Modeling of pre-transplantation liver viability with spatial-temporal smooth variable selection

Abstract

Background and objectiveLiver viability assessment plays a critical role in liver transplantation, and the accuracy of the assessment directly determines the success of the transplantation surgery and patient's outcomes. With various factors that affect liver viability, including pre-existing medical conditions of donors, the procurement process, and preservation conditions, liver viability assessment is typically subjective, invasive or inconsistent in results among different surgeons and pathologists. Motivated by these challenges, we aimed to create a non-invasive statistical model utilizing spatial-temporal infrared image (IR) data to predict the binary liver viability (acceptable/unacceptable) during the preservation. MethodsThe spatial-temporal features of liver surface temperature, monitored by IR thermography, are significantly correlated with the liver viability. A spatial-temporal smooth variable selection (STSVS) method is proposed to define the smoothness of model parameters corresponding to different liver surface regions at different times. ResultsA case study, using porcine livers, has been performed to validate the efficacy of the STSVS method. The comparison results show that STSVS has the better overall prediction performance compared to the past state-of-the-art predictive models, including generalized linear model (GLM), support vector machine (SVM), LASSO, and Fused LASSO. Moreover, the significant predictors identified by the STSVS method indicate the importance of edges of lobes in predicting liver viability during the pre-transplantation preservation. ConclusionsThe proposed method has the best performance in predicting liver viability. This ‘real-time’ prediction method may increase the utilization of donors’ livers without damaging tissues and time-consuming, yet imprecise feature assessment.