4.7. Light/Dark-Box Behavioral Assay

At the ages of 0.5, 1, 1.5, 2, and 3 months (for Lrat^−/− and Lrat^+/+ animals) and 5 months (for Lrat ^+/− and Lrat^+/+), a vision-based behavioral assay was performed. A customized light/dark-box was used with dimensions of 100 × 50 × 40 cm (length × width × height). Half of the box was darkened. The box was placed in the same position in the room during every measurement to prevent possible light/shade interference. The animals were placed in the light area of the box and filmed for twenty minutes. Deep learning was used to extract key features. In short, a Faster Recursive Convolutional Neural Networks (Faster R-CNN) was used to locate and track the rat’s head. The Faster R-CNN was developed using the resnet18 architecture and trained on 658 randomly sampled, annotated video frames. After the detector was trained, it was deployed on each video. A transition zone, dark zone, and light zone were determined. The transition zone was defined as a circle centered at the doorway base with a radius of 1.25 times the doorway’s width. The rat was tracked in the light/dark box, and per frame, the rat’s location was tracked. After tracking the rat’s heads in each video, the data were processed using Matlab. A random subset of videos was selected to manually extract all parameters and compare the data to the values extracted via the Faster R-CNN-based algorithm. No significant differences were found between the manually extracted and automatically extracted parameters, confirming the automatic analysis’s robustness.