Dialysis patients’ hospital readmission prediction using artificial neural networks

Popular Abstract

Using pictures of patient data to predict hospitalization

Predicting patient hospitalization has been a major objective of health care providers due to its impact on patients’ life quality. These benefits are demonstrated in driving down health care cost in addition to lowering mortality rates and complications due to unplanned hospitalizations.

Like any other patients, dialysis patients experience unplanned hospitalizations which endangers their life and increase their medical expenditure. Artificial intelligence, on the other hand, has shown its potential in different life aspects with remarkable results. Artificial intelligence models have already been integrated in various contexts in our daily life. This integration is evident in internet search engines, maps and translators, to name a few. To predict dialysis patients’ upcoming hospitalization events artificial intelligence models can be used as credible and accurate tools.

Several artificial intelligence methods have been developed which showed exceled performance in different fields. Here, we have chosen to work with a method called artificial neural networks which are inspired on the basic biological architecture of the human brains. Artificial neural networks comprise from neurons that can take several inputs into account, process them and produce an appropriate output. These neurons, and the network they comprise, can be trained based on already available data and applied to almost any circumstance.

One type of artificial neural network architecture has shown a transcendent performance in image analysis, and since its inception, it has come a long way in improving its capacity. In fact, this architecture surpassed humans in identifying different object in numerous given images. This model is called a convolutional neural network since its processes images by looking into a small part of the image at a time along the entire image. We have reasoned that if the patient data was presented to a convolutional network as an “image” its capacity could be harnessed in patient hospitalization prediction.

Due to a limited computational power, a relatively small, with limited data and non-complex convolutional network was used. This network showed it can match the prediction power of more traditional artificial neural networks with more data to investigate. This indicates that convolutional networks can be applied with comparable or better results to other artificial intelligence models when the data is presented in an appropriate manner (as an image). Consequently, applying a bigger, more complex, and with more data as inputs, convolutional networks can be successfully used in dialysis patients’ hospitalization prediction thus subsequently improving their life quality.


Advisor: Mattias Ohlsson
Master´s Degree Project 30 credits in Bioinformatics 2016
Department of Biology, Lund University

• author: Hatem, Gad
• supervisor: Mattias Ohlsson
• organization: Degree Projects in Bioinformatics
• course: BINP41 20161
• year: 2016
• type: H2 - Master's Degree (Two Years)
• subject: Biology and Life Sciences
• language: English
• id: 8891391
• date added to LUP: 2016-09-12 09:02:04
• date last changed: 2016-09-12 09:02:04