DeepBrainIPP

DeepBrainIPP automates morphological quantification of mouse brain structures from T2-weighted magnetic resonance imaging (MRI) using deep learning.

Key Features:

• Automated Skull-Stripping: Deep learning-based skull-stripping isolates brain volumes from T2-weighted MRI images to improve segmentation input quality.
• Data Augmentation Strategy: An augmentation protocol expands an in-house dataset from 172 images to 4,040 samples for robust model training.
• Segmentation of Large Brain Regions: Atlas/template registration is applied after skull-stripping to segment large brain regions.
• Segmentation of Small Structures: A dedicated model directly segments small structures such as the paraflocculus in the cerebellum, outperforming registration-based methods.
• Performance Metrics: Reported average Dice scores are 0.96 for skull-stripping and 0.89 for small-structure segmentation, with residual volume errors of 2.18% and 5.32%, respectively.

Scientific Applications:

• Neurobiological Research: Quantitative morphological analysis of mouse brain structures for neurobiological studies.
• Brain Development Studies: Analysis of morphological changes relevant to brain development in mouse models.
• Disease Modeling: Morphological phenotyping to support studies of neurological disease models in mice.
• Genetic Phenotyping: Detection and quantification of anatomical phenotypes in genetically modified mouse lines.
• In vivo and Ex vivo Imaging: Application to both in vivo and ex vivo MRI samples.

Methodology:

Integrates deep learning models with traditional image processing, including deep learning-based skull-stripping, atlas/template registration for large-region segmentation, and a dedicated deep-learning model for small-structure segmentation.