Deformability Cytometry Clustering with Variational Autoencoders
Seith, D. D.; Combs, C. T.; Siwy, Z. S.
Show abstract
Mechanobiology has shown great success in revealing complex cellular dynamics in various pathologies and physiologies. Most methods for assessing a cells mechanical properties, however, generally extract only a few physical constants such as Youngs modulus. This can limit the potential for accurate classification given the wide variety of rheological properties of cells, there are many ways for cells to differ. While it was recently shown that deep learning can classify cells more accurately than traditional approaches, it is not clear how this may be extended to unsupervised classification. In this work, we showcase the potential for a deep learning model to classify cells in an unsupervised fashion using a blend of physical properties. We introduce the combination of a variational autoencoder and a previously described clustering loss for classifying cells in an unsupervised fashion.
Matching journals
The top 8 journals account for 50% of the predicted probability mass.