Event Indicator Function classifier for identifying cell tracking errors and phenotypes

Abstract

Biologists estimate parameters such as division time, death time, time to differentiate into specialized cell in order to model cell behavior and to develop novel ways to fight diseases such as Cancer, HIV and others. One of the critical steps of such analysis of cells in video microscopy is to follow each of the cells through their generations and collect relevant information. Variability of cell density and dynamics in different video, hamper portability of existing automated cell tracking systems across videos. These errors have to be identified and corrected using human assistance to achieve tracker portability across videos. In this paper, we propose Event Indicator Function (EIF) classifier to predict the tracking errors and cell phenotypes (division and death) frame-by-frame using a set of features (metrics) that are collected during tracking. EIF classifier models the metrics using empirical thresholds to identify the errors and phenotypes. Finally, EIF classifier performance has been evaluated on variety of microscopic videos that differ both in cell density and dynamics, illustrated results show the significance of the proposed classifier.