Search

Kidney transplantation (KT) is an optimal treatment for end-stage renal disease (ESRD). Currently, short-term KT outcomes are indeed excellent, but long-term successful outcomes are still difficult to achieve, and improving them is crucial for kidney recipients. An early and accurate prediction of long-term graft survival helps healthcare practitioners to create a more personalized treatment plans for patients and facilitates the performance of clinical trials. In this study, we propose a machine learning framework to early predict graft survival after five years of KT and determine the most influential parameters that affect the survival. Our dataset was collected from Charles Nicolle Hospital in Tunis in Tunisia and it included pre, peri, post KT aspects. We utilized four machine learning algorithms to select the most important features: the least absolute shrinkage and selection operator logistic regression (Lasso-LR), Random Forrest (RF), Decision Tree (DT), and Chi-square (Chi-sq). We utilized three Scikit-learn functions to implement those algorithms: SelectFromModel (SFM), Recursive Feature Elimination (RFE), and SelectKBest (SKB). Five algorithms were utilized to builds prediction models based on the data groups resulted from the feature selection step: logistic regression (LR), k-nearest neighbors (KNN), extreme gradient boosting (XGB), and artificial neural network (ANN). We evaluated the models using five performance measures: accuracy, sensitivity, specificity, F1 measure, and area under the curve (AUC). XGBoost resulted the best model with the highest AUC (89.7%). It was based ten features selected by RF algorithm and SFM function. The accuracy, sensitivity, specificity, and F1 of the best model were 91.5%, 91.9%, 87.5%, and 89.6%, respectively. This study proposes a novel approach for investigating long-term allograft survival while considering the complex relationship between all KT aspects and long-term outcomes. Our framework can be used as a decision support system for Nephrologists to early detect graft status, which helps in developing safer recommendations for kidney patients and consequently obtaining positive KT outcomes and mitigating the risks of graft failure.

Any biomedical device requiring power from a source other than the human body or gravity is considered an active device. Currently available active biomedical devices encompass an enormous variety of technologies, ranging from large imaging machines to miniature implantable stimulators. These devices are vulnerable to cybersecurity threats, especially for devices capable of communication with an internet network. An attack exploiting these vulnerabilities can cause a variety of consequences, including data theft, denial-of-service, and serious patient harm. The chapter provides a comprehensive review of cyberattacks on biomedical devices in a hierarchical layered framework (e.g., sensing, communication, and control) with three specific attacks as case studies: (1) MRI unit-based attack, (2) infusion pump-based attack, and (3) implantable medical device attack.