PatchPerPixMatch

PatchPerPixMatch performs automated 3D searches to identify neuronal morphologies in multi-color flip-out (MCFO) light microscopy datasets from Drosophila Gen1 GAL4 lines for matching segmentation fragments to target neurons.

Key Features:

• Automated 3D Search Capability: Performs full 3D searches within MCFO acquisitions of Gen1 GAL4 expression patterns instead of relying on 2D projection images.
• Instance Segmentation and Morphology Matching: Automates instance segmentation of MCFO imagery and searches for target neuron morphologies by minimizing an objective function that covers the target with well-fitting segmentation fragments.
• Computational Efficiency and Robustness: Executes the 3D search with computational efficiency and demonstrates robustness to inaccuracies in automated neuron instance segmentation.
• Intersectional Labeling Approach: Leverages intersectional labeling strategies such as the split-GAL4 system to exploit sparse and specific neuronal labeling for identification.

Scientific Applications:

• Large-scale neuron identification: Enables automated identification of specific neuron morphologies across large Drosophila MCFO datasets, including approximately 30,000 morphologies from ~20,000 MCFO acquisitions and ~3,500 Gen1 GAL4 lines.
• Neural circuit mapping: Facilitates mapping and comparison of identified morphologies to resources such as the Drosophila hemibrain for neurocircuitry analysis.

Methodology:

PatchPerPixMatch first performs automated instance segmentation on MCFO imagery and then searches for target morphologies by optimizing an objective function that selects segmentation fragments to cover the target morphology.