Artificial Intelligence Could Reduce False Positives In Lung Cancer Screenings
Preliminary results show artificial intelligence developed by the University of Pittsburgh and UPMC can reduce false positive test results for lung cancer.
The current method for diagnosing lung cancer early is called a low-dose CT scan. It's a 3D x-ray test that takes a few seconds and detects shadows, or nodules, in the lungs that could potentially be cancerous. However, it's so sensitive that it has a false positive rate of 96 percent.
"If this were Las Vegas, we would just bet that none of [the nodules] were cancer and we'd be right 96 percent of the time," said senior author David Wilson, co-director of the Lung Cancer Institute at UPMC Hillman. "But obviously that would defeat the purpose of early diagnosis, and the four out of 100 people who have lung cancer would be out of luck.
False positives can cause unnecessary stress to the patient, and additional tests can be expensive and time consuming.
A machine-learning algorithm developed by Wilson and other local researchers processes a patient's low-dose CT scan and other information that could signal cancer risk. It also analyzes the number nodules and how many blood vessels are near the nodules.
The artificial intelligence then creates a probability of how likely the patient is to have cancer, and under a certain threshold, cancer is ruled out.
The researchers compared the results from the AI to actual diagnoses of patients. The study shows the algorithm can rule out 30 percent of false positives without missing a single case of cancer.
"Hypothetically speaking, we can take the results of the CT scan and put it into the machine learning algorithm and get the results very quickly, within 24 hours," Wilson said.
Wilson said there's more work to do to determine how reliable the artificial intelligence is, though the initial results are promising. The next step is to use the technique on a larger patient pool outside the hospital system to make sure the algorithm stays accurate.
Lung cancer screenings are recommended for people between 55 and 80 years old, who also have a 30 pack-year history of smoking. That means a person smoked one pack a day for 30 years, or two packs a day for 15 years.
WESA receives funding from UPMC and the University of Pittsburgh.