Ultrastructural Analysis of Mitochondria and Neurons in the Ventral Nerve Cord

Built transformer-based instance segmentation pipelines for large-scale SEM volumes to analyze neuronal and mitochondrial structures.

← Home

Transformer-based segmentation enabled large-scale 3D mapping of mitochondria within Drosophila motor neurons.

To address the challenge of analyzing mitochondrial morphology and distribution across the vast and complex architecture of neurons, I developed automated segmentation pipelines for large-scale SBF-SEM datasets. This figure demonstrates our ability to reconstruct an entire Drosophila motor neuron (blue) along with every individual mitochondrion (various colors), capturing both the intricate neuronal arborization and the diverse mitochondrial populations within it. This advance allows our lab to systematically quantify mitochondrial organization across neuron types at high resolution, providing a platform for investigating how neuronal energy demands are met at the subcellular level.

Dorsal Midline Neurons Example Segmentation

Sample of dorsal midline neurons

As we continue to produce segmentations, our cohort of citizen scientists are helping us build a consensus dataset of polished segmentations. This work allows us to redistribute segments with high variance to attain a stronger signal.

Contributions

Led analysis of large-scale 3D SEM imaging data using transformer models to understand energy demands of cell types of interest.
Built an instance segmentation pipeline for Drosophila VNC neurons and mitochondria.
Adapted Flood Filling Networks with vision transformers (SAM2) to reconstruct branching and mitochondrial networks.
Coordinated a crowdsourcing initiative engaging thousands to generate consensus training datasets.
Analyzed mitochondrial distribution and morphology across motor neurons.