Sign in Subscribe
Artificial intelligence

Edge AI Teledermatology for Efficient Skin Disease Diagnosis

Strategic Insights

— 1 min read

Researchers have proposed a novel framework for edge-based inference using knowledge distillation on mobile devices, addressing the challenges of deploying high-performance AI models in teledermatology. The framework, which applies knowledge distillation to compress a high-performance teacher model into a small student model, has been evaluated on the ISIC 2019 and Fitzpatrick17k-C datasets, demonstrating its potential for efficient inference on mobile devices with limited resources. The study's findings highlight the practicality of the proposed framework, with a compression rate of 55.88 on the ISIC 2019 dataset and an inference time reduced by approximately 352 times.

Key Takeaways:

  • Researchers from the Computer Science Department at Binus University have developed a novel framework for edge-based inference using knowledge distillation on mobile devices for teledermatology.
  • The framework applies knowledge distillation to compress a high-performance teacher model into a small student model, enabling efficient inference on mobile devices with limited resources.
  • The proposed framework has been evaluated on the ISIC 2019 and Fitzpatrick17k-C datasets, demonstrating its potential for efficient skin disease diagnosis.
  • Key contributions include a novel framework for edge-based inference, evaluation of pre-trained models on the specified datasets, and practical deployment on mobile devices.
  • The experiments demonstrate that the scenario of distilling RegNetY32GF into MobileNetV2 results in the most efficient model, maintaining both accuracy and model size.
  • The prototype evaluation shows the practicality of the proposed framework with a compression rate of 55.88 on the ISIC 2019 dataset and an inference time reduced by approximately 352 times.
  • The research concludes that the proposed framework enables efficient inference on mobile devices with limited resources, making it a promising approach for teledermatology.

Statistics:

  • Compression rate: 55.88 on the ISIC 2019 dataset
  • Inference time reduction: approximately 352 times
  • Number of datasets used for evaluation: 2 (ISIC 2019 and Fitzpatrick17k-C)
  • Number of pre-trained models evaluated: 1 (RegNetY32GF)
  • Number of student models generated: 1 (MobileNetV2)
  • Model size: maintained accuracy and size using knowledge distillation

Sources:

  • Diverse Representation Knowledge Distillation for Efficient Edge AI Teledermatology in Skin Disease Diagnosis (IEEE Access, 2025)
  • Computer Science Department, Binus University (Faculty website)
  • NewsRx LLC (News Report, July 11, 2025)
  • IEEE (Publisher of IEEE Access journal)
  • Binus Graduate Program (Faculty website)
  • DOI: 10.1109/ACCESS.2025.3581225 (Free version of the journal article)

Read more