مقاله انگلیسی رایگان در مورد شبکه عصبی مصنوعی برای پیش بینی وضعیت عملکردی عضو

مقاله سال 2016

 مجله امگا – Omega

1. Introduction

 Organ transplants are one of the most viable treatment options for patients with organ failures and may also be their only option. Coupled with the cost of the operation and the lack of readily available organs, the need for pairings of organ donors and recipients which result in successful transplants is critical. However, datasets on donors and recipients may contain a vast amount of information, both in number of records and in attributes of the donor, recipient, and their relationship. The need to parse this data is therefore, acute and any single person attempting to perform a prediction may result in heavy bias as the consideration of the important attributes may be difficult, both due to their number and the way in which it may be difficult to determine which attributes contribute towards the outcome of the transplant. There exists a need, therefore, to not only perform accurate predictions on a complex dataset, but also to parse this dataset in some way so as to reduce it to a manageable form. Investigations into this have been performed using different approaches such as by Oztekin et al. (2011), who analyzed a lung transplant dataset using decision trees, and reinforced by studies as that by Zhuang et al.(2009) who demonstrate the effectiveness of data mining and machine learning for decision making by medical practitioners. Meisel and Mattfeld (2010) also supported this idea by identifying key areas in which operational research and data mining can work synergistically to create innovative approaches towards solutions for problems concerning decision making.

As the data here is ill understood, however, a prediction method that can cope with this lack of knowledge must be utilized. Artificial neural networks (ANN) are one tool that are capable of being trained on a dataset containing attributes of which relationships may be rather complex and are capable of performing accurate predictions on a testing set. ANNs might suffer from over-fitting, however, and may be sensitive towards data that contains conflicting observations, and therefore the dataset would be preprocessed by data envelopment analysis (DEA), a linear programming method for determining the relative efficiency of a set of observations. This would hypothetically allow a reduction in the dataset used for training the ANN without greatly impacting its performance.

ANNs consist of layers of neurons with distinct weights separating the neuron connections which allow the ANN to be trained on a complex dataset and determine its own understanding of the relationship of the attributes. The medical field is one particularly well-suited application of ANNs, for example the predictions of organ transplants by Dvorchik et al. (1996), cancer diagnosis by Abass (2002), or other clinical applications as shown by Dybowski and Gant (2001).