screening.