Full text
Transforming BioImage Analysis: The AI4Life Approach Caterina Fuster-Barceló, PhD – caterina.fusterbarcel[email protected] Cell Physics – 09th October 2025
2025 - present 2022 - 2025 BioImage.IO Chatbot SAMJ
What’s
Accessibility of AI/Deep Learning Methods BioImage Model Zoo Repositories: Training and opportunities:
Open Calls: 1, 2, 3… Go! Phase 1: Review Phase 2: Consultations Phase 3: Projects
Open Calls: 1, 2, 3… Go! 1 2 3 72 applications 16 reviewers 8 projects 51 applications 17 reviewers 7 projects 28 applications 13 experts 10 consultations 8 projects
Wound-healing time-lapse of epithelial monolayers The data was provided by Madeline Lovejoy and Rafael Garcia-Mata (PhD) at Garcia-Mata Lab, University of Toledo (USA). The data: The problem: Study the migration of the cells to close the inflicted wound
A closer look The data was provided by Madeline Lovejoy and Rafael Garcia-Mata (PhD) at Garcia-Mata Lab, University of Toledo (USA). Confluent 2D epithelial monolayer; pipette-tip scratch, fluorescent timelapse Compare cell lineages – need per-cell segmentation and tracking plus morphology and motility stats. Challenge: high confluence, variable contrast, edge effects at wound, identity swaps.
From N2V to CAREamics – tooling n2v and CAREamics are both libraries in python for image denoising, superresolution and restoration. n2v is no longer maintained. Move to CAREamics! ⚠ n2v is not compatible with TensorFlow 2.16 ⚠
How to use CAREamics? https://github.com/CAREamics/careamics-examples/blob/main/applications/n2v/2D/MouseNuclei_Noise2Void.ipynb config = create_n2v_configuration( experiment_name="mouse_nuclei_n2v", data_type="array", axes="SYX", patch_size=(64, 64), batch_size=16, num_epochs=10, ) careamist = CAREamist(source=config) careamist.train( train_source=train_data, )
How to use CAREamics? https://github.com/CAREamics/careamics-examples/blob/main/applications/n2v/2D/MouseNuclei_Noise2Void.ipynb config = create_n2v_configuration( experiment_name="mouse_nuclei_n2v", data_type="array", axes="SYX", patch_size=(64, 64), batch_size=16, num_epochs=10, ) careamist = CAREamist(source=config) careamist.train( train_source=train_data, )
How to use CAREamics? https://github.com/CAREamics/careamics-examples/blob/main/applications/n2v/2D/MouseNuclei_Noise2Void.ipynb config = create_n2v_configuration( experiment_name="mouse_nuclei_n2v", data_type="array", axes="SYX", patch_size=(64, 64), batch_size=16, num_epochs=10, ) careamist = CAREamist(source=config) careamist.train( train_source=train_data, )
Denoising Segmenting Tracking CellPose The Pipeline CellPose
Segmenting CellPose
Why segmentation before tracking? ✨ We need to know exactly where they are at each time-frame and how much they occupy (pixel-wise) . ✨ We need to know which cell it is individually, as in instance segmentation.
Why segmentation before tracking? ✨ We need to know exactly where they are at each time-frame and how much they occupy (pixel-wise) . ✨ We need to know which cell it is individually, as in instance segmentation. Semantic Instance
Cellpose Pachitariu, M., Rariden, M., & Stringer, C. (2025). Cellpose-SAM: superhuman generalization for cellular segmentation. It’s everything: a GUI, a library, a model and even a dataset. from cellpose import io, models, train model = models.CellposeModel()
Cellpose Pachitariu, M., Rariden, M., & Stringer, C. (2025). Cellpose-SAM: superhuman generalization for cellular segmentation. Option 1: Use a Cellpose model as-is. 💡Tips: Calibrate to the size of your nuclei. Try on different magnifications, stainings, …
From segmented cells to trajectories Firsts things first: the masks. ✨ Input: centers are detected from masks. ✨ Linking: LAP tracker connects objects frame-to-frame by distance. ✨ Gap-closing: reconnects short disappearances. ✨ Segment splitting: link spots for cell division
From segmented cells to trajectories Firsts things first: the masks. ✨ Input: centers are detected from masks. ✨ Linking: LAP tracker connects objects frame-to-frame by distance. ✨ Gap-closing: reconnects short disappearances. ✨ Segment splitting: link spots for cell division
Displaying tracks
Denoising Segmenting Tracking CellPose The Pipeline CellPose
Sharing
FAIR Science Reuse Make sure that your workflow is reproducible and your data can be reused. Interoperability Data needs to be integrated with applications or workflows for analysis, storage, and processing. Findability Metadata and data should be easily to find for both humans and computers. Accessibility Knowing how the data can be accessed, even including authentication and authorisation is crucial.
A zoo for DL models in microscopy … segmentation denoising image-classification image-reconstruction semantic-segmentation image-restoration 2d 3d pytorch tensorflow whole-slide-imaging
Model Development Model Packaging Upload to bioimage.io Model Accessibility Model Deployment Model Fine Tuning Developers End-users The life cycle of a model in the BioImage Model Zoo
Consuming Fine-tuning Exporting …
🐘 happy-elephant 📊Statistics ⬇ Downloads: 109 👁 Views: 218 🏷Tags nuclei segmentation epithelial cells fluorescence microscopy 2D cell-segmentation cellpose 🔗Links GitHub Repository
