Artificial intelligence, trained by Google, detects lung cancer more precisely than doctors

During the diagnosis, the model developed by the company, allowed to detect 5% more cases of the disease.

Artificial intelligence, trained by Google, detects lung cancer more precisely than doctors
Artificial intelligence, trained by Google, detects lung cancer more precisely than doctors.


Over the past three years, Google's teams have been using artificial intelligence in the health sector - from eye disease diagnostics to cancer detection. The company described one of its studies, which showed that AI can detect lung cancer more accurately than physicians. This is stated in a message on the company's website.

According to the World Health Organization, more than 1.7 million people die every year from lung cancer, making this cancer the most deadly of all types of the disease worldwide, more than breast cancer or prostate cancer, and is the sixth most widespread the cause of death in the world. The company emphasizes that intervention is more successful if the disease is detected at an early stage, but statistics say that in most cases, cancer is diagnosed at later stages.

In the last decade, screening methods have been invented to identify people at high risk for lung cancer. Google has focused on improving this technology.

In 2017, the company began to explore how to solve existing problems using AI technologies. Using advances in 3D modeling, Google has made progress in modeling lung cancer prediction, as well as in creating the basis for future clinical trials.

Radiologists usually look at hundreds of 2D images in one CT (computed tomography), but cancer can be so small that it's hard to detect. Google reports that it has been able to create a model that can not only generate a general forecast of cancer of lung cancer but also identify thin malignant tissues in the lungs (nodes). The model can also take into account information from previous scans that is useful for predicting lung cancer since the growth rate of suspicious nodes may indicate its destructive nature.

The model created by specialists of the company was trained on the data of surveys using computer tomography (in total - 45 856 cases). The results were verified on the second set of data and compared to the effects of six certified professionals.

Using data from only one survey, AI found 5% more cases than doctors, with a simultaneous decrease in the number of false positives of more than 11%.

At the moment, the company is negotiating with partners all over the world to continue research and deploy clinical validation.