AI could avoid further tracking of lung nodules

A team of researchers from the Noordwest Ziekenhuisgroep in Alkmaar, the Netherlands, retrospectively examined 300 patients receiving a follow-up CT examination for an accidentally detected lung nodule. Preliminary results from more than 100 patients showed that almost 20% could have been safely released from follow-up based on volumetric nodule analysis by a commercial AI software application.

“We can conclude that AI volumetry leads to 1 in 5 earlier exit from lung cancer screening follow-up,” said presenter Dr. Inge Gimbel.

Lung cancer remains the leading cause of cancer death, accounting for nearly 1.8 million deaths annually, according to Gimbel. Effective screening and early detection are key to reducing mortality.

However, the frequent detection of lung nodules leads to an increasing number of CT scans being performed. In the United States alone, an estimated 1.5 million patients are diagnosed with accidental lung nodules each year, Gimbel said. Moreover, only a small percentage of these nodules will be malignant.

“So in order to save money, it’s very important to know which patients can be safely discharged earlier,” she said.

According to the 2015 British Thoracic Society guidelines for the management of pulmonary nodules, patients with low-risk nodules are recommended to have at least two years of follow-up based on 2D measurements or a follow-up period of at least one year based on 3D measurements.

“As these volume measurements are automatically performed by the AI, it provides an opportunity for early discharge of these patients, which will lead to a reduction in the number of CT scans, hospital appointments, patient load and of radiation dose, and overall also results in reduced costs,” Gimbel said.

In their recent study, the researchers sought to determine whether using AI to automatically detect and measure lung nodules could result in earlier exit from cancer screening follow-up. They retrospectively assembled a cohort of 300 patients who received CT screening for monitoring lung nodules in 2020.

Veye Lung Nodules (Aidence) version 3.9.3 was then applied to CT studies, but volume measurements were not used to exit patients from screening follow-up earlier, Gimbel said. The software was set to an operating point of 0.85, along with a 4 mm filter to detect all nodules 4 mm and larger.

In Vienna, Gimbel shared preliminary results from 108 patients in the study. Of these patients, AI software volume measurements could have been used to exit 20 (18.5%) from follow-up.

Based on estimates of the number and cost of CT scans for lung cancer screening in the United States, AI software could save about $80 million each year, according to Gimbel.

“Thus, the use of AI volumetrics is of great importance to reduce costs,” she said.

Gimbel acknowledged that the results so far have been preliminary and that no firm conclusions can yet be drawn. Furthermore, this was a retrospective study with a short follow-up period. And AI can also produce false positive results and incorrect measurements, she said.

“So you need a radiologist to verify those measurements,” she said.

Copyright © 2022

Comments are closed.