Every day algorithms take part in our lives in some way or together, and almost always do so to influence us sometimes – even negative-without us even noticing, so it's not bad when we see one with the potential to save lives. On Monday, researchers from Google's artificial intelligence department, together with a group of health researchers, announced a deep learning algorithm that could detect lung cancer with a 94.4% success rate.
The findings were published in the journal Nature Medicine, which noted that, apart from having a high accuracy rate, the algorithm could perform better than radiologists in some circumstances. According to the study, the system achieved that success rate in 6,716 cases received from the National Lung Cancer Screening Trial and similar accuracy in 1,139 independent clinical cases.
The researchers carried out two studies, one where a previous scanner was available and one that was not. In the first scenario, the deep learning algorithm – trained with CT scans of people with lung cancer, without cancer and with modules that had become cancerous – had an identification rate above six radiologists, and in the second, the results were between people and even an algorithm.
"This whole process of experimentation is as a student at school," said Dr Daniel Tse, Google's project manager, at the New York Times. "We use a large set of training data that will teach you to learn on your own what cancer is and when it will be cancer in the future." We did a final examination with some data I had never seen after spending a lot of time training it, and the final result was excellent. "
But the results of this study are the first steps of algorithmic identification. It's still far from detailed enough to be implemented widely in healthcare centers that perform cancer screening, but our start is promising. "A CT scan of a smoker's lung is so bad that it's hard to make it wrong," says Dr. Eric Topol, director of the Scripps Research Institute in California, told the New York Times.
Many technology companies, including Google, already use algorithms as detection tools in their large-scale platforms, particularly for moderation tasks. And these automated systems still fail very much. But researchers working on this technology to detect lung cancer recognize the risk of operating a system of this type without fully confirming its effectiveness and also ensuring t that control mechanisms to protect the system and its usage control
"We are working with organizations around the world to get an idea of how we can put this technology into practice in clinics in a productive way," said Dr Tse at the New York Times. "We don't want to rush this."
To be aware of more news, follow us on Twitter @GizmodoES