DDeep3M

DDeep3M performs deep learning–based segmentation of biomedical images to extract structures such as vessels, somata, and brain tumors across electron microscopy, 3D optical microscopy, and MRI modalities.

Key Features:

• Docker-Based Deployment: Packaged in a Docker container to encapsulate the computational environment.
• Multi-Scale Image Segmentation: Validated on microscale electron microscopy volumes, mesoscale 3D optical microscopy image stacks of mouse brains (recall/precision and Dice indexes > 0.96 for vessels and somata), and macroscale MRI for brain tumor segmentation validated with the BraTS dataset.
• Performance Metrics: Reports high accuracy with recall/precision scores and Dice indexes exceeding 0.96 for mesoscale vessel and somata segmentation.
• Architecture: Builds upon the CDeep3M neural network architecture.
• Benchmarking: Compared against three existing models across datasets spanning micro- to macro-scales.

Scientific Applications:

• Brain Tumor Segmentation: Application to MRI brain tumor segmentation validated using the BraTS dataset.
• Neuroanatomical Studies: Segmentation of neural structures in mouse brain 3D optical microscopy image stacks, including vessels and somata.
• Electron Microscopy Segmentation: Segmentation of structures in microscale electron microscopy data volumes.

Methodology:

Builds on the CDeep3M architecture, is packaged in Docker, and was compared against three existing models across electron microscopy, 3D optical microscopy of mouse brains, and BraTS MRI datasets.