cellmirage 0.1.2

Efficient and scalable image registration (alignment) for multiplexed imaging data

MIRAGE

Efficient and scalable image registration (alignment) for multiplexed imaging data. Currently only available for GPU.

Installation

git clone https://github.com/dpeerlab/MIRAGE.git
cd mirage
pip install -e .

Example

Also, check out the example in tests/quickstart.ipynb:

Before alignment:

After alignment:

# load package
import mirage

# Load images
image = mirage.tl.get_data("sample2_image.tiff")
reference = mirage.tl.get_data("sample2_reference.tiff")
# Images must be 2D numpy arrays (grayscale only) and scaled to 0-1.

# Initialise model
mirage_model = mirage.MIRAGE(
    images=image,
    references=reference,
    bin_mask=bin_mask,
    pad=12,
    offset=12,
    num_neurons=196,  # more for larger images
    num_layers=2,  # more for larger images
    pool=1, 
    loss="SSIM"
)

# Train model
mirage_model.train(batch_size=256, num_steps=256, lr__sched=True, LR=0.005)

# Apply transformation
mirage_model.compute_transform()
image_aligned = mirage_model.apply_transform(image)

# Inspect results pre/post alignment of a smale mesh
mesh = mirage.tl.Mesh(x=80, y=160, pad=35)
mirage.pl.plot_before_after(reference, image, image_aligned, mesh=mesh)