University of Twente Student Theses
Developing Interactive System for Enhanced Communication and Feedback between Medical Experts and Explainable AI models
Zhao, Xun (2025) Developing Interactive System for Enhanced Communication and Feedback between Medical Experts and Explainable AI models.
This is the latest version of this item.
|
PDF
44MB |
| Abstract: | The adoption of Artificial Intelligence (AI) in medical imaging has the potential to enhance clinical decision-making, but the lack of transparency in AI models remains a major challenge. However, the lack of transparency in AI models, often described as their "black-box" nature, poses significant challenges to their adoption in clinical settings. Explainable AI (XAI) addresses this issue by providing interpretable outputs that help medical professionals understand and trust AI-generated predictions. This thesis focuses on the development and evaluation of an interactive XAI system designed to enhance communication between medical experts and AI models, specifically for hip fracture detection. The study employs a user-centered design approach and iterative evaluations to refine the system, leveraging input from radiologists and clinicians at hospital ZGT. Two AI models, YOLO and PIP-Net, are compared to assess their performance using metrics of classification accuracy, Intersection over Union (IoU) for the clarity of their explainable outputs. The findings highlight the strengths of the binary-class YOLO model in delivering accurate and clinically relevant results, while also addressing challenges such as data imbalance and annotation styles that affect model performance and user interface usability. By incorporating an expert-in-the-loop framework, the system enables a continuous feedback cycle where medical experts provide critical insights to refine AI outputs. Future research should focus on improving annotation consistency by reconciling varying annotation styles, integrating annotations into PIP-Net’s training process, and refining dataset composition to reduce imbalances. Additionally, expanding the system to support DICOM images and real-time interactive feedback will further align AI outputs with clinical needs. |
| Item Type: | Essay (Master) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 02 science and culture in general |
| Programme: | Interaction Technology MSc (60030) |
| Link to this item: | https://purl.utwente.nl/essays/105208 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page