Automatic Identification of Fusion Events in TIRF Microscopy Image Sequences

Abstract

This paper presents a novel computer vision system for automated identification of vesicle-plasma membrane fusion events in image sequences obtained from Total Internal Reflection Fluorescence (TIRF) microscopes. Identification of such events is important in order to better understand the process of exocytosis in cells. Manual analysis of thousands of images is painstakingly slow and subjective since the events are hard to recognize, even for experts. The proposed identification method assembles an image sequence in a 3D stack and extracts connected regions representing candidate events. Each candidate is then described by a set of novel domain specific descriptors. Similarity scores between genuine fusion events and fusion candidates are calculated in the PCA (Principal component analysis) eigenspace. The system's performance was evaluated on large TIRF movies as well as on simulated data. The results illustrate the ability of the proposed algorithm to find the majority of fusion events while maintaining a low false positive rate. To our knowledge this paper is the first to report a detailed analysis of fully automatic annotation of large TIRF image sequences.