Graphs are ubiquitous data structures that eloquently express complex relationships, while simultaneously capturing underlying patterns that exist within data. These properties make graphs preferred for modeling complex systems, such as the dynamics of mitochondria. Traditionally, analysis of mitochondrial dynamics required manual inspection of microscopy imagery by specialists, which was a tedious and error prone process. These manual and strenuous tasks combined with the dramatic increase in the volumes of microscopy imagery being generated has created a strong demand for automated approaches for modeling subcellular organelles. Yet, modeling mitochondria is far from a trivial task because they do not adhere to any predetermined shape, but rather they are amorphous, diffuse structures. In this work, we provided a novel-graph based methodology for modeling and analyzing the morphology and dynamics of mitochondria. Ultimately, in our pursuit of gaining more insights regarding the dynamics of mitochondria, we have developed temporal and spatial anomaly detection techniques and novel strategies for conducting mitochondrial dynamics classification.
