Detection of descending neurons across animals in fluorescence microscopy data

Abstract

Neurons that descend from the brain to the spinal cord or other motor centers are important for controlling movement and behavior. Detecting and mapping these neurons can provide valuable insights into the neural circuits involved in motor control and decision-making. However, the process of detecting these neurons can be challenging, particularly when trying to generalize across different animal models. The proposed approach utilized auxiliary learning to train a model to learn shared representation across different animals, followed by test time adaptation to fine-tune the model for accurate neuron detection in a new animal. The auxiliary learning involves training a detection task as auxiliary task to assist the primary segmentation task. The project evaluated the proposed approach on fluorescence microscopy data from Drosophila. The results demonstrated the effectiveness of the proposed approach in detecting descending neurons with high accuracy, outperforming baselines with the same base model but without auxiliary learning architecture. Moreover, the proposed approach is shown to be robust to different animals and has higher generalizability. This project has potential applications in neuroscience research for understanding the functional connectivity of descending neurons in different animal models. The proposed approach can also be extended to other imaging modalities and neuroscience applications that require cross-animal generalization.