[3.10-AP] Implement AI-Driven Propagation and Auto-Adjustment of Segments Across Slices

[sedghi]

Purpose
This feature adds an AI-driven tool to automatically propagate and adjust segmentations across sequential image slices. By utilizing encoder-decoder AI models like Meta’s Segment Anything Model (SAM), this tool enables efficient slice-by-slice segmentation, automatically adapting segments to match anatomical variations. This functionality is implemented on the client side using WebGPU and supports multiple ONNX-compatible models, making it extensible for custom AI-driven segmentation workflows.

Why This Matters
Manual segmentation across multiple slices can be tedious and time-consuming, especially in 3D imaging. This AI-powered tool streamlines the process, reducing manual effort and improving segmentation accuracy by dynamically adjusting segments as they propagate. This is particularly valuable for clinicians and researchers working with complex anatomical structures that vary across slices, allowing for faster and more consistent segmentation.

Key Changes

• AI-Driven Slice Propagation: Automatically extends and adjusts segments to subsequent slices, improving efficiency in multi-slice segmentation.
• Client-Side AI with WebGPU: Runs AI models directly on the client side using WebGPU, enabling high-performance, real-time segmentation without server dependency.
• Extensible Model Support: Supports various ONNX-compatible encoder-decoder models, allowing users to integrate custom AI models beyond Segment Anything.

Impact on Users and Developers

• For Users: The AI-driven propagation tool makes segmentation workflows faster and more accurate, as segments are adjusted automatically to align with anatomical changes across slices.
• For Developers: This feature offers an extensible framework for integrating ONNX-compatible AI models, enabling custom segmentation solutions and simplifying the deployment of AI on the client side.