dimble

Nimble Digital Imaging for Medicine

Pipeline

Completed

• Near lossless and easy conversion from DICOM and back
• Support for fast and random access of metadata
• Extremely fast and zero-copy loading to CPU/GPU
• Safe: no codegen/exec based on the metadata
• Support for ITK file formats [ref], including NIfTI

WIP

All relevant data types including uint16, f16, bf16, complex64 and complex128

• Currently supports f32, trivial to support other datatypes

Bindings for Python and conversion to NumPy/CuPy/JAX/Torch tensors

• Currently supports loading to Torch tensors (easily extensible)

Installation

# using ssh
git clone [email protected]:StrongCompute/dimble.git
# OR using https
git clone https://github.com/StrongCompute/dimble.git

cd dimble

make install
make validate_install

Usage

import dimble

# convert to dimble
dimble.dicom_to_dimble('xray.dicom', 'xray.dimble')

# load a dimble file's pixel data
dataset = dimble.load_dimble('xray.dimble', fields=["7FE00010"], device="cpu")

# load a dimble file's pixel data sliced to a 224x224 chunk offset by 100 in each dimension
dataset = dimble.load_dimble('xray.dimble', fields=["7FE00010"], device="cpu", slices=[slice(100,100+224), slice(100,100+224)])

# convert back to dicom
dimble.dimble_to_dicom("xray.dimble", "xray.dicom")

Developing

make install-dev

Testing

make install-dev
make test