Publication: XAI for convolutional neural network based on chest x-ray image classification
Loading...
Date
2024-07
Authors
Tang, Yoke Joo
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
This study investigates the integration of Explainable Artificial Intelligence (XAI) with Convolutional Neural Networks (CNNs) for classifying chest X-ray
images to enhance transparency and reliability in clinical contexts. Traditional CNNs, while effective in medical image analysis, face challenges due to their complexity and "black-box" nature, which limits trust and understanding. The specific objectives include developing an XAI framework for CNNs in chest X-ray classification, comparing different XAI models, and evaluating their performance in medical imaging. The methodology involves data collection and pre-processing to ensure quality, followed by training a CNN to recognize patterns in the data. XAI techniques like LIME, CAM, and SHAP were integrated into the trained CNN to provide interpretable results. The accuracy of the model was calculated through few testing. The explanation is the evaluate through four aspects, which is understandability, interpretability, accuracy and usefulness from a total of 12 respondents. In conclusion, the LIME model received the most positive feedback, with the highest understandability and interpretability score compared to SHAP and CAM, which is crucial for clinical acceptance. Although LIME's accuracy was slightly unstable with varied lung areas, satisfaction ratings were similar across models, showing minimal practical differences. Both LIME and CAM scored well in usefulness, outperforming SHAP, but overall satisfaction indicated a need for improved practical utility.