Abstract:
This thesis, titled "Explainable Artificial Intelligence Techniques for Clinical Decision Support Systems," investigates the integration of explainable artificial intelligence (XAI) to enhance the transparency and usability of deep learning models in clinical decision support systems (CDSS) in order to make well-informed decisions. Clinical Decision Support Systems (CDSS) play a crucial role in enhancing healthcare delivery by aiding in the accuracy and efficiency of medical decisions. However, the reliance on deep learning models introduces complexities due to their 'black box' nature, which obscures their decision-making processes and hinders user trust and understanding. This thesis addresses these challenges by implementing explainable artificial intelligence (XAI) techniques, focusing on transforming black box models into interpretable 'white box' models. Specifically, decision trees are utilized as surrogate models, which clarify the decision-making process by linking deep learning model outputs with their corresponding input features in a transparent manner. This approach not only maintains the high predictive accuracy of deep learning models but also enhances their interpretability. By employing feature and subset selection techniques, the research further refines the interpretability, ensuring that only the most significant features contribute to decision-making, thereby simplifying the model without sacrificing performance. The integration of these XAI techniques into CDSS bridges the gap between the advanced capabilities of deep learning and the necessity for transparency, making these systems more accessible and trustworthy for healthcare professionals. This thesis demonstrates that through the use of XAI, it is possible to retain the benefits of deep learning while providing clear, understandable insights into the models' operations, fostering greater acceptance and reliability in clinical settings. Additionally, this research highlights the potential for XAI to facilitate ongoing improvements and updates to CDSS, ensuring that these systems evolve in line with emerging medical knowledge and practices, further enhancing patient outcomes and healthcare efficiency.
