Skip to content
A tool which implements Convolutional Neural Networks to train a predictive model to classify skin lesions.
Python HTML JavaScript
Branch: master
Clone or download
Latest commit 064024b Apr 28, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
tfjs-models/Cancer Add files via upload Dec 15, 2018
LICENSE Create LICENSE Apr 28, 2019 Update Dec 19, 2018
cancer_classes.js Add files via upload Dec 10, 2018 Add files via upload Dec 10, 2018
predict.js Add files via upload Dec 10, 2018 Add files via upload Dec 10, 2018
skinpredict.html Add files via upload Dec 10, 2018

Skin Lesion Classifier

A predictive model that uses the HAM10000 dataset, trained on MobileNet to classify skin lesions into seven categories. Our model has been deployed on a web app.

We have trained the model locally using native Keras, then we converted it to TensorflowJS, allowing for a smooth web deployment. Our model's accuracy score settles around 82%, this is mainly due to a small and unbalanced dataset.

Our training process is documented in the preprocessing and model python files. The TFJS files, Python files, as well as our web deployment files are all accessible here.

Our model was trained on a Paperspace P5000 Notebook; each epoch ran for roughly 15 minutes, totalling to 2 hours and 30 minutes for 10 epochs.


  1. (Marsh's Skin Lesion Analyzer) - []
  2. (HAM10000 Dataset) - []
  3. (Skin Lesion Classification) - []

Blog post on this:

You can’t perform that action at this time.