電気インピーダンス計測と畳み込みニューラルネットワークを併用したイースト細胞の死細胞率と体積濃度の可視・定量化

抄録

A neural network using CNN was constructed to visualize the dead cell rate and volume concentration of yeast cells, and accuracy verification was performed using test data by simulation. Regarding the construction of CNN, the normalized imaginary impedance Ψ was used as the input. When actually learning, it was observed that the model loss converged around 1000 learning times. In addition, when the dead cell rate and volume concentration were visualized using test data, it was confirmed that the accuracy of 1 or 2 cell regions was high, but that of 3 or 4 cells was low. In addition, when the accuracy is quantitatively verified, the image error shows a value as close to 0 as possible in the most accurate data, and the highest image error is IQ = 0.064 [-] for the dead cell rate and IQ = 0.012 [-] for the volume concentration. From this result, it was found that this method is useful when the number of cell regions is small, but there is room for improvement when the number of cell regions is large.