Unsupervised Deep Learning with Stacked Autoencoders for Cell Line Classification in Microscopic Images

This paper focuses on classifying cell line images using a deep neural network based on stacked autoencoders. The cells are classified into three types: MSC (Mesenchymal Stem Cells), TAF (Tumor-Associated Fibroblasts), and SK-BR-3 (Human Breast Cancer Cells). The implemented methods are: the deep neural network extracts features from the images using its architecture, which consists of two stacked autoencoders with hidden layer sizes of 200 and 100, and a SoftMax layer is used as the final layer to classify the cells. The results achieved with this method are an average accuracy of 96.7% for classifying MSC, TAF, and SK-BR-3 cells. In conclusion, this approach avoids the need for image segmentation, which can be difficult for certain images, that need labeling, and the accuracy of the method depends only on the hidden layer sizes and the number of autoencoders used.