This use case cover a Skin cancer classification into 7 types . A skin cancer is a very extensive studied and important dissease in the healt care world with millions of affected people. This type of AI diagnostic allow to p revent and detect possible skin cancer in patients , helping doctors to classify and detect easily it.
Dataset
We use the Skn Cancer MNIST dataset from kaggle. This dataset is unbalanced as we can see in the nex image:
We apply a data augmentation to the lower classes to try to balance all them and have a minimum of images per each class, the distribution after increase the number of images of minority classes are represented in the following plot:
You can access to the new dataset using perceptilabs github.
Model
| Layer | Configuration |
|---|---|
| Input Layer | |
| MobileNetV2 | include_top=false, pretrained=imagenet |
| Dense | Activation=Relu, Neurons=512 |
| Dense | Activation=Softmax, Neurons=7 |
| Output Layer |
Workspace
Statistics View
Accuracy plot