Documentation Index
Fetch the complete documentation index at: https://docs.atomscale.ai/llms.txt
Use this file to discover all available pages before exploring further.
Optical image analysis automatically segments features from microscopy videos and tracks how their shape evolves over time. The pipeline uses a foundation segmentation model to identify regions of interest in each frame, then computes morphological metrics across the full video.
Overview
The optical image pipeline processes data in three stages:
- Frame normalization: corrects for haze, uneven illumination, and low contrast
- Video segmentation: a deep learning model segments features frame-by-frame, maintaining temporal consistency across the video
- Metric computation: morphological properties are measured for each segmented feature per frame, then tracked as timeseries with statistical anomaly scoring
Key Metrics
| Metric | What It Tells You |
|---|
| Perimeter | Arc length of segmented feature boundaries, in pixels. |
| Circularity | How close a feature’s shape is to a perfect circle, computed as P²/(4πA). A value of 1.0 indicates a perfect circle. |
| Edge roughness | Deviation of the feature boundary from its convex hull, as a fraction of the convex hull perimeter. Higher values indicate more irregular edges. |
| Hausdorff distance | Maximum boundary displacement between consecutive frames, in pixels. Captures the largest local shape change between frames. |
Adding Data
Upload optical microscopy video files through the data page. Analysis begins automatically once the upload completes.Viewing Results
Once processing completes, the workspace shows several sections:Video player: Scrub through the processed video frame by frame. The player displays metadata including FPS and total frame count.Analysis region: Shows the segmentation overlay on the current frame, highlighting the regions being tracked by the pipeline.Timeseries chart: Plots morphological metrics (perimeter, circularity, edge roughness, Hausdorff distance) across all video frames. Each metric includes EMA-smoothed mean and standard deviation bands, along with anomaly scores that flag frames with unusual shape changes.A status badge indicates pipeline progress. If results need to be regenerated, use the refetch button to re-run analysis.Analysis Pipeline
The optical image pipeline processes video through three stages.Stage 1: Frame Normalization
Raw video frames are corrected for haze, uneven illumination, and low contrast. Brightness drift is removed across the video, and feature boundaries are enhanced to improve segmentation quality.Stage 2: Video Segmentation
A deep learning segmentation model processes frames sequentially, producing instance masks that track features across time. For long videos, frames are processed in batches with overlap to maintain tracking consistency at batch boundaries.Stage 3: Metric Computation
For each segmented feature in each frame, the pipeline computes perimeter, circularity, edge roughness (boundary deviation from the convex hull), and Hausdorff distance (maximum boundary displacement between consecutive frames). Statistical deviation metrics and anomaly scores are computed over time to flag frames with unusual shape changes.
