LabelProp: A semi-automatic segmentation tool for 3D medical images

LabelProp is a tool that provides a semi-automated method to segment 3D medical images with multiple labels.It is a convenient implementation of our peer-reviewed method designed to assist medical professionals in segmenting musculoskeletal structures on scans based on a small number of annotated slices (Decaux et al., 2023).LabelProp leverages deep learning techniques, but can be used without a training set.It is available as a PyPi package and offers both a command-line interface (CLI) and an API.Additionally, LabelProp provides two plugins, namely 'napari-labelprop' and 'napari-labelprop-remote', which facilitate training and inference on a single scan within the multi-dimensional viewer Napari.It is available on GitHub with pretrained weights (https://github.com/nathandecaux/napari-labelprop) Statement of needSegmenting musculoskeletal structures from MR images is crucial for clinical research, diagnosis, and treatment planning.However, challenges arise from the limited availability of annotated datasets, particularly in rare diseases or pediatric cohorts (Conze et al., 2020).While manual segmentation ensures accuracy, it is labor-intensive and prone to observer variability (Vdineanu et al., 2022).Existing semi-automatic methods based on point and scribbles require minimal interactions but often lack reproducibility (Chanti et al.,