OCT_Segmentation_a1 / README.md
farrell236's picture
Deploy native Gradio OCT inference app
8fca640 verified
|
Raw History Blame Contribute Delete
2.01 kB
---
title: OCT Segmentation A1
emoji: 🏢
colorFrom: green
colorTo: purple
sdk: gradio
sdk_version: 6.31.0
python_version: "3.12"
app_file: app.py
pinned: false
license: apache-2.0
short_description: MYXOS OCT Segmentation AI-READI (agent1)
---
# MYXOS OCT Segmentation · Agent 1
Visualize retinal OCT boundaries using a model trained on AI-READI.
1. Upload one structural OCT DICOM volume, or its contiguous split parts.
2. Choose the scanner family and processor. Spectralis is selected initially;
**Detect from DICOM** is also available.
3. Run **Full volume** inference, or choose a fast **Selected B-scan preview**.
4. Browse the boundary-overlay and original-OCT tabs. Use the toolbar to zoom,
pan, fit, or view fullscreen; adjust visible boundaries and display settings.
5. Choose NPZ predictions or the current overlay PNG, then download the result.
Supported scanners: Heidelberg Spectralis, Topcon Maestro2, Topcon Triton, and
Zeiss Cirrus. This Space runs on CPU with batch size 1 initially; full-volume
inference can take longer than a slice preview. The model uses its saved FP32
inference configuration. Preview mode omits smoothing between B-scans.
NPZ files contain `positions[boundary, B-scan, A-scan]` in original zero-based
axial pixels and `boundary_keys`. Slice previews also include `frame_indices`.
Inference runs on this Space's server. Uploads and exports are temporary;
**Clear volume** removes the staged volume and its results. Idle viewer sessions
expire after two hours, and Gradio clears cached files on a two-hour schedule.
This model is specialized to AI-READI structural OCT exports. The boundaries
retain supplied vendor label distinctions and are research predictions.
The retained checkpoint achieved 1.427501 native axial pixels MAE (3.733009 µm)
on 2,392 frozen validation volumes. Those historical results measure agreement
with supplied automatic annotations; they do not establish clinical performance
or numerical equivalence on this hosting hardware.