Robust examination of cell apoptosis timing in presence of noisy environment

The process of programmable cell death, i.e., apoptosis, physiologically occurs during development and aging and as a homeostatic mechanism to maintain cell populations in tissues. Apoptosis also happens as a defense mechanism in immune reactions or when cells are damaged by disease or external stimuli (drugs). Due to its complexity and the fact that apoptosis fate resolves in a very short time (a few hours in general), apoptosis mechanisms have been extensively studied only recently with the advent of advanced time-lapse microscopy. Timing related to apoptosis stages is strongly correlated to many factors including cell type, drug dose, cell microenvironment, and related cross-talks whose knowledge is too little to predict apoptosis duration. Such times are of fundamental importance since they linked with drug efficacy, immunotherapy treatment, cancer-immune interaction effectiveness. In light of this, by exploiting video analysis, deep learning algorithms, and multiple linear regression, we presented a platform to examine the apoptosis and blebbing times with very high accuracy and precision levels. More in detail, we artificially generated, through a computer vision analysis platform, synthetic apoptosis videos with randomly variated apoptosis timing profiles. By using a pre-trained Convolutional Neural Network (CNN) architecture within the so-called transfer learning procedure, we encoded each frame of the video into a list of numerical descriptors. Automatic examination of apoptosis timing profiles was then accomplished by training a multivariate linear regression (MLR) model. An extended version of the work will present further advancement of this research by considering real videos of dying cells and additional confounding effects.