ScanAID

• 

  init

• 

  processed

• 

  upload

• 

  submited

Inspiration

Breast cancer is a terrible problem with a large dataset, permitting it to be addressible. Cancer can be easily handled through early detection, but difficulty of wait teams mean this may be infeasible for many. AI can fix this.

What it does

Processes medical images - for now, breast cancer - and returns a quick medical advice form, highlights the cancer, and then gives a full report on top of this

How we built it

Challenges we ran into

• Efficiently handling data
• Finding the right data
• Model hallucinations
• Speed

Accomplishments that we're proud of

• Actually developing a demo able POC
• Efficient data augmentation and information structuring.

What we learned

• Never surrender
• Medical images are hard to understand without sufficient finetuning.
• Deceptive complexity

What's next for ScanAID

• More diverse data training'
• Integrating attention mechanisms for better focus
• Integrating video processing

Built With

• llm
• mistral
• multi-modal-ai
• pixtral
• python