MODIFICATION OF AUCMEDI FRAMEWORK FOR THE APPLICATION OF POSTHOC BASED EXPLAINABLE ARTIFICIAL INTELLIGENCE IN MEDICAL IMAGE ANALYSIS

• Main Author: Ihsan Fadhiilah, Fikri
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/86196
• Institution: Institut Teknologi Bandung

Deep Learning shows remarkable performance in medical image analysis, such as object segmentation and computer-based diagnosis. However, deep learning models are often considered "black boxes" due to their complex and opaque decision-making processes. Explainable Artificial Intelligence (XAI) aims to explain model predictions, thereby increasing understanding and trust in their results. The AUCMEDI framework helps simplify medical image analysis with deep learning and standardize model evaluation by incorporating XAI. However, AUCMEDI is still limited in supporting the addition of new post-hoc XAI methods. Therefore, the XAI functionality in the AUCMEDI framework was modified to flexibly facilitate the addition of post-hoc XAI methods, with indications such as minimal changes to internal components when integrating post-hoc XAI methods and user freedom in configuring XAI methods. The modification process used a domain engineering approach and hot spot-based framework development by applying the strategy design pattern and the Separation of Concerns (SoC) design principle. The modification was successfully implemented and tested using unit testing and coverage testing, resulting in a coverage value of 90 percent from 700 lines of XAI functionality code. Additionally, changes to components when integrating other post-hoc XAI methods were significantly fewer compared to before the modification process.