New AI tool detects melanoma with 99% accuracy

New AI tool detects melanoma with 99% accuracy, outperforming existing models.

Why it matters

  • Early detection of melanoma can save lives.
  • AI could assist doctors in diagnosing skin cancer more accurately and quickly.

By the numbers

  • 99.01% accuracy rate on ISIC 2020 dataset.
  • Outperformed four other models, with the closest competitor at 98.2% accuracy.
  • Trained on over 10,000 images from HAM10000 and 33,000 images from ISIC 2020.

The big picture

  • The tool combines two deep learning models to segment and classify skin lesions.
  • Potential to expand to other cancers like breast or lung cancer.
  • Future plans include automating the process and integrating patient health records.

What they're saying

  • Some users question if the AI outperforms human pathologists.
  • Concerns about false positives and overdiagnosis.
  • Optimism about the potential for early cancer detection.

Caveats

  • The system currently requires manual steps between models.
  • Further validation is needed against human pathologists and in real-world settings.

What’s next

  • Development of a fully automated system.
  • Integration of patient health records to improve accuracy.
  • Potential expansion to detect other cancers.