Towards Domain-Specific Explainable AI: Model Interpretation of a Skin Image Classifier using a Human Approach

Abstract

Machine Learning models have started to outperform medical experts in some classification tasks. Meanwhile, the question of how these classifiers produce certain results is attracting increasing research attention. Current interpretation methods provide a good starting point in investigating such questions, but they still massively lack the relation to the problem domain. In this work, we present how explanations of an AI system for skin image analysis can be made more domain-specific. We apply the synthesis of Local Interpretable Model-agnostic Explanations (LIME) with the ABCD-rule, a diagnostic approach of dermatologists, and present the results using a Deep Neural Network (DNN) based skin image classifier.

Survey of explainable artificial intelligence techniques for biomedical imaging with deep neural networks

“…A DNN was used as skin image classifier and LIME with the ABCDrule for model explainability [212]. The ABCD-rule of dermoscopy is known to outperform other methods [212].…”

“…A DNN was used as skin image classifier and LIME with the ABCDrule for model explainability [212]. The ABCD-rule of dermoscopy is known to outperform other methods [212]. For explainability, the LIME model was fused with a previously introduced human medical algorithm by the authors [212].…”

“…The ABCD-rule of dermoscopy is known to outperform other methods [212]. For explainability, the LIME model was fused with a previously introduced human medical algorithm by the authors [212].…”

LIMEcraft: handcrafted superpixel selection and inspection for Visual eXplanations

“…Based on (Stieler et al, 2021)'s work, it can be presumed that changes such as rotation and shift of the skin lesion should not change the prediction of the model. On the other hand, changing the color and boundaries may change the features of the lesion.…”

Detecting Deepfake Images Using Deep Learning Techniques and Explainable AI Methods

“…The authors of the paper [33] proposed a toolbox called "Neuroscope" that addresses the issue by providing state-of-the-art visualization algorithms as well as freshly modified methods for semantic segmentation of CNNs. One of the approaches [34] proposed a domain-specific explanation system for skin image analysis. The authors applied Local Interpretable Model-Agnostic Explanations (LIME) and presented the results using a Deep Neural Network (DNN) based image classifier, but for the field of deepfake, LIME is still unused.…”