APPLICATION OF 2D PERFUSION ANGIOGRAPHY TO ASSESS THE EFFECTIVENESS OF VASCULAR EMBOLIZATION: ANALYSIS USING THE XGBOOST ALGORITHM

Abstract

Introduction. Interventional radiology is a cornerstone of minimally invasive medicine, yet objective assessment of embolization efficacy remains difficult because conventional visual evaluation of angiographic images is subjective and operator-dependent. Two-dimensional (2D) perfusion angiography, derived from digital subtraction angiography (DSA), enables quantitative assessment of hemodynamic parameters, including time to peak (TTP) contrast enhancement, reflecting the timing of peak contrast enhancement, and mean transit time (MTT), reflecting the duration of contrast transit through the perfused tissue. Aim. To quantify changes in TTP and MTT before and after splenic artery embolization and, as an exploratory analysis, to assess whether a machine-learning model can automatically distinguish pre- from post-embolization perfusion states. The clinical efficacy of the procedure was additionally evaluated at 12-month follow-up. Materials and methods. Paired pre- and post-embolization perfusion data from nine patients were analysed. TTP and MTT were compared using the paired Student’s t-test. An XGBoost (Extreme Gradient Boosting) classifier was trained to distinguish pre- from post-embolization states using TTP and MTT values measured at individual time points and evaluated by patient-grouped 3-fold cross-validation, with leave-one-patient-out cross-validation as a robustness check. Results. Embolization was associated with significant increases in TTP (from 4.58 ± 0.45 to 6.25 ± 0.43 s) and MTT (from 3.25 ± 0.30 to 4.61 ± 0.36 s; both p<0.001), indicating marked attenuation of perfusion within the embolized territory. At 12 months, significant clinical improvement was observed, including reduced spleen volume and increased platelet count (both p<0.001). The XGBoost model achieved an accuracy of 0.94 and an AUC-ROC of 0.94, with TTP as the principal discriminating feature. Conclusions. In this small proof-of-concept cohort, 2D perfusion angiography quantified embolization-related hemodynamic changes, and machine learning demonstrated preliminary feasibility for the automated discrimination of pre- and post-embolization perfusion states. Larger, externally validated studies with standardized acquisition protocols are required before clinical application.