Google AI Predicts Lung Cancer Better Than Radiologists

Google engineers have developed a new AI program capable of diagnosing lung cancer in patients more accurately than most human doctors.


In a study featured in the journal Nature Medicine, researchers trained a deep learning program to detect the malignancy with a success rate of 94.4 percent.

While Google AI is still considered a work in progress, it offers a brief glimpse of what the technology holds for the future of medicine.


By feeding AI programs with large amounts of data, the technology can be trained to identify different medical conditions that would otherwise be too difficult or too time-consuming for human doctors to detect.

These artificial neural networks make use of algorithms to help fine-tune their searches and allow them to learn as they do their work. The more information these AI are fed, the better they become at diagnosing illnesses.

So far, Google has already developed deep learning systems to help ophthalmologists diagnose eye disease in diabetes sufferers, as well as help pathologists, examine microscope slides to identify malignancies.

"We have some of the biggest computers in the world," said Dr. Daniel Tse, a researcher at Google and one of the authors of the study.

"We started wanting to push the boundaries of basic science to find interesting and cool applications to work on."


In the new study, Tse and his colleagues fed Google AI with CT scans of people suspected of having lung cancer.

Doctors have long recommended the diagnostic to detect cancer risks for people who have a long history of smoking. Screening has also been shown to help lower the risk of lung cancer deaths, according to previous research.

CT scans can identify definite cancers in patients, but they can also locate specific spots that are at risk of developing into malignancies. Doctors can use this information to group their patients based on their cancer risks and determine whether they need follow-up tests to monitor cancer growth.

However, these medical tests have their limitations. Not only can they miss tumors in patients, but they can also misread benign spots for cancerous ones. This can lead to doctors recommending highly invasive and risky procedures such as biopsies or surgeries.

To help make diagnosis more accurate, the Google researchers developed a neural network with several layers of processing. They then fed it with CT scans of patients with already defined diagnoses. Some of these individuals had lung cancer, while some did not. Some were also identified with nodules, which later developed into malignancies.


Of the total 6,716 cases of patients with known diagnoses, the deep learning program was able to make a 94 percent accurate reading.

The AI was later set against expert radiologists to find out whether it could do a better job than human doctors even no prior scan was made available. The machine edged out the physicians, making fewer false positives and false negatives than its human counterparts.

However, when the researchers provided an earlier scan, the results were a lot closer, both for the deep learning program and the human doctors.

With the AI capable of processing huge amounts of data, it could detect subtle patterns in cancer diagnoses that humans simply might not be able to identify.

Google's cancer-detecting AI is still in its development stage, but it is already being considered as a positive step in making diagnoses more accurate.

https://www.techtimes.com/articles/243519/20190521/google-ai-predicts-lung-cancer-better-than-radiologists.htm