Chest radiograph deep learning model detects PRISm for opportunistic screening

A chest radiograph deep learning model detected preserved ratio impaired spirometry (PRISm) with AUROC 0.892 (95% CI 0.875-0.906), sensitivity 71.6%, and specificity 87.9% in a retrospective internal test set (n=10,917 paired exams).

• Retrospective single-center study using 54,654 chest radiograph–spirometry pairs from 38,470 Japanese health checkup participants; internal testing set n=10,917.
• For all pulmonary dysfunction subtypes (FEV₁ decline, FVC decline, combined decline, airflow limitation) the model achieved AUROC >0.8.
• Single-center retrospective design without external validation limits generalizability; health checkup cohort may not reflect symptomatic populations.

RadPigeon summaries are original and for information only. They are not clinical advice.