QLayers

Quantitative layer based analysis for renal magnetic resonance imaging.

Installation

The easiest way to install qlayersis via pip:

pip install qlayers

or if you're a conda user:

conda install qlayers -c conda-forge

Alternatively, you can install qlayersfrom source in pips editable mode:

git clone https://github.com/alexdaniel654/qlayers.git
cd qlayers
pip install -e .

Quick Start

Documentation for qlayers can be found here. This includes a full API reference, tutorials and examples of how to use qlayers to generate layers and apply them to quantitative maps. The tutorials can also be run through binder if you want to have a go with qlayers without installing anything. However, if you want to get started on your machine with the minimum lines of code, the snippet of code below should get you going.

import nibabel as nib
from qlayers import QLayers

mask_img = nib.load("kidney_mask.nii.gz")
t2star_img = nib.load("t2star_map.nii.gz")

qlayers = QLayers(mask_img, pelvis_dist=10)
qlayers.add_map(t2star_img, "t2star")

df = qlayers.get_df(format="wide")
df.groupby("layer").median().loc[:, "t2star"].plot(
    xlabel="Depth (mm)", ylabel="$T_2^*$ (ms)"
)

Theory

Background

The premise behind qlayers was first proposed by Pruijm et al and is based on the idea to segment the kidney into layers based on each voxels distance from the surface of the kidney. The average of a quantitative parameter can be calculated for each layer producing profiles of, for example, T₂^* with depth. The outer and inner layers are analogous to the cortex and medulla respectively while the gradient of the profile is representative of the cortico-medullary difference. qlayers extends this idea by allowing the user to define layers based on a 3D mask and apply the layer to any quantitative parameter.

Generating Layers

Layers are generated via the process outlined in the figure below.

a i. Shows the mask thats input to the QLayers class. This mask then has any holes smaller than fill_ml filled as these are most likely cysts and therefore not cortical surfaces, a ii. The mask is then converted from a voxel representation to a mesh surface representation, b i, this mesh is then smoothed because anatomical scans of the kidneys often have a low through-plane resolution, b ii. The distance from the centre of each voxel in the kidneys to the closest surface on the mesh is then calculated, b iii. As the tissue adjacent to the renal pelvis is not representative of the medulla, this is automatically excluded from the resulting depth maps. Fist the pelvis is automatically segmented, c i, and the distance from each voxel in the kidneys to the pelvis calculated as above, c ii. Voxels closer than a specified threshold pelvis_dist are then excluded from the depth maps, c iii. Finally, a layer image is generated by quantising the depth map to a desired layer thickness, typically 1 mm although shown with 5 mm layers here for illustrative purposes, d.

Applying Layers to Quantitative Data

If the space parameter of the QLayers object is set to layers, when a quantitative map is added to the QLayers object, it is resampled to the same resolution and orientation as the layers. If the space parameter is set to map then the layers are resampled to the resolution and orientation of the quantitative map. In both cases, Pandas DataFrames can be generated with the quantitative value, depth and layer each voxel is in. These DataFrames can then be used for further calculations such as generating profiles or linear regressions to explore the cortico-medullary difference. Some example voxels are shown in the table below.

Citing 3DQLayers

If you have used 3DQLayers in your research, please cite the following conference abstract:

Daniel AJ, Francis ST. Volumetric Layer Based Analysis for Quantitative Renal MRI. In: Proc. Intl. Soc. Mag. Reson. Med. 33. Singapore; 2024:2748.

Alternatively, if you want to cite a specific version of this software, each release has an individual DOI on Zenodo, the DOI for the latest release can be found here.

Contributing

We welcome contributions to QLayers, a full contributing guide can be found here.