This repository contains the implementation of SplineDist, a machine learning framework for automated cell segmentation with spline curves. The manuscript is accepted at ISBI 2021. The code in its current state allows reproducing the paper experiments but is still in development. We are currently working to package it in a cleaned and optimized form. In the meantime, we encourage interested end-users to contact us for more information and assistance.
If you prefer to try out SplineDist from a user interface, we also have a napari plugin.
SplineDist has been designed for performing instance segmentation in bioimages. Our method has been built by extending the popular StarDist framework. Our repository relies on the high-quality StarDist repository. We encourage the user to explore StarDist repository for further details on the StarDist method.
While StarDist models objects with star-convex polygonal representation, SplineDist models objects as parametric spline curves. As our representation is more general, it allows to model non-star-convex objects as well, with the possibility of conducting further statistical shape analysis.
To install and set up SplineDist, follow these steps:
-
Clone the repository and navigate to the SplineDist directory:
git clone [email protected]:uhlmanngroup/splinedist.git cd SplineDist
-
Create and activate a new environment:
mamba create -n splinedist python=3.8 mamba activate splinedist
-
Install Anaconda packages:
mamba install anaconda
-
Install SplineDist:
python3 -m pip install .
Three walkthrough notebooks have been included in this repository for data-exploration, training, and inference tasks.
The synthetic dataset used in the SplineDist manuscript can be found here.This dataset contains synthetic images with mostly star-convex and some non-star convex cell-like objects.
Some pretrained SplineDist models are available here. These models are trained on open-source datasets of fluorescence microscopy images and Haematoxylin & Eosin stained histology images.