This amazing AI app can detect cancer with 97% accuracy

Source: Deep Learning on Medium


One of the things I love about machine learning is how young the market is and how many great business opportunities are still available.

It was only 7 years ago that AI researchers started using this neat little trick called ‘convolution’ to teach computer how to see. We can now build apps that can perceive the world, analyze what’s in front of them, and help us do things we could never do before.

This opens up a brand new market for thousands of new apps and services, and millions of new customers.

You could literally be the one to launch the next killer app and become a world-leading entrepreneur!

All you need is a great idea.

So here is a great example to get your creative juices flowing.

SkinVision is an app created by a Dutch company that analyzes skin cancer from photographs taken with your smartphone.

Within 30 seconds after taking the picture, you receive a cancer risk assessment and in case of a high-risk, instant treatment options.

The app has a clinical detection accuracy of 97%.

To put this in perspective, GPs generally achieve 60% accuracy, dermatologists score 75%, and the very best dermal specialists achieve 92%.

This app blows them all out of the water. It performs on a superhuman level, better than the best medical specialists.

The SkinVision service has over 1 million users, and it has already successfully detected 27,000 high-risk skin cancer cases.

So how does it work?

Image analysis relies on Convolutional Neural Networks, or CNNs. Many industry-strength networks like VGG16 or ResNet are publicly available and can be retrained for any task.

The SkinVision team is probably using a Computer Vision pipeline that uses color filters, thresholding, blob filtering, and shape detection to analyze the photos.

The SkinVision team probably uses multiple CNNs, trained on thousands of medical cases, to analyze the shape, texture, and color of melanomas. Each CNN has a Dense Classifier at the end to produce a 3-way verdict: low, medium, or high risk.

It’s not rocket science.

You could have built this app yourself.

All it takes is dedication, lots of work, and a great idea to start with.

Do you have a great machine learning app idea? Write a comment and tell me about it!