Year: 2025
Type: dissertation
Abstract: Healthcare systems rely on the determination of International Classification of Diseases (ICD) codes for their operation.So far, medical code tagging required a specialist to read large amounts of unstructured medical text, making it a time-consuming and labor-intensive approach.Recently, many approaches have been proposed that involve using state-ofthe-art Natural Language Processing (NLP) techniques, to automate this process and reduce costs.Inspired by the success of Deep Learning and Classifier Chains -a mathematical model that takes into account label inter-dependencies -we present a novel approach that integrates these two methods in order to automate the multi-label classification of medical text.Our approach explicitly models label inter-dependencies, a key characteristic of ICD codes, and demonstrates its impact on improving classification performance.Using discharge summaries from the MIMIC-III database, we implement an application-driven pipeline combining advanced natural language processing (NLP) techniques with ensemble methods to optimize chain order.Comparisons with both traditional and advanced multi-label classification models demonstrate the practical benefits of our approach in terms of accuracy and computational efficiency.These findings validate the initial hypothesis and effectively connect machine learning research with practical, real-world applications.
Authors: Μαρούλη, Μαρία Γ., Μαρούλη, Μαρία Γ.
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7112585283
