DermaScan

DermaScan homapage
skin picture upload
Skin disease prediction
Home remedies
Skin health score
DermaScan System architecture
AI ML architecture and flow

💡 Inspiration

Globally, millions of individuals experience persistent dermatological challenges such as acne, eczema, pigmentation, and dermatitis. Despite this prevalence, dermatological care remains largely inaccessible, costly, and inconsistent.

We identified a fundamental opportunity: Can artificial intelligence replicate a dermatologist’s diagnostic intelligence with real-time precision and accessibility?

That insight gave rise to DermaScan — an AI-driven dermatology and skin analytics platform that merges advanced computer vision with multimodal reasoning.

Our vision is to create a seamless intersection of medical-grade analysis and consumer-level usability, where AI provides clinically relevant insights through an interface designed for trust and empathy.

✨ What It Does

DermaScan transforms a standard smartphone camera into a real-time dermatological scanner. It captures facial imagery, performs computational analysis using deep learning models, and outputs a detailed skin health assessment, including diagnostic predictions, glow analytics, and personalized care recommendations.

Within seconds, users receive an interpretive report supported by AI reasoning and quantitative analysis.

🧠 Core Features

AI-Driven Dermatological Classification Convolutional Neural Network (CNN) models trained on annotated dermatology datasets detect early-stage conditions such as acne, eczema, and fungal infections. The model generates condition probabilities and evidence heatmaps for interpretability.

Facial Glow and Texture Analysis Implements tone histogram analysis, symmetry measurement, and Gray-Level Co-occurrence Matrix (GLCM) texture evaluation to derive a composite Glow Index, representing skin vitality and uniformity.

Adaptive Skincare Recommendation Engine Through integration with Google Gemini 2.5 Flash, DermaScan applies multimodal reasoning to produce tailored skincare and lifestyle recommendations based on image data, environmental context, and user profiles.

Skin Health Analytics Dashboard Provides a quantitative overview of dermatological health progression using time-series visualizations and anomaly detection algorithms.

Self-Learning Insight Layer Employs incremental model retraining on anonymized user data to improve generalization across demographics and lighting conditions.

🧩 How We Built It

Frontend (UI/UX)

Stack: React.js, Next.js, Tailwind CSS, Framer Motion
Architecture: Component-based modular design optimized for scalability and dynamic rendering
Design Philosophy: Minimal, clinical-grade interface with smooth transitions and low cognitive load
Core Features: Real-time image capture interface, live inference visualization, and asynchronous AI response rendering

Backend (Infrastructure)

Framework: Flask (Python) with RESTful API endpoints
Functionality: Handles inference requests, Gemini reasoning queries, and data persistence
Database: MongoDB Atlas for structured user profiles and historical scan logs
Security: JWT-based authentication and AES-256 encryption for health data

AI and Machine Learning Stack

Model Architecture:
- CNN model for disease detection trained on multi-ethnic dermatological datasets
- Image preprocessing pipeline using OpenCV for illumination correction, face region extraction, and normalization
- Texture and tone analysis using histogram equalization and spatial filtering
- Gemini 2.5 Flash integration for text synthesis, reasoning, and contextual insights
Frameworks & Libraries: TensorFlow, Keras, OpenCV, NumPy, Pandas, Scikit-learn
Optimization: Model quantization and TensorFlow Lite conversion for edge inference with minimal latency
Training Strategy: Data augmentation, dropout regularization, and adaptive learning rate schedulers for stability

Deployment Architecture

Frontend: Deployed on Vercel for SSR optimization and edge network caching
Backend: Hosted on Render with GPU-backed inference scaling
Database: Cloud-hosted MongoDB cluster with automated backups
CI/CD Pipeline: GitHub Actions for model versioning, API updates, and containerized deployment via Docker

⚔️ Challenges We Faced

Managing lighting and pose inconsistencies — solved using OpenCV-based photometric normalization and augmented training data
Ensuring dataset fairness across diverse skin tones using an expanded Fitzpatrick dataset
Reducing inference latency through model pruning and GPU-optimized threading
Designing a user interface that preserves clinical precision while maintaining consumer-level usability
Implementing stable integration with Gemini for context-driven NLP interpretation

🏆 Accomplishments

Developed a fully functional AI dermatology pipeline capable of live inference
Combined diagnostic modeling, glow scoring, and skincare recommendation systems into one cohesive framework
Built a deployable multi-layer architecture ready for telemedicine integration
Optimized inference latency below 500ms on mid-tier hardware
Achieved interpretability via Grad-CAM visualizations for transparent AI predictions

🎓 What We Learned

Medical AI requires rigorous fairness testing and ethical data governance
Combining AI-driven insights with empathetic UX significantly improves trust and adoption
Cross-domain collaboration between machine learning, design, and clinical knowledge accelerates innovation
Explainability and user interpretability are critical for medical AI applications

🚀 What’s Next

DermaScan is evolving into a complete AI dermatology and telehealth framework integrating diagnostics, consultations, and predictive analytics.

Upcoming Features

AI-assisted doctor consultations with automated Gemini health summaries
Predictive modeling for skin condition forecasting and progress analytics
Augmented reality overlays for live skin visualization
IoT-based environmental factor monitoring (UV, hydration, air quality)
Federated learning pipeline for privacy-preserving global model improvement

🌈 Vision

We envision AI dermatology as a ubiquitous, accessible, and clinically reliable service — bridging the gap between healthcare and personal wellness.

DermaScan aims to establish a global skin intelligence infrastructure, enabling anyone to understand, monitor, and optimize their skin health using computational analysis and AI reasoning.

This is not just diagnostic automation — it’s dermatological intelligence, re-engineered through data and empathy.