a Two-stage Strategy for Skin Cancer Classification Based on Dermoscopic Images
- Resource Type
- Conference
- Authors
- Kang, Xiaoyu; Li, Heng; Chen, Jianpin; Chai, Xinyu
- Source
- 2022 3rd International Conference on Computer Science and Management Technology (ICCSMT) ICCSMT Computer Science and Management Technology (ICCSMT), 2022 3rd International Conference on. :110-113 Nov, 2022
- Subject
- Computing and Processing
Deep learning
Computer science
Analytical models
Computational modeling
Transforms
Data models
Skin
skin cancer classification
data imbalance
deep learning
classification strategy
- Language
Data imbalance in skin lesions datasets are common problems of skin cancer classification tasks based on deep learning. This study proposes a two-stage classification strategy based on an advanced benchmark multi-classification model, which transforms the "multi-category" task into a "main class and non-main class binary" task and a "multi-category within non-main classes" task to smooth the class distribution of skin cancer datasets. This strategy can significantly improve the final classification performance compared to the direct multi-classification task, and can provide a reference for solving similar data imbalance problems.