🏥 KARR Clinic – AI-Powered Skin Cancer Diagnosis

📌 Overview

KARR Clinic presents a solution that leverages AI to support early skin cancer diagnosis. This project uses a machine learning model that classifies skin cancer lesions as benign or malignant using NIH-provided image datasets and clinical data.

By integrating AI into the diagnostic workflow, we aim to assist clinicians in making faster and more accurate assessments, ultimately improving patient outcomes.

💡 Inspiration

We were inspired by the global challenge of timely and accurate skin cancer diagnosis. With early detection being key to survival, we saw an opportunity to build a tool that could empower healthcare professionals by combining the power of AI with real clinical data.

Skin cancer is often visually inspected, and misclassification can delay treatment. We wanted to explore how machine learning could aid in rapid and reliable detection—especially in under-resourced areas where dermatologists may not be readily available.

⚙️ What it does

KARR Clinic’s solution takes a dermoscopic image as input and classifies it as benign or malignant using a trained convolutional neural network (CNN). We also incorporate clinical metadata (age, gender, lesion location) to enhance the accuracy of the model.

Users can upload an image through a simple web interface, and the backend runs inference using our trained model to return results in real time.

🏗️ How we built it

Frontend: Developed using Next.js and styled with Tailwind CSS, providing a user-friendly UI for image upload and result display.
Backend: Built with Node.js and Express, hosting API endpoints that handle model inference requests.
Machine Learning: We used TensorFlow to train a CNN on the NIH skin lesion dataset, applying data preprocessing, augmentation, and binary classification.
Data: Images were normalized and resized. Clinical data was integrated using feature engineering and appended to model input.

🧗 Challenges we ran into

Data Quality: Cleaning and preprocessing medical data is tricky. We spent time balancing the dataset and removing mislabeled entries.
Integration: Ensuring smooth communication between frontend, backend, and the ML model took careful design and debugging.
Time Constraints: Building a full-stack AI application in a limited timeframe meant we had to prioritize features and stay agile.
Model Performance: Preventing overfitting while maintaining accuracy across lesion types required extensive tuning.

🏅 Accomplishments that we're proud of

Built a working AI model that can classify skin cancer lesions with strong performance.
Integrated clinical data into the model pipeline, improving its real-world utility.
Deployed a responsive, clean frontend that connects seamlessly to the backend.
Created a full-stack project that bridges machine learning and web development in a meaningful healthcare context.

📚 What we learned

The power of AI in healthcare—and the ethical responsibility that comes with it.
How clinical features can dramatically improve prediction models when used correctly.
Building an end-to-end system under a strict deadline requires clear planning and team collaboration.
Learned how to optimize image-based classification models for real-world applications.

🚀 What's next for KARR Clinic

🧠 Integrate model explainability with Grad-CAM or SHAP to visualize predictions.
☁️ Deploy to a cloud platform (e.g., AWS or Vercel) for real-world use.
🔒 Implement secure handling of medical data and user privacy (HIPAA compliance).
🧬 Expand to multi-class classification (e.g., differentiating between types of malignant lesions).
📱 Develop a mobile-friendly version to improve accessibility in remote areas.

Developed with 💙 by Team KARR for HackUSF. This project is for educational purposes only and is not intended for clinical use.