AI Alzheimer Analyzer: ConvNeXt CNN with MLflow MLOps

🧠 This project delivers an end to end medical AI diagnostic system for Alzheimer's disease detection using brain MRI scans, combining computer vision with MLOps and CI/CD practices for production-ready healthcare applications.

🧠 Transfer learning is done based on Meta AI's ConvNeXt (2022 SOTA CNN), a Vision Transformer-inspired architecture using pure convolutions. It is fine-tuned for rapid adaptation to 4-class Alzheimer's staging (Non Demented, Very Mild Demented, Mild Demented, Moderate Demented).

🧠 Progressive unfreezing with staged learning rate reduction for safe transfer learning. Automatic Mixed Precision (AMP) is enabled for efficient training resource. Robust MLOps integration via MLflow and Databricks enables experiment tracking, model version control, and API model serving.

🧠 Multiple model formats are supported and underwent benchmark for their inference speed. Available formats are PyTorch (native), ONNX (for optimized cross-platform inference), and MLflow (for standardized deployment). Besides that, Grad-CAM heatmaps are applied for explainable AI, highlighting brain regions influencing predictions to build clinical trust.

🧠 Lastly, a complete CI/CD workflow via GitHub Actions are enabled for automated unit testing and deployment to Hugging Face Spaces. The Gradio web app provides intuitive interaction with ONNX Runtime for fast inference and Databricks API as fallback.

Tech Stack:
- Language & Framework: Python, PyTorch, ONNX Runtime, Gradio
- Platform: Kaggle, Databricks, Hugging Face
- MLOps: Databricks, MLflow (experiment tracking, model registry, API serving)
- CI/CD: Github actions (for pytest, Hugging Face Spaces)