Purpose

Accurate three-dimensional reconstruction of the orbit is fundamental to surgical planning, biomechanical modelling, and understanding pathological changes in the orbit. However, the medial wall (lamina papyracea) and orbital floor are sub-millimetric in thickness and lie below the resolution limit of routine clinical CT, where the partial volume effect causes them to appear discontinuous or absent on conventional bone-window segmentation. This results in incomplete reconstructions and inaccurate orbital volumes. We describe and validate a “deductive” segmentation method that overcomes the partial volume effect by inferring bone position from adjacent tissue interfaces.

---

Methods

Ten anatomical skulls were imaged, with both right and left orbits analysed (n = 20). Soft-tissue equivalence within the orbital cavity was reproduced using 2% alginate to replicate post-mortem orbital contents and reduce air-related imaging artefact. CT acquisition followed a clinically standard protocol with thin-slice 0.8 mm isotropic resolution. Reconstruction was performed in 3D Slicer using multi-class segmentation, deducing the position of the medial wall and floor from the sinus–bone–soft-tissue interface rather than relying on direct bone signal. Outputs were compared to ground-truth manual measurements.

---

Results

Deductive segmentation showed strong agreement with ground truth across the principal orbital regions, with intra-class correlation R² = 0.897 (p 0.05).

---

Conclusion

Deductive segmentation provides a clinically practical, reproducible method for reconstructing the thin orbital walls despite the partial volume effect, supporting its application in oculoplastic surgical planning and patient specific 3D visualisation.

---

Additional Authors

First name	Last name	Base Hospital / Institution
Claire	Daniel	Moorfields Eye Hospital
Christos	Bergeles	Kings College London
Asit	Arora	Kings College London
Jimmy	Uddin	Moorfields Eye Hospital

↑ Back to top