Electronic health records (EHRs) contain information about a patient's healthcare, such as medications and medical history. Natural Language Processing (NLP) has many applications and is increasingly used to improve various aspects of patient care, research, and healthcare administration. Also, Fuzzy logic and fuzzy models have found several applications in healthcare, particularly in situations where traditional binary logic may not be suitable due to the inherent uncertainty and imprecision in medical data. To utilize the information from EHRs, the new combination model is designed to complete the medical notes classification task and provide trustworthiness interpretations. The whole procedure is described below: First, two multi-sense word embedding models are applied to extract valuable information from medical notes and complete the classification task, which is compared with Machine Learning (ML) and Deep Learning (DL) models. In the second stage, after reviewing the fuzzy logic system applications in the medical area, the main gaps in research are identified and the new combination model is generated. Finally, the proposed combination model provides the interpretations based on the classification results, which is compared with the results from the latest feature extraction models. There are three experiments conducted to make sure the multi-sense word embedding models are capable of assigning more weights to one word and classifying the text with good performance; identifying the combination point with the fuzzy logic system to improve the trustworthiness of the interpretation; evaluating the interpretation results with various levels, such as word, sentence, and document level. All the datasets are adopted from public open resources, such as Kaggle, which do not require ethical review. In this research, the proposed model which is combined with multi-sense word embedding model (PFT) and Fuzzy logic system, completed the medical text classification task and provided the trustworthiness interpretation. Furthermore, this thesis delivers additional knowledge to the existing literature through the experiments conducted on reliable datasets and the model's quality while also suggesting future improvements. This work also addresses the controversy regarding whether human factors should be involved in the medical data evaluation phase.
Date of Award | 15 May 2024 |
---|
Original language | English |
---|
Supervisor | Anju Johnson (Main Supervisor), Minsi Chen (Co-Supervisor) & Grigoris Antoniou (Co-Supervisor) |
---|