Brainchop: Providing an Edge Ecosystem for Deployment of Neuroimaging Artificial Intelligence Models

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Neuroscience

Background:

• Deep learning models show promise in medical imaging.
• Lack of efficient distribution platforms limits model sharing.
• Existing solutions often rely on cloud-based processing, raising privacy concerns.

Purpose of the Study:

• To introduce brainchop, a web application for applying deep learning models to neuroimaging data.
• To provide a platform for efficient and private local processing of medical images.
• To facilitate the distribution and utilization of AI models for various neuroimaging tasks.

Main Methods:

• Developed a fully functional web application using pure JavaScript.
• Integrated optimized helper functions for volume conforming and component filtering.
• Leveraged end-user graphics cards for rapid model inference.
• Included integrated visualization tools for validation, annotation, and editing.

Main Results:

• Brainchop enables rapid (seconds) brain extraction, tissue segmentation, and regional parcellation.
• The application ensures data privacy by processing on the user's local machine.
• Facilitates distribution of models for tasks like registration and abnormal tissue identification (tumors, lesions).

Conclusions:

• Brainchop offers an efficient, browser-based solution for applying deep learning to neuroimaging.
• The platform enhances accessibility and privacy for AI-driven medical image analysis.
• Encourages broader adoption and development of AI models in neuroscience research through its open-source nature.