notes.wiki

Drawing the map of neuronal circuits at microscopic resolution is important to explain how brain works.

The first step it's to find de neurons' soma position in a reasonable time using a reasonable amount of computational resources. NeuroGPS is a strong algorithm that can find neurons across different brain areas and it's demonstrated its high robustness to the shape, size and spatial distribution of neurons.

In this work an heterogeneous dataset it's provided to test the algorithm using the F1 metric. It's find that using a precise pair of values treshold - minimum radius in NeuroGPS it's possibile to obtain an high F1-score that means an high precision in locating neurons. On the other side, it is also important to instruct the algorithm to find on its own what treshold and minimum radius parameters it has to use for identifying neurons in new datasets of 3D images. The LOSO function it's a kind of idea.

NeuroGPS works well with a tuning approach, but it requires too much time of computation: the conclusion is that it's impossibile to use NeuroGPS for a complete scanning of an human brain, or to do a large scale work in these terms.

The aim it's to improve algorithms that can automatically finds cell in 3D images stacks and get better results to help research.