CS-AF: A cost-sensitive multi-classifier active fusion framework for skin lesion classification

Convolutional neural networks (CNNs) have achieved the state-of-the-art performance in skin lesion analysis. Compared with single CNN classifier, combining results of multiple classifiers via fusion approaches shows to be more effective and robust. Since datasets are usually limited statistically biased, while designing an approach, it is important consider not only each classifier on training/validation dataset, but also relative discriminative power (e.g., confidence) regarding individual sample testing phase, which calls for active approach. Furthermore, analysis, data certain classes benign lesions) abundant makes them over-represented majority, some other cancerous deficient underrepresented minority. It crucial precisely identify samples from (i.e., terms amount data) minority class lesions). In words, misclassifying a severe or less should cost money, time even lives). To address such challenges, we present CS-AF, cost-sensitive multi-classifier framework classification. experimental evaluation, prepared 96 base (of 12 architectures) ISIC Challenge 2019 research dataset. Our show that our consistently outperforms both static competitors accuracy total costs.