A machine learning based approach for quantitative evaluation of cell migration in Transwell assays based on deformation characteristics

Many pathological and physiological processes, including embryonic development, immune response and cancer metastasis, involve studies on cell migration, and especially detection methods, for which it is difficult to satisfy the requirements for rapid and quantitative evaluation and analysis. In view of the shortcomings in simultaneously quantifying the number of migrated cells and non-migrated cells using Transwell assays, we propose a novelty approach for the evaluation of cell migration by distinguishing whether the cells have migrated based on the regularity of the cell morphology changes. Traditionally, the status of living cells and dead cells are detected and analyzed by machine learning using some common morphological characteristics, e.g., area and perimeter of the cells. However, the accuracy of detecting whether cells have migrated or not using these common characteristics is not high, and the characteristics are not appropriate for our studies. Therefore, from the point of view of mechanism analysis for the migration behavior, we examined the regularity of different morphology changes of migrated cells and non-migrated cells, and thus discovered the distinguishable morphological characteristics. Then, two deformation characteristics, deformation index and taper index are proposed. Then, a machine learning based algorithm that can identify migrated cells according to the proposed deformation characteristics was devised. In addition, images of migrated cells and non-migrated cells were obtained from the Transwell assays. This algorithm was trained, and was able to successfully identify migrated cells with an accuracy of 84% using the proposed morphological characteristics. This method greatly improves the identification accuracy when compared with the identification of traditional characteristics of which the accuracy was about 54.7%. This machine learning based method might be employed as a potential tool for cell counting and evaluation of cell migration with the aim of reducing time and improving automation compared with the traditional method. This method is effective, rapid, and incorporate advances in artificial intelligence which could be used for adapting the current evaluation methods.