Morphologically Based Cell Classification in Mixed Cultures / 医用生物力学

ABSTRACT

Objective To make quantitative analysis on collected cell images combined with machine learning integrated clustering algorithm, so as to explore a method for fast recognition and classification of cells in mixed cultures based on morphology. Methods The morphometric properties of A549 and 3T3 cells in vitro were characterized by immunostaining, the fluorescent images were then analyzed with CellProfiler to extract the parameters of cell morphology. The parameters were loaded into CellProfiler Analyst to be trained with machine learning algorithm, and a rule was developed to form a generalization capability for cell classification in mixed cultures. Results The accuracy of the training classifier was 81-24%, and the binary classifications of A549 and 3T3 cells could be realized. Conclusions The method of machine learning is very effective in parameter clustering. The application of machine learning into cell image recognition can provide pre-judgment for rapid pathological examination of tissue sections, thereby reducing the workload of doctors and improving the accuracy of diagnosis.