DhanushS-11/Cancer-Image-Classification-with-AutoML-Driven-Neural-Architecture-Optimization
A deep learning system that automatically designs optimal CNN architectures using Neural Architecture Search to classify lung - colon cancer from histopathology images. Achieves 99.72% accuracy across five cancer types with robust regularization. PyTorch-based solution ready for medical imaging deployment with exceptional generalization performane
What's novel
A deep learning system that automatically designs optimal CNN architectures using Neural Architecture Search to classify lung - colon cancer from histopathology images. Achieves 99.72% accuracy across five cancer types with robust regularization. PyTorch-based solution ready for medical imaging deployment with exceptional generalization performane
Code Analysis
11 files read · 3 roundsA 5-class lung/colon cancer histopathology image classifier using a simple random-search NAS over a small CNN architecture space, served via a Flask API with Grad-CAM heatmaps and hardcoded textual explanations, wrapped in a React frontend.
Strengths
The Grad-CAM implementation is correct and well-structured, and the React frontend is clean and functional with good UX patterns (drag-and-drop, loading states, error handling). The NAS pipeline, while simple, is a legitimate end-to-end implementation with proper logging, timing, and model checkpointing.
Weaknesses
The classification report is fabricated (duplicated classes, all 1.00 metrics), the 'textual explanations' are hardcoded strings unrelated to model reasoning, the README overstates the NAS as exploring 'thousands of architectures' when only 15 random configs are tested, and there are zero tests in the entire project. The accuracy numbers are inconsistent across files (99.04%, 96.3%, 100%), and the final trained model actually performed worse than the NAS-phase model.
Score Breakdown
Signal breakdown
Innovation
Craft
Traction
Scope
Evidence
Commits
8
Contributors
1
Files
47
Active weeks
2
Repository
Language
Python
Stars
2
Forks
0
License
—