Automated model-based lesion tracking in CT: colorectal liver metastases as a use case for development and...

To develop and evaluate an automated CT liver lesion-tracking algorithm that matches lesions over time, detects new metastases, and reports per‑lesion confidence to support response assessment.

The study included 87 adults with unresectable colorectal liver metastases (CRLM) who had baseline and 8-week follow-up contrast-enhanced CT. Three radiologists generated a consensus reference. We developed a machine learning-driven, automated model-based lesion tracking (Auto-MBT) that provides per-lesion matching confidence. Performance was compared with: (1) deformable registration + overlap; (2) deformable registration + Auto-MBT; and (3) affine registration + Auto-MBT. Analyses were stratified by lesion size (< 1 cm, 1-3 cm, overall) and count (≤ 5, 6-10, > 10 per scan), and the triage utility of confidence scores was assessed by blinded adjudication. A publicly available melanoma dataset was used for external testing.

On the CRLM test set (35 pairs), affine + Auto-MBT matched 458/464 lesions (precision/recall 99%), detected 28/30 new lesions (90%/93%), and identified 42/44 disappeared lesions (93%/96%). Overall matching F1 was 0.989; affine + Auto-MBT outperformed deformable + overlap (0.797) and deformable + Auto-MBT (0.959). Confidence scores were higher for radiologist-accepted matches than rejected matches (0.72 versus 0.54) supporting triage utility. On the external set (23 pairs), overall matching F1 was 0.994, affine + Auto-MBT outperformed deformable + overlap (0.768) and deformable + Auto-MBT (0.971).

Auto-MBT accurately tracks CRLM and flags new lesions across sizes and counts, with per‑lesion confidence to triage, not replace, clinician review, enabling more comprehensive and accurate therapy response assessment.

The automated tracking framework accurately matched liver lesions, including new and subcentimeter lesions, improving longitudinal assessment.

Manual tracking of colorectal liver metastases (CRLM) on routine CT images is time-consuming and subject to inter-observer variability. The algorithm demonstrates robust lesion‑level tracking performance (F1 > 0.9) in high lesion‑count CRLM, including matched, new, and disappeared lesions. Automated total-tumor tracking complements Response Evaluation Criteria in Solid Tumors‒RECIST by quantifying all lesions, extending evaluation beyond limited target metastases.