Rationale Extraction and Crohn’s Disease Detection from Computed Tomography Enterography Reports

  • Author / Creator
    Dai, Jiayi
  • Building predictive models with higher predictive performance is the common pursuit in text classification tasks. In almost all domains of text classification problems, the current state-of-the-art models (e.g., Bi-LSTM and BERT) are based on deep neural networks that learn language representations with deep and sophisticated neural architectures. However, the lack of interpretability limits the real-world applications of these deep models, especially in life-critical domains. The desire for the interpretability of neural networks that aims to provide predictions along with explanations has been rapidly emerging. Rationale extraction, which is one best practice of explainable artificial intelligence (XAI) in building explainable neural classifiers, learns with only instance-level supervision to identify discriminative features as explanations for predictions; it can be applied in the medical domain to provide explainable disease diagnostic predictions. The scope of the dissertation is on rationale extraction and predictive models, with a focus on detecting Crohn's disease from CT enterography radiology reports. Specifically, the work of the dissertation: 1. explores rationale extraction as a tool for knowledge acquisition from CT enterography reports, 2. introduces IBDBERT, an inflammatory bowel disease (IBD)-specific BERT large language model, which achieves the state-of-the-art classification accuracy in detecting Crohn's disease from CT enterography reports in comparison to CNN, Bi-LSTM and both generic and domain-specific BERT models and 3. constructs the first ensemble architecture of rationale extraction by imitating human interaction.

  • Subjects / Keywords
  • Graduation date
    Fall 2023
  • Type of Item
  • Degree
    Master of Science
  • DOI
  • License
    This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.