### Description

NetMets provides a metric for comparing the spatial geometry and connectivity of two interconnected graphs, like those produced using neuron and vascular/microvascular centerline segmentation. Networks can be provided in either SWC or OBJ format and can have interconnected components (such as cycles).

NetMets also provides tools for visualizing differences between graphs, including colormaps indicating differences between geometric features and edge mapping to indicate differences in connectivity.

### Usage

NetMets requires two files describing the graphs to be compared, along with a ** sigma** value describing the metric tolerance:

`>> NetMets -s SIGMA [GRAPH_A] [GRAPH_B]`

The default output describes, in order, the False Positive Rate (FPR) and False Negative Rate (FNR) for the geometry and connectivity:

` gFPR gFNR cFPR cFNR`

` 0.1083 0.1850 0.3220 0.2131`

The resulting metrics can be visualized using the `-g` flag, and a list of all options is requested using:

`>> NetMets -h`

### Related Work

Alternative metrics include the DIADEM Metric, which can be applied to tree-like structures (like individual neurons), and Path2Path, a proposed algorithm to measure the similarity between the geometric shape of two neurons.

### Future Directions

*Please contact me if you are interested in contributing.*

I am currently happy with the geometry metric. It is fast to compute, intuitive to visualize, and accurately captures structures of interest to biologists.

The connectivity metric has the potential to become time-consuming for highly-connected networks. For example, when a segmentation contains clusters of small false-positive nodes, evaluating connections across these clusters can become impractical. I believe this can be fixed by further using the geometry metric to improve edge mapping, thereby limiting the number of potential sister edges that can map to any given edge.

### Acknowledgements

This software was produced in collaboration with the *Farsight Project* as well as Chris Bjornsson and Jonathan Taylor. I would also like to thank James Burck from the Open Connectome Project for making several important changes to the code.