Body axis identification (dorsal-ventral / anterior-posterior)

[pskeshu]

Motivation

Body axes become morphologically visible from comma stage onward:

• AP axis: elongation direction (head is slightly narrower with denser nuclear signal from pharynx)
• DV axis: perpendicular, ventral side is the concave surface of the comma curve

Knowing the axes helps with longitudinal consistency — the embryo rotates within the eggshell between timepoints. Without axis tracking, it's ambiguous whether a morphological change between T41 and T42 reflects stage progression or embryo rotation.

Use cases

• Rotation disambiguation: "the embryo looks different" → did the stage change, or did it rotate 30°?
• Twitching detection: late pretzel embryos twitch (rapid brief movements). Axis tracking could distinguish twitching from actual stage progression.
• Consistent classification: same embryo from different orientations should classify the same way

Constraints

With histone-only (nuclear) fluorescence, axis identification relies on morphological asymmetry, not cell-specific markers. Reliable from comma stage onward. Not useful during early/bean when the embryo is roughly symmetric.

Output

An additional field in perception output or session state:

orientation: {ap_angle_deg: float, dv_angle_deg: float}

Tracked over time to detect rotation events vs stage transitions.

Priority

Defer until fold-stage accuracy is improved (#4). The accuracy bottleneck is more impactful to solve first.