Fully Automated Contour-Regulated Deformable Registration Framework for Liver CT Perfusion
Abstract
Purpose
To develop and validate a contour-regulated automated registration framework for correcting motion artifacts and aligning time-resolved scans in liver CT perfusion (CTp) series.
Methods
Forty-six CTp datasets from 23 patients were analyzed; each having a pre-contrast CT scan (craniocaudal coverage: >20cm), 60-80 post-contrast cine scans (4cm), and 8-10 post-contrast helical scans (10cm). A 2D nnU-Net model was trained using 23 pre-contrast CT scans with manually delineated liver contours and applied to all cine and helical series for liver auto-segmentation and evaluated on the rest 23 scans. The auto-segmented contours were used to regulate a two-stage registration consisting of an affine registration followed by a TransMorph-based deformable registration. The gastrointestinal air was filled with soft tissue intensity before registration. Multiple models with different contour-regulation weights were trained on 36 randomly selected cases and tested on the remaining 10 cases.
Results
Liver auto-segmentation achieved mean Dice similarity coefficients of 0.92 for helical and 0.89 for cine scans by comparing with manually delineated contours at randomly-selected time points (helical:3, cine:6). Registration accuracy was assessed by comparing with pre-registration using mean surface distance (MSD) and 95th percentile Hausdorff distance (HD95) on liver landmarks. The best-performing model, with an auto-determined contour-regulation weight of 0 or 0.8, reduced the MSD from 5.1 ± 1.4 mm to 1.0 ± 0.3 mm (P < 0.01) and HD95 from 9.7 ± 2.2 mm to 3.0 ± 0.9 mm (P < 0.01) in cine scans. In helical scans, MSD decreased from 3.2 ± 0.9 mm to 1.0 ± 0.3 mm (P < 0.01) and HD95 from 6.6 ± 1.9 mm to 2.9 ± 1.0 mm (P < 0.01).
Conclusion
The proposed contour-regulated, two-stage registration framework substantially improved temporal alignment in dynamic liver CTp images, providing a robust foundation for accurate perfusion analysis and subsequent clinical assessment.