Screening for lung fibrosis using serum surfactant protein-D, KL-6, and a deep learning algorithm on chest radiographs: a prospective observational study.
Authors
Affiliations (5)
Affiliations (5)
- Department of Respiratory Medicine and Allergology, Sapporo Medical University School of Medicine, South 1-West 16, Chuo-ku, Sapporo, Hokkaido, 0608543, Japan. [email protected].
- Department of Diagnostic Radiology, Sapporo Medical University School of Medicine, South 1-West 16, Chuo-ku, Sapporo, Hokkaido, 0608543, Japan.
- Scholarly Communication Center Planning and Development Office, Sapporo Medical University, South 1-West 16, Chuo-ku, Sapporo, Hokkaido, 0608543, Japan.
- Department of Respiratory Medicine and Allergology, Sapporo Medical University School of Medicine, South 1-West 16, Chuo-ku, Sapporo, Hokkaido, 0608543, Japan.
- Department of Clinical Research Education and Management, University of Ryukyus Graduate School of Medicine, 207 Uebaru Nishihara Town, Okinawa, 9030215, Japan.
Abstract
Early identification of lung fibrosis remains difficult. In Japan, the serum biomarkers surfactant protein-D (SP-D) and KL-6 are commonly used to monitor interstitial lung diseases (ILD) in clinical practice, but their potential role in the early detection of lung fibrosis has not yet been fully clarified. Although chest radiography is also considered a possible tool for identifying subclinical pulmonary fibrosis, detecting early-stage disease remains challenging. A deep learning-based software, BMAX, was recently developed to identify fibrosing ILD on chest radiographs. Its capability to detect lung fibrosis in a health-checkup setting requires validation. Study participants were randomly recruited from individuals undergoing routine health examinations. All participants underwent chest radiography and serum SP-D and KL-6 testing. Those with elevated biomarker levels (≥ 110 ng/mL for SP-D and ≥ 500 IU/mL for KL-6) or radiographic abnormalities were advised to undergo further evaluation with chest computed tomography (CT). Lung fibrosis on CT was assessed independently by one pulmonologist and one thoracic radiologist. BMAX assigned a confidence score for lung fibrosis (ranging from 0 to 1) on each radiograph. In participants who underwent CT, the sensitivity and specificity of BMAX (using a confidence score > 0.3 as the threshold), SP-D, and KL-6 for detecting lung fibrosis were evaluated. Among the 2,751 individuals enrolled, 228 were recommended for CT, and 81 underwent the scan. Lung fibrosis was identified on chest CT in 8 of the 81 participants. The positivity rates for SP-D, KL-6, and BMAX (confidence score > 0.3) were 5.9%, 2.4%, and 5.9%, respectively. SP-D showed a sensitivity of 1.000 and a specificity of 0.315, while KL-6 showed a sensitivity of 0.750 and a specificity of 0.753. BMAX demonstrated a sensitivity of 1.000 and a specificity of 0.904. SP-D and KL-6 may be useful screening biomarkers for lung fibrosis in health checkup settings, offering high sensitivity and moderate positivity rates. BMAX also appears promising as a standalone screening tool for detecting lung fibrosis on chest radiographs.