Classification of healthy and cancerous cells using optical rheology and machine learning

The optical stretcher (OS) probes single suspended cells with forces around 100 pN. It has been shown, that in the physiological regime of the OS carcinoma cells tend to be softer than their healthy counterparts. Yet, a clear characterization based on rheological data could not be achieved, even though its correlation with cancerous traits is strongly suggested.
We use the high throughput of the OS to perform machine learning based discrimination of individual cells with a breast cancer model. We performed around 30,000 single cell stretching experiments on epithelial cells (MCF-10A), carcinoma cells (MDA-MB 436) and fibroblasts (NIH 3T3), and extracted 33 morphological and rheological parameters to use them for classification with random forest and support vector machine algorithms. Our approach allows us to distinguish among the three cell types with an average sensitivity of 71%. These intial results do not only elevate the importance of the mechanics of cells during tumor progression, but also promise live cell sorting for scientific and medical applications.