This app uses a tensorflow machine learning model (see https://github.com/DefineDan/ml-skin-cancer-classification-model) converted to a tf-lite model to classify skin lesions in 7 different cancer categories. The app allows the user to upload an image from gallery or to take a picture with the camera and classify it. The app will display the type and the confidence as well as info about the diagnoses.
Currently no images are stored, since it is very sensible data.
- crop (e.g. fixed zoom, edge detection, user by crop), and edit (e.g. white balance) incoming images, to extract lesion information only
- give advice on how to take image (e.g. camera mask)
- display pobability distribution for other classes
- tweek tflite model prediction parameters
- close tf model after longer inactivity
- overall design
- more/better info for diagnoses
One could add:
- account and login
- history of uploaded images
- api for health care providers