Skip to content

Workflow

The following diagram gives on overview of the BioSegment workflow. Users interact with a frontend using their browser. They can visualize a dataset, edit annotations and create segmentations using AI models. The BioSegment backend handles the tasks given by the frontend and fetches the datasets from disk storage. For long-running tasks like conversion and fine-tuning, separate workers are used.

Placeholder

Data

  • Dataset
    • Electron-microscopy data
    • example formats: pngseq, tif3d
  • Classes of interest
    • example classes of interest: mitochondria, endoplasmatic reticulum...
  • Segmentations
    • Attribution for each part of a dataset to an interest
    • mostly ground-truth or machine made
  • Annotations
    • Stroke or area of a part of the dataset that is attributed
    • made by a human
  • AI models
    • able to take EM data and an annotation and create a segmentation
    • can be pretrained and further fine-tuned with additional annotations
    • e.g. UNet

Actors

  • Scientist
    • A domain expert that wants to visualize and annotate EM data with a specialized tool
  • AI engineer
    • Implements and pretrains AI models

User flow

Example of the user flow for a scientist when interacting with a BioSegment frontend.

Placeholder