Skin Cancer Detection using AI
summary
Prototype app classifies ISIC 2018 lesion images using a Pix2PixGAN for background removal and a DenseNet-based CNN for benign vs malignant diagnosis, moving accuracy from ~80% to about 90–93% after adding segmentation and binary compression; models train on Azure VMs, convert via ONNX/Core ML and TensorFlow Lite, and run on-device with Xamarin packaging. Highlights: Pix2PixGAN background removal, DenseNet classifier, edge deployment (Core ML / TFLite).
AI-written summary · about
chapters (11)
- 0:00Introduction of presenters
- 2:02Hackathon project overview
- 2:48Project goals and scope
- 4:40Model development process
- 6:41Understanding skin cancer
- 9:00Initial pipeline and research
- 10:36Overcoming model challenges
- 14:25Segmentation and GANs
- 17:58Performance metrics analysis
- 20:56Final pipeline and deployment
- 23:52Q&A and concluding remarks
related talks
25mComputer Vision in Melanoma Detection
37mAI in medicine: where do we stand?
29mHow AI Is Transforming Medical Imaging
17mDermatology AI API for healthcare related platform
24mBreast cancer detection with IBM PowerAI Vision
29mCancer diagnostics enabled by machines: finding and classifying single tumor....
55mDeep Learning in Medical Image Diagnostics
01h 02mArtificial Intelligence for Early Detection and Diagnosis of Cervical Cancer