Abstract
Melanoma skin lesion classification and segmentation are two critical tasks for computerized diagnostic systems. These tasks are challenging due to the inherently fuzzy nature and diverse shape variations of skin lesions in dermoscopy images. A joint learning approach has the potential to significantly improve the outcomes of both tasks. This is because there is a strong correlation between skin lesion classification and segmentation. This paper proposes a novel unified multi-task learning approach that facilitates joint segmentation of skin lesion boundaries while simultaneously classifying skin lesion abnormalities. The proposed method incorporates a new regularization technology called joint learning, which promotes mutual benefit between the two tasks by learning shared features and preventing the dominance of one task over the other. The proposed method was evaluated on a publicly dataset called ISIC 2016, involving two different types of skin cancers, benign and melanoma. The experimental results showed that the proposed multi-task learning method has the potential to enhance the overall diagnosis performance of skin cancer analysis. The proposed method achieved an F1 score of 79.94% compared to the single-task learning model's performance of 78.26% while preserving the segmentation capabilities. This study represents a significant stride toward potential generalizability under diverse clinical applications.
| Original language | English |
|---|---|
| Pages (from-to) | 321-325 |
| Number of pages | 5 |
| Journal | IET Conference Proceedings |
| Volume | 2023 |
| Issue number | 11 |
| DOIs | |
| State | Published - 2023 |
| Externally published | Yes |
| Event | 2023 International Conference on Green Energy, Computing and Intelligent Technology, GEn-CITy 2023 - Hybrid, Iskandar Puteri, Malaysia Duration: 10 Jul 2023 → 12 Jul 2023 |
Bibliographical note
Publisher Copyright:© The Institution of Engineering & Technology 2023.
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 3 Good Health and Well-being
ASJC Scopus subject areas
- General Engineering
Fingerprint
Dive into the research topics of 'Joint Segmentation and Recognition of Melanoma Skin lesions via Multi-Task Learning'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver