Medical imaging pipelines that refuse to stutter

What it does

FAST is a C++ framework for building medical image processing pipelines. It ingests everything from DICOM and NIfTI to whole-slide images and live Clarius ultrasound feeds, then runs OpenCL-accelerated algorithms and neural network inference across multiple backends. Visualization happens in separate threads so the UI stays responsive while the GPU crunches marching cubes or volume renders.

The interesting bit

The framework treats data objects as abstractions that exist simultaneously on CPU, GPU, and other storage areas, keeping them coherent automatically. That means you can stream temporal data from a scanner or movie file through a TensorRT or OpenVINO model without rewriting the pipeline logic for static versus dynamic input.

Key highlights

• Abstracts away explicit memory management by maintaining coherent data objects across processors.
• Streams static and temporal data through the same pipeline code, including live ultrasound and webcam feeds.
• Supports multiple neural network backends (TensorRT, OpenVINO, TensorFlow, ONNX Runtime) and formats (ONNX, UFF, SavedModel, etc.).
• Ships with OpenCL implementations of common algorithms like marching cubes, non-local means, and seeded region growing.
• Offers Python bindings and Qt integration for embedding into existing applications.

Verdict

FAST is worth a look if you are building real-time medical imaging or digital pathology tools in C++ or Python and need GPU acceleration without writing your own memory management layer. If you are not in the medical imaging space, the domain-specific format support and visualization widgets are likely overkill.