If lung cancer is so deadly, it's partly because it's hard to diagnose – but this new Google artificial intelligence program has proven to be a lifesaver.

According to a new study by Google and Northwestern Medicine, their new deep learning system was able to outperform radiologists in detecting malignant lung nodules.

If the system becomes more widely available in a clinical setting, it could improve the accuracy of early diagnosis of lung cancer, which could lead to earlier treatment and thousands of lives saved.

The in-depth learning system was compared to radiologists evaluating low-dose computed tomography (CT) scans in patients, some of whom had had biopsy-confirmed cancer within one year.

In most comparisons, the model worked as well – and in some cases, even better – than radiologists.

LOOK: You can now use these "Bee Savior" pocket cards to save hungry bees on the sidewalk

The system has also produced fewer false positives and fewer false negatives, which could result in fewer unnecessary follow-up procedures and fewer missed tumors.

The paper was published in Nature Medicine earlier this week.

Lung cancer is the most common cause of cancer deaths in the United States. The number of deaths in 2018 is estimated at 160 000. Extensive clinical trials in the United States and Europe have shown that chest screening can identify cancer and reduce mortality rates. However, high error rates and limited access to these screenings mean that many lung cancers are usually detected at an advanced stage when they are difficult to treat.

MORE: First smartphone app to detect ear infections in children could avoid unnecessary visits to a doctor

In-depth learning is a technique that teaches computers to learn by example. The deep learning system uses both the main scanner and, where appropriate, an anterior scanner of the patient as input. Previous CT scans are useful for predicting cancer risk from lung cancer because the growth rate of suspicious lung nodules may indicate cancer. The computer was trained using low dose, fully identified, biopsy-confirmed thoracic computed tomography (CT).

"Radiologists typically examine hundreds of two-dimensional images or" slices "in a single scanner, but this new machine-learning system can visualize the lungs in a very large three-dimensional image," said the Dr. Mozziyar Etemadi, co-author of the study. research assistant in anesthesiology at the Feinberg School of Medicine at Northwestern University and in engineering at the McCormick School of Engineering.

"The AI ​​in 3D can be much more sensitive in its ability to detect early lung cancer than the human eye that looks at 2D images," he added. "It's technically" 4D "because it's not just a scanner, but two (the current analyzer and last year) in time.

MORE: Bedside bioprinter may soon use burned patient cells to print new skin directly on wounds

"To build AI in order to visualize CT scanners in this way, you need a huge computer system at the Google scale. The concept is new but its very design is also because of its scale.

The new system identifies both a region of interest and a high probability of lung cancer.

The model outperformed six radiologists when no CT scan was available and as efficient as the radiologists when it existed before.

CHECK-OUT: Smart Caption glasses allow members of the deaf audience to directly watch theater artists

"The system can categorize a lesion with more specificity. Not only can we better diagnose cancer, we can also say that if a person does not have cancer, it could save him from an invasive, expensive and risky lung biopsy, "said Etemadi.

Google scientists developed the in-depth learning model and applied it to 6,716 unidentified CT scanners provided by Northwestern Medicine to validate the accuracy of its new system. Scientists discovered that the artificial intelligence system was able to spot sometimes tiny malignant lung nodules with a model AUC of 0.94 cases.

Shravya Shetty, Technical Manager at Google, said, "This area of ​​research is extremely important because lung cancer has the highest death rate among all cancers and the adoption of lung cancer screening presents many challenges.

LOOK: Surgeons successfully implant the world's first 3D printed coast – and plan to do more in the future

"Our work examines ways in which AI can be used to improve the accuracy and optimize the screening process, so as to facilitate the implementation of screening programs," Shetty added. "The results are promising and we look forward to continuing our work with our partners and peers."

The authors point out that these findings need to be clinically validated in large patient populations, but they claim that this model could help improve the management and outcomes of lung cancer patients.

"Most of the software we use as clinicians is designed for patient care, not for research," said Etemadi. "My entire team has spent more than a year putting effort into extracting and preparing data to help this exciting project.

"The ability to collaborate with world-renowned scientists at Google, using their unprecedented computing capabilities to create something with the potential to save tens of thousands of lives a year is truly a privilege."

Reprinted from Northwestern Now

Be sure and share the good news with your friends on social media …