GLAM: Generative Lung Architecture Modeling (glamPipe)

Introduction

GLAM: Generative Lung Architecture Modeling (glamPipe) is a flexible pipeline designed for microscopy and bio-medical images. It offers a suite of tools for 3D image segmentation, 3D mesh generation (.stl files), and preparation of training sets for diffusion process-based image generation.
In combination with training and predicting of the diffusion model, the pipeline supports computationally generating 3D meshes that can be bio-printed to enlarge the imaged dataset for drug and treatment testing.
Diffusion model: https://github.com/ida-mdc/diff3d

Features

• 3D Image Segmentation: Utilize pre-trained model for precise segmentation of 3D images of tissue - done on printable size patches of the input images.
• 3D Mesh Generation: Convert segmented images into high-quality 3D meshes in STL format, suitable for 3D printing.
• Diffusion Process Preparation: Prepare your dataset for training with diffusion process-based generative models, enabling the creation of synthetic biological structures.

Installation

Create a conda/virtual env and activate it: conda create -n glamPipe python conda create -n glamPipe python=3.11 clone the repository and run the following command in its root directory:
pip install .

Command-Line Arguments

This tool supports various command-line arguments to customize its behavior. Below is a detailed description of each argument:

• -obp, --output-base-path (Required): Specifies the output path where result directories will be created or found if segmentation was previously done. This argument is required.

• -s, --is_segment: Enables the segmentation process. When set, path to original images and the segmentation model must be provided.

• -m, --is_mesh: Enables mesh creation from segmentation results (probability maps).

• -g, --is_prep_for_diffusion: Prepares the segmented images as a training set for diffusion processes.

• -porg, --path-originals: Path to the original images. Required if -s is set for segmentation.

• -sdd, --segmentation-dir-date: Specifies the date of the segmentation directory to use. Should not be provided when performing new segmentation as it implies segmentation results already exist.

• -c, --condition: Sets the experimental condition for the images being processed. Valid choices are emphysema, healthy, or fibrotic. If None is left all input images will be processed together. Relevant only for segmentation.

• -tm, --threshold-method: Chooses the thresholding method for segmentation. Options are triangle, otsu, li, yen. Default is triangle. Relevant for segmentation and mesh creation.

• -psm, --path-segmentation-model: Path to the segmentation model file. Required if -s is set for segmentation.

• -dvs, --default-voxel-size: Specifies the default voxel size (from the microscopy settings) as a list of three floats - zxy. Default is [0.0000022935, 0.0000013838, 0.0000013838].

• -dmsp, --default-mesh-size-in-pixels: Specifies the default mesh size in pixels as a list of three integers - zxy. Default is [64, 256, 256].

• -gs, --gaussian-sigma: Specifies the Gaussian sigma for smoothing as a list of three floats - zxy. Default is [1.2, 0.8, 0.8].

License

MIT License