In this project you will learn to measure features in images of skin lesions, and predict the diagnosis (for example, melanoma) from these features in an automatic way. You will do things like:
- Use Github to maintain your Python code
- Explore the dataset
- Segment images by hand
- Implement methods to measure image features
- Explore and transform these features
- Use the features with machine learning classifiers to predict the lesion diagnosis
- Perform experiments to evaluate different parts of your method
- Write a report about your findings in LaTeX
A dermatologist in your city asks if you can investigate whether some features of skin lesions can be reliably measured by image analysis algorithms. The dermatologist is concerned about news about AI, and explicitly wants to use a simple and interpretable algorithm.
Your goal is to do a small investigation, where you process the images to measure different types of features, train machine learning classifiers to predict the lesion diagnosis, and report on your results.
You must also deliver a trained classifier, that the dermatologist will evaluate on external images, that you do not have access to.