Toward Explainability of Machine Learning in Medical Imaging: Generalizability, Separability, and Learnability
Open AccessThe applications of Deep Learning (DL) for medical imaging have become increasingly popular in recent years. During my studies of applications of Machine Learning (ML) and DL methods in medical imaging, I realized that there is a trade-off between accuracy and explainability for these methods. Although some DL methods have better performances, they are more difficult to understand and to explain. The lack of explainability limits the acceptance of DL applications by clinicians. The requirement of explainability and the DL applications for medical imaging that I have investigated thus have stimulated my research interest in eXplainable Artificial Intelligence (XAI). Explainability has multiple facets, and there is to date no unified definition. For explainable ML, I have primarily addressed these aspects: the separability of data, cluster validation, Generative Adversarial Network (GAN) evaluation, generalizability of the Deep Neural Network (DNN), learnability of DL models, and transparent DL. The study of explainable ML had been motivated by several completed applications for medical-object detection and segmentation. Studies of medical image analysis and the XAI contain very rich questions. My research aims to contribute to medical image analysis by focusing on the performance (accuracy) and explainability of applications, using ML and DL. The long-term goals of these works are to help make DL-based Computer-Aided Diagnosis (CAD) systems be transparent, understandable, and explainable and to win the trust of end-users; eventually, these new techniques can be widely accepted by clinicians to improve medical diagnosis and treatment outcomes.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.