Scientists teach AI to diagnose brain cancer

In a new research effort, the international team of scientists found that the machine learning algorithm could be a powerful tool for medicine. The group published their work in Nature, training and training successful AI in identifying different brain tumors with the most impressive accuracy.

To determine between different types of brain cancer, the team needs some criteria that computers can use to distinguish between them. With more than 100 types of brain tumors already in the medical record, the identification process can be very difficult even for a human doctor.

The researchers used a DNA process called methylation as a biological "fingerprint" and taught AI the difference between those "fingerprints" with specific types of cancer.

Like all machine learning algorithms, AI needs a basic knowledge to compare. The team said the figures for 2,800 cancer patients are the starting point, allowing 91 different signs of the tumor to be identified. They then asked the computer to determine the type of tumor in more than 1,000 known samples and found that AI's judgment was inconsistent with human diagnosis in some cases.

However, not only does the computer predict incorrectly in identification; doctors have misdiagnosed about 12% of previous studies, and AI is correctly diagnosed. The impressive accuracy of this system makes it a powerful tool for the medical community.

The researchers wrote: "For a broader approach, we designed a free online sorting tool, its use does not require any additional data processing. Our results provide a plan for the creation of AI-based tumor classification facilities using machine learning to learn about different cancer entities, including the underlying pathological changes. tumors ".