Segmentation - RodentDataAnalytics/mwm-ml-gen GitHub Wiki

After a project has been created or loaded the user is able to proceed to the Segmentation process.

Contents

1. Segmentation Overview
2. The Segmentation Panel

Segmentation Overview

As described in the publication of Gehring, Tiago V., et al., compared to previous approaches that attempted to categorise the behaviour of animals by using the whole swimming paths, this approach first splits the animal trajectory inside the Morris Water Maze into segments and then tries to classify each one of them. In that way the difficulty of classifying long animal trajectories exhibiting many different kinds of behaviour into one strategy is overcome.

The segmentation process works as follows: each trajectory is divided into N segments of length d. Each segment overlaps with previous segments (by an h percentage) to reduce the classification variance due to unfavourable segmentations.

As rule of the thumb the the segmentation length should roughly be between the bounds 2xR and 2.7xR, where R equals the arena radius (in cm). Segments with length less than 2xR are difficult to be categorised to a class because they are very small, while segments with length more than 2.7xR are possible to fall under multiple classes. Segmentation overlap is advised to be between 70% to 90%.

The Segmentation Panel

• Segmentation Length: Specifies the length d of the segments (small variations may apply due to the the discrete nature and spacing of available data points).

• Segment Overlap: Specifies the percentage of overlap (h) between the segments. h needs to be positive and less than 1 (for example 0.7 equals to 70% overlap). This field accepts comma-separated values in order to run the segmentation process more than once.

• Segmentation (Button): If the above information is provided correctly then the program reads the animal trajectories and segments them. Moreover, it computes eight features of the trajectories and of the segments. When the process is finished all the values are stored inside the project folder (refer to the section The Project Folder for more information of how the various files are arranged. The Default Segmentation will list all the segmentations of this project. In case a segmentation is not shown press the Refresh button. The default segmentation specifies with which segmentation the user wants to proceed and its name specifies the defined length and overlap (for example segmentation_configs_250_09.mat means that the trajectories were segmenting using segment length of 250cm and overlap 90%).

Note: A dummy segmentation will be present as soon as a new project has been created with 0 segmentation length and overlap. This segmentation refers to the whole swimming paths of the animals.