| Abstract [eng] |
The dissertation presents a three-dimensional model of a phospholipid membrane, describing the simulation of electrochemical impedance spectroscopy (EIS) measurements in the presence of arbitrarily distributed membrane defects. Using the finite element method (FEM), cases with various membrane defect sizes, densities, and scattering types are simulated. The influence of membrane damage on its electrochemical properties is studied by examining the admittance phase curves (EIS spectra). Machine learning methods are used to build models for estimating quantitative membrane properties from EIS data. The defect clustering phenomenon and its influence on the EIS spectra of membranes is investigated, several theoretical clustering models and quantitative clustering evaluation metrics are examined. Regression models are used to evaluate clustering properties from EIS spectra, and also to investigate the compatibility of clustering models with real defect locations observed in atomic force microscopy (AFM) images of membranes. The task of automatic detection of membrane defects in microscopy images is also considered. Several different algorithms are proposed and tested with real AFM images, and the impact of defect detection accuracy on simulated EIS spectra is studied. The modeling results are validated using experimentally obtained microscopy and spectroscopy data. |
