Dermo

Your Personal Dermatology Smart Assistant -Get instant, accurate skin, mole, and cancer analysis with personalized recommendations from our AI-powered dermatology platform.

Comment

Inspiration

Dermo was inspired by the growing need for accessible and accurate skin lesion diagnosis tools. With skin cancer being one of the most common types of cancer worldwide, we recognized that unlike regular doctor or dentist checkups, dermatologist visits are often reactive rather than preventive, with many individuals delaying or avoiding consultation due to cost, inconvenience, or uncertainty about whether their condition requires professional attention. While existing AI models in this space are often monetized services with limited accessibility and questionable accuracy, we aimed to create a more trustworthy alternative by first addressing the lack of comprehensive, publicly available datasets—creating the largest publicly available skin disease dataset—and then developing an accurate, accessible platform that helps individuals determine whether they should seek further professional medical diagnostics, encouraging timely consultation when necessary while reducing unnecessary visits for benign conditions.

What it does

Dermo is an advanced skin lesion diagnosis platform that allows users to choose between specialized models based on their specific needs:

  • Mole Analysis Model: Specifically designed to analyze moles and predict whether they are cancerous or benign, helping users quickly determine if they should seek professional medical attention for potential skin cancer.
  • Comprehensive Skin Condition Model: Provides a detailed analysis of various skin conditions, identifying 22 different skin disorders with high accuracy. This model, trained on our curated largest publicly available skin disease dataset, offers a broader diagnostic capability for users with general skin concerns.

Both models operate through a user-friendly web interface that:

  • Accepts easy image uploads for quick analysis
  • Provides clear, understandable results with confidence scores
  • Offers educational information about detected conditions
  • Delivers both macro and micro-level analysis of skin lesions
  • Provides severity ratings for condition urgency
  • Recommends appropriate clinical treatments based on medical guidelines
  • Suggests safe at-home treatments for managing symptoms while waiting for professional care
  • Includes follow-up recommendations based on the severity and type of condition

How we built it

Dermo is built using a modern, full-stack architecture:

Backend

  • Custom Dataset Creation: Developed and curated the largest publicly available skin disease dataset with 22 different skin conditions (>380,000 images)
  • Python-based ML Pipeline: Core system built with Python, leveraging PyTorch for training a model on over 380,000 data points, the largest publicly available data set for dermatology.
  • Multiple Specialized Models: Implemented separate models for mole analysis and comprehensive skin condition diagnosis depending on user needs
  • FastAPI Framework: Built a robust API layer with automatic documentation and type checking, allowing communication between front and backends
  • Advanced Image Processing: Used OpenCV and Albumentations for sophisticated image preprocessing and augmentation
  • Treatment Recommendation Engine Powered by Groq: Implemented the Meta llama-3.3 model through Groq to generate treatment recommendations to patients based on likelihood and ailment diagnosis.
  • Severity Assessment System: Created a standardized rating system to evaluate the urgency of skin conditions

Frontend

  • React with TypeScript: Modern, type-safe frontend for a responsive and intuitive user experience
  • Tailwind CSS: Utility-first CSS framework for a clean, responsive design
  • Vite Build System: Fast development and optimized production builds
  • Real-time Processing: Immediate image analysis and result visualization
  • Secure File Handling: Safe and efficient image upload and processing
  • Interactive Results Display: Clear presentation of diagnoses, severity ratings, and treatment options
  • Treatment Information Cards: Visually appealing cards showing clinical and at-home treatment options

Infrastructure

  • Cloud Deployment: Backend deployed on Render, frontend on Vercel
  • Database Management: PostgreSQL for production, SQLite for development
  • API Integration: Connected to ISIC dataset and medical knowledge bases
  • Scalable Architecture: Designed to handle multiple concurrent users efficiently

Challenges we ran into

  1. Data Quality: Ensuring models work across different skin types and lighting conditions
  2. Dataset Creation: Curating and organizing the largest publicly available skin disease dataset
  3. Computing Resources: Training models on massive datasets with limited computational power
  4. Medical Validation: Aligning predictions with medical standards while maintaining accuracy
  5. Integration: Connecting multiple ML models with the web interface seamlessly
  6. File Management: Maintaining and hosting large medical image files efficiently with minimal latency

Accomplishments that we're proud of

  • Created the largest publicly available skin disease dataset
  • Developed a robust multi-model system for comprehensive skin lesion analysis
  • Achieved high accuracy in preliminary diagnoses
  • Created an intuitive and accessible user interface
  • Successfully integrated medical knowledge base with AI predictions
  • Built a scalable architecture that can handle real-world usage
  • Contributed to the advancement of AI in dermatology by providing valuable training data
  • Successfully trained a custom model on our comprehensive dataset, achieving state-of-the-art performance

What we learned

  • Advanced techniques in computer vision and deep learning
  • Best practices in medical AI system development
  • Importance of data preprocessing and augmentation
  • Effective ways to combine multiple ML models
  • Web development with modern frameworks and tools
  • Medical image processing and analysis
  • API design and integration
  • User experience considerations for medical applications
  • Data curation and organization for medical imaging
  • Ethical considerations in medical AI development
  • Distributed computing and model optimization techniques

What's next for Dermo

  1. Dataset Expansion: Adding more diverse skin types and conditions with automated quality checks
  2. Model Improvements: Implementing transfer learning for rare conditions and optimizing for edge devices
  3. Medical Integration: Partnerships with healthcare providers and integration with telemedicine platforms
  4. Infrastructure Optimization: Implementing edge computing and enhancing caching systems for better performance
  5. UI/UX Enhancements: Refining the user interface with accessibility features, personalized dashboards, and interactive educational content

Built With

Share this project:

Updates