Short-term memory, also referred to as the working memory, is the transient memory of task-related information in competitive environments 18. Short-term and long-term memory are evaluated after a maze test. Food is placed at the ends of some of the arms, and animals’ spatial memory performance can be quantified as quantities such as the number of times each arm is accessed, the amount of time spend in each arm, the number of times that baited and non-baited arms are accessed, and the sequence of accessed arms 6. Therefore, a RAM is used in this work to explore the behavioral characteristics of rodents with TBI. TBI and Alzheimer’s disease share many common pathological features, such as deposits of β-amyloid protein (Aβ) in the brain, neurofibrillary tangles, neuronal synapse loss and cognitive impairment 17. Approximately 10 million people suffer from TBI each year 16. TBI, also referred to as head injury or intracranial injury, is an epidemic that arises from brain damage caused by an external force. Many studies have shown that traumatic brain injury (TBI) is a high risk factor for Alzheimer’s disease 15. RAM assessment methods have also been successfully implemented for humans 14. For example, animals’ spatial memory is improved after experiencing enriched environments 12 or exercise stimulation 13. RAMs reveal features that can be used to identify Alzheimer’s disease 10 or frailty 11 associated with spatial learning impairment in animals, and exhibit various advantages when assessing spatial learning improvement and progress. Among behavioral tests, radial-arm mazes (RAMs) have been widely used to quantify the spatial learning situations of healthy and unhealthy animals 6, 7, 8, 9 due to their convenience and high reproducibility.
Devices and methods for assessing learning and memory functions have been extensively used since their development in the early twentieth century 5. In the field of epidemiology and molecular genetics, rodents have long been used to investigate basic mechanisms of spatial learning and memory 1, 2, 3, 4. However, it took the YOLO counterpart an hour or so to locate a rat contained in the frames, which highlights the contribution of this work.
In other words, this work slightly outperformed the YOLO counterpart in terms of the accuracy in both cases, which indicates the robustness of this work. The YOLO counterpart gave an accuracy beyond 97% in the absence and in the presence of the intruder. For comparison purposes, the same experiments were carried out using a pre-trained YOLO v2 model. The maze was intruded by an experimenter to further test the robustness, and the accuracy slightly fell to 98%. Due to simple operations and a high robustness against the noise spots formed by the droppings of the rat, it took just minutes to process more than 9000 frames, and an accuracy above 99% was reached as well. The IR images were binarized and then duplicated for subsequent intersection and opening operations.
An infrared (IR) night-vision camera was hung above the maze to capture IR images of the rat. This paper aims to develop a position tracking algorithm by which a rat in a radial arm maze can be accurately located in real time.