pathoscope-AI

web interface for anaylsis

🚀 Inspiration

The project was inspired by the urgent need for accessible and rapid histopathology analysis. Traditional diagnosis of cancer via microscopic tissue slides is time-intensive and often subjective. As pathology workloads increase, there's a critical need to support professionals with tools that accelerate, standardize, and explain results—especially in low-resource settings. Patho-Scope AI was born from this gap.

🔍 What It Does

Patho-Scope AI is an intelligent diagnostic workbench that:

Accepts histopathology images (lung and colon tissues)
Classifies them into one of five trained categories:

  - Colon Adenocarcinoma
  - Colon Benign
  - Lung Adenocarcinoma
  - Lung Benign
  - Lung Squamous Cell Carcinoma

Displays a confidence score
Uses Google Gemini to explain results in human-readable language
Automatically logs all analysis data to a MongoDB-based “lab journal” for easy review and audit

🛠️ How We Built It

Dataset: 25,000 labeled histopathology images from Kaggle (Lung & Colon Cancer Histopathological dataset)
Training: Used Google Colab with TensorFlow/Keras to fine-tune a DenseNet121 model. Model saved in .keras format.
Backend: Python Flask
Frontend: HTML/CSS/JavaScript, fully responsive SPA
AI Explanation: Google Gemini API generates interpretive summaries
Database: MongoDB Atlas stores predictions, explanations, and images (as Base64)

⚠️ Challenges We Ran Into

Training 25K images efficiently on limited compute; resolved via Google Colab
Achieving fast, low-latency inference with TensorFlow in a Flask environment
Balancing simplicity with completeness in UI design
Connecting frontend with AI inference backend smoothly using vanilla JS

🏆 Accomplishments That We're Proud Of

Seamlessly integrated image classification and AI-driven explanation
Created a fully functional web-based diagnostic interface
Achieved consistent classification performance with real-time feedback
Stored all logs in a searchable, persistent "lab journal"
Made the system mobile-responsive and user-friendly

📚 What We Learned

Practical deployment of TensorFlow models in production environments
Leveraging LLMs like Gemini for medical explanation
Efficiently converting and storing image data as Base64
Designing for fluid diagnostic workflows used by actual medical professionals