A computer uses a stethoscope to scan lungs

Using AI in Lung Cancer Screening

A team of Google engineers teamed up with researchers at Northwestern University Feinberg School of Medicine to create a computer program that spotted lung cancer more accurately than expert radiologists. Their results were published in Nature Medicine.1

Developing AI screening technology

The Google artificial intelligence (AI) team created what is called a deep learning model. Deep learning is a technique that teaches computers to learn by example. In this case, the Google engineers developed its AI using more than 42,000 lung cancer screenings produced using low-dose computed tomography (LDCT).

In this case, the AI detected malignant (cancerous) nodules in the lungs with 11% fewer false positives and 5% false negatives compared to trained doctors who reviewed the same scans.

The research team speculates that their algorithm was more accurate due to the limitations of the human eye. Doctors must review an LDCT scan in a series of hundreds of two-dimensional images or “slices.” AI is able to “view” the lungs as a single, three-dimensional image, and simultaneously compare that single image to any previous scans. The system was also able to identify “areas of interest” where there was a high likelihood of cancer developing.2

How lung cancer screening works

Doctors screen for lung cancer using a test called a spiral low-dose computed tomography (LDCT). The LDCT takes several x-ray pictures of the chest at different angles and exposes the patient to a much lower amount of radiation than a standard CT (computed tomography) test. LDCT is the only test proven through research to reduce the number of deaths from lung cancer and can detect lung cancer better than chest x-rays because the LDCT produces a more detailed picture.

The radiologist will look over LDCT pictures to spot nodules, or small, round masses of tissue in the lungs. Nodules can be caused by cancer, infections, scar tissue, and other conditions, but cancerous nodules look different from non-cancerous ones.

Who should be screened?

Screening tests can be performed before symptoms are noticeable to try and catch cancer in its earliest, and most treatable, stages. Lung cancer caused an estimated 160,000 deaths in the U.S. in 2018, more deaths than any other type of cancer, in part because it is usually found after it has spread. Studies show that LDCT screening reduces mortality by 20-43%.1

Screening tests for lung cancer are recommended for people who are thought to be at higher risk of developing the disease. Those considered to be high risk include:

Pack years is a term used to help quantify the equivalent of how many years a patient has smoked. It is calculated by multiplying the number of packs of cigarettes smoked per day by the number of years the person has smoked. For example, smoking one pack per day for one year equals 1 pack year, whereas smoking two packs per day for two years equals 4 pack years.

The promise of better lung cancer detection

The engineers and doctors involved in this experiment hope that more accurate screening will lead to earlier lung cancer detection, which could result in earlier, more successful treatment. Fewer false positives will also help encourage the adoption of wider, more consistent screening, lower medical costs, and reduce unnecessary health care “scares.”

By providing your email address, you are agreeing to our privacy policy. We never sell or share your email address.

More on this topic


Join the conversation

or create an account to comment.