Automated neuron tracking inside moving and deforming C. elegans using deep learning and targeted augmentation

Area of Science:

• Neuroscience
• Computational Biology
• Machine Learning

Background:

• Accurate neuronal activity readout from 3D functional imaging is crucial for understanding brain function.
• Segmenting and tracking neurons in 3D is challenging in behaving animals due to brain movement and deformation.
• Traditional methods rely on extensive manual annotation of neural data, which is labor-intensive.

Purpose of the Study:

• To develop an automated method for synthesizing neural annotations to reduce manual effort.
• To improve the efficiency of analyzing neuronal activity in 3D functional imaging.
• To enable the study of neuronal dynamics in freely moving animals.

Main Methods:

• Introduced 'targeted augmentation' to automatically generate artificial annotations from a few manual ones.
• Developed 'Targettrack' which learns brain deformations to synthesize annotations for new postures.
• Created a graphical user interface for end-to-end application of the method.

Main Results:

• Significantly reduced the need for manual annotation and proofreading of 3D neural data.
• Successfully applied Targettrack to recordings with neurons labeled as key points or 3D volumes.
• Uncovered rich patterns in interneuron dynamics, including switching neuronal entrainment, in freely moving animals exposed to odor pulses.

Conclusions:

• Targettrack offers an efficient solution for annotating 3D functional imaging data, particularly in dynamic biological systems.
• The method facilitates the analysis of neuronal activity in behaving animals, providing new insights into neural dynamics.
• Automated annotation synthesis is a promising approach for advancing neuroscience research.