A Showcase for xAI
in Clinical Practice & Education
exAID enables medical professionals to understand and use AI's diagnosis as an aid in clinical practice. Additionally, the extensive model & data analysis helps future dermatologists & medical researchers to better understand patterns in data & developers to identify biases in the model.
- Conceptual & Textual Diagnosis Justification
- Visual Concept Localization
- Data Analysis
- Model Analysis
Features
exAID offers a Diagnostic Mode for clinical practice and an Educational Mode for educational & development purposes.
Diagnostic Mode
The Diagnostic Mode of the exAID framework provides accurate, AI-powered diagnostic predictions for many use cases in medicine. It is intended for the support of medical professionals in clinical settings.
exAID enables arbitrary AI models to intuitively justify their decisions using conceptual clues defined by medical professionals in easy-to-understand full-text explanations.
Precise Concept Localization Maps (CLMs) highlight the region the AI associates with a given diagnostic concept in the image. These localization maps support doctors in finding relevant diagnostic clues more efficiently.
Educational Mode
The Educational Mode allows to quantify dataset-wide concept influences of single concepts to the AI's decision, providing useful insights into the AI's reasoning processes.
The data analysis features allow practicing and prospective dermatologists to manually inspect complete data collections. Patterns in the data can be better understood through selective filtering, highlighting and sorting of images.
The framework enables Deep Learning Developers to thoroughly analyse the AI’s conceptual use throughout layers. Moreover, dimensionality reduction techniques help visualizing model’s representation of provided data at different stages.
Our Team
exAID is being developed by a team of researchers at German Research Center for Artificial Intelligence GmbH (DFKI). The use case of dermatoscopic lesion classification is developed with kind support from Department of Dermatology of Universitätsklinikum Münster and Department of Dermatology of Universitätsklinikum Düsseldorf.
Related Publications
Lucieri, A., Bajwa, M. N., Braun, S. A., Malik, M. I., Dengel, A., & Ahmed, S. (2020). On interpretability of deep learning based skin lesion classifiers using concept activation vectors. Proceedings of the 2020 International Joint Conference on Neural Networks (IJCNN 2020). IEEE, New York.
Lucieri, A., Bajwa, M. N., Dengel, A., & Ahmed, S. (2020). Explaining AI-based Decision Support Systems using Concept Localization Maps. Neural Information Processing (ICONIP 2020). Springer, Cham.
Contact Us
Feel free to contact us if you have any problems or questions concerning our framework or service.
Address
Trippstadter Straße 122,67663 Kaiserslautern,
Germany