Cell Tracking Dashboard
Github page

Cell Tracking Dasbhoard

Motivation

In live-cell imaging, analyzing the movements of cells is crucial for many applicative scenarios such as biochemistry, bioinformatics, cell biology, and genetics. Since following cells by hand is extremely tedious and time consuming, we can design algorithms to do that for us. However, no algorithm is perfect which introduces the additional challenge to evaluate an algorithm accuracy to understand which algorithm we can trust. For this purpose we are developing the Cell Tracking Visualization Dashboard, a web tool simplifying the analysis of cell tracking algorithm by single out and visualize error links from the predicted data compare to ground truth data.

Design

There are three main views allowing the user to analyze cell tracking algorithms' performance at different level of granularity

Overview
The first objective is providing an overview of the algorithms' results. Through the Overview dashboard users can identify problematic Fields of Views (FOV), overall algorithms' performance, and the best perfoming algorithm for each FOV. To this end, the first view focuses on displaying number of error links produced by different algorithm for each FOV. FOVs are ordered by sum of number of error links.
Single Algorithm
The second objective is providing a detailed overview of a single algorithm's performance. Through the Single Algorithm dashboard a user can study each single error link occured in eac FOV of the dataset. Each FOV is represented by a card in the main view. Linking errors are explicitly represented in the domain space of the images. Each point indicates the location at which the error is occurring (i.e., where the predicted track agrees with the ground truth track). The line indicates the error link committed (i.e., the wrong connection between two points belonging to two different ground truth tracks). Each card shows additional statistics regarding the FOV such us: the FOV index, the number of linking errors, the percentage of linking errors (i.e., number of errors / total number of linking * 100), the total number of cells in the FOV, and the total number of links. The last two values in particular provide a good indication of how challenging a FOV is since the more dense a FOV is (i.e., the more cells are imaged in the same FOV) the more likely linking errors are to occur.
Single FOV
The third dasbhoard allows us to evaluate the detailed algorithm's performance on a single FOV. This view is organized into two components. The first component is a lineage forest representing all occurrences of a mitosis event. Ground truth data is used to generate the geometry of each lineage tree. The leftmost point of a lineage tree indicates the time at which the cell enters the FOV. The rightmost point of a lineage tree indicates the time at which the cell exists the FOV. Each branch in the tree represents a mitosis event. Each branch is colored in black if it contains no linking errors. Otherwise, a linear color scale used used to encode the number of linking errors committed on each ground truth cell track. This way, a user can easily identify problematic cell tracks. The second component allows the user to inspect specific cell tracks. This component displays all cell tracks belonging to the FOV under study. The main purpose of this view is to allow the user to investigate each single linking error and the causes that may have originated the error. To this end, the view also displays the original image set providing explicit representation of the cells from which tracks have been computed. By means of a slider on the top of the cell track view the user can change the time frame thus changing the image displayed. While moving the slider, error links are automatically filtered in order to show only error links that appeared at a time frame less or equal than the one represented in the image.