Frequently Asked Questions
What is ScarsAI?
ScarsAI, or the Skin Cancer Automatic Recognition System, is a CNN image classification model created with to classify images of moles into either benign or malignant skin cancer.
Or, in other words, it takes in images of moles and attempts to predict whether they are harmful (malignant) or not (benign).
Or, in other words, it takes in images of moles and attempts to predict whether they are harmful (malignant) or not (benign).
How accurate is ScarsAI?
ScarsAI in training is about 85% accurate, but in real-life application accuracy can vary depending on quality of input information. The dataset it was trained on is relatively restrictive for the goal of the model.
What should the input image be?
The input image to the model should be a close-up picture of human skin. The model works best if the image has good resolution and is taken in a well-lit environment. Also try and minimise background noise. Users on mobile devices can directly take a picture from the website using the device's camera. The model does not function if there is no mole or lesion present in the image.
What is the model trained on?
The model is trained on this dataset from the ISIC archive.
Where can I find the source code to this project?
This project is open source, so all the code is avaliable on this GitHub page. The authors can be found on the About page.
Can this website be used in a medical environment?
It is strongly recommended that this website never be used in a real medical setting. The model is not accurate enough to properly diagnose skin cancer. If you believe you have skin cancer, please consult a doctor for proper diagnosis.