Deep learning model automates acetabular cup measurement on postop CT using robotic reference

A deep-learning model (PointNet++) measured acetabular cup orientation on postoperative CT with MAE 4.48° (anteversion) & 3.89° (inclination), outperforming manual (8.91°/8.70°, p<0.05). Internal validation only.

• In internal validation, 71% of anteversion and 81% of inclination predictions were within 5° of the robotic reference.
• Lewinnek classification correct in 81/94 hips vs 57/94 with manual measurement.
• External validation is required before broader clinical use.

Automated summary

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