Early hemodynamic assessment and longitudinal imaging for predicting graft outcomes post-transplant

ziqian wu, Songxiu Li, Siyu Ouyang, J Hu, Yuan Qiang, Jie Ding, Qiu Guo, Jidong Gao, Wei Wei, Ke Ren

The early identification of early allograft dysfunction (EAD) and the long-term prediction of graft-related adverse event-free survival (GRAEFS) are crucial for effective post-transplant management. The study encompassed two complementary analyses: (1) an early hemodynamic assessment, which evaluated the diagnostic efficacy of hepatic artery ultrasound parameters for the identification of EAD; and (2) a longitudinal imaging prognostic analysis, which developed radiomics and Vision Transformer Base/16-based (Vit b/16) nomograms utilizing Delta features extracted from serial grayscale ultrasound images at six key time points relative to day 1, for the prediction of 3-year GRAEFS. The performance of the models was externally validated and compared with clinical baseline models. A hepatic artery resistive index (RI) below 0.55 within 4 days post-transplant is linked to EAD, showing a high negative predictive value (92.2%) for ruling out EAD. The Vit b/16 nomogram excelled in predicting 3-year GRAEFS, with AUCs of 0.951, 0.847, and 0.831 in training, internal, and external validation cohorts, respectively, surpassing the radiomics nomogram. In external validation, it outperformed the best clinical model (AUC: 0.831 vs. 0.808, <i>p</i> = 0.048) with a net reclassification improvement of 0.307 (<i>p</i> = 0.028). Both models were stable in sensitivity analyses excluding tumor recurrence. Interpretability showed the radiomics nomogram emphasized tissue heterogeneity, while the Vit b/16 nomogram focused on liver margin and intrahepatic vascular structures. This study shows that a drop in the hepatic artery RI soon after surgery is linked to EAD, and a Vit b/16-based nomogram from ultrasound images effectively predicts 3-year GRAEFS. These results need prospective validation before clinical use. Early post-transplant, the hepatic artery shows a ‘high flow, low resistance’ state, with a resistance index below 0.55 within 4 days serving as a potential early graft dysfunction indicator.Both the radiomics model and the Vit b/16 deep learning model effectively predicted graft-related adverse event-free survival, with significant differences in survival curves (all Log-rank tests <i>p</i> &lt; 0.05).Model analysis reveals that radiomics nomograms prioritize the ‘Entropy’ feature for tissue heterogeneity, while Vit b/16 nomograms emphasize the liver’s edge and intrahepatic vascular structures. Early post-transplant, the hepatic artery shows a ‘high flow, low resistance’ state, with a resistance index below 0.55 within 4 days serving as a potential early graft dysfunction indicator. Both the radiomics model and the Vit b/16 deep learning model effectively predicted graft-related adverse event-free survival, with significant differences in survival curves (all Log-rank tests <i>p</i> &lt; 0.05). Model analysis reveals that radiomics nomograms prioritize the ‘Entropy’ feature for tissue heterogeneity, while Vit b/16 nomograms emphasize the liver’s edge and intrahepatic vascular structures.