ARNAI Artifact Removal Boosts Spine Segmentation on Postoperative Radiographs

Adding ARNAI to AI segmentation pipeline increased mean DSC from 0.814 to 0.870 and reduced L4-L5 Cobb angle error by ~70% (4.7° MAE) on internal test set, though external validation showed limited agreement.

• Internal test set: With ARNAI, implant-containing radiographs rejected decreased 64.21% (95 to 34) and L4-L5 SCA MAE fell from 16.2°–15.6° to 4.7°.
• External test set: Rejected radiographs dropped 65.45% (110 to 38), but L4-L5 SCA MAE reduction (to 9.5°–9.7°) was not significant after correction; agreement with experts remained limited.

Automated summary

RadPigeon summaries are original and for information only. They are not clinical advice.