SAM-med3d features

Description

SAM-med3d features are derived from segment anything model - medical 3D (SAM-med3d), a foundation model for 3D medical image segmentation. doi: https://doi.org/10.1109/tnnls.2025.3586694

Purpose

SAM-med3d features provide:

Deep learning-based segmentation masks
Pre-trained feature representations
Segmentation confidence metrics
Alternative segmentation approaches

Currently we extract 384 CLS token features from the SAM-med3d model. these are black-box features learned by the model during training on large-scale medical image datasets.

Current status

This feature category stores segmentation outputs from SAM-med3d:

3D segmentation masks
Probabilistic segmentation maps
Segmentation confidence scores

Future development

Planned extensions include:

Fine-tuned SAM-med3d models for organoid data
Feature extraction from learned representations
Comparison with traditional segmentation methods
Integration with other foundation models