Effect of Superpixel Aggregation on Explanations in LIME -- A Case Study with Biological Data

End-to-end learning with deep neural networks, such as convolutional neural\nnetworks (CNNs), has been demonstrated to be very successful for different\ntasks of image classification. To make decisions of black-box approaches\ntransparent, different solutions have been proposed. LIME is an approach to\nexplainable AI relying on segmenting images into superpixels based on the\nQuick-Shift algorithm. In this paper, we present an explorative study of how\ndifferent superpixel methods, namely Felzenszwalb, SLIC and Compact-Watershed,\nimpact the generated visual explanations. We compare the resulting relevance\nareas with the image parts marked by a human reference. Results show that image\nparts selected as relevant strongly vary depending on the applied method.\nQuick-Shift resulted in the least and Compact-Watershed in the highest\ncorrespondence with the reference relevance areas.\n