Stanford CS 229: Machine Learning

Taught by Andrew Ng


Class project: Vision-Based Classification of Skin Cancer Using Deep Learning

Developed a deep convolutional neural network (CNN) using TensorFlow that acheives 78% balanced accuracy for Melanoma Classification. Raw skin lesion images are pre-processed and lesion segmentation is automated using OpenCV prior to being fed into the CNN. The CNN is a fine-tuned, VGG-16 network, thats pre-trained on the ImageNet datset. The last convolutional layer is trained on the ISIC 2016 dataset which classifies images of skin lesions as one of two classes: malignant or benign for Melanoma.

Check out my Final Project Paper