A pytorch-based deep learning toolkit for medical image analysis. MedicalVision aims to provide a light wrap upon pytorch, which can further reduce the time on developing new algorithms for medical image analysis task, such as classification, registration and segmentation etc.
As the deep learning goes popular in computer vision, many deep learning based works/architectures have been proposed to handle traditional medical image analysis tasks (classification, registration and segmentation). Unfortunately, to the best my knowledge, there is not a simple yet effective toolkit based on pytorch which can enable fast prototyping. In the daily work, I suffer from writing DataLoaders for various medical image datasets and reproducing the algorithms introduced in some paper. To make life easy, the MedicalVision toolkit is created which aims to provide:
- Dataloaders for famous medical image dataset
- Common losses and metrics used in the state-of-the-art models
- Model zoo, including pretrained state-of-the-art models
- ...
The MedicalVision toolkit is still under development. The following popular deep learning models are coming soon:
- 2D U-net
- 3D U-net