Automated CT Lesion Tracking Achieves 99% Precision in Matching Colorectal Liver Metastases Over Time

An automated CT lesion-tracking algorithm matched 458/464 liver metastases (99% precision/recall, F1 0.989), detected 28/30 new lesions, and flagged 42/44 disappeared lesions. Confidence scores separated correct from incorrect matches (0.72 vs 0.54). External validation on melan…

• Affine+Auto-MBT outperformed deformable registration+overlap (F1 0.797) and deformable+Auto-MBT (F1 0.959).
• Per-lesion confidence scores: radiologist-accepted matches had mean 0.72 vs rejected 0.54, supporting triage.
• External testing on a melanoma dataset yielded overall matching F1 0.994.

Automated summary

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