I’ve released a Keras 3-based AI library for medical image processing. Inspired by ‘MONAI’, it is designed to streamline deep learning workflows for medical imaging, offering pre-built models, data augmentation techniques, and specialized utilities for tasks like segmentation, and classification.
• Library: GitHub - innat/medic-ai: AI Toolkit for Healthcare Imaging
• Website: medicai
• 3D classification guide: [medicai] 3D Image Classification | Kaggle
• 3D segmentation guide: [medicai] 3D Image Segmentation | Kaggle
• Convert .nii to TFRecord: Generate 3D .nii to TFRecord Dataset | Kaggle
The library currently supports models like SwinUNETR and focuses on 3D medical imaging. All preprocessing techniques are implemented in pure TensorFlow, enabling seamless integration with the tf.data API for efficient data pipeline execution. With Keras 3’s multi-backend support, the library is designed for future adaptability across different deep learning frameworks.