Deep Learning for Skin Disease Diagnosis - [Resource]
Han Seung Seog recently published on reddit about a demo website that he created to diagnose 176 skin diseases. While he didn't provide any link to github or outside resources, I was able to find his repo.
So what is this all about?
His model is based on a deep learning algorithm and the architecture is comprised of ResNet152 and VGG19 - for a convolutional neural network. He points out that the model is the successor of one of his previous models, for onychomycosis.
Han Seung used 300,000 images to train the model, divided in 179 classes and 176 skin disorders. These images have been collect from 4 Korean University hospitals. As per Han Seung:
"The web-based test platform provides 3 differential diagnosis after analyzing image. Because there are many false positive diagnosis, the diagnosis predicted by the CNN should not be used as a confirmative diagnosis." [source]
If you want to use this for test purposes - not actual diagnosis - please make sure to take the photo in a bright good light without shades. I haven't tested it myself but I'd be curious to know if someone did and if it performed at least decent.
You can test the model by following the link below:
To stay in touch with me, follow @cristi
Cristi Vlad Self-Experimenter and Author
For more accurate results there need to be more variables taken into consideration other than image analysis. Using Machine learning is already a huge step and an excellent tool to make sense of data collection. Thanks for sharing
That's right. but if it works as 'marketed', it should serve as useful tool to doctors.
Definitely an excellent addition, no doubt there :)
post very useful for me, thank you @cristi, sharing posts very good, creative ideas in issuing posts very unique and useful, I really liked it, it made my mind more deeply.
what?