bowang-lab/MedSAM
4,316 stars·589 forks·Jupyter Notebook·Apache-2.0·17 viewswww.nature.com/articles/s41467-024-44824-z↗
MedSAM
MedSAM is a deep learning framework designed for automating the segmentation of anatomical structures in 2D and 3D medical imagery. It provides specialized tools for fine-tuning pretrained segmentation weights on custom medical datasets and evaluating the accuracy of those predictions against ground truth labels.
The project focuses on adapting the Segment Anything Model architecture for medical use, enabling the isolation of specific anatomical structures through prompt-guided methods such as bounding boxes and point prompts.
The system covers a full medical AI workflow, including data engineering for intensity normalization and spatial resampling, as well as model management for checkpoint conversion and weight extraction. It supports training via single or distributed GPU pipelines to process large-scale imaging datasets.
Features
- Medical Image Segmentations - Automates the identification and isolation of specific anatomical structures in 2D and 3D medical scans.
- Image Segmentations - Implements a deep learning framework for partitioning 2D and 3D medical images into anatomical foreground and background pixels.
- Segmentation Model Training - Optimizes model weights using single or distributed GPU clusters to learn complex image patterns in medical scans.
- SAM-Based Implementations - Implements the Segment Anything Model architecture to automate anatomical structure segmentation in medical imagery.
- Medical Imaging Training Frameworks - Offers a PyTorch-based training framework optimized for processing large-scale medical imaging datasets on single or distributed GPUs.
- Prompt-Guided Anatomical Segmentations - Provides prompt-guided segmentation of anatomical structures using bounding boxes and point prompts.
- Segmentation Evaluation Metrics - Computes quantitative performance metrics by comparing medical image predictions against expert ground truth labels.
- Prompt-Based Segmentations - Enables isolation of anatomical structures using bounding boxes and point prompts to guide the segmentation model.
- Model Fine-Tuning - Implements a training pipeline for optimizing pretrained segmentation weights on custom medical datasets.
- Vision Model Fine-Tuning - Optimizes a Segment Anything Model on custom medical datasets to improve segmentation accuracy for specific anatomy.
- Segmentation Model Testing - Quantifies segmentation accuracy by comparing model predictions against expert ground truth labels using standard metrics.
- Distributed GPU Training - Ships a distributed GPU training pipeline to synchronize gradients across multiple cards for large-scale medical datasets.
- Weight Remapping - Implements checkpoint weight transformation to ensure seamless compatibility between training and inference environments.
- Medical Imaging Preprocessing Tools - Transforms various medical imaging file types into unified data structures for downstream preprocessing.
- Model Checkpoint Converters - Includes utilities to transform trained model weights into compatible formats for streamlined deployment and inference.
- Weight Extraction - Provides tools for extracting and standardizing model weights from checkpoints to ensure correct loading during inference.
- Medical Segmentation Metrics - Computes accuracy metrics by comparing medical image segmentation predictions against expert ground truth labels.
- Medical Dataset Formatting - Converts raw medical images and labels into optimized binary formats for efficient supervised learning.
- Medical Image Normalization - Normalizes raw medical imaging files by adjusting window levels and resampling dimensions for ML pipelines.
- Image Dimension Standardizations - Standardizes voxel size and image dimensions across diverse medical datasets for consistent model processing.
- DICOM Windowing Adjusters - Adjusts window levels and width to normalize contrast and brightness of medical images across different scanners.
Star history
How this analysis was created: This summary and feature list are AI-generated from collected project material and can contain mistakes. Stars, license and language are imported from GitHub. Inclusion does not mean that we have tested or audited this project. Check the source documentation for any feature you depend on. Learn more on our About page.
