Home » Skin Lesion Classification: Transformation-based approach to CNNs

Skin Lesion Classification: Transformation-based approach to CNNs

flow.png

We create a stacked CNN classifier for Skin Lesions that takes multiple transformations of the input data during training and evaluation. Our method achieves ~97% accuracy fused from three networks performing at ~93% accuracy. This increase is more dramatic for sensitivity going from ~78% to ~90%.

GitHub

https://github.com/charlieLehman/isic_cnn

Related Publication

C. Lehman, M. Halicek, G. AlRegib, “Skin Lesion Classification: A Transformation Based Approach to Convolutional Neural Networks”, submitted to IEEE International Workshop on Machine Learning and Signal Processing (MLSP2017), Tokyo, Japan, September 2017

%d bloggers like this: