Physiological Characteristics and Uptake Patterns of Microparticles for Predictive Cancer Diagnosis Using Machine Learning

Early and accurate diagnosis of cancer is critical for the success rate of treatment and can be the difference between life and death. However, the current conventional cancer diagnostic methods are limited. Imaging-based methods for example, such as morphological analysis of tissues (histopathology) or cells (cytology) require a visible change to the tissue, which by then may be too late. Thus, there is a need for new, faster, and more flexible approaches to aid early diagnosis and help physicians identify the optimal treatment protocols for each patient based on their individual condition and cancer stage. To meet this need, computational tools such as image processing, machine and deep learning have been used in the last decade and are dramatically changing the field. Here, we show how mechanical and physiological parameters of cancer cells combined with machine learnings models can result in high accuracy for both needs. Cancer cells were found to use phagocytic-like uptake very extensively, in correlation with cell deformability and malignancy. The mechanical deformation of cells results from the combination of several parameters, such as the membrane elasticity and fluidity, cytoskeleton rigidity and dynamical rebuilding of the cytoskeleton. By analyzing patterns of microparticles uptake by cancer cells and introducing them to machine learning classifier models we were able to classify with high accuracy the source of the subpopulations, distinguishing between malignancy levels and drug resistance. Our findings present a new approach that has high potential for a simple, fast and accurate method for the early diagnosis of cancer.