TrailMap

TRAILMAP is a specialized software package designed to extract axonal data from cleared brain samples using a 3D convolutional neural network. It was originally developed for analyzing iDISCO-cleared samples imaged via lightsheet microscopy but is versatile enough to identify various filamentous structures in 3D volumes. The tool supports both inference with a pre-trained model and transfer learning for custom datasets, making it adaptable to diverse neuroimaging research needs.

Key Features

• Axon Segmentation — Identifies axonal projections in 3D brain imaging data using a deep learning model.
• Transfer Learning Support — Allows training on custom annotated datasets to adapt to new data types or structures.
• 3D Volume Processing — Handles entire brain volumes by processing them in manageable chunks to accommodate hardware constraints.
• Data Preparation Utilities — Includes scripts for labeling, edge detection, and generating training/validation sets from annotated volumes.
• GPU Acceleration — Leverages CUDA and cuDNN for efficient model inference and training on Nvidia GPUs.

Philosophy

TRAILMAP emphasizes accessibility and practicality, providing detailed guidance for users ranging from machine learning novices to experienced researchers, with a focus on enabling custom model adaptation through transfer learning.