Segment-and-Classify: ROI-Guided Generalizable Contrast Phase Classification in CT Using XGBoost

TL;DR

This work tackles automated CT contrast phase classification by leveraging ROI-focused features extracted from TotalSegmentator and a lightweight XGBoost classifier, aiming for robust generalization across institutions. The approach is trained on WAW-TACE and externally validated on VinDr-Multiphase and C4KC-KiTS, outperforming 3D CNN baselines in arterial/venous and delayed phases while maintaining near-perfect performance for non-contrast. The combination of ROI-based feature extraction and a gradient-boosted classifier provides a computationally efficient, interpretable alternative to deep learning models with strong cross-domain transfer. This has practical implications for automated protocol management and reduced radiologist workload in multi-institution workflows.

Abstract

Purpose: To automate contrast phase classification in CT using organ-specific features extracted from a widely used segmentation tool with a lightweight decision tree classifier. Materials and Methods: This retrospective study utilized three public CT datasets from separate institutions. The phase prediction model was trained on the WAW-TACE (median age: 66 [60,73]; 185 males) dataset, and externally validated on the VinDr-Multiphase (146 males; 63 females; 56 unk) and C4KC-KiTS (median age: 61 [50.68; 123 males) datasets. Contrast phase classification was performed using organ-specific features extracted by TotalSegmentator, followed by prediction using a gradient-boosted decision tree classifier. Results: On the VinDr-Multiphase dataset, the phase prediction model achieved the highest or comparable AUCs across all phases (>0.937), with superior F1-scores in the non-contrast (0.994), arterial (0.937), and delayed (0.718) phases. Statistical testing indicated significant performance differences only in the arterial and delayed phases (p<0.05). On the C4KC-KiTS dataset, the phase prediction model achieved the highest AUCs across all phases (>0.991), with superior F1-scores in arterial/venous (0.968) and delayed (0.935) phases. Statistical testing confirmed significant improvements over all baseline models in these two phases (p<0.05). Performance in the non-contrast class, however, was comparable across all models, with no statistically significant differences observed (p>0.05). Conclusion: The lightweight model demonstrated strong performance relative to all baseline models, and exhibited robust generalizability across datasets from different institutions.

Figures (6)

• Figure 1: Contrast enhancement timing of selected organs from the TotalSegmentator list. Organs are sorted by onset of enhancement. Horizontal lines denote time post-contrast injection ($t=0$) during which each organ is best visualized. Dashed vertical lines indicate standard CT phases timings: non-contrast, arterial ($\sim$30s), venous ($\sim$70s), and delayed ($\sim$180s).
• Figure 2: Arterial-phase scan of Patient #1033 from the VinDr-Multiphase dataset. XGBoost correctly classified the scan as arterial, while all 3D models predicted delayed, and ts_phase misclassified it as venous. This case is particularly challenging, as the aorta shows only subtle enhancement—possibly corresponding to a late arterial phase—a feature that was overlooked by the 3D models.
• Figure 3: Delayed-phase scan of Patient #349 from the WAW-TACE dataset. The image is limited to the abdomen, with no visualization of the urinary bladder; however, residual contrast is visible in the small bowel (blue box), and the inferior vena cava (IVC) (red box).
• Figure S1: Phase classification performance on the VinDr-Multiphase Dataset, illustrated by confusion matrices and ROC curves; (a) XGBoost, (b) ResNet3D 18-layer (r3d_18), (c) Mixed Convolution Network 18-layer (mc3_18), and (d) R(2+1)D 18-layer (r2plus1d_18).
• Figure S2: Phase classification performance on the C4KC-KiTS Dataset, illustrated by confusion matrices and ROC curves; (a) XGBoost, (b) ResNet3D 18-layer (r3d_18), (c) Mixed Convolution Network 18-layer (mc3_18), and (d) R(2+1)D 18-layer (r2plus1d_18).