Artificial Intelligence beats specialists in cancer diagnosis

A study in the US has found that an artificial intelligence based system was able to outperform specialist doctors at diagnosing lung cancer from imaging.

There are still many things that artificial intelligence can't do well, but it's perfectly suited to pattern recognition and classification problems such as image analysis. Now a team of researchers at Google working with Northwestern University in Illinois have produed a study showing that AI can outperform not just the average person but also specialist doctors at disgnosing lung cancer from imaging.

The researchers used a self-learning system and trained it on real images of over 42,000 lung CT scans from almost 15,000 patients. The images were pre-classified into two groups: Those that went on to develop cancer, and those that didn't. With just these facts, the AI was able to quickly develop a model to determine which group a supplied image fell into.

After training the AI, the group put it head-to-head with a team of six expert radiologists who specialised in analysing CT scans, and the AI turned out to be more effective than the experts. It was also able to classify better with a single CT scan than a human doctor, who would need follow-up scans to reach the same accuracy.

One major benefit of having an AI perform these tests is that it can be done quickly and automatically, flagging up suspected cancer in scans to doctors so that those patients can be seen with higher priority. The results are expected to increase cancer detection rates by around 5% and reduce false positives from interpreting CT scans by 11%. The results were published in Nature Medicine, and the next step for the technology would be to use it in a controlled clinical trial.

Source: BBC News