Deepweed: automated individual segmentation and tracking for duckweed lineage analysis.

Abstract

Motivation: Aquatic duckweeds, including Wolffia australiana, are the smallest and fastest growing flowering plants on the planet, with numerous emerging biotechnological applications. Duckweed populations grow through rapid clonal budding, but existing phenotypic characterization and image analysis tools typically operate only at the population level, tracking neither individual plants nor their lineage structure. Deepweed addresses this gap, combining automated individual segmentation, temporal tracking, and clonal lineage reconstruction from time-lapse imaging of duckweed.

Results: Deepweed processes time-lapse duckweed imaging datasets using deep-learning-based instance segmentation and multi-object tracking to assign persistent identities to individual fronds and infer mother-daughter relationships. The workflow outputs lineage graphs, individual-level morphological metrics, and population-level growth summaries, supported by lineage-annotated time-lapse visualizations. Applied to dense Wolffia populations, in both freely growing or controlled microfluidic environments, our framework can track hundreds of Wolffia plants simultaneously, resolves overlapping individuals, and reconstructs multi-generation lineage structure, enabling analyses that are inaccessible with population-level phenotyping methods.

Availability: Deepweed is implemented in Python using PyTorch, scikit-image, OpenCV, and btrack. Source code, trained models, example datasets, and documentation are freely available at https://github.com/SamOliveiraLab/Deepweed. The Deepweed code is archived at https://doi.org/10.5281/zenodo.18844282 and the datasets are archived at https://doi.org/10.5281/zenodo.18843917.