Lung cancer: artificial intelligence improves screening

With artificial intelligence, lung cancer screening is proving more effective. Indeed, a study shows that the rate of false positives is considerably reduced.

Lung cancer is the deadliest in the world. Screening for the disease is essential in order to detect it early and improve the patient's chances of survival. However, the current screening method is not very effective. Indeed, it has a false positive rate of 96%. We are talking about a false positive when a test indicates that a person is sick when in fact it is not the case.

Researchers at the University of Pittsburgh (Pennsylvania, USA) and UPMC Hillman Cancer Center have found a way to dramatically reduce the false positive rate: artificial intelligence. The study is published in the journal Thorax.

25% of positive results but only 4% of cancers

This is the first time that artificial intelligence has been used to sort benign nodules and malignant (cancerous) nodules in lung cancer screening. Generally, doctors use a low-dose scanner to screen for cancer in high-risk individuals. In the United States, a quarter of scanners indicate a positive result, about 25%, while less than 4% of patients have cancer.

"A positive test creates anxiety, increases health care costs and follow-up tests are not risk free," says study co-author Panayiotis (Takis) Benos. "For 96% of people with benign nodules, these procedures are useless, so we try to use the data to determine which ones are benign and which ones are malignant."

Nearly a third of false positives avoided

The researchers in the study analyzed the results of low-dose scans of 218 patients at the UMPC Hillman Cancer Center, who were at high risk of cancer. These patients knew if they had benign nodules or if they were really sick. Next, the researchers integrated this data into an automatic learning algorithm. It is an artificial intelligence process, which involves giving a computer the "ability" to learn, to improve its performance from data. The algorithm in question was able to calculate the probabilities of cancer. Below a certain threshold, he indicated that there was no disease.

Comparing the assessment of artificial intelligence to the diagnoses made by doctors, the researchers found that they could have avoided 30% of patients with benign nodules to undergo further examinations. And that, without missing a single case of cancer. The effectiveness of this technique is being evaluated over a larger population.

Video: MD vs. Machine: Artificial intelligence in health care (December 2019).