Abstract Rationale Accurate longitudinal tracking and volume quantification of lung nodules are critical for risk stratification in lung cancer screening but are time-intensive and subject to inter-reader variability. Automated methods can improve tumor tracking and volume measurement reproducibility, and enhance early malignancy detection through objective growth rate thresholds. We developed and tested an automated pipeline for longitudinal nodule tracking and volumetric analysis to determine whether computed tumor growth rates discriminate malignant from benign nodules. Methods We developed a Python-based pipeline integrating ANTsPy deformable registration (Symmetric Normalization), geometric centroid-based tracking (30 mm radius), and native-space volumetry using voxel-counting and smoothed mesh reconstruction. The pipeline was applied to a curated Durham VA Medical Center dataset comprising 27 high-risk patients with 200+ longitudinal CT scans (median 7.4 scans/patient) processed with ClearRead vessel suppression. All timepoints were registered to each patient’s earliest scan. Nodules were automatically detected, tracked across timepoints, and filtered by temporal presence (≥50% of scans). Tumor growth rates (exponential model) were computed using the tumgr R package. High-growth outliers were defined statistically (Q3 + 1.5 × IQR; corresponding to 0.003572/day, or volume-doubling time 194 days) and correlated with pathologically confirmed cancer diagnosis using Fisher’s exact test. Results The pipeline tracked 374 nodules across 27 patients (mean 13.9 per patient). Of these, 209 nodules (56%) had ≥3 timepoints enabling longitudinal modeling, and 165 (44%) had 2 measurements. Growth rates were successfully computed for 254 nodules (68%), including 152 with ≥3 timepoints and 102 with 2 timepoints. Median growth rate was 0.000821/day (Q1: 0.000447, Q3: 0.001697), corresponding to a median volume doubling time of 844 days. Nodules with markedly rapid growth (0.003572/day, equivalent to VDT 194 days) were classified as high-growth outliers and showed a 57% malignancy rate (8/14), compared with 22% (25/113) in all other nodules (p = 0.0026). At the patient level, 43% (3/7) of those with at least one high-growth nodule had confirmed lung cancer versus 5% (1/20) without (p = 0.0419). Automated analysis required 15 minutes versus ∼4 hours manually. Thus, identifying high-growth outliers provides an objective, quantitative threshold linked to malignancy risk and potential clinical action. Conclusions Our automated longitudinal tracking pipeline with integrated tumor growth rate analysis enables scalable identification of high-risk pulmonary nodules. Integration of vessel-suppressed imaging with deformable registration and native-space volumetry offers a framework for risk stratification that could reduce unnecessary surveillance while expediting identification of aggressive lesions, enhancing precision in lung cancer screening programs. This abstract is funded by: U.S. Department of Veterans Affairs, Biomedical Laboratory Research and Development (VA BLR&D), Grant I01BX004121-05A1
Al-Shakhshir et al. (Fri,) studied this question.