Various Medical Aspects of Liver Transplantation and its Survival Prediction using Machine Learning Techniques

Abstract

Objective: Prognosis models play a significant role in forecasting the patient's survival in Organ transplantation. To review the impact of machine learning methods in predicting the of survival of patients who undergoes Liver Transplantation using a Multilayer Perceptron Artificial Neural Network model with an extensive discussion of all the medical aspects is the key objective of this paper. Methods/Analysis: Medical practitioners studied various parameters during pretransplantation for predicting the survival of a patient. This study considered those parameters and reviewed whether these parameters have any vital part in the survival rate of a patient after Liver Transplantation (LT). This study also compared various scores including Model for End Stage Liver Disease (MELD) score, Emory score and Child score that are used in survival prediction. Currently the medicinal specialists estimate the outcome of LT with MELD score. We employed a detailed learning about the health aspects of LT and various machine learning techniques used in this area. In order to perform the experimentation, the dataset was congregated from the United Network for Organ Sharing transplant database (n = 65534). With the three layer architecture, the model trains the attributes of donors, recipient and transplantation using back propagation algorithm. 10-fold cross validation was applied in each training and test set before training. During the training process, the appropriate donor-recipient pairs were found out and obtained the best liver patient survival in transplantation. Findings: We conducted a comprehensive study about LT for the liver patient survival prediction. We proposed a Multilayer Perceptron Artificial Neural Network model to predict the survival rate after LT with 99.74% accuracy using United Network Organ Sharing registry. We also compared the performance of proposed model with existing models and proved that proposed model produced more accuracy than other models. Novelty/Improvement: The multilayer perceptron model succeeded clinical scores in terms of high accuracy, sensitivity and specificity. Machine learning techniques show better performance than conservative numerical methods in donor, recipient and transplantation attributes which are used to predict the survival. Due to less expensive and producing reliable solutions with rich datasets, machine learning techniques have been succeeded the conventional statistical methods and medical scoring systems. The proposed model predicts a promising accuracy for the prediction of best survival rates after.