Technical Note: PYRO-NN: Python reconstruction operators in neural networks.

Area of Science:

• Medical imaging
• Deep learning
• Computational physics

Background:

• Deep learning is increasingly applied to medical image reconstruction.
• Existing methods often lack efficient, integrated CT reconstruction frameworks.
• Workarounds are used for solvable reconstruction problems.

Purpose of the Study:

• To present PYRO-NN, a generalized framework for embedding computed tomography (CT) reconstruction operators into deep learning environments.
• To address the lack of efficient, integrated CT reconstruction within deep learning pipelines.
• To provide a tool for developing end-to-end trainable neural networks for medical image reconstruction.

Main Methods:

• PYRO-NN is built on the Tensorflow deep learning framework.
• It includes CUDA-accelerated projectors and back-projectors as Tensorflow layers.
• A high-level Python API facilitates FBP and iterative reconstruction experiments with real CT data.

Main Results:

• The framework enables the design of end-to-end neural network pipelines with integrated CT reconstruction.
• Algorithms and tools are referenced and compared to non-deep learning frameworks.
• Baseline experiments demonstrate the framework's capabilities.

Conclusions:

• PYRO-NN facilitates the setup of end-to-end trainable neural networks within Tensorflow for medical image reconstruction.
• It aims to advance reproducible research in medical physics.
• The toolkit empowers the community to leverage deep learning for improved medical image reconstruction.