Application of Machine Learning Methods for Processing and Analyzing Data from Probe and Optical Microscopy of Biological Objects

Abstract

This review systematizes modern machine learning (ML) methods for solving key image processing tasks in optical microscopy (OM) and atomic force microscopy (AFM). Traditional microscopy data analysis methods face problems of subjectivity, low throughput, and difficulty in quantifying multimodal information. The review discusses the use of ML for the following main tasks: segmentation of objects on OM and AFM images of biological structures; classification of samples based on morphological features; regression analysis for predicting the properties of biological objects; multiparametric data analysis combining information from various visualization channels (height, adhesion, etc.). Special attention is paid to the capabilities of ML to identify complex correlations between surface topography and functional characteristics, which is difficult in traditional analysis. The review also discusses key issues, including the need to create specialized datasets, the problem of interpretability of models, and issues of reproducibility of results.