Related Experiment Video
Updated: Oct 22, 2025
Author Spotlight: A 3D Digital Model for the Diagnosis and Treatment of Pulmonary Nodules
Published on: May 19, 2023
Deep Learning for Detection of Pulmonary Metastasis on Chest Radiographs
Eui Jin Hwang1, Jeong Su Lee1, Jong Hyuk Lee1
1From the Department of Radiology, Seoul National University College of Medicine, 101 Daehak-ro, Jongno-gu, Seoul 03080, Korea (E.J.H., J.S.L., J.H.L., W.H.L., J.H.K., K.S.C., T.W.C., T.H.K., J.M.G., C.M.P.); Department of Radiology, Namwon Medical Center, Namwon, Korea (W.H.L.); Graduate School of Medical Science and Engineering, Korea Advanced Institute of Science and Technology, Daejeon, Korea (K.S.C.); and Department of Radiology, Naval Pohang Hospital, Pohang, Korea (T.H.K.).
A deep learning-based computer-aided detection (CAD) system significantly improved the detection of new lung metastases on chest radiographs for cancer patients. This AI tool enhanced diagnostic yield without increasing false positives, aiding surveillance where CT access is limited.
Area of Science:
- Radiology
- Artificial Intelligence
- Oncology
Background:
- Pulmonary metastasis surveillance is crucial for cancer patients.
- Chest radiography is an accessible imaging modality, especially when CT is limited.
- Computer-aided detection (CAD) systems offer potential to enhance radiographic interpretation.
Purpose of the Study:
- To evaluate the efficacy of a deep learning (DL)-based CAD system in improving the diagnostic yield of newly visible lung metastasis on chest radiographs.
- To assess the impact of DL-CAD on the false-referral rate for lung metastasis detection.
Main Methods:
- A retrospective diagnostic cohort study compared chest radiographs interpreted with (CAD-assisted) and without (conventional) a regulatory-approved DL-CAD system.
- Data included patients referred from medical oncology, with propensity score matching for age, sex, and primary cancer.
- Generalized estimating equations were used to compare diagnostic yield and false-referral rates.
Main Results:
- The CAD-assisted group showed a significantly higher diagnostic yield for newly visible lung metastasis (0.86%) compared to the conventional group (0.32%; P = .004).
- The false-referral rate in the CAD-assisted group (0.34%) was not inferior to the conventional group (0.25%), meeting the non-inferiority margin.
- The study analyzed 2916 radiographs in the CAD-assisted group and 5681 in the conventional group.
Conclusions:
- Deep learning-based CAD systems can enhance the diagnostic yield for detecting new lung metastases on chest radiographs in cancer patients.
- The implementation of DL-CAD demonstrates improved detection accuracy without a significant increase in false referrals.
- DL-CAD serves as a valuable tool for lung metastasis surveillance, particularly in resource-limited settings.
