Melanoma is the most dangerous form of skin cancer, which can be treated from early detection. Often, the skin can display moles with abnormal coloring/shape which may or may not be a sign of Melanoma.
What it does
This program is trained on preliminary Melanoma and normal skin images. The user can just upload a picture of his/her skin segment and get fast feedback on whether or not he/she should see the doctor.
How I built it
This program uses unsupervised Machine Learning to classify the skin images. The images are classified through a linear kernel Single Vector Machine.
Challenges I ran into
Machine Learning was very difficult to implement on images that are so similar. It was hard to find public and normalized data sets to perform the analyzes
Accomplishments that I'm proud of
With only 100 images I was able to get the Machine Learning algorithm to perform at 70% accuracy. This rate will improve as the image processing methods can be fine tuned and much more images can be added to the training data set.
What I learned
I learned how to use sklearn to implement machine learning. This is the first time I implemented an image classification method. I learned a lot about image processing in python.
What's next for SkinDoc
Once I can get opencv installed on a remote server this can be easily deployed to a remote server. SkinDoc needs more training data sets to improve its accuracy.