Concept-Attention Whitening for Interpretable Skin Lesion Diagnosis

arXiv:2404.05997v1 Announce Type: new
Abstract: The black-box nature of deep learning models has raised concerns about their interpretability for successful deployment in real-world clinical applications. To address the concerns, eXplainable Artificial Intelligence (XAI) aims to provide clear and understandable explanations of the decision-making process. In the medical domain, concepts such as attributes of lesions or abnormalities serve as key evidence for deriving diagnostic results. However, existing concept-based models mainly depend on concepts that appear independently and require fine-grained concept annotations …

abstract applications artificial artificial intelligence arxiv attention box clear clinical concept concepts concerns cs.cv decision deep learning deployment diagnosis domain explainable artificial intelligence intelligence interpretability making medical nature process type world xai