OmniScanner

Jupyter Notebook hosted from Intel Tiber

Project Story

Inspiration

AI adoption is accelerating rapidly, and we saw this as an urgent opportunity to apply it to cancer research. OmniScanner harnesses Intel Tiber Cloud’s AI accelerators to deliver rapid, web-based screening for multiple cancer types using a custom CNN.

What We Learned

Deep Learning & Imaging: Fine-tuned a pre-trained CNN on medical scans; implemented real-time augmentation for varied view angles.
Cloud AI & Optimization: Leveraged Intel oneAPI for sub-second inference on Tiber Cloud GPUs.
Full-Stack Development: Built a TypeScript Next.js + shadcn/ui frontend; created a Django REST backend.
DevOps & Collaboration: Automated deployments via GitHub Actions; mediated regular code reviews and CI checks to keep the team in sync.

How We Built It

Frontend
- Next.js App Router (TypeScript) & Tailwind via shadcn/ui
Backend
- Django REST Framework + Simple JWT for /api/scan/
AI Inference
- Deployed and ran code using Intel Tiber servers to optimize runtime
- Used pre-processing techniques such as tensors, rotations, and dimensionality reduction like PCA to fine-tune and increase dataset complexity

Challenges

Data Quality: Limited annotated scans required robust augmentation and transfer learning.
Performance: Achieving high F1 and accuracy scores with the confusion matrix.
Time Constraints: With only 10 hackathon hours, we focused on stomach cancer first and stubbed additional types. Training a CNN is a time intensive process.
Team Coordination: Synchronized frontend, backend, and model work through peer code reviews and enforced CI.

Despite tight timelines, OmniScanner proves that cloud-native AI can democratize early cancer detection and empower clinicians worldwide.