Impact of a deep learning-based computer-aided detection system (BMAX) on CT findings and clinical parameters in screening-detected interstitial lung abnormalities: a cross-sectional study.
7 affiliations
- 1Department of Respiratory Medicine, Kumamoto University Hospital, Faculty of Life Sciences, Kumamoto University, Kumamoto, Japan.
- 2Department of Respiratory Medicine, Kumamoto Kenhoku Hospital, Kumamoto, Japan.
- 3Department of Respiratory Medicine, National Hospital Organization Kumamoto Saishun Medical Center, Kumamoto, Japan.
- 4Division of Respiratory Medicine, Saiseikai Kumamoto Hospital, Kumamoto, Japan.
- 5Department of Respiratory Medicine, Japanese Red Cross Kumamoto Hospital, Kumamoto, Japan.
- 6Department of Radiology, Kindai University Faculty of Medicine, Osaka, Japan.
- 7Department of Radiology, Kurume University School of Medicine, Fukuoka, Japan.
Abstract
BMAX is a deep learning-based computer-aided detection (CAD) system designed to identify chronic fibrosing interstitial lung disease (ILD) on chest radiographs. While its utility in clinical settings is recognized, the association between BMAX scores and specific computed tomography (CT) findings or clinical parameters in patients with interstitial lung abnormalities (ILAs) remains unclear. This study investigated the correlation between BMAX scores and specific CT findings and clinical parameters in patients with ILAs. In this observational, cross-sectional study, we enrolled 362 patients with ILAs detected during health screening. Patients were classified into high (≥0.3, n=286) or low (<0.3, n=76) BMAX groups. CT findings and clinical parameters were compared between groups. Correlations between BMAX scores and relevant biomarkers, respiratory function, and imaging features were evaluated. Patients in the high-BMAX group had greater rates of fibrotic ILA (90.2% vs. 63.2%, P<0.001), usual interstitial pneumonia (UIP) patterns (14.7% vs. 0.0%, P<0.001), traction bronchiectasis (87.8% vs. 60.5%, P<0.001), and honeycombing (17.8% vs. 0.0%, P<0.001) than patients in the low-BMAX group. The high-BMAX group also exhibited statistically lower, although clinically preserved, pulmonary-function parameters, including percent predicted forced vital capacity (%FVC; 96.7% vs. 103.1%, P<0.001) and percent predicted diffusing capacity for carbon monoxide (%DLco; 88.1% vs. 104.4%, P<0.001), and elevated levels of disease severity markers [Krebs von den Lungen-6 (KL-6) and surfactant protein D (SP-D), both P<0.001] compared to the low-BMAX group. Correlation analysis confirmed that BMAX scores were significantly associated with these biomarkers, pulmonary-function parameters, and fibrosing CT patterns. BMAX scores are associated with fibrotic characteristics and indicators of ILA severity, suggesting their potential utility for identifying high-risk fibrotic imaging features in patients with ILA.
Related
- Pulmonary hypertension optimal screening in interstitial lung disease (PHOS-ILD): a scoring system for the detection of pulmonary hypertension in patients with interstitial lung disease.
- Association of current HbA1c with concurrent multiorgan screening abnormalities and the incremental value of routine clinical characteristics in type 2 diabetes: a single-center cross-sectional study in Wenzhou, China.
- Validity of the screening for early endometriosis detection (SEED) screening tool: protocol for a cross-sectional study
- Equivocal and definite interstitial lung abnormalities as risk factors for trastuzumab deruxtecan-related interstitial lung disease in breast cancer
- Impulse oscillometry-derived respiratory mechanical abnormalities and fibrotic disease progression in interstitial lung disease: a single-center cohort study
BibTeX
@article{furukawa2026impact,
title = {{Impact of a deep learning-based computer-aided detection system (BMAX) on CT findings and clinical parameters in screening-detected interstitial lung abnormalities: a cross-sectional study.}},
author = {Tsuguhiro Furukawa and Kimitaka Akaike and Hidenori Ichiyasu and Yuiko Masuda and Hiroko Okabayashi and Shohei Hamada and Aiko Masunaga and Kodai Kawamura and Kazuhiro Iyonaga and Takeshi Johkoh and Kiminori Fujimoto and Kazuya Ichikado and Takuro Sakagami},
journal = {Journal of thoracic disease},
year = {2026},
month = sep,
doi = {10.21037/jtd-2026-1912},
url = {https://europepmc.org/article/MED/42996926},
note = {CC BY-NC-ND},
}
RIS
TY - JOUR
TI - Impact of a deep learning-based computer-aided detection system (BMAX) on CT findings and clinical parameters in screening-detected interstitial lung abnormalities: a cross-sectional study.
AU - Tsuguhiro Furukawa
AU - Kimitaka Akaike
AU - Hidenori Ichiyasu
AU - Yuiko Masuda
AU - Hiroko Okabayashi
AU - Shohei Hamada
AU - Aiko Masunaga
AU - Kodai Kawamura
AU - Kazuhiro Iyonaga
AU - Takeshi Johkoh
AU - Kiminori Fujimoto
AU - Kazuya Ichikado
AU - Takuro Sakagami
PY - 2026
DA - 2026/09/24
JO - Journal of thoracic disease
DO - 10.21037/jtd-2026-1912
UR - https://europepmc.org/article/MED/42996926
AB - BMAX is a deep learning-based computer-aided detection (CAD) system designed to identify chronic fibrosing interstitial lung disease (ILD) on chest radiographs. While its utility in clinical settings is recognized, the association between BMAX scores and specific computed tomography (CT) findings or clinical parameters in patients with interstitial lung abnormalities (ILAs) remains unclear. This study investigated the correlation between BMAX scores and specific CT findings and clinical parameters in patients with ILAs. In this observational, cross-sectional study, we enrolled 362 patients with ILAs detected during health screening. Patients were classified into high (≥0.3, n=286) or low (<0.3, n=76) BMAX groups. CT findings and clinical parameters were compared between groups. Correlations between BMAX scores and relevant biomarkers, respiratory function, and imaging features were evaluated. Patients in the high-BMAX group had greater rates of fibrotic ILA (90.2% vs. 63.2%, P<0.001), usual interstitial pneumonia (UIP) patterns (14.7% vs. 0.0%, P<0.001), traction bronchiectasis (87.8% vs. 60.5%, P<0.001), and honeycombing (17.8% vs. 0.0%, P<0.001) than patients in the low-BMAX group. The high-BMAX group also exhibited statistically lower, although clinically preserved, pulmonary-function parameters, including percent predicted forced vital capacity (%FVC; 96.7% vs. 103.1%, P<0.001) and percent predicted diffusing capacity for carbon monoxide (%DLco; 88.1% vs. 104.4%, P<0.001), and elevated levels of disease severity markers [Krebs von den Lungen-6 (KL-6) and surfactant protein D (SP-D), both P<0.001] compared to the low-BMAX group. Correlation analysis confirmed that BMAX scores were significantly associated with these biomarkers, pulmonary-function parameters, and fibrosing CT patterns. BMAX scores are associated with fibrotic characteristics and indicators of ILA severity, suggesting their potential utility for identifying high-risk fibrotic imaging features in patients with ILA.
SN - 2072-1439
SN - 2077-6624
ER -