CAIPI in Practice: Towards Explainable Interactive Medical Image Classification

Would you trust physicians if they cannot explain their decisions to you? Medical diagnostics using machine learning gained enormously in importance within the last decade. However, without further enhancements many state-of-the-art methods are not suitable for medical application. The most important reasons insufficient data set quality and black-box behavior of algorithms such as Deep Learning models. Consequently, end-users correct model’s corresponding explanations. latter is crucial trustworthiness domain. research field explainable interactive searches that address both shortcomings. This paper extends CAIPI algorithm provides an interface simplify human-in-the-loop approaches image classification. enables end-user (1) investigate (2) prediction explanation, (3) influence quality. After optimization with only a single counterexample per iteration, model achieves accuracy $$97.48\%$$ on MNIST $$95.02\%$$ Fashion MNIST. approximately equal procedures. Besides, reduces labeling effort by $$80\%$$ .