Machine learning-based techniques in liver transplantation: a systematic review

Background and aims: Liver transplantation or hepatic transplantation is the replacement of adiseased liver with another person’s healthy liver. Liver transplantation is a treatment option forthose with severe conditions due to end-stage liver disease and also in cases of acute liver failure.Artificial intelligence and its primary subfield, machine learning, have begun to see widespreadapplication in medicine, including liver transplantation. Numerous machine learning (ML) modelsare proposed by researchers around the world to achieve this purpose. Our study aims to investigatethe application of machine learning techniques in liver transplantation.Method: A systematic search was carried out in the Medline (PubMed), Scopus, Web of Science(WOS), Cochrane databases using a combination of the key terms liver, transplantation and machinelearning and their alias from January ۲۰۱۰ to February ۲۰۲۳. Then, the title, abstract, andfull text of extracted articles were screened using the PRISMA checklist. Studies dealing with theapplication of ML algorithms coupled with liver transplantation were included. The informationregarding developed models was gathered from studies that were reviewed.Results: Of ۱۲۱۵ retrieved articles, ۸۹۷ studies were excluded after the title and abstract screening.Finally, ۴۱ articles were determined as eligible studies that met our inclusion criteria.Machine learning methods used include logistic regression (n=۹, ۱۴%), Support vector machine(n=۸, ۱۲.۵%) technique, neural networks (n=۷, ۱۱%), Random Forests (n=۷, ۱۱%), Decision Tree(n=۷, ۱۱%), Bayesian network (n=۶, ۹.۲۵%), linear regression (n=۶, ۹.۲۵%), Convolutional NeuralNetwork (CNN) (n=۴, ۶.۲۵ %), Gradient Boosting trees (XGBoost) (n=۳, ۴.۷۵%), KNN (n=۳,۴.۷۵%), K-means (n=۳, ۴.۷۵%), and Markov Model (n=۱, ۱.۵%). Most studies (n=۳۲) used morethan one machine learning technique or a combination of different techniques to make their models.In most studies, researchers succeeded in estimating post-transplantation survival rate (n=۲۱)or predicting recipient-donor matching in liver transplantation based on characteristics of donorrecipients and past experiences with liver donors and recipients (n=۱۰) by developing machinelearning algorithms. Additionally, UNOS was mentioned as the most preferable data source in thestudies that were reviewed. In most studies performance of machine learning algorithms was excellentand promising.Conclusion: The information currently available on liver transplants is huge. Machine learningcontext has led to the use of other non-traditional techniques more suitable, showing better performance,to conduct predictions regarding liver transplant data.

کلیدواژه ها:

machine learning ، artificial intelligence ، liver transplantation ، liver failure