Cancer Classifier
Published:
Skin Cancer Early Detection Classifier
This project addresses the urgent need to reduce the burden on healthcare systems and improve patient outcomes through early skin cancer detection. By enabling patients to proactively monitor their skin health via photographic assessments and questionnaires, we aim to significantly shorten the diagnosis timeline and facilitate quicker medical interventions, potentially decreasing mortality and morbidity associated with late-stage skin cancer diagnoses.
Data Source
We utilized the Skin Cancer (PAD-UFES-20) dataset from Kaggle, which contains both image data and patient information.
Approach
Our preliminary skin cancer assessment combines two parts:
- A questionnaire
- A photo assessment
This comprehensive approach provides more information for assessing skin cancer risk compared to using the classifier alone.
Models Used
For text data:
- Support Vector Machine (SVM)
- Random Forest (RF)
- XGBoost (XGB)
- K-Nearest Neighbors (KNN)
- LightGBM (LGBM)
For image data:
- ResNet50
- VGG16
- Big Transfer (BiT)
- ConvMixer
Web Application
We deployed our model as a web application, allowing users to upload their images and answer the questionnaire to receive a risk assessment.
Experience the web app: Skin Cancer Classifier Web App