Auto-Micro

🔬 Project Overview

Auto-Micro is a modular AI platform for automatic microscopy image analysis. It processes raw microscope or whole-slide images (WSI) and delivers fast, explainable predictions through a microservice architecture.

The system is designed to:

• Handle high-resolution medical images (e.g., histopathology, microbiology)
• Support on-premise deployment with no cloud dependency
• Provide end-to-end automation: from image ingestion to model inference
• Enable real-time interaction through a React frontend and REST APIs

Auto-Micro combines deep learning, MLOps, and Kubernetes orchestration to streamline digital pathology workflows.

🧫 Clinical Background

Microscopy-based diagnostics are critical in detecting infections, cancer, and tissue abnormalities. However, manual examination of slides is:

• Time-consuming
• Prone to human error
• Inconsistent between experts

Auto-Micro supports clinicians and researchers by offering a scalable, reproducible, and interpretable AI pipeline for analyzing complex image data.

🎯 Research Objective

Auto-Micro is designed to:

• Automate the preprocessing and patch extraction from large WSI files
• Train and serve deep learning models for classification and segmentation tasks
• Provide an interactive dashboard to monitor data and models
• Ensure traceability and reproducibility for clinical research
• Run completely on local infrastructure with Kubernetes (Minikube)

🧠 Core Features

• Microservices Architecture
  Independent FastAPI services for data handling, training, and inference

• On-Prem MLOps
  Continuous integration, training, and model deployment inside local clusters

• Explainable AI
  Grad-CAM visualizations and patch-level prediction summaries

• Modular Storage with MinIO
  Efficient handling of WSIs, patch sets, annotations, and model artifacts

• Interactive Frontend
  Built with React for uploading, monitoring, and viewing model results

🧪 Methodology

The pipeline consists of four main stages:

1. Data Acquisition & Processing

• Real-time upload from microscopes or bulk import of WSIs
• Patch extraction with configurable tile sizes

2. Model Training

• Models built in PyTorch with support for CNNs and Vision Transformers

3. Inference & Deployment

• Best-performing model selected based on validation metrics
• Deployed as an API endpoint within Kubernetes using FastAPI
• Responses returned in real time with heatmaps and class scores

4. Visualization & Monitoring

• Web-based dashboard with React
• Shows patch predictions, Grad-CAM overlays

🛠️ Technologies Used

• PyTorch, Scikit-learn, OpenCV
• FastAPI, Docker, Kubernetes (Minikube)
• MinIO for object storage
• React frontend for interaction and visualization

💡 Why Auto-Micro?

This project showcases:

• Real-world MLOps in an on-premise setting
• Microservice engineering for scalable AI pipelines
• Expertise in medical image processing and WSI handling
• Strong emphasis on explainability, compliance, and modularity