How Machine Learning Can Expedite Melanoma Detection

There are around 5 million new cases of melanoma every year. The deadly skin cancer is hard to keep up with, and the occurrence rate has grown immensely over the last 30 years. Researchers at IBM’s T.J. Watson Research Center have been working to expedite melanoma detection and diagnosis by using machine learning to analyze dermascopy images. Ultimately, they intend to help dermatologists identify melanoma faster, so that they can prevent it from spreading to other parts of the body.

Watson and the International Skin Imaging Collaboration (ISIC) are partnering to form a community alliance that encourages researchers to contribute and grow to Watson’s technology. The ISIC has a large open source archive of images that serves as a resource for dermatologists to advance their diagnostic skills, and allows developers to create and test algorithms for skin cancer triage. Watson and ISIC are hosting challenges for researchers to submit algorithms and techniques for melanoma detection. Since 2015, they’ve received 125 submissions from various universities and other institutions that have used the archive to create innovative approaches for skin cancer diagnosis. The challenges follow a basic protocol:

1.  Distinguish the lesion from the normal skin.
2.  Determine “clinically interesting patterns” that physicians typically look for when diagnosing a lesion.
3.  Identify the lesion’s disease.

Watson’s machine learning technology then pools these different techniques together to generate different combinations for specific tasks. “It’s a combination of deep-learning, hand-crafted features, unsupervised learning, and many different techniques put together to address this challenge,” Watson researcher Noel Codella told medGadget. “We see in our studies and others that this type of technique always seems to outperform the use of single methods alone. And the optimal combination of those techniques change depending on the task at hand.”