Skip to main content
SIR 2026 Event Main banner

SIR 2026

Scientific Session

Embolization

Session: Portal Hypertension 1

170 - Predicting Adequate Future Liver Remnant Hypertrophy Following Liver Venous Deprivation (LVD) in Colorectal Liver Metastases: A Radiomic-Based Approach

Monday, April 13, 2026
3:09 PM - 3:18 PM ET
Purpose: To investigate whether clinical characteristics and radiomic features extracted from the non-embolized future liver remnant (FLR) on pre-embolization CT scans are associated with kinetic growth rate (KGR) following liver venous deprivation (LVD).

Materials and Methods: A retrospective analysis was conducted on 22 patients with Colorectal Liver Metastases (mean age 51.1 ± 11.5 years; 73% male), who underwent LVD between September 2018 and August 2023. Embolization was performed using microparticles and coils for the portal vein and n-BCA with Amplatzer plugs for the hepatic vein via percutaneous approach. Pre- and post-procedure FLR volumes were quantified using CT volumetry. The standardized liver volume was estimated by the Vauthey formula, and KGR was calculated from FLR volumetric changes. The non-embolized FLR was segmented (excluding major vessels) using MIM software, and 107 original radiomic features (first order and gray level) were extracted using PyRadiomics. Associations between radiomic and clinical variables with continuous KGR were assessed using Spearman correlation. Categorical variables were evaluated with Wilcoxon rank-sum tests. An exploratory multivariable linear regression was performed to assess the independent association of clinical and radiomic variables with KGR.

Results: 18/22 patients (82%) achieved adequate hypertrophy (KGR > 2%/week). After LVD, the median degree of hypertrophy was 16% and the median KGR was 3.9%/week. Embolization particles were ≤ 100 µm in 11/22 (50%) and 250–400 µm in the rest. No clinical variable, including particle size, ALT, AST, INR, or total bilirubin, correlated with KGR (all p > 0.05). Pre-procedure BMI showed a negative correlation with KGR (ρ = –0.46, p = 0.034). Of 107 radiomic features, the strongest univariable correlations were sphericity (ρ = –0.51, p = 0.016), SmallDependenceHighGrayLevelEmphasis (ρ = +0.47, p = 0.027), and ShortRunLowGrayLevelEmphasis (ρ = –0.47, p = 0.027). In multivariable modeling, sphericity (β = –1.27, p = 0.002) and SmallDependenceHighGrayLevelEmphasis (β = +2.04, p < 0.001) remained significant.

Conclusion: This preliminary analysis suggests that radiomic features of the non-embolized FLR may have predictive value for liver regenerative capacity in patients planned for hepatectomy as shape-based measures such as sphericity and textural gray-level features demonstrated moderate associations with KGR. These findings indicate that radiomic signatures could be incorporated to develop machine learning models that could aid in patient selection and optimizing outcomes.