MedicAI

The medicai is a Keras based library designed for medical image analysis using machine learning techniques. Its core strengths include:

• Backend Agnostic: Compatible with tensorflow, torch, and jax.
• User-Friendly API: High-level interface for transformations and model creation.
• Scalable Execution: Supports training and inference on single/multi-GPU and TPU-VM setups.
• Essential Components: Includes standard metrics and losses, such as Dice.
• Optimized 3D Inference: Offers an efficient sliding-window method and callback for volumetric data

🛠 Installation

PyPI version:

pip install medicai

Installing from source GitHub:

pip install git+https://github.com/innat/medic-ai.git

🍁 Available Features

The medicai library provides a range of features for medical image processing, model training, and inference. Below is an overview of its key functionalities.

Image Transformations

medicai includes various transformation utilities for preprocessing medical images:

• Basic Transformations:
  • Resize: Adjusts the image dimensions.
  • ScaleIntensityRange: Normalizes intensity values within a specified range.
  • CropForeground: Crops the image to focus on the region of interest.
  • Spacing: Resamples the image to a target voxel spacing.
  • Orientation: Standardizes image orientation.
  • NormalizeIntensity: Normalize the intensity of tensors based on global or channel-wise statistics.
  • SignalFillEmpty: Fills nan, positive infinity, and negative infinity values in specified tensors with a given replacement.
• Augmentations for Robustness:
  • RandCropByPosNegLabel: Randomly crops based on positive and negative label ratios.
  • RandRotate90: Randomly rotates images by 90 degrees.
  • RandShiftIntensity: Randomly shifts intensity values.
  • RandFlip: Randomly flips images along specified axes.
  • RandomSpatialCrop: Randomly crops a region of interest (ROI).
• Pipeline Composition:
  • Compose: Chains multiple transformations into a single pipeline.

Models

Currently, medicai focuses on 3D models for classification and segmentation:

• SwinTransformer – 3D classification task.
• SwinUNETR – 3D segmentation task.

Inference

• SlidingWindowInference – Processes large 3D images in smaller overlapping windows, improving performance and memory efficiency.

💡 Guides

Segmentation: Available guides for 3D segmentation task.

Task	GitHub	Kaggle
Covid-19
BTCV
BraTS
Spleen

Classification: Available guides for 3D classification task.

Task (Classification)	GitHub	Kaggle
Covid-19

🤝 Contributing

Please refer to the current roadmap for an overview of the project. Feel free to explore anything that interests you. If you have suggestions or ideas, I’d appreciate it if you could open a GitHub issue so we can discuss them further.

1. Install medicai from soruce:

!git clone https://github.com/innat/medic-ai
%cd medic-ai
!pip install keras -qU
!pip install -e .
%cd ..

Add your contribution and implement relevant test code.

1. Run test code as:

python -m pytest test/

# or, only one your new_method
python -m pytest -k new_method

🙏 Acknowledgements

This project is greatly inspired by MONAI.

📝 Citation

If you use medicai in your research or educational purposes, please cite it using the metadata from our CITATION.cff file.