DermaLanding
Research
Hackathon Project

SkinDeep

SkinDeep is a bias-free dermatology triage tool built at a hackathon. It addresses health equity by using a diverse skin-type dataset (Fitzpatrick 17k) to reduce racial bias in AI dermatology. Features lighting quality checks, confidence scoring, and community image donation for fairer model training.

Tech Stack

5 technologies
Frontend
1
Streamlit

Fast, responsive web interface

Backend
2
Python

Core backend and ML pipeline

Pandas

Data handling for Fitzpatrick 17k dataset

AI / ML
2
PyTorch

Transfer learning with EfficientNet for skin lesion classification

NumPy / Pillow

Image processing and histogram analysis for lighting checks

Key Features

Skin condition recognition trained on diverse skin types

Lighting quality check for accurate diagnosis on darker skin

Transparent confidence scoring

Community image donation for fairer AI training

Built and deployed within 36-hour hackathon

Stack Summary

Frontend
Streamlit
Backend
PythonPandas
AI / ML
PyTorchNumPy / Pillow