segevmunitz/astrocytes_morphology_detection

How segevmunitz/astrocytes_morphology_detection works

AstroSeg builds complete GFAP-positive astrocyte instances from microscopy, linking each predicted cell’s nucleus, soma, and processes. The principal path prepares manifest-backed images and annotations, trains a multi-head model, predicts image patches, then reconstructs individual cells and measurements/QC. Separate branches support human mask import/review, pseudo-labeling, the older semantic/binary pipeline, and Cellpose refinement. Edges are limited to README statements and shown source references/calls; unsampled implementation details are not inferred.

Main entry points: predict_instances.py Patch inference · instance_trainer.py Instance trainer · instance_dataset.py Instance patches · workflow.py Mask import

Components 26

Data preparation

Manifest schema src/astroseg/io/manifest.py
Microscopy I/O src/astroseg/io/ome_tiff.py
Channel selection src/astroseg/preprocessing/channels.py
Instance patches src/astroseg/datasets/instance_dataset.py
Instance targets src/astroseg/preprocessing/instance_targets.py
Semantic patches src/astroseg/datasets/astrocyte_dataset.py

Model training

Instance trainer src/astroseg/training/instance_trainer.py
Multichannel U-Net src/astroseg/models/multichannel_unet.py
Training losses src/astroseg/training/losses.py
Checkpoints src/astroseg/training/checkpoints.py

Instance prediction

Patch inference src/astroseg/inference/predict_instances.py
Cell reconstruction src/astroseg/postprocessing/astrocyte_instances.py
Nucleus detection src/astroseg/preprocessing/nucleus_detection.py
Cell morphology src/astroseg/postprocessing/skeleton.py
Image measurements src/astroseg/analysis/morphology.py
QC overlays src/astroseg/visualization/overlays.py

Annotation workflows

Mask import src/astroseg/annotations/workflow.py
Instance annotation workflow src/astroseg/annotations/instance_workflow.py
Mask review GUI src/astroseg/gui/mask_review.py
Pseudo-labeling src/astroseg/annotations/pseudo_labels.py

Other model paths

Binary U-Net src/astroseg/models/unet.py
Cellpose transfer src/astroseg/cellpose/transfer.py
Cellpose model src/astroseg/models/cellpose_transfer.py

Other components

Microscopy user
Dataset artifacts
Instance results

Connections 32

Microscopy user → Manifest schema: defines dataset README.md
Manifest schema → Instance patches: indexes samples src/astroseg/datasets/instance_dataset.py
Instance patches → Microscopy I/O: loads images src/astroseg/datasets/instance_dataset.py
Instance patches → Instance targets: builds targets src/astroseg/datasets/instance_dataset.py
Instance patches → Channel selection: prepares channels
Microscopy user → Instance trainer: trains model README.md
Instance trainer → Multichannel U-Net: optimizes src/astroseg/training/instance_trainer.py
Instance trainer → Training losses: computes loss src/astroseg/training/instance_trainer.py
Instance trainer → Checkpoints: saves weights src/astroseg/training/instance_trainer.py
Instance trainer → Instance patches: consumes batches
Multichannel U-Net → Patch inference: predicts heads
Patch inference → Nucleus detection: uses nuclei
Patch inference → Cell reconstruction: reconstructs cells
Microscopy user → Patch inference: predicts images README.md
Patch inference → Checkpoints: loads weights
Patch inference → Instance results: writes predictions README.md
Cell reconstruction → Instance results: labels instances README.md
Cell reconstruction → Cell morphology: derives morphology
Image measurements → Cell morphology: uses skeleton src/astroseg/analysis/morphology.py
Microscopy user → Mask import: imports masks README.md
Mask import → Microscopy I/O: loads microscopy src/astroseg/annotations/workflow.py
Mask import → QC overlays: creates QC overlay src/astroseg/annotations/workflow.py
Microscopy user → Mask review GUI: reviews masks README.md
Instance annotation workflow → Mask import: uses import workflow src/astroseg/annotations/instance_workflow.py
Microscopy user → Pseudo-labeling: creates pseudo-labels README.md
Microscopy user → Binary U-Net: runs baseline README.md
Microscopy user → Cellpose transfer: refines Cellpose README.md
Cellpose transfer → Cellpose model: transfers weights
Semantic patches → Microscopy I/O: loads images src/astroseg/datasets/astrocyte_dataset.py
Semantic patches → Channel selection: prepares inputs src/astroseg/datasets/astrocyte_dataset.py
Microscopy user → Dataset artifacts: stores artifacts README.md
Instance results → Dataset artifacts: persists outputs README.md

Drawn by GitDiagram on November 12, 2026 from the repository's file tree, README and source files. Read as Markdown ·