SkinIntel

SkinIntel
Result
Search for Hospital and Cancer Center
Upload Image
Overview

Inspiration

Skin cancer is one of the most common and yet highly treatable forms of cancer—if detected early. However, access to dermatologists is limited in many areas, and even when available, early warning signs can be overlooked. We wanted to bridge this gap by combining the power of AI with clinically proven diagnostic frameworks to provide accessible, accurate, and interpretable skin lesion analysis to anyone, anywhere.

What it does

SkinIntel is an AI-powered platform designed to detect skin cancer from dermatoscopic images using advanced deep learning techniques integrated with the ABCDE rule of dermatology. It provides:

Real-time skin lesion classification (benign vs. malignant)
Detailed ABCDE breakdown for transparency
Personalized risk assessment based on user data
Recommendations for next steps
A healthcare provider locator with optimized routing to the nearest dermatologists or cancer centers

How we built it

We built SkinIntel using a hybrid tech stack:

Frontend: Vite + React + Tailwind CSS for a responsive, accessible UI
Backend: FastAPI in Python to manage API endpoints and data flow
AI/ML Models: PyTorch and TensorFlow to implement CNNs (ResNet-50, Inception-v3) and U-Net for segmentation
Data: Trained on HAM10000, ISIC, NIH, and proprietary clinical images
Integration: Google Maps API for real-time healthcare locator, Dijkstra's algorithm for routing
Deployment: Dockerized for smooth local and cloud deployment

Challenges we ran into

Data Quality & Balance: Managing class imbalance in skin lesion datasets required careful preprocessing and augmentation.
Model Interpretability: Ensuring the AI's decisions were explainable to both clinicians and non-experts was a core focus that took extensive iteration.
UI Accessibility: Meeting WCAG 2.1 standards while keeping the interface modern and intuitive was a delicate balance.
Integration Complexity: Integrating various components like AI models, survey logic, Google Maps API, and healthcare routing was technically challenging but rewarding.

Accomplishments that we're proud of

Achieving 91.3% accuracy, 93.2% specificity, and 0.944 AUC-ROC, outperforming industry benchmarks
Successfully merging clinical heuristics (ABCDE) with deep learning for enhanced interpretability
Designing a user-friendly interface that works seamlessly across all devices
Developing a fully working prototype with real-time lesion analysis and specialist locator

What we learned

AI alone is not enough—interpretability, accessibility, and user trust are equally important in healthcare applications
Real-world datasets present real-world challenges—messy, imbalanced, and varied—but they also lead to more robust models
Cross-disciplinary collaboration between AI developers, UX designers, and medical professionals yields more meaningful and usable solutions

What's next for SkinIntel

Roadmap

Mobile App Launch: Develop native iOS and Android versions to increase accessibility EHR Integration: Enable direct integration with electronic health records for seamless clinical workflows Global Reach: Add multi-language support and regional skin type calibration for broader usability Time-Series Analysis: Track lesion evolution over time for enhanced diagnostic insights Telehealth Features: Connect users directly with dermatologists for virtual consultations Privacy-First Learning: Implement federated learning to improve model accuracy while preserving user privacy Expand Use Cases: Extend AI support to detect other dermatological conditions like psoriasis, eczema, and acne API Marketplace: Launch a developer API ecosystem to support third-party healthcare and research integration