Novel AI technology successfully detects parenchymal, airway abnormalities in IPF

A novel artificial intelligence-based system successfully quantified lung lesions and airway volumes on CT imaging among patients with idiopathic pulmonary fibrosis, according to data published in Annals of the American Thoracic Society.

“We newly developed a novel AI-based system named AIQCT that immediately enables quantification of the volumes of 10 parenchymal patterns as well as of airways. The accuracy of the software was evaluated by comparison of visual scores and by similarity analysis, especially for reticulation, honeycombing and bronchi,” Tomohiro Handa, MD, PhD, associate professor of advanced medicine for respiratory failure at the Graduate School of Medicine at Kyoto University, Japan, and colleagues wrote. “When applied to the [high-resolution CT] images of 120 patients with IPF, bronchial and normal lung volumes were independently associated with survival after adjusting for the GAP stage.”

The researchers first applied the AI software in 304 high-resolution CT scans from patients with diffuse lung diseases. AIQCT automatically categorized and quantified 10 types of parenchymal patterns and airway volumes. To validate this system, researchers compared the area percentages of each lung lesion quantified by AIQCT with those from the visual scores using 30 plain high-resolution CT images with lung diseases. The researchers also performed 3D analysis for similarity with ground truth using high-resolution CT images from 10 patients with IPF and applied AIQCT to 120 patients with IPF who underwent high-resolution CT scanning and analyzed associations between the measured airway volumes and survival.

Correlations between the AIQCT system and the visual scores were moderate to strong depending on the parenchymal pattern, with a correlation coefficient of 0.44 to 0.95. The Dice similarity coefficients between AIQCT data and ground truth were 0.67 for reticulation, 0.76 for honeycombing and 0.64 for bronchi, the researchers reported.

During a median follow-up period of 2,184 days, 66 deaths occurred and one patient underwent lung transplantation.

Bronchial volumes (adjusted HR = 1.33; 95% CI, 1.16-1.53) and normal lung volumes (aHR = 0.97; 95% CI, 0.94-0.99) were independently associated with survival after adjustment for the gender-age-lung physiology IPF stage, according to the study.

“This study revealed that bronchial and normal lung volumes evaluated by the AIQCT system were independent prognostic parameters for IPF,” the researchers wrote. “Further investigation is necessary to elucidate the differences in the characteristics between CT software packages that quantify lung parenchymal lesions.”

Comments are closed.