Automatic concept extraction from biomedical material

Abstract

Natural Language Processing is a vibrant field of computer science that provides computers with the ability of understanding human language. In the field of medical data, there is a demanding need to lower the amount of documents clinicians and researchers need to manage in order to learn new concepts to improve their day-today practice. The research presented in this thesis aims at the design and evaluation of an algorithm based on neural networks that will extract the relevant entities from biomedical papers in order to reduce the amount of time needed for reading papers. Of all the topics in medicine that can take advantage of this thesis, the one it has been chosen in particular is the one of intervertebral discs. One of the reasons is the availability of experts on the topic in the current university. Moreover, it is a very interesting field as cells that form part of this structure have different properties based on their location. This makes it indeed a complex task to retrieve the relevant information because depending on the considered region some properties will be prominent whereas in other they might not be that relevant. The methodology used in the process it has been to use some off-the-shelf libraries already implemented in Java as a baseline and then use python to code a new architecture modifications to allow the algorithm to detect the relevant named entities. The results are compared with the gold standard obtained from the experts in the field and the conclusions are drawn from the observations.