Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- 188fc77f-4a63-45bf-a2b0-059e96b8a409
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a study to improve melanoma classification using computer vision, with Ardiansyah, Zulfikar Amri; Shidik, Guruh Fajar; and Soeleman, M. Arief as activity participants. The study addresses skin cancer, which affects the skin and may occur in benign or malignant cases. The activity was undertaken in the context of increasing skin cancer cases and the expanding application of computer vision in the medical domain.
The study builds on earlier research that explored convolutional neural network methods, including MobileNetV2. To improve the method, the activity employed the MobileNetV3 convolutional neural network architecture together with data augmentation techniques and the AdamW optimizer. The scope covered three melanoma datasets: the ISIC Archive, DermIS, identified as the Dermatology Information System, and MED-NODE. These datasets comprised both benign and malignant cases.
The reported finding was an improvement in classification accuracy compared with previous studies, which reported a maximum accuracy of 75%. The study reported accuracy reaching up to 97.5%. It also reported that the combination of MobileNetV3, augmentation, and AdamW optimization produced a 22.5% accuracy increase on the MED-NODE dataset. On this basis, the activity concluded that the selected convolutional neural network architecture, augmentation, and optimization approach had a substantial impact on melanoma classification accuracy.
Through this activity, Universitas Dian Nuswantoro contributed to health-related research concerning melanoma classification and the application of computer vision in the medical domain. The activity is aligned with the stated goal concerning health research.