Biblio

This work proposed a unified approach to increase the explainability of the predictions made by Convolution Neural Networks (CNNs) on medical images using currently available Explainable Artificial Intelligent (XAI) techniques. This method in-cooperates multiple techniques such as LISA aka Local Interpretable Model Agnostic Explanations (LIME), integrated gradients, Anchors and Shapley Additive Explanations (SHAP) which is Shapley values-based approach to provide explanations for the predictions provided by Blackbox models. This unified method increases the confidence in the black-box model’s decision to be employed in crucial applications under the supervision of human specialists. In this work, a Chest X-ray (CXR) classification model for identifying Covid-19 patients is trained using transfer learning to illustrate the applicability of XAI techniques and the unified method (LISA) to explain model predictions. To derive predictions, an image-net based Inception V2 model is utilized as the transfer learning model.

Most augmented reality applications connect virtual information to anchors, i.e. physical places or objects, by using spatial overlays or proximity. However, for industrial use cases this is not always feasible because specific parts must remain fully visible in order to meet work or security requirements. In these situations virtual information must be displayed at alternative positions while connections to anchors must still be clearly recognizable. In our previous research we were the first to show that for simple scenes connection lines are most suitable for this. To extend these results to more complex environments, we conducted an experiment on the effects of visual interruptions in connection lines and incorrect occlusion. Completion time and subjective mental effort for search tasks were used as measures. Our findings confirm that also in 3D scenes with partial occlusion connection lines are preferable to connect virtual information with anchors if an assignment via overlay or close proximity is not feasible. The results further imply that neither incorrectly used depth cues nor missing parts of connection lines make a significant difference concerning completion time or subjective mental effort. For designers of industrial augmented reality applications this means that they can choose either visualization based on their needs.