Automated Neuron Tracking Inside Moving and Deforming C. elegans Using Deep Learning and Targeted Augmentation

Recent breakthroughs now enable the imaging of neurons within freely moving animals, presenting an exciting frontier in neuroscience. However, a significant challenge arises in decoding circuit activity from these images, particularly when dealing with organisms like worms, in which the brain undergoes dynamic movements and deformations within a flexible body. Until now, the scientific community has grappled with the lack of tools to effectively tackle this issue.

In response to this challenge, a collaborative team of scientists from École Polytechnique Fédérale de Lausanne (EPFL) and Harvard, has pioneered an innovative AI method to track neurons inside animals with dynamic and deformable brains. The research was led by Sahand Jamal Rahi (EPFL) and Aravi Samuel (Harvard) and spearheaded by Ph.D. students Cori Park (Harvard) and Mahsa Barzegar-Keshteli (EPFL). The results of their study, published in Nature Methods,* showcase a groundbreaking approach based on a convolutional neural network (CNN), a powerful type of artificial intelligence trained to recognize and understand patterns in images.

The intricacy arises when identifying and tracking neurons during a dynamic sequence of an animal's brain activity, demanding extensive manual labeling due to the organism's diverse appearances over time caused by various body deformations. To address this hurdle, the researchers developed an advanced CNN incorporating 'targeted augmentation.'

This novel technique automatically generates reliable annotations from a limited set of manual annotations, allowing the CNN to effectively learn internal brain deformations and create annotations for new postures. This significantly diminishes the need for laborious manual annotation and verification.

The versatility of this new method is evident in its capability to identify neurons represented as individual points or 3D volumes in images. The researchers tested the method on the roundworm C. elegans, a widely used model organism in neuroscience which has 302 neurons. Employing the enhanced CNN, the scientists measured activity in the worm's interneurons, which act as bridges between neurons. The findings unveiled complex behaviors in these interneurons, such as altered response patterns when exposed to different stimuli like periodic bursts of odors.

The researchers have made their CNN accessible, offering a user-friendly graphical interface that integrates targeted augmentation. This comprehensive pipeline spans from manual annotation to final proofreading, providing a significant reduction in the manual effort required for neuron segmentation and tracking. This advance not only streamlines the process but also triples the analysis throughput compared to full manual annotation.

In essence, this pioneering method holds immense potential to expedite research in brain imaging, fostering a deeper understanding of neural circuits and behaviors. It represents a transformative stride toward advancing our knowledge of the intricate workings of the brain.

-------------------------------------
* Park, C.F., M. Barzegar-Keshteli, K. Korchagina, A. Delrocq, V. Susoy, C.L. Jones, A.D.T. Samuel & S.J. Rahi, "Automated neuron tracking inside moving and deforming C. elegans using deep learning and targeted augmentation," Nat Methods (2023). https://doi.org/10.1038/s41592-023-02096-3.